Introduction to Signal Processing

Lecture 1: Introduction

Filip Elvander

Dept. Information and Communications Engineering

Aalto University
0.
In this lecture

• Welcome to the course!

• Filip Elvander, professor of signal processing
• Originally from Sweden, lived in Belgium prior to coming to Finland
• Some practical administrative things.
• Course overview
• What is the course about?
• Why do we need signal processing?
• What can you do after the course?
• Introduction to signals

February 24, 2025

0.
Course administration

• Course literature
• Digital Signal Processing (Fourth Edition), John G. Proakis and
Dimitris G. Manolakis (publisher: Pearson).
• Course structure
• 11 lectures
• 2 exercise sessions per week. The sessions are not mandatory, but
you are encouraged to attend.
• Sets of problems for the exercise sessions can be found on
MyCourses.
• TAs: Anton Björkman, Linda Fabiani, and Ilmari Ayres.
• Examination
• Written exam: April 11 from 13:00 to 16:00 in TU2 - 2005.
• Hand-in assignments throughout the course: these are mandatory
and give bonus points that adds to the result from the exam. Note:
this applies only to the regular exam on April 11, 2025.
Assignments are handed in on MyCourses. Check instructions and
deadlines.
• Contact information
• Filip Elvander, [email protected]
February 24, 2025
0.
Course overview

Discrete-time signals • Signals and systems in discrete time.

(Ch. 2)
• Frequency-domain analysis:
Discrete-time systems • Signals in terms of oscillating
(Ch. 2) components, systems in terms
of how they respond to these.
Frequency analysis of signals
• Convenient and very useful for
(Ch. 4 and 7)
many applications
z-transform • Design and implementation of
(Ch. 3) systems (”filters”) that will modify a
signal the way you like.
Freq. analysis of systems and filters
(Ch. 5 and 7) • Prerequisites:
• Concepts of signals and
Implementation of systems
systems in continuous time.
(Ch. 9)
• Fourier analysis.
Filter design • Elementary calculus with
(Ch. 10) complex numbers.

February 24, 2025

0.
Course overview

Discrete-time signals Today:

(Ch. 2) • Introduction!
Discrete-time systems • Where do signals and signal
(Ch. 2) processing show up?
• Examples from some different fields.
Frequency analysis of signals
(Ch. 4 and 7) • Elementary signals.
• Reading: Chapter 1.
z-transform
(Ch. 3)

Freq. analysis of systems and filters

(Ch. 5 and 7)

Implementation of systems
(Ch. 9)

Filter design
(Ch. 10)

February 24, 2025

0.
What is a signal?
A signal is a function that conveys information about a phenomenon. Or more
prosaic:
• a signal is a function of one or more independent variables.
• Possible variables: time, position on earth, pixel coordinates
• Possible signals: sound, radar, medical images, video streams, stock
prices

0.4

0.3

0.2

0.1

-0.1

-0.2

-0.3
Hurricane Katrina (NOAA). 0 0.5 1
t (s)
1.5 2 2.5

February 24, 2025

0.
What can you do with signal processing?
Signal processing is concerned with three things:
• Building mathematical models for signals
• ... when you can measure what you are speaking about and
express it in numbers you know something about it; but when you
cannot measure it, when you cannot express it in numbers, your
knowledge is of a meagre and unsatisfactory kind. – Lord Kelvin
• Extract useful information from signals
• decode speech from a mobile phone call
• search for exoplanets
• find out if you’re a sci-fi-fan based on your Netflix views
• Modify signals or their information content
• separate mother and baby ECGs
• noise-reduction in hearing aids
• virtual and augmented reality

February 24, 2025

0.
Guitar notes
8
• Music and speech can be divided into
7 -40

tonal and non-tonal (noise-like) parts.

6 -60

Power/frequency (dB/Hz)
• Tonal: pitch, timbre. Non-tonal:

Frequency (kHz)
5
-80

texture. 4
-100
3
• Signal processing: separate the two. -120
2

8
1 -140

7 -40 0
100 200 300 400 500 600 700 800
Time (ms)
6 -60 Tonal content Play .

Power/frequency (dB/Hz)
Frequency (kHz)

8
5 -40
-80
7
4 -60
6
-100

Power/frequency (dB/Hz)
Frequency (kHz)
3
5 -80

-120
2 4
-100

3
1 -140
-120
2
0
100 200 300 400 500 600 700 800 1 -140

Time (ms)
0
100 200 300 400 500 600 700 800
Guitar note Play . Time (ms)

Sound examples from Roland Badeau. Non-tonal content Play .

February 24, 2025
0.
Radio astronomy

February 24, 2025

0.
Messier 87 supermassive black hole

• Event Horizon Telescope: global

network of radio telescopes.
• First image ever of a black hole
(2019).
• Signal processing: fuse together the
data, translate radio waves to
information (the image)

Messier 87∗ .
Event Horizon Telescope.

February 24, 2025

 0.
Gravitational waves
• Very difficult detection problem:
space-time contracts on the order of
a hair’s width.
• Look for signatures predicted by
General Relativity.
• Signal processing: match theoretical
signatures to data, clean out noise.

Abbott et al. (2016), Observation of Gravitational Waves from a Binary Black

Hole Merger, Physical Review Letters.

February 24, 2025

0.
Echo cancellation
• In laptops and phones, loudspeaker and microphone are very close to
each other ⇒ echoes and feedback.
• Signal processing: identify and cancel loudspeaker sound from
microphone.

Power/frequency (dB/Hz)
4
Frequency (kHz)

-20 4
-40
-20
-60
2
3.5
-80
-100 -40
-120 3

Power/frequency (dB/Hz)
0 -140

Frequency (kHz)
0.5 1 1.5 2 -60
2.5
Time (secs)

Power/frequency (dB/Hz)
4
Frequency (kHz)

-20
-80
-40 2
-60
2 -80
1.5 -100
-100
-120
0 1 -120
0.5 1 1.5 2
Time (secs)
Power/frequency (dB/Hz)

4 0.5 -140
Frequency (kHz)

-50
0
2 0.5 1 1.5 2
-100
Time (s)

0 -150
0.5 1
Time (secs)
1.5 2
Echo removed Play

Loudspeaker signal. Play Speech Play

Microphone signal. Play

February 24, 2025
0.
13dB miracle

• In storing and streaming of information (e.g., audio), we need to reduce

the data size (compression). Play
• How can we do this without introducing annoying artefacts?
• Signal processing: exploit psychoacoustics for efficient coding.
10 -40 10 -40

9 9

-60 -60
8 8

7 7

Power/frequency (dB/Hz)

Power/frequency (dB/Hz)
Frequency (kHz)

Frequency (kHz)
-80 -80
6 6

5 5 -100
-100

4 4

3 -120 3 -120

2 2
-140 -140
1 1

0 0
0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
Time (s) Time (s)

White (flat) noise Play . Shaped noise Play .

Johnston and Brandenburg, 1991.

February 24, 2025

0.
Electrocardiogram separation

Picture from DOI: 10.1038/s41597-020-0538-z

February 24, 2025

0.
EMG – prosthetics control

Electromyogram (EMG) can be used for

controlling prosthetic limbs. From a signal
processing point of view, this requires
1. Pre-processing (noise reduction and
feature extraction),
2. Detection (is there a signal or not?)
3. Classification (what is the desired
movement?)

Picture from DOI: 10.1109/ACCESS.2019.2891350

and DOI: 10.1016/j.eswa.2012.02.192

February 24, 2025

0.
Financial markets

• Financial time series (stock and

commodity prices, exchange rates,...)
are also a type of signals.
• Difference from earlier examples:
man-made
• Signal processing: get rich(???).
Stock exchanges use prediction
algorithms to try to catch ”flash
Black Monday (1987).
crashes” early as to halt trading.

2010 flash crash.

February 24, 2025

0.
Goals of the course

• Be able to analyze signals (we will come to which type)

• understand how they behave in time
• understand properties connected to frequency
• Be able to analyze systems (things that act on signals)
• the systems we will look at are often referred to as ”filters”.
• Be able to implement and to design systems
• create your own tools for manipulating signals

February 24, 2025

0.
Types of signals

• Signal dimension = number of independent variables

• 1D: typically functions x(t) where t is time.
• 2D: temperature on Earth depending on your GPS coordinates.
• Multi-channel signals: more than one ”measurement stream”.
 
• 2-channel: stereo sound s(t) = sL (t) sR (t) . Each of the
channels are 1D-signals (functions of time).
• 3-channel: color video 
s(x, y,t) = sR (x, y,t) sB (x, y,t) sG (x, y,t) . Each color channel is
3D, here function of pixel coordinates (x, y) and time t.
In this course, the focus is on signals that are single-channel and 1D.

February 24, 2025

0.
In this course: discrete-time 1D signals
The signal x(n) only exists for n = . . . , −2, −1, 0, 1, 2, . . ., that is
x = {. . . , x(−2), x(−1), x(0), x(1), x(2), . . .}.
Typically, x is obtained from sampling an analog (continuous-time) signal xa (t)
at regular intervals T :
x(n) = xa (T n) , n = . . . , −2, −1, 0, 1, 2, . . .

1 1

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0

-0.2 -0.2

-0.4 -0.4

-0.6 -0.6

-0.8 -0.8

-1 -1
0 5 10 15 20 0 5 10 15 20
t n

February 24, 2025

0.
Our signal building blocks
Two simple signals that you will see all the time:
(
1 , n=0
Impulse: δ (n) = = {. . . , 0, 0, 1, 0, 0, . . .}
0 , n ̸= 0

1.4 1.4

1.2 1.2

1 1

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0

-0.2 -0.2
-5 -4 -3 -2 -1 0 1 2 3 4 5 -5 -4 -3 -2 -1 0 1 2 3 4 5

Figure: Impulse, δ [n]. Figure: Step, u[n].

February 24, 2025
0.
Our signal building blocks
Two simple signals that you will see all the time:
(
1 , n=0
Impulse: δ (n) = = {. . . , 0, 0, 1, 0, 0, . . .}
0 , n ̸= 0
(
1 , n≥0
Step: u(n) = = {. . . , 0, 0, 1, 1, 1, . . .}
0 , n<0

1.4 1.4

1.2 1.2

1 1

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0

-0.2 -0.2
-5 -4 -3 -2 -1 0 1 2 3 4 5 -5 -4 -3 -2 -1 0 1 2 3 4 5

Figure: Impulse, δ [n]. Figure: Step, u[n].

February 24, 2025
0.
The impulse
You can build any discrete-time signal by scaling and delaying impulses:

δ (n) = {. . . , 0, 0, 1, 0, 0, . . .} ⇒ aδ (n − 1) = {. . . , 0, 0, 0, a, 0, . . .}

For a general signal:

x(n) = {. . . , a−2 , a−1 , a0 , a1 , a2 , . . .}

= . . . + a−2 {. . . , 1, 0, 0, 0, 0, . . .} + a−1 {. . . , 0, 1, 0, 0, 0, . . .}
+ a0 {. . . , 0, 0, 1, 0, 0, . . .}
+ a1 {. . . , 0, 0, 0, 1, 0, . . .} + a2 {. . . , 0, 0, 0, 0, 1, . . .} + . . .
= . . . + a−2 δ (n + 2) + a−1 δ (n + 1) + a0 δ (n) + a1 δ (n − 1) + a2 δ (n − 2) + . . .
∞
= ∑ am δ (n − m).
m=−∞

This idea might seem like a silly thing but will be very useful.

February 24, 2025

0.
The step
The step function can be used to truncate a the negative time indices of a
signal:
(
x(n) , n ≥ 0
u(n)x(n) =
0 , n<0
or to window out a desired part of it
(
x(n) , p ≤ n < q
(u(n − p) − u(n − q)) x(n) =
0 , n < p or n ≥ q

1.5
7
1
6 0.5
0
5 -0.5
-5 -4 -3 -2 -1 0 1 2 3 4 5
4 1.5

3 1
0.5
2 0
-0.5
1 -5 -4 -3 -2 -1 0 1 2 3 4 5
1.5
0
1
-1 0.5
0
-2
-0.5
-5 -4 -3 -2 -1 0 1 2 3 4 5 -5 -4 -3 -2 -1 0 1 2 3 4 5

February 24, 2025

0.
The step
The step function can be used to truncate a the negative time indices of a
signal:
(
x(n) , n ≥ 0
u(n)x(n) =
0 , n<0
or to window out a desired part of it
(
x(n) , p ≤ n < q
(u(n − p) − u(n − q)) x(n) =
0 , n < p or n ≥ q

1.5
7
1
6 0.5
0
5 -0.5
-5 -4 -3 -2 -1 0 1 2 3 4 5
4 1.5

3 1
0.5
2 0
-0.5
1 -5 -4 -3 -2 -1 0 1 2 3 4 5
1.5
0
1
-1 0.5
0
-2
-0.5
-5 -4 -3 -2 -1 0 1 2 3 4 5 -5 -4 -3 -2 -1 0 1 2 3 4 5

February 24, 2025

0.
Our system building blocks

x(n) ax(n)
a

x(n) x(n − 1)
z−1

x1 (n)

x2 (n) x1 (n) + x2 (n)

+

x(n)
x(n)
x(n)

February 24, 2025

0.
An example system

x(n) y(n)
b0 + +

z−1 z−1

b1 −a1

• The system manipulates the signal x(n) and creates the output y(n).
• This looks fairly complicated: can we say something about what the
system does to the signal?
• If x(n) is a sound signal, how does y(n) sound?
• Central part of the course: analyze the behavior of systems, and to
design new systems.
• As we will see, it is convenient to talk about this in terms of frequency.
February 24, 2025
0.
ECG – powerline interference

1000 1.2
ECG

800 1

600 0.8

400 0.6

200 0.4

0 0.2

-200 0
0 0.5 1 1.5 10 20 30 40 50 60 70
time (s) frequency (Hz)

February 24, 2025

0.
ECG – powerline interference

1000 1.2
ECG

800 1

600 0.8

400 0.6

200 0.4

0 0.2

-200 0
0 0.5 1 1.5 10 20 30 40 50 60 70
time (s) frequency (Hz)

February 24, 2025

0.
Sinusoids and periodicity
A continuous-time sinusoidal signal is given by

xa (t) = A cos(2πF0 t + ϕ) , F0 frequency in Hz.

We can sample xa (t) with sampling frequency Fs = 1/T , giving the

discrete-time signal

x(n) = xa (nT ) = xa (n/Fs ) = A cos(2πF0 /Fs n + ϕ) = A cos(2π f0 n + ϕ),

for n = . . . , −2, −1, 0, 1, 2, . . ., where f0 = F0 /Fs is the relative frequency.

Periodicity
A continuous-time signal xa (t) is periodic with period T ∈ R if xa (t + T ) = xa (t)
for all t ∈ R.
A discrete-time signal x(n) is periodic with period T ∈ N if x(n + T ) = x(n) for
all n ∈ N.
Note: all continuous-time sinusoids are periodic, by in discrete-time the
frequency has to be a rational number.

February 24, 2025

0.
Sinusoids
Let’s take two different cont. time signals xa (t) and xk (t) as
xa (t) = A cos(2πF0 t + ϕ) , xk (t) = A cos(2πFk t + ϕ)
where Fk = F0 + kFs , with k = ±1, ±2, . . .. Sampling xk (t) with sampling
frequency Fs yields
x(n) = xk (n/Fs ) = A cos(2π(F0 + kFs )/Fs n + ϕ)
= A cos(2πF0 /Fs n + ϕ + nk2πFs /Fs )
= A cos(2πF0 /Fs n + ϕ)
= xa (n/Fs )
= A cos(2π f0 n + ϕ).
Exactly the same as samples from xa !
• xa (t) and xk (t) both correspond to the same discrete-time signal with
relative frequency f0 = F0 /Fs .
• xk (t) is called an alias of xa (t).
• In fact, there are infinitely many aliases as k = ±1, ±2, . . . yield exactly
the same x(n).
• Conclusion: −Fs /2 ≤ F0 < Fs /2 or equivalently −1/2 ≤ f0 ≤ 1/2.
February 24, 2025
0.
Aliasing

0.8

0.6

0.4

0.2

-0.2

-0.4

-0.6

-0.8

-1
0 1 2 3 4 5 6 7 8

Figure: Continuous time.

February 24, 2025

0.
Aliasing

1 1

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0

-0.2 -0.2

-0.4 -0.4

-0.6 -0.6

-0.8 -0.8

-1 -1
0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8

Figure: Continuous time. Figure: Discrete time.

February 24, 2025

0.
Aliasing

1 1

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0

-0.2 -0.2

-0.4 -0.4

-0.6 -0.6

-0.8 -0.8

-1 -1
0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8

Figure: Continuous time. Figure: Discrete time.

February 24, 2025

0.
Aliasing
In general, consider a continuous-time signal
L
xa (t) = ∑ Aℓ cos(2πFℓt + ϕℓ )
ℓ=1

and sample it with sampling frequency Fs to get x(n) = xa (n/Fs ).

• if −Fs /2 ≤ Fℓ < Fs /2 for all ℓ, then xa (t) is uniquely defined by the
discrete samples x(n), n = . . . , −2, −1, 0, 1, 2, . . . This means that we can
reconstruct xa (t) from x(n).
• if |Fℓ | > Fs /2 for some ℓ, then this is not longer possible. Fℓ can be
replaced with another frequency F̃ℓ : −Fs /2 ≤ F̃ℓ < Fs /2 that yields exactly
the same x(n) (but this corresponds to a different cont.-time signal!)
• This is called aliasing and is what you see on a video where a car wheel
seems to spin backwards Link .
• Fs /2 is called the Nyquist rate: the limit on the frequencies you can
reconstruct.
• In practice, frequencies higher than Fs /2 will cause distortion.

February 24, 2025

0.
Aliasing

-20 -20
xa xa
x x
-40 Fs/2
-40 Fs/2

-60 -60

-80 -80
dB

dB
-100 -100

-120 -120

-140 -140

-160 -160
0 5 10 15 20 0 500 1000 1500 2000
frequency (kHz) frequency (Hz)

Original (blue) Low-pass filtered (blue) Down-sampled (red)

February 24, 2025

0.
Where does an aliased frequency show up?
If a continuous-time signal xa (t) with frequency F0 is sampled with sampling
frequency Fs with F0 > Fs /2, then this will ”look” exactly as some other
frequency in the interval (0, Fs /2). Which is that?
• Any frequency F when sampled at Fs looks the same as F + kFs , for any
integer k. Note: k can be negative!
• In order to find the frequency that F0 is aliased to, we need to find the
corresponding k such that F0 + kFs ∈ (−Fs /2, Fs /2). If you get a negative
frequency, you simply flip it to the corresponding positive one. Why is
this true?
Algorithm:
1. Is F0 > Fs /2? Then there will be aliasing.
2. Find k such that F0 + kFs ∈ (−Fs /2, Fs /2). Call this Fk = F0 + kFs .
3. The aliased frequency is F̃0 = |Fk |.

February 24, 2025

0.
Summary of today

• Signals and signal processing show up everywhere!

• Brief introduction to our most important ”basic” signals: the impulse, the
step, and sinusoids.
• Brief introduction of the standard ”system blocks”.
• The concept of frequency for continuous-time and discrete-time signals.
• Reading instructions: Chapter 1 Introduction. For next lecture
(Thursday), have a look at Chapter 2.
• Exercises for chapters 1 and 2 can be found on the course webpage.

February 24, 2025