Department of Bioengineering

Biomedical Engineering

Lecturer: Prof. Dario Farina

Signals and Control II

Signals

Year 2
2020, Autumn Term

Author: Last Update:

Binghuan Li December 27, 2020
Acknowledgement
I would like to express my deepest appreciation to Prof. Dario Farina for his help in
reviewing these notes with the highest profession.

I would like to extent my sincere thanks to Rea Tresa, Ben Ford, and Elisa Soliani for
their insightful comments. Their work significantly optimized the structure.

Special thanks to Haroon Chughtai. His notes greatly enlightened me when preparing
this new set.

Update Notes
1. Section 5.3.2 : Added the derivation of time shifting property.

2. Section 5.5 : Corrected math typos, rearranged the example and improved the read-
ability.

3. Section 5.5.1 : Added a description of LTI systems.

4. Section 5.6 : Refined the description of magnitude and phase spectra. An example
is adopted from the textbook.

1
CONTENTS

Contents
1 Definition, Classification and Properties of Signals 4
1.1 Definition of signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2 Continuous and Discrete-time Signals . . . . . . . . . . . . . . . . . . . . . 4
1.2.1 Digital Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Deterministic and Stochastic Signals . . . . . . . . . . . . . . . . . . . . . 5
1.4 Periodic Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.5 Signal Energy and Power . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.5.1 Energy and Power of a Generic Signal . . . . . . . . . . . . . . . . 6

2 Types of Signals 8
2.1 Periodic Complex Exponential Signals in Continuous-time Domain . . . . . 8
2.2 Periodic Complex Exponential Signals in Discrete-time Domain . . . . . . 8
2.3 The Unit Impulse in Discrete-time Domain . . . . . . . . . . . . . . . . . . 8
2.4 The Unit Step in Discrete-time Domain . . . . . . . . . . . . . . . . . . . . 9
2.5 The Unit Step in Continuous-time Domain . . . . . . . . . . . . . . . . . . 10
2.5.1 Convolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

3 Simple Operations on Signals 13

3.1 Transformations of the Time Variable . . . . . . . . . . . . . . . . . . . . . 13
3.2 Amplitude Transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.3 Linear Combination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.4 Multiplication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.5 Scalar Products and Norms . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.5.1 Scalar Product and Norm of Vectors . . . . . . . . . . . . . . . . . 14
3.5.2 Scalar Product and Norm of Discrete-time Signals . . . . . . . . . . 14
3.5.3 Scalar Product and Norm for Continuous-time Signals . . . . . . . 15
3.6 Characterising Similarity/Di↵erence Between Signals . . . . . . . . . . . . 15
3.6.1 Measuring the Similarity . . . . . . . . . . . . . . . . . . . . . . . . 15
3.6.2 Normalized Cross-correlation Function fc (✓) . . . . . . . . . . . . . 15
3.6.3 Measuring the Di↵erence . . . . . . . . . . . . . . . . . . . . . . . . 16

4 Fourier Series 17
4.1 Orthonormal Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
4.2 Fourier Basis Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
4.3 Fourier Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

5 Fourier Transform 21
5.1 From Fourier Series to Fourier Transform . . . . . . . . . . . . . . . . . . . 21
5.2 The Continuous-time Fourier Transform . . . . . . . . . . . . . . . . . . . 22
5.3 Properties of Fourier Transform . . . . . . . . . . . . . . . . . . . . . . . . 23
5.3.1 Linearity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
5.3.2 Time shifting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
5.3.3 Conjugation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
5.3.4 Dual property . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
5.3.5 Time scaling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
5.3.6 Parseval’s relation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
5.3.7 Di↵erentiation in time . . . . . . . . . . . . . . . . . . . . . . . . . 26

2
CONTENTS

5.3.8 Convolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
5.4 More basic Fourier transforms . . . . . . . . . . . . . . . . . . . . . . . . . 27
5.4.1 Impulse . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
5.4.2 Complex exponential . . . . . . . . . . . . . . . . . . . . . . . . . . 27
5.4.3 Cosine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
5.4.4 Sine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
5.5 Example of application of properties of the Fourier transform . . . . . . . . 27
5.5.1 LTI systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
5.6 Magnitude and Phase Spectra . . . . . . . . . . . . . . . . . . . . . . . . . 29
5.7 Fourier Transform of Periodic Signals . . . . . . . . . . . . . . . . . . . . . 30
5.7.1 Fourier transform of a train of impulses . . . . . . . . . . . . . . . . 31

6 Sampling Theorem 32
6.1 Sampling Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
6.2 Reconstruction Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
6.3 Nyquist-Shannon Sampling Theorem . . . . . . . . . . . . . . . . . . . . . 34

3
1 DEFINITION, CLASSIFICATION AND PROPERTIES OF SIGNALS

1 Definition, Classification and Properties of Signals

1.1 Definition of signals
• Signals describe physical phenomena as patterns of variations of some form.

• Mathematically, signals are functions of one or more independent variables.

• For example, a signal s(t) can be a function of the continuous independent variable
time t 2 [↵, ] A two-dimensional signal f (x, y) can be a function of two spatial
coordinates x, y.

1.2 Continuous and Discrete-time Signals

• Signals can be a function of the continuous time variable, in which case we will
use the notation x(t) with t 2 R,
or of the discrete time variable, in which case we will use the notation x[n] with
n 2 Z. (Fig.1)

Fig. 1: Continuous signal x(t) and discrete signal x[n]

• Discrete-time signals are often (but not necessarily) a sampling of continuous-time

signals.
x[n] = xc (nT ) 1 < n < +1
T is sampling period.

• A discrete-time signal can be represented as a sequence of numbers, or, a vector.

1.2.1 Digital Signals

• When we speak of a digital signal we often mean one that has been sampled
(captured at regular points in time) and quantisised.

• When one refers to a 12-bit signal, they are referring to the number of amplitude
quantisation levels.

• Sampling a continuous signal may be done without losing any information from
the original signal. Conversely, quantisation always implies losing information.
(Fig.2)

We focus on the signals of one independent variable!

4
1 DEFINITION, CLASSIFICATION AND PROPERTIES OF SIGNALS

Fig. 2: Sampling and quantisation

1.3 Deterministic and Stochastic Signals

• Deterministic: a signal that can be predicted exactly (an analytical formulation
exists).

– Example: x(t) = sin(2⇡t)

• Stochastic: a signal that cannot be predicted exactly before it has “occurred”;

any signal that conveys information to us when we observe it.

– Example: Thermal noise across a resistor, EEG traces, etc.

• We can often meaningfully describe the statistical properties of stochastic signals

by building a model of their generation (stochastic processes).

We will mainly deal with deterministic signals in this course!

1.4 Periodic Signals

• A periodic continuous-time signal x(t) has the property that there is a positive value
of T for which x(t) = x(t + T ) for all values of t (similar definition for discrete-time
signals).

• A periodic signal has the property that it is unchanged by a time shift of T , we will
say that x(t) is periodic with period T . (Fig.3)

Fig. 3: A periodic signal with period T

5
1 DEFINITION, CLASSIFICATION AND PROPERTIES OF SIGNALS

1.5 Signal Energy and Power

For a continuous-time signal x(t) for t1  t  t2 and for a discrete-time signal x[n] for
n1  n  n2 , energy and power can be represented as follows:
Z t2
Energy(continuous time) = |x(t)|2 dt
t1
Z t2
1
Power(continuous time) = |x(t)|2 dt
t2 t1 t1
n2
X
Energy(discrete time) = |x[n]|2
n1
n2
X
1
Power(discrete time) = |x[n]|2
n2 n1 n1

We get the conclusion above from the calculation for electrical power and energy.
Let v(t) and i(t) represent the voltage and current across the resistor of resistance
R.

• The instantaneous power across the resistor is the product v(t)i(t), which is
proportional to v 2 (t).

• The total energy Z t2

1 2
v (t) dt
t1 R

• The average power Z t2

1 1 2
v (t) dt
t2 t1 t1 R

Similar properties can be applied to any continuous-time signals and discrete-time

signals.

• Often, the signals are directly related to physical quantities capturing power and
energy in a physical form.

• These properties are important characteristics of signals, even if in some cases do

not reflect physical energy or power.

1.5.1 Energy and Power of a Generic Signal

Extend the range to: 1 < t < +1 or 1 < n < +1

• In continuous time: Z +1
Energy = |x(t)|2 dt
1
Z T
1
Power = lim |x(t)|2 dt
T !1 2T T

6
1 DEFINITION, CLASSIFICATION AND PROPERTIES OF SIGNALS

• In discrete time:
+1
X
Energy = |x[n]|2
1

XN
1
Power = lim |x[n]|2
N !+1 2N + 1 N

We will use the mathematical definitions above, regardless of the direct physical meaning
of each term!

7
2 TYPES OF SIGNALS

2 Types of Signals
2.1 Periodic Complex Exponential Signals in Continuous-time
Domain
x(t) = ej!0 t

• Periodic, period T = 2⇡
|!0 |
.

• The signal x(t) = e j!o t

has the same period.

• The complex exponential defined above is closely related to the sinusoidal signal:
A j j!0 t A j j!0 t
x(t) = A cos(!0 t + ) = e e + e e
2 2
2⇡
which has the same period T = |!0 |
.

• The complex exponentials and sinusoidal signals have infinite energy and finite
power.

– Example: for the signal x(t) = A cos(2⇡!0 t + ) with the period T1 ,

A2
Power =
2

2.2 Periodic Complex Exponential Signals in Discrete-time Do-

main
x[n] = ej!0 n
And the sinusoidal signal becomes
A j j!0 n A j j!0 n
x[n] = A cos(!0 n + ) = e e + e e
2 2
• n 2 Z. Thus, x[n] is the same signal for !0 + 2⇡k with k 2 Z. The frequency
of oscillation in discrete time exponentials does not increase monotonically but is
limited to 2⇡.

• x[n] is not always periodic.

2.3 The Unit Impulse in Discrete-time Domain

In discrete-time domain, the unit impulse is the simplest signal: (Fig.4)

(
1, n = 0
[n] =
6 0
0, n =
Fig. 4: The unit impulse

8
2 TYPES OF SIGNALS

The discrete-time signal impulse delayed by the integer k is as follows: (Fig.5)

(
1, n = k
[n k] =
6 k
0, n =

Fig. 5: The unit impulse delayed by the integer k

• For any discrete-time signal x[n], directly from the definition of [n], we have

x[n] [n k] = x[k] [n k]

This implies: any signal multiplied by the unit impulse is zeroed for all time
samples, apart from the integer time where the unit impulse is centered.

• From the property above, we have:

1
X 1
X
x[n] [n k] = x[k] [n k]
n= 1 n= 1

1 *
1, when k = n
X
= x[k] [n k]
n=1
= x[k]

• Any arbitrary discrete-time signal can be expressed as the sum of scaled and de-
layed impulses:
+1
X
x[n] = x[k] [n k]
k= 1

2.4 The Unit Step in Discrete-time Domain

Unit step in discrete-time domain is defined as: (Fig.6)

( +1
1, n 0 X
u[n] = = [n k]
0, n < 0 k=0

Fig. 6: Unit step in discrete-time domain

• The unit impulse is the discrete time derivative of the unit step.

[n] = u[n] u[n 1]

9
2 TYPES OF SIGNALS

2.5 The Unit Step in Continuous-time Domain

Unit step in continuous-time domain is defined as: (Fig.7)

(
1, t > 0
u(t) =
0, t < 0

Fig. 7: Unit step in continuous-time domain

• The continuous-time unit step function has a discontinuity in t = 0, it is not

di↵erentiable. To solve the issue, we define:
du (t)
(t) =
dt
– The area of (t) is equal to 1 at any value of . (Fig.8, right)
– As ! 0, u (t) ! u(t). (gradient ! 0) (Fig.8, left)
– As ! 0, the impulse (t) becomes of shorter duration and higher amplitude:
(t) = lim !0 (t)

Fig. 8: Continuous approximation to unit step

• From the definition above, the following property holds: (Fig.9, left)
Z t
u(t) = (⌧ )d⌧
1

• For continuous-time, the unit impulse in the continuous-time domain shifted by the
time delay is (t ).(Fig.9, right)

Fig. 9: Unit impulse shifted by the time delay

10
2 TYPES OF SIGNALS

• In continuous-time domain: (similar to

discrete-time domain)

x(t) (t) ⇡ x(0) (t)

As (t) ! (t): better approximation

x(t) (t) = x(0) (t)

More generally: with time shifting

x(t) (t t0 ) = x(t0 ) (t t0 )

Fig. 10

• We also obtain:
Z 1 Z 1
x(t) (t t0 )dt = x(t0 ) (t t0 )dt
1 1
Z 1
= x(t0 ) (t t0 )
1
= x(t0 )

• This implies that any arbitrary continuous-time signal x(t) can be represent as:
Z 1
x(t) = x(⌧ ) (t ⌧ )d⌧
1

2.5.1 Convolution
We define the following transformation between two signals (Convolution):
Z 1
y(t) = x(⌧ ) h(t ⌧ )d⌧ = x(t) ⇤ h(t)
1

+1
X
y[n] = x[k] h[n k] = x[n] ⇤ h[n]
k= 1

For any continuous-time signal and any discrete-time signal:

x(t) = x(t) ⇤ (t)

x[n] = x[n] ⇤ [n]

by extension for arbitrary delays:

x(t t0 ) = x(t) ⇤ (t t0 )

x[n k] = x[n] ⇤ [n k]

11
2 TYPES OF SIGNALS

Summary of properties of the unit impulse

• For multiplication:
x(0) · (t) = x(t) · (t)
x[0] · [n] = x[n] · [n]

x(t0 ) · (t t0 ) = x(t) · (t t0 )
x[k] · [n k] = x[n] · [n k]

• For convolution:
x(t) = x(t) ⇤ (t)
x[n] = x[n] ⇤ [n]

x(t t0 ) = x(t) ⇤ (t t0 )
x[n k] = x[n] ⇤ [n k]

12
3 SIMPLE OPERATIONS ON SIGNALS

3 Simple Operations on Signals

3.1 Transformations of the Time Variable
• Time delay

• Time reversal

• Time scaling

Fig. 11: Transformations of the time variable

3.2 Amplitude Transformation

y(x) = Ax(t) + B
where A, B are constants.

3.3 Linear Combination

• In continuous-time domain:

y(t) = a1 x1 (t) + a2 x2 (t) + ... + aN xN (t)

• In discrete-time domain:

y[n] = a1 x1 [n] + a2 x2 [n] + ... + aN xN [n]

where ai 2 C, i = 1, 2, ..., N are real or complex numbers.

3.4 Multiplication
• In continuous-time domain:
y(t) = x1 (t) · x2 (t)

• In discrete-time domain:
y[n] = x1 [n] · x2 [n]

This implies instantaneous multiplication for each time instant or each discrete time
sample.

13
3 SIMPLE OPERATIONS ON SIGNALS

3.5 Scalar Products and Norms

3.5.1 Scalar Product and Norm of Vectors
~ and B:
• The scalar product between two 3-D vectors A ~ projection of A
~ on B.
~

~·B
A ~ = Ax · Bx + Ay · By + Az · Bz = |A|
~ · |B|cos(
~ )

• For vectors, the Euclidean norm, or norm-2, is the length of vectors in Euclidean
space: p q
~ 2= A
||A|| ~·A ~ = A = A2x + A2y + A2z

3.5.2 Scalar Product and Norm of Discrete-time Signals

• Scalar product for discrete-time signals:
1
X
hx1 [n], x2 [n]i = x1 [n]x⇤2 [n]
n= 1

• Norm-2 for discrete-time signals:

v ! 12
u1 1
p uX X
||x[n]||2 = hx[n], x[n]i = t |x[n]|2 = |x[n]|2
1 1

norm-2 is the square root of the energy of the signal

• Norm-p for discrete-time signals:

1
! p1
X
||x[n]||p = |x[n]|p , for 1  p < 1
1

– p = 1:
1
X
||x[n]||1 = |x[n]|
1

– p ! 1: infinity norm / maximum norm:

||x[n]||1 = max|x[n]|
n

• Norms are measures of the signal “strength”. Each norm is a di↵erent way of
measuring signal strength.
e.g. the norm-2 is associated to the energy.

• The space of signals with a finite norm p is called Lp space.

14
3 SIMPLE OPERATIONS ON SIGNALS

3.5.3 Scalar Product and Norm for Continuous-time Signals

• Scalar product for continuous-time signals:
Z +1
hx1 (t)x2 (t)i = x1 (t)x⇤2 (t)dt
1

• Norm-2 for continuous-time signals:

sZ ✓Z ◆ 12
p +1 1
2
||x(t)||2 = hx(t), x(t)i = |x(t)|2 dt = |x(t)|
1 1

• Norm-p for continuous-time signals:

✓Z +1 ◆ p1
p
||x(t)||p = |x(t)| dt ; ||x(t)||1 = max|x(t)|
1 n

3.6 Characterising Similarity/Di↵erence Between Signals

3.6.1 Measuring the Similarity
The scalar product is a measure of the similarity between two signals:
A~·B~ = |A|
~ · |B|cos(
~ )
Normalize the scalar product to the lengths of the vectors: if cos( ) = 1, the two vectors
have the same direction.
A~·B ~
cos( ) =
~ · |B|
|A| ~
The normalized scalar product for vectors indicates how similar the two vectors are.
• Normalized scalar product for vectors between continuous-time signals:
hx1 (t), x2 (t)i
||x1 (t)||2 ||x2 (t)||2
• Normalized scalar product for vectors between discrete-time signals:
hx1 [n], x2 [n]i
||x1 [n]||2 ||x2 [n]||2
3.6.2 Normalized Cross-correlation Function fc (✓)
• In practical conditions, the signals are corrupted by noise.

Fig. 12: Signals under non-ideal condition

15
3 SIMPLE OPERATIONS ON SIGNALS

• A possible estimate of the delay between the two signals is the time interval by
which we need to shift one of the signals so that it is maximally similar to
the other.
Normalized cross-correlation function between x1 (t) and x2 (t):
hx1 (t), x2 (t ✓)i
fc (✓) =
||x1 (t)||2 ||x2 (t)||2

R1
1
x1 (t)x2 (t ✓)dt
= qR qR
+1 2 dt · +1
1
|x 1 (t)| 1
|x2 (t ✓)|2 dt
• The ✓ value corresponding to the maximum of f (✓) is an estimate of the delay
between the two signals.

Example: Measure muscle fiber conduction velocity

By using cross-correlation function, we are able to find the time delay between two
EMG signals. By measuring the distance between two electrodes, we can calculate
the conduction velocity.

3.6.3 Measuring the Di↵erence

• An alternative way to measure the similarity between two signals is to compute the
strength of their di↵erence (i.e., norm).
• For example, using the norm-2 as measure of strength, we can define the mean
squared error (MSE) between signals :
M SE(✓) = ||x1 (t) x2 (t ✓)||22
Z +1
= |x1 (t) x2 (t ✓)|2 dt
1

– M SE(✓) is a function of the shift ✓.

– Minimum value of ✓ best estimates the delay.
– It is the energy of the error signal.

16
4 FOURIER SERIES

4 Fourier Series
4.1 Orthonormal Functions
• Orthonormal functions has the following property:
(
0, i 6= k
h i (t), k (t)i =
1, i = k

The N signals { i (t)}i=1...N with this property is referred to as an orthonormal

set of signals.

Think the orthogonal unitary vectors defining coordinate axes (i, j, k) in Euclidean
space!

• Consider a generic signal that can be described by a linear combination of a set

of orthonormal signals:
N
X
x(t) = ai i (t)
i=1

where ai are unknown complex or real numbers.

• The coefficients ai can be determined by projecting the signal into each function
{ i (t)}i=1...N :
N
X
hx(t), k (t)i = h ai i (t), k (t)i
i=1
Z 1 N
X
⇤
= ai i (t) k (t)dt
1 i=1
N
X Z 1
⇤
= ai i (t) k (t)dt
i=1 1

= ak

– Scalar product is a linear operator.

– The coefficients provide all the information in the signal: if we know ak , we
know the signal.
– If the signal x(t) belongs to a larger space, the projected signal will be an
approximation of the original signal with minimum MSE.

Example: Haar basis function

Given the signal: 8
>
<1, 0  t < 12
(t) = 1, 12 < t  1
>
:
0, otherwise

17
4 FOURIER SERIES

The following signals define a set of orthonormal basis functions:

rk = (2r t k) for r = 0, 1, 2... and k = 0, 1, 2, ..., 2r 1

Fig. 13: Haar basis functions

4.2 Fourier Basis Functions

Fourier basis functions are:
1 1 2⇡i
i (t) = p ej!i t = p ej T t
T T
t is defined in the time interval [0, T ].

These functions have the following properties:

• periodic, period = T
i

• period is a function of the fundamental period T

• frequencies are the multiples of fundamental frequency 1

T

• orthonormal, when 0  t  T :

– i 6= k,
Z T Z T
⇤ 1 2⇡i
j 2⇡k
h i (t), k (t)i = i (t) k (t)dt = ej T
t
e T
t
dt = 0
0 T 0

– i = k, Z Z
T T
2 1
h i (t), k (t)i = | i (t)| dt = dt = 1
0 T 0

18
4 FOURIER SERIES

4.3 Fourier Series

Fourier series: can be used to represent any periodic signal with finite energy in a single
period.
X1 Z T
j 2⇡k t 1 +2 2⇡k
x(t) = ck e T with ck = x(t)e j T t
k= 1
T T
2

Derivation
If we assume:
+1
X
x(t) = ai i (t)
i= 1
+1
1 X 2⇡k
=p ak e j T t
T k= 1

with the coefficients ak :

Z T
⇤
ak = hx(t), k (t)i = x(t) k (t)dt
0
Z T
1 j 2⇡k t
=p x(t) e T dt
T 0

An equivalent expression is the Fourier series

+1
X Z T
j 2⇡k t 1 j 2⇡k t
x(t) = ck e T with ck = x(t)e T dt
k= 1
T 0

• The infinite set of orthonormal functions of the Fourier series describes any periodic
signal with finite energy in a single period.

Fourier series with a finite number of terms:

+N
X 2⇡k
x(t) ⇡ xN (t) = ck e j T
t

k= N

Error of approximation decreases as the number of terms increases,

+N
X 2⇡k
EN (t) = x(t) xN (t) = x(t) ck e j T
t

k= N

Z T Z T
2
lim |EN (t)| dt = 0, if |x(t)|2 dt < 1
N !1 0 0
Fourier series converges as the number of terms increases:

19
4 FOURIER SERIES

Fig. 14: Example of convergence of the Fourier series of a periodic square wave

20
5 FOURIER TRANSFORM

5 Fourier Transform
5.1 From Fourier Series to Fourier Transform
To find a representation of any finite energy signal, not necessarily periodic:
set the periodical of a period signal to infinity.

(
1, |t| < T1
x(t) = T
0, T1 < |t| < 2

The Fourier series of the periodic signal

x(t) above is:
1
X 2⇡k
x(t) = ck e j T
t

k= 1

with
Z
1 T1 2⇡k
ck = x(t)e j T t
T T1
2 sin(k!0 T1 )
=
k!0 T
1 2 sin(!T1 )
=
T ! !=k!0

where !0 = 2⇡T
is the frequency of the
first harmonic.

Plot ck against !. As T increases, the frequency becomes smaller, the points on

the plot become closer.

Generalize the example above:

The signal x(t) defined in the interval t 2 [ T1 , T1 ]. We build the corresponding
periodic signal x̃(t), which equals to x(t) in one period. As T ! 1, x̃(t) ! x(t).

21
5 FOURIER TRANSFORM

1
X 2⇡k
x̃(t) = ck e j T
t

k= 1

with
Z T
1 2
j 2⇡k t
ck = x̃(t)e T dt
T T
2
Z +1
1 jk!0 t
= x(t)e dt
T 1
| {z }
X(k!0 )
1
= X(k!0 )
T
x(t) ! X(!) is known as Fourier transform.

1
X 2⇡k
x̃(t) = ck e j T
t

k= 1
X1
1 2⇡k
= X(k!0 )ej T t
k= 1
T
1
X
1
= X(k!0 ) ejk!0 t !0
2⇡ k= 1

As T ! 1, x̃(t) ! x(t), !0 ! 0:
Z 1
1
x̃(t) = x(t) = X(!) ej!t d! (1)
2⇡ 1

X(!) ! x(t) is known as inverse Fourier transform.

5.2 The Continuous-time Fourier Transform

Fourier transform: Z +1
j!t
X(!) = x(t) e dt
1

Inverse Fourier transform:

Z 1
1
x(t) = X(!) ej!t d!
2⇡ 1

• Fourier transform is a mathematical transformation employed to transform signals

between time domain and frequency domain.

• Fourier transform is a reversible operation.

• Fourier transform is a linear transformation: it is defined by an integral

22
5 FOURIER TRANSFORM

• For each value of !, the Fourier transform is a complex number representing the
projection of the signal on the complex exponential function ej!t .

X(!) = hx(t), ej!t i

• The Fourier transform exists for signals in the L2 space. These signals can
be expressed as the combination of functions ej!t .

Compare Fourier series and Fourirer transform:

Fourier series:
X1 Z
j 2⇡k t 1 T j 2⇡k t
x(t) = ck e T with ck = x(t) e T

k= 1
T 0

Fourier transform: Z +1
1
x(t) = X(!) ej!t d!
2⇡ 1
Z +1
j!t
X(!) = x(t) e dt
1

The Fourier series represent periodic signals with discrete frequencies; the Fourier
transform represents non-periodic signals with continuous frequencies.

5.3 Properties of Fourier Transform

5.3.1 Linearity
FT {a1 x1 (t) + a2 x2 (t)} = a1 X1 (!) + a2 X2 (!)

Derivation
Given
Z +1 Z +1
j!t j!t
X1 (!) = x1 (t) e dt and X2 (!) = x2 (t) e dt
1 1

Z +1
FT {a1 x1 (t) + a2 x2 (t)} = a1 x1 (t) + a2 x2 (t) e j!t dt
1
Z +1 Z +1
j!t j!t
= a1 x1 (t)e dt + a2 x2 (t)e dt
1 1
= a1 X1 (!) + a2 X2 (!)

5.3.2 Time shifting

j!t0 j!t0
FT {x(t t0 )} = e FT {x(t)} = e X(!)

23
5 FOURIER TRANSFORM

Derivation
Z +1
j!t
FT {x(t t0 )} = x(t t0 ) e dt
1

Let r = t t0 :
Z +1
j!(r+t0 )
FT {x(t t0 )} = x(r) e dr
1
Z +1
j!r j!t0
= x(r) e e dr
1
Z +1
j!t0 j!r
=e x(r) e dr
1

Let r = t:
Z +1
j!t0 j!t
FT {x(t t0 )} = e x(t) e dt
1
| {z }
Fourier Transform
j!t0
=e FT {x(t)}
j!t0
=e X(!)

5.3.3 Conjugation
FT {x⇤ (t)} = X ⇤ ( !)
if x(t) is real, X( !) = X ⇤ (!).

Derivation
The Fourier transform is:
Z +1
j!t
X(!) = x(t) e dt
1

Take the complex conjugate:

Z +1
⇤
X (!) = x⇤ (t) ej!t dt
1

Change ! ! !:
Z +1
⇤
X ( !) = x⇤ (t) e j!t
dt
1
= FT {x⇤ (t)}

5.3.4 Dual property

FT FT
if x(t) ! X(!), then X(t) ! 2⇡x( !)

24
5 FOURIER TRANSFORM

FT FT
Example: (t) ! 1, 1 ! 2⇡ (!)

Derivation
The Fourier transform is:
Z +1
j!t
X(!) = x(t) e dt
1

Change ! ! t, to avoid confusion, also change t ! u:

Z +1
X(t) = x(u) e jtu du
1

The inverse Fourier transform of x( !) is:

Z +1
1 1
FT {x( !)} = x( !)ej!t d!
2⇡ 1

Change ! ! u, Z +1
1 1 jut
FT {x(u)} = x(u)e du
2⇡ 1

This yield dual property:

FT
X(t) ! 2⇡x( !)

5.3.5 Time scaling

1 !
FT {x(at)} = X( )
|a| a
a is a non-zero real number

From the above property,

FT {x( t)} = X( !)

Derivation
Fourier transform of x(at) is:
Z +1
j!t
FT {x(at)} = x(at)e dt
1

Replace at ! u, then t = ua , dt = a1 du
Z +1
1 j! u
FT {x(u)} = x(u)e a du
|a| 1
1 !
= X( )
|a| a

25
5 FOURIER TRANSFORM

5.3.6 Parseval’s relation

Z +1 Z +1
2 1
|x(t)| dt = |X(!)|2 d!
1 2⇡ 1
2
the function |X(!)| is termed the energy-density spectrum of the signal.

Derivation
Start from the L.H.S:
Z +1 Z +1
2
|x(t)| dt = x(t)x⇤ (t)dt
1 1
Z +1 ✓ Z +1 ◆
1
= x(t) X ⇤ (!)e j!t
d! dt
1 2⇡ 1
Z +1 ✓ Z +1 ◆
1 ⇤ j!t
= X (!) x(t)e dt d!
2⇡ 1 1
Z +1
1
= X ⇤ (!)X(!)d!
2⇡ 1
Z +1
1
= |X(!)|2 d!
2⇡ 1

5.3.7 Di↵erentiation in time

⇢
dx(t)
FT = j!X(!)
dt
⇢ n
d x(t)
FT = (j!)n X(!)
dtn

5.3.8 Convolution
FT {x(t) ⇤ h(t)} = X(!)H(!)

Derivation
Z +1 Z +1
j!t
FT {x(t) ⇤ h(t)} = x(⌧ )h(t ⌧ )e dt
1 1

By the change of variables t ⌧ = ↵,

Z +1 Z +1
FT {x(t) ⇤ h(t)} = x(⌧ )h(↵)e j!(t+↵) d⌧ d↵
1 1
Z +1 Z +1
j!⌧
= x(⌧ )e d⌧ h(↵)e j!↵ d⌧
1 1
= X(!)H(!)

Convolution in the time domain is equivalent to multiplication in the Fourier do-

main.

26
5 FOURIER TRANSFORM

5.4 More basic Fourier transforms

5.4.1 Impulse
j!T
x(t) = (t T ) X(!) = e
In particular for T = 0, X(!) = 1.

5.4.2 Complex exponential

x(t) = ej!0 t X(!) = 2⇡ (! !0 )
j!0 t
x(t) = e X(!) = 2⇡ (! + !0 )

5.4.3 Cosine
x(t) = cos(t) X(!) = ⇡[ (! !0 ) + (! + !0 )]

5.4.4 Sine
⇡
x(t) = sin(t) X(!) = [ (! !0 ) (! + !0 )]
j

5.5 Example of application of properties of the Fourier trans-

form

vR
I

vin vc

For the circuit above, the input/output relation is characterized by the following
di↵erential equation:
dvc
vc + RC = vin
dt
By taking Fourier transform (using linearity property and di↵erentiation in time
property):
1
Vc (!) + j!RCVc (!) = Vin (!) ! Vc (!) = Vin (!)
1 + j!RC
• The Fourier transforms are coefficients indicating the weights of each complex
exponential signal e j!t composing the signals.
• The above relation therefore tells us how the circuit processes the input signal
by changing the weights of the coefficients.

27
5 FOURIER TRANSFORM

• In the frequency domain, the circuit acts as a factor that multiplies each
coefficient in a frequency-dependant way.
1
Let H(!) = 1+j!RC
, this represents the frequency response of the circuit.

Vc (!) = H(!)Vin (!)

Let the signal input vin (t) = ej!0 t , when applying the dual property:

Vin (!) = 2⇡ (! !0 )

Therefore:
2⇡ 2⇡
Vc (!) = (! !0 ) = (! !0 ) = H(!0 )2⇡ (! !0 )
1 + j!RC 1 + j!0 RC
Take inverse Fourier transform:
j!0 t
vc (t) = H(!0 )e

Since H(!0 ) is a complex number,

H(!0 ) = |H(!0 )| · e|j\H(!

{z
0)
} ! vc (t) = |H(!0 )|ej(!0 t+\H(!0 ))
| {z }
maginitude phase spectra

The frequency response of the circuit changes the magnitude and the phase of the
complex exponential, NOT the frequency.

If the input is an impulse signal vin (t) = (t), the response of the circuit in the
frequency domain is:

Vc (!) = H(!)Vin (!) = H(!), since Vin (!) = 1

And the response to the impulse in the time domain is:

1 1 t 1 t
FT {H(!)} = e RC u(t) = e ⌧ u(t) = h(t)
RC ⌧
From the example above, there is an input-output relation in the fre-
quency domain.
Y (!) = H(!)X(!)

5.5.1 LTI systems

The relation Y (!) = H(!)X(!) holds for a larger class of systems: LTI systems.

For any continuous-time signals:

Z +1
x(t) = x(⌧ ) (t ⌧ )d⌧
1

28
5 FOURIER TRANSFORM

Assume there is a system T {·}:

• associates an output y(t) to an input x(t) : y(t) = T {x(t)}.

• linear, acts on the signal in the same way over time.

For any signal x(t):

⇢Z +1 Z +1 ⇢
y(t) = T {x(t)} = T x(⌧ ) (t ⌧ )d⌧ = x(⌧ )T (t ⌧ )d⌧ d⌧
1 1

If h(t) = T { (⌧ )} (response of the system to the impulse response):

Z +1
T { (t ⌧ )} = h(t ⌧ ), then : y(t) = x(⌧ )h(t ⌧ )d⌧ = x(t) ⇤ h(t)
1

The output of a linear, time-invariant (LTI) system is the convolution of the input
with the impulse response, i.e. the system is fully defined by the impulse response.

5.6 Magnitude and Phase Spectra

In general, X(!) is a complex function of !:

X(!) = a(!) + jb(!) = |X(!)|ej\X(!)

• |X(!)| is magnitude, it describes the basic frequency content of a signal, i.e. the
relative magnitudes of the complex exponentials that make up x(t).

• \X(!) is phase angle, it determines the di↵erent look of signals, even if the
magnitude remains unchanged.

Examplea
A ship encounters the superposition of three wave trains, each of which can be
modeled as a sinusoidal signal.
1 2
x(t) = 1 + cos(2⇡t + 1) + cos(4⇡t + 2) + cos(6⇡t + 3)
2 3
With fixed magnitudes for these sinusoids, the amplitude of their sum may
be quite small or very large, depending on the relative phases. The
implications of phase for the ship, therefore, are quite significant.

29
5 FOURIER TRANSFORM

a
Example adopted from Signals and Systems, 2nd Edition, P424

Specifically, for the circuit example above:

1 1
|H(!0 )| = p =q
1 + ! 2 (RC)2 1 + ( !!c )2

!
\H(!0 ) = tan 1 (!RC) = tan 1 ( )
!c

5.7 Fourier Transform of Periodic Signals

If x(t) is a periodic signal:
+1
X
FT {x(t)} = 2⇡ ck (! k!0 )
k= 1

where Z + T2
1 j 2⇡k t
ck = x(t)e T dt
T T
2

Derivation
Fourier series of a periodic signal x(t) with period T is:
+1
X Z + T2
j 2⇡k t 1 j 2⇡k t
x(t) = ck e T with ck = x(t)e T dt
k= 1
T T
2

30
5 FOURIER TRANSFORM

By applying linearity property:

⇢ X
+1
2⇡k
X(!) = FT ck e j T
t

k= 1
+1
X
= ck FT {ejk!0 t }
k= 1
+1
X
= 2⇡ ck (! k!0 )
k= 1

5.7.1 Fourier transform of a train of impulses

+1
X +1
X
FT 2⇡
x(t) = (t nT ) ! X(!) = (! k!0 )
n= 1
T k= 1

where
2⇡
!0 =
T

Derivation
+1
X
x(t) = (t T)
n= 1

From the definition above, x(t) is periodic with period T :

+1
X Z + T2
j 2⇡k t 1 j 2⇡k t 1
x(t) = ck e T with ck = (t)e T dt =
k= 1
T T
2
T

Thus:
+1
2⇡ X 2⇡
X(!) = (! k!0 ) with !0 =
T n= 1 T

31
6 SAMPLING THEOREM

6 Sampling Theorem
To obtain a discrete-time signal from a continuous-time signal, we need a C/D converter.

Fig. 15: C/D converter

The mathematical expression for a C/D converter is:

x[n] = xc (nT ) 1 < n < +1

1
where T is sampling period, fs = T
is sampling frequency.

• In general, the C/D transformation cannot be inverted.

• Infinite continuous signals can reproduce a given sequence of samples,

An ideal C/D converter applies T property, so that the sampling can be done without
loosing information.

6.1 Sampling Process

Impulse train modulator s(t) is:
+1
X
s(t) = (t nT )
1

The sampled signal xs (t) is obtained by multiplying the impulse train modulator by the
continuous-time signal xc (t):

xs (t) = xc (t) s(t)

+1
X
= xc (t) (t nT )
n= 1
+1
X
= xc (nT ) (t nT )
n= 1

Sampled signal, xs (t), is still defined in continuous-time, but it contains all information
in the sampled discrete-time domain.

32
6 SAMPLING THEOREM

Apply the Fourier transform to xs (t):

Xs (!) = FT {xc (t)} · F T {s(t)}

1
= Xc (!) ⇤ F T {s(t)}
2⇡
+1
X
1
= Xc (!) ⇤ (! k!s )
T n= 1
+1
1 X
= Xc (! k!s )
T n= 1

2⇡
Fig. 16: (a)Original signal;(b) Sampled signal
where sampling frequency !s = !0 = T
.

For sampled signals: !N is the signal bandwidth

• if !s 2!N , the replicas in the periodization do not overlap

• if !s < 2!N , the replicas overlap, also known as aliasing1 .

Fig. 17: (c) !s 2!N ; (d) aliasing: !s < 2!N

6.2 Reconstruction Process

Ideal low-pass filters can be used to reconstruct the signals.

Fig. 18: Low-pass filter system

1
In Latin, ”alias” means ”others”

33
6 SAMPLING THEOREM

1. Multiply the sampled signals by the function Hr (!)

Xc (!) = Xs (!) · Hr (!)

Fig. 19: Hr (!)

2. Due to the convolution property:

Xc (!) = FT {xs (t) ⇤ hr (t)}

3. Apply inverse Fourier transform:

xc (t) = xs (t) ⇤ hr (t)

+1
X
= xc (nT ) (t nT ) ⇤ hr (t)
n= 1
+1
X
= xc (nT )hr (t nT )
n= 1

+1
X sin(⇡(t nT )/T )
= xc (nT )
n= 1
⇡(t nT )/T

with
sin(⇡t/T ) Fig. 20: Reconstruction
hr (t) =
⇡t/T

6.3 Nyquist-Shannon Sampling Theorem

Nyquist-Shannon sampling theorem provides the condition under which the C/D trans-
formation can be inverted without loosing information.

Let xc (t) be a band-limited signal with Xc (!) = 0, |!| !N . Then xc (t) is uniquely
determined by its samples x[n] = xc (nT ), if
2⇡
!s = 2!N
T
where 2!N is the minimal sampling rate and referred to as the Nyquist rate.

xc (t) can be reconstructed by generating a periodic impulse train in which

successive impulses have amplitudes that are successive sample values. This
impulse train is then processed through an ideal lowpass filter. The resulting
output signal will be exactly equal xc (t).

34