Signals and Systems

Continuous-time Fourier series - CTFS

Branko Jeren, Damir Seršić, Zvonko Kostanjčar, Ana Sović Kržić,

Marko Subašić

University of Zagreb
Faculty of Electrical Engineering and Computing

Creative Commons Licence–Non commercial–No derivatives 3.0 v1.0

(UNIZG FER) CTFS 1 / 31

Contents

1 CTFS – introduction

2 CTFS – basic properties

3 Gibbs phenomenon

4 The relationship between signal duration and spectrum

(UNIZG FER) CTFS 2 / 31

Continuous-time Fourier series
Continuous-time Fourier series – CTFS
The coefficients of the continuous-time Fourier series of a periodic
continuous signal f with a basic period of T0 are:
Z
1 −jk 2π t
Fk = f (t)e T0 dt, ∀k ∈ Z. (1)
T0 T0

They exist when (1) is defined and finite ∀k ∈ Z, and are called the
spectrum of signal f . The reconstruction of the signal from its spectrum F
is
∞
jk 2π t
X
f (t) = Fk e T0 , ∀t ∈ R. (2)
k=−∞

If the coefficients (1) exist and the reconstruction (2) converges, then f
and F constitute a Fourier transform pair: F = CTFS(f ), f = ICTFS(F ).
|F | is the amplitude spectrum and ∠F is the phase spectrum of signal f .

(UNIZG FER) CTFS 3 / 31

Continuous-time Fourier series - convergence

There are different families of signals for which the CTFS is defined.
We will discuss two significant representatives.
Finite-power periodic signals
If f is a periodic continuous signal with finite energy over the period:
Z
|f (t)|2 dt < ∞,
T0

then the continuous-time Fourier series converges in the mean squared

error sense:
Z T0 K
X 2
f (t) − Fk e jkω0 t dt → 0 for K →∞
0 k=−K

(UNIZG FER) CTFS 4 / 31

Continuous-time Fourier series - convergence

Signals that satisfy Dirichlet conditions

If f is a periodic continuous signal which is:
(a) absolutely integrable over the period
Z
|f (t)|dt < ∞,
T0

(b) has a finite number of maxima and minima over the period,
(c) has a finite number of discontinuities over the period,
then the CTFS converges for all points except for discontinuities
(f (td+ )+f (td− ))
(f (td+ ) 6= f (td− )), where it converges to 2 .

(UNIZG FER) CTFS 5 / 31

Continuous-time Fourier series – example
CTFS example
Determine the coefficients of the continuous-time Fourier series of the
periodic signal in the figure:
f (t )
A

0
-T0 -T0 +τ 0 τ T0 T0 +τ
t→
Fk
A τ
Solution: T0

Aτ
F0 = T0 ,
0 0τ
kω N
jkω0 τ 0 2π k→
Aτ sin 2
Fk = T0 kω0 τ e− 2 for k = ±1, ±2, . . . T0
2 -6π τ -4π τ -2π τ 0 2π τ 4π τ 6π τ ω→
∠Fk
(UNIZGπFER) CTFS 6 / 31
Example continued – line spectrum
the coefficients of the Fourier series are complex values and the set
{Fk }|∞
k=−∞ can be graphically depicted by separate stem-plots of
their amplitude and phase.
Two plots represent the line spectrum of the periodic signal f (t).

 Aτ

T0 , k=0
|Fk | = kω
sin 20
τ
 Aτ

T0 kω0 τ , k = ±1, ±2 . . . ,
2

sin kω20 τ ω0 τ
∠Fk = ∠A + ∠ kω0 τ
−k
2
2
|Fk | is the amplitude spectrum,
∠Fk is the phasea spectrum of a periodic signal
a
the phase of a real number is zero if the number is positive, and π or −π if
it is negative

(UNIZG FER) CTFS 7 / 31

Example continued – line spectrum
f (t ), t = 1, T0 = 6, w0 = 2p / 6
A

0
-T0 -T0 +t 0 t T0 T0 +t
t
Fk
A t
T0

0 
0 2p k
T0
-6p t -4p t -2p t 0 2p t 4p t 6p t w
Fk
p
2p
T0
0

-p
0 k

-6p t -4p t -2p t 0 2p t 4p t 6p t w

(UNIZG FER) CTFS 8 / 31

Example continued – construction of the phase spectrum
w0 t
-k , t = 1, T0 = 6, w0 = 2p / 6
2
3p
p
0
-p
-3p
0 k
sin(k w0 t / 2)

k w0 t / 2
3p
p
0
-p
-3p
0 k
w0 t sin(k w0 t / 2)
-k +
2 k w0 t / 2
3p
p
0
-p
-3p
0 k
æ w0 t sin(k w0 t / 2) ö÷
çç-k + ÷
çè 2 k w0 t / 2 ÷ømodulo 2 p
3p
p
0
-p
-3p
0 k

(UNIZG FER) CTFS 9 / 31

Example continued – conclusion

The spectrum of a periodic signal is discrete, and the spacing between

the spectrum frequencies is ω0 = 2π/T0 .
Our amplitude spectrum is an even function, and our phase spectrum
is an odd function (explained later, this holds true for all real-valued
signals).
The envelope of the amplitude spectrum of a rectangular periodic
signal is of the form sin(w
w
)
.
The width of a rectangular signal (τ ) and the width of the main lobe
of the spectrum (4π/τ ) are inversely proportional.
Since modulus 2π of the phase is calculated (atan2), the result is
wrapped in interval [−π, π].

(UNIZG FER) CTFS 10 / 31

Continuous-time Fourier series – example

CTFS example – even signal

Determine the coefficients of the continuous-time Fourier series of the
periodic signal in the figure
g(t )
A

0
-T0 -τ 0 τ T0
2 2 t→
Gk
A τ
T0
Solution:

Aτ 0 N
G0 = T0 , 0 2π k→
T0
kω0 τ
Aτ sin 2 ω→
Gk = T0 kω0 τ for-6πk τ= ±1, -2π.τ. .
-4π τ ±2, 0 2π τ 4π τ 6π τ
2 ∠Gk
π

(UNIZG0FER) CTFS
N 11 / 31
2π
CTFS even signal – line spectrum
g(t )
A

0
-T0 -τ 0 τ T0
2 2 t→
Gk
A τ
T0

0 N
0 2π k→
T0
-6π τ -4π τ -2π τ 0 2π τ 4π τ 6π τ ω→

∠Gk
π

0 N
2π
T0
-π
0 k→
-6π τ -4π τ -2π τ 0 2π τ 4π τ 6π τ ω→

Gk
A τ
T0
2π
T0

0
0 k→

-6π τ -4π τ -2π τ 0 2π τ 4π τ 6π τ ω→

(UNIZG FER) CTFS 12 / 31

Spectra of continuous-time rectangular periodic signals

Our spectra:

Aτ sin kω20 τ sin kω20 τ

|Gk | = kω0 τ
, ∠Gk = ∠A + ∠ kω0 τ
T0 2 2

Aτ sin kω20 τ sin kω0 τ

ω0 τ
|Fk | = kω0 τ
, ∠Fk = ∠A + ∠ kω τ2 −k
T0 2 2
0 2

τ


of rectangular signals g (t) and f (t) = g t − 2 , ∀t ∈ R.
τ
We conclude that a delay of corresponds to a linear phase shift of
2
the spectrum by − kω20 τ , while the amplitude spectrum remains
unchanged.

(UNIZG FER) CTFS 13 / 31

Contents

1 CTFS – introduction

2 CTFS – basic properties

3 Gibbs phenomenon

4 The relationship between signal duration and spectrum

(UNIZG FER) CTFS 14 / 31

CTFS – basic properties

Parseval’s relation
For periodic signals of finite energy over a single period, i.e. for
finite-power periodic signals, the following holds:
Z ∞
1 X
Pf = |f (t)|2 dt = |Fk |2 .
T0 T0
k=−∞

CTFS of real signals

For real periodic signals f for which CTFS exists, the following holds:
the spectrum is conjugate symmetric, i.e. F−k = Fk∗ ,
the amplitude spectrum is even and the phase spectrum is odd
|Fk | = |F−k | i ∠Fk = −∠F−k .

Hence: f (t) = F0 + ∞
P
k=1 2|Fk | cos(kω0 t + ∠Fk ).

(UNIZG FER) CTFS 15 / 31

CTFS – basic properties

CTFS of real even signals

For real, even, and periodic signals f for which the CTFS exists, the signal
spectrum is real.

Best approximation in terms of least energy of the error

The continuous signal
K
2π
jk T t
X
fˆK (t) = Fk e 0

k=−K

is the best approximation of the signal f in terms of minimum squared

2π
jk T t
error in the space of signals spanned by elementary signals {e 0 }K
k=−K

(UNIZG FER) CTFS 16 / 31

CTFS – continuous-time signal example
CTFS – example
Determine the coefficients of the Fourier series of the depicted periodic
signal (A = 1, ω1 = 1, and T0 = 5π 4 )

f (t )
A A cos(w1t )

0
-T0 - p 0
p T0 t
2w1 2w1

Solution. The coefficients of the Fourier series are:

Z T0 Z π
1 2 1 2ω1
Fk = f (t)e −jkω0 t dt = Acos(ω1 t)e −jkω0 t dt
T0 − T0 T0 − π
2 2ω1
 
2Aω1 πω0
= cos k , for k = 0, ±1, ±2, . . .
T0 (ω12 − k 2 ω02 ) 2ω1
(UNIZG FER) CTFS 17 / 31
Continuous-time signal example – signal spectrum

Fk
2A
T0 w1

0
-10 -5 0 5 10
k

The inverse transform of the line spectrum coefficients reconstructs the

original signal f in the time domain.

(UNIZG FER) CTFS 18 / 31

Signal approximation with a finite number of harmonics
The signal f can be approximated by the signals fˆK , using different
numbers of harmonics, K = 1, 3, 7, 35
K K  
X X 2Aω1 πω0
fˆK (t) = Fk e jkω0 t = cos k e jkω0 t
T0 (ω12 − k 2 ω02 ) 2ω1
k=−K k=−K

K =1 K =3
A A

0 0
-T0 0 T0 t  -T0 0 T0 t 
K =7 K = 35
A A

0 0
-T0 0 T0 t  -T0 0 T0 t 

(UNIZG FER) CTFS 19 / 31

Continuous-time signal example

Conclusions:
For our continuous-time signal example we have:
K
X
∀t ∈ R, f (t) ≈ fˆK (t) = Fk e jkω0 t
k=−K

For K → ∞ we get a perfect approximation of the signal f .

Question: does this hold universally?

Let us check a periodic, piecewise continuous signal with a finite number

of discontinuities over a single period.

(UNIZG FER) CTFS 20 / 31

Contents

1 CTFS – introduction

2 CTFS – basic properties

3 Gibbs phenomenon

4 The relationship between signal duration and spectrum

(UNIZG FER) CTFS 21 / 31

Gibbs phenomenon - rectangular signal example
Let us consider a rectangular even periodic signal f of length τ , period T0
and amplitude one.

The Fourier series coefficients are:

τ τ sin kω20 τ
F0 = , Fk = for k = ±1, ±2, . . .
T0 T0 kω20 τ
The CTFS approximation of signal f with a finite number of
harmonics is
K
ˆ τ X τ sin kω20 τ jkω0 t
fK (t) = + e
T0 T0 kω20 τ
k=−K
k6=0

On the next slide, approximations with K = 1, 3, 9, 35; T0 = 2 and

τ = 1 are depicted.

(UNIZG FER) CTFS 22 / 31

Gibbs phenomenon - rectangular signal example
K=1 K=3
1.09 1.09
1 1

0.5 0.5

0 0
-0.09 -0.09
-2 0 2 -2 0 2
t t
K=9 K=35
1.09 1.09
1 1

0.5 0.5

0 0
-0.09 -0.09
-2 0 2 -2 0 2
t t
(UNIZG FER) CTFS 23 / 31
Gibbs phenomenon

The example demonstrates an overshoot (i.e. a ripple) in the vicinity

of the discontinuity, with a maximum value which does not diminish
with increasing K .
This is called the Gibbs phenomenon.
For K → ∞, the value of the overshoot remains constant (cca 9%),
but its width approaches zero.
For ∀t, except for discontinuity points, the CTFS reconstruction
approaches the value of the original signal (K → ∞).
At discontinuities (f (td+ ) 6= f (td− )) it converges to:

(f (td+ ) + f (td− ))
.
2

(UNIZG FER) CTFS 24 / 31

Gibbs phenomenon - approximation error
The approximation error is eK (t) = f (t) − fˆK (t), ∀t ∈ R
K=1 K=3

0.5 0.5

0.09 0.09
0 0
-0.09 -0.09

-0.5 -0.5
-2 0 2 t -2 0 2 t

K=9 K=35

0.5 0.5

0.09 0.09
0 0
-0.09 -0.09

-0.5 -0.5
-2 0 2 t -2 0 2 t

(UNIZG FER) CTFS 25 / 31

Gibbs phenomenon - the energy in the error signal over a
single period

|eK (t)|2 dt over a single period

R
The energy of the error signal T0
tends to zero (K → ∞).
Therefore, the difference between f and its approximation by infinite
Fourier series has zero energy, in spite of having non-zero amplitude
at a discrete set of discontinuities.

(UNIZG FER) CTFS 26 / 31

Contents

1 CTFS – introduction

2 CTFS – basic properties

3 Gibbs phenomenon

4 The relationship between signal duration and spectrum

(UNIZG FER) CTFS 27 / 31

The relationship between the width of a rectangular pulse
and the width of the main lobe of the spectrum
For an even periodic rectangular signal
g(t )
A= 1
t

0
-T0 -t 0 t T0
2 2 t

we determine the spectrum for 0 < τ < T0 , for τ = T0 and for τ = 0

For 0 < τ < T0 the coefficients of the Fourier series are (previously
calculated)
1
Aτ τ 1
G0 = = τ = , for k = 0,
T0 T0 T0
Aτ sin kω20 τ 1 sin kω20 τ
Gk = = , for k = ±1, ±2, . . .
T0 kω20 τ T0 kω20 τ

(UNIZG FER) CTFS 28 / 31

The relationship between the width of a rectangular pulse
and the width of the main lobe of the spectrum
For τ = T0 , the signal g (t) = g (t + mT0 ) = T10 , ∀t ∈ R, and the
2π
coefficients of the Fourier series are (ω0 = T 0
):

Aτ 1
G0 = = , for k = 0,
T0 T0
Aτ sin kω20 τ 1 sin kπ
Gk = kω τ
= = 0, for k = ±1, ±2, . . .
T0 2
0 T0 kπ

For τ = 0, the rectangular

P∞ pulse becomes the Dirac delta function, so
g (t) = combT0 (t) = m=−∞ δ(t − mT0 ), and the coefficients of the
Fourier series are, for ∀k ∈ Z:
Z T0 Z T0
1 2
−jkω0 t 1 2 1
Gk = f (t)e dt = δ(t)e −jkω0 t dt =
T0 − T0 T0 − T0 T0
2 2

(UNIZG FER) CTFS 29 / 31

The relationship between the width of a rectangular pulse
and the width of the main lobe of the spectrum
å
¥
g(t ) = combT0 (t ) = m =-¥
d(t - mT0 )

(1) (1) (1)

0
-T0 -T0/2 0 T0/2 T0 t 
Gk
1/T0

0 
0 k
2p/T0 w
g(t )
1/t

0
-T0 -T0/2 -t/2 t/2 T0/2 T0 t 
Gk
1/T0
2p/T0
0
-2p/t 0 2p/t k
w

g(t )

1/T0
0
-T0 -T0/2 0 T0/2 T0 t 
Gk
1/T0
(UNIZG FER) CTFS 2p/T0 30 / 31
Literature

A.V. Oppenheim and A.S. Willsky, with S.H. Nawab (1997.), Signals and
Systems, Prentice-Hall
Chapters: 3.3, 3.4, 3.5.

(UNIZG FER) CTFS 31 / 31