Introduction to Signal Processing

Lecture 3: Frequency analysis of signals

Part 1 – continuous time

Filip Elvander

Dept. Information and Communications Engineering

Aalto University
0.
Today’s lecture

Discrete-time signals • We leave time-domain and enter

(Ch. 2) frequency domain.
Discrete-time systems • Fourier analysis: describe signals by
(Ch. 2) using sinusoids.
• Concepts: Fourier transform, inverse
Frequency analysis of signals
Fourier transform, spectrum.
(Ch. 4 and 6.1 (Ch. 7 later))
• Today: we look at these concepts in
z-transform continuous time.
(Ch. 3)
• Next lecture: what happens in
Freq. analysis of systems and filters discrete time, and the interface
(Ch. 5 and 7) between continuous and discrete
time.
Implementation of systems
• Reading: Chapter 4.
(Ch. 9)

Filter design
(Ch. 10)

March 3, 2025
0.
From last week: eigenfunctions of LTI systems
Recall, input complex exponential (α ∈ R and ω ∈ R):
x(n) = en(α+ jω) , n = . . . , −2, −1, 0, 1, 2, . . . ,
Then, for an LTI system T ,
∞
y(n) = T (x(n)) = x(n) ∑ h(k)e−k(α+ jω) .
k=−∞
Even simpler, let α = 0:
∞
y(n) = T (x(n)) = x(n)H(ω) , H(ω) = ∑ h(k)e−k jω .
k=−∞

If we have a sum of sinusoids, by linearity of the system

P P
x(n) = ∑ c p en jω p
⇒ y(n) = T (x(n) = ∑ H(ω p )c p en jω .p

p=1 p=1

• LTIs give us the same sinusoids back, with complex scaling

depending on frequency.
• What if we can describe signals in general with sinusoids?
March 3, 2025
0.
Two different types of building blocks
Last week we decomposed a signal into a sequence of impulses
∞
x(n) = ∑ x(k)δ (n − k).
k=−∞

By linearity and time-invariance we then had that

∞
y(n) = T (x(n)) = ∑ x(k)h(n − k) = x(n) ∗ h(n).
k=−∞

Now, we will instead try to decompose a signal into sinusoids:

x(n) = ∑ ck en jωk ,
k
which by LTI would allow us to write
y(n) = T (x(n)) = ∑ H(ωk )ck en jωk .
k

• h(n) impulse response of the system

• H(ω) frequency response of the system. We will return to this in later
lectures.
March 3, 2025
0.
Frequency analysis

• Visible light can be decomposed into

different colors. This is called
analysis, and can be performed
using a prism.
• The split-up colors are called the
spectrum.
• We can also build any light we want
by adding appropriate colors together.
This is called synthesis and can also
be done using a prism.
• Light color ↔ frequency of
electromagnetic wave.
• Let’s do the same thing for signals!

March 3, 2025
0.
Today in a nutshell: linear algebra
T
Let x = x1 x2 . . . xN ∈ RN . We can interpret the numbers x1 , . . . , xN as


coordinates in the canonical basis:

 T
x = x1 e1 + x2 e2 + · · · + xN eN , ek = . . . 0 1
|{z} 0 ...
element k

• eTk eℓ
= 0 if k ̸= ℓ and eTk ek
= 1 ⇒ orthogonal ⇒ linearly independent.
• N linearly independent vectors ⇐⇒ basis for RN .
• Orthogonality: we can get the coordinates by simple multiplication,
ek T x = x1 ek T e1 +x2 ek T e2 + . . . + xk ek T ek + . . . + xN ek T eN = xk .
| {z } | {z } | {z } | {z }
=0 =0 =1 =0
Let a1 , a2 , . . . , aN be some other orthogonal basis for RN . That means
∃c1 , . . . , cN ∈ R : x = c1 a1 + . . . + cN aN .
Again, we can get the coefficients ck easily:
aTk x
ak T x = x1 ak T a1 + . . . + xN ak T aN = ak T ak ck ⇒ ck =
aTk ak
Today, we will do the same thing, but for signal spaces instead of RN : find a
basis that we can represent the signal in, and find the coefficients.
March 3, 2025
0.
Spoiler alert
In linear algebra, for any invertible matrix A ∈ RN×N , the operation

b = A−1 x

is a change-of-basis or transform: b is the representation of x in the basis of

the columns of A as x = Ab.
• The choice of basis depends on the application: what are you going to
use it for?
• In our case, it would be nice to be able to work with sinusoids because
of LTI systems.
• We are in luck! We can create a basis and a corresponding transform.
• x ↔ signal, A−1 ↔ Fourier transform.
• A transform lets us keep all the information of a signal, but gives a
representation that is easier to work with.

March 3, 2025
0.
WARNING

• All lot of today’s material has to do with infinite series and integrals over
the real line.
• Questions about convergence are important: operations like
interchanging summation and integration are only valid sometimes.
• This lecture is an overview of the application of Fourier techniques to
signals, but mathematical details are left out.
• The book is a bit sloppy in this respect: for a more exact treatment, see
your course on Fourier analysis.
• The Fourier transform has many important symmetry properties: for
these, see the last section of Chapter 4.

March 3, 2025
0.
Four types of signals
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 2 4 6 8 10 0 2 4 6 8 10
t t
Continuous and periodic. Continuous and aperiodic.
5 5

4 4

3 3

2 2

1 1

0 0

-1 -1

-2 -2

-3 -3

-4 -4

-5 -5
0 5 10 15 20 0 5 10 15 20
n n
Discrete and periodic. Discrete and aperiodic.
March 3, 2025
0.
Four types of signals
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 2 4 6 8 10 0 2 4 6 8 10
t t
Continuous and periodic. Continuous and aperiodic.
5 5

4 4

3 3

2 2

1 1

0 0

-1 -1

-2 -2

-3 -3

-4 -4

-5 -5
0 5 10 15 20 0 5 10 15 20
n n
Discrete and periodic. Discrete and aperiodic.
March 3, 2025
0.
Sinusoids
Consider a continuous-time (complex) sinusoid xa (t) according to

xa (t) = ce j2πF0 t , t ∈ R.

• F0 ∈ R is the frequency, c ∈ C is the (complex) amplitude.

• xa (t) is clearly periodic: xa (t) = xa (t + Tp ) for any t ∈ R were Tp = 1/F0 .
• Tp is the period of the signal.

March 3, 2025
0.
Sums of sinusoids
Let’s now say that xa (t) is a sum K sinusoids according to
K
xa (t) = ∑ ck e j2πF t
k

k=1

• Fk frequency for sinusoid k ⇒ corresponding period is Tk = 1/Fk .

• Can xa (t) be a periodic signal, i.e., is there Tp so that xa (t) = xa (t + Tp )
for all t ∈ R?

K K K
?
xa (t + Tp ) = ∑ ck e j2πF (t+T ) = ∑ ck e j2πF t e j2πF T
k p k k p
= xa (t) = ∑ ck e j2πF t .
k

k=1 k=1 k=1

The only way to get equality for all t is to have e j2πFk Tp = 1, i.e., Fk Tp ∈ Z (Z is
the set of integers) for k = 1, . . . , K.
Fk Fk Tp N
⇒ = = k , for some Nk , Nℓ ∈ Z
Fℓ Fℓ Tp Nℓ

It follows directly that there has to exist some F0 such that Fk = F0 Nk , Nk ∈ Z.

March 3, 2025
0.
Non-harmonic mixture

• For an arbitrary set of sinusoids, the signal will not repeat itself.
• Example below: the period Tp of the ”slowest” sinusoid is superimposed
in all plots.

March 3, 2025
0.
Harmonic mixture

• For a harmonic mixture, all frequencies Fk are Fk = kF0 for some F0 .

• Equivalently, the period of each component is Tk = Tp /k, for some Tp .
• F0 is the fundamental frequency, and Tp is the fundamental period.

March 3, 2025
0.
Harmonically related sinusoids
We have shown that the only sums of sinusoids that are periodic, are those
that are harmonically related:
K
xa (t) = ∑ ck e j2πkF t ,
0
for some F0 .
k=1

• The period of the signal is Tp = 1/F0 .

• F0 is called the fundamental frequency.
• The individual sinusoids are called harmonics.

March 3, 2025
0.
Two immediate questions

1. What happens if instead of K harmonics, we have infinitely many? That

is, can we have something like
∞
xa (t) = ∑ ck e j2πkF0 t .
−∞

2. If it is so that any harmonic series is a periodic signal, can any periodic

signal be written as a harmonic series?
We have two ”it depends” answers:
1. It depends on how you choose ck .
2. It depends what you mean by ”be written as”, and on how ”nice” the
signal xa (t) is.
For proper answers, you have to check your Fourier analysis course. In our
context we can say:

In many interesting and useful cases, YES!

March 3, 2025
0.
Electrocardiogram

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)

March 3, 2025
0.
Electrocardiogram

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)

March 3, 2025
0.
Continuous-time periodic signals
Let xa (t) be a continuous-time periodic signal with period Tp . If the signal is
”nice” enough, then we can(!!) represent it as a Fourier series:
∞
xa (t) = ∑ ck e j2πkF0 t , F0 = 1/Tp .
k=−∞

What are the coefficients ck ? For ℓ ∈ Z(= {0, ±1, ±2, . . .}),
Z Tp Z Tp ∞ Z Tp ∞
xa (t)e− j2πℓF0 t dt = ∑ ck e j2πkF0 t e− j2πℓF0 t dt = ∑ ck e j2π(k−ℓ)F0 t dt
0 0 k=−∞ 0 k=−∞
∞ Z Tp
”nice”
= ck
∑ e j2π(k−ℓ)F0 t dt
k=−∞ 0
∞ Z Tp
”nice”
= ck
∑ e j2π(k−ℓ)F0 t dt = Tp cℓ .
k=−∞ |0 {z }
=Tp if k = ℓ, otherwise 0

Z Tp
1
⇒ ck = xa (t)e− j2πkF0 t dt , k = 0, ±1, ±2, . . .
Tp 0

March 3, 2025
0.
Continuous-time periodic signals
The set e j2πF0 kt | k ∈ Z are orthogonal functions on the interval [0, Tp ]:


Z Tp
(
Tp if k = ℓ
e j2πkF0 t e− j2πℓF0 t dt =
0 0 if k ̸= ℓ.
Recall from linear algebra that if u1 , u2 , . . . , uN form an orthonormal basis for
RN , then for any vector x ∈ RN ,
N
x= ∑ ck uk , where ck = uTk x.
k=1

Our case is very similar!

∞ Z Tp
1
xa (t) = ∑ ck e j2πkF0 t , where ck = xa (t)e− j2πkF0 t dt
k=−∞ Tp 0

• Signal space with infinite dimension ⇒ infinitely many basis functions.

• A Fourier series has infinitely many terms. What does the equality
xa (t)” = ” ∑∞ j2πkF0 t actually mean?
k=−∞ ck e
• We mean that the series converges to xa (t).
March 3, 2025
0.
Quick facts on convergence
Convergence of the Fourier series to xa (t):
N
partial sums SN (t) = ∑ ck e j2πkF0 t → xa (t) as N → ∞.
k=−N

But we can mean different things with the convergence ” → ”.

In energy
R Tp
If 0 |xa (t)|2 dt < ∞ then
Z Tp
|SN (t) − xa (t)|2 dt → 0 , N → ∞.
0

Note: there can still be t ∈ [0, Tp ] so that SN (t) − xa (t) ̸→ 0.

Almost everywhere
Dirichlet conditions:
R Tp
1. 0 |xa (t)| dt < ∞,
2. xa (t) is of bounded variation: has countably many discontinuities and
countably many minima/maxima,
Then, SN (t) − xa (t) → 0 for all t where xa (t) is continuous.
March 3, 2025
0.
Signal decomposition
n  o
The set of pairs kF0 , |ck |2 | k ∈ Z is called the power spectrum of xa (t),
motivated by
Z Tp ∞
1
Parseval’s identity: |xa (t)|2 dt = ∑ |ck |2 ⇒ |ck |2 power at kF0 .
Tp 0 k=−∞

1 0.025
| ck | 2
0.8

0.6 0.02

0.4

0.2 0.015

-0.2 0.01

-0.4

-0.6 0.005

-0.8

-1 0
0 2 4 6 8 10 -6 -4 -2 0 2 4 6
t F (Hz)

Cont.-time periodic signal. Power spectrum.

March 3, 2025
0.
Signal decomposition
n  o
The set of pairs kF0 , |ck |2 | k ∈ Z is called the power spectrum of xa (t),
motivated by
Z Tp ∞
1
Parseval’s identity: |xa (t)|2 dt = ∑ |ck |2 ⇒ |ck |2 power at kF0 .
Tp 0 k=−∞

1 0.15

0.8

0.6

0.4
0.1
0.2

-0.2
0.05
-0.4

-0.6

-0.8

-1 0
0 1 2 3 4 5 6 7 -6 -4 -2 0 2 4 6
F (Hz)

March 3, 2025
0.
Signal decomposition
n  o
The set of pairs kF0 , |ck |2 | k ∈ Z is called the power spectrum of xa (t),
motivated by
Z Tp ∞
1
Parseval’s identity: |xa (t)|2 dt = ∑ |ck |2 ⇒ |ck |2 power at kF0 .
Tp 0 k=−∞

1 0.15

0.8

0.6

0.4
0.1
0.2

-0.2
0.05
-0.4

-0.6

-0.8

-1 0
0 1 2 3 4 5 6 7 -6 -4 -2 0 2 4 6
F (Hz)

March 3, 2025
0.
Signal decomposition
n  o
The set of pairs kF0 , |ck |2 | k ∈ Z is called the power spectrum of xa (t),
motivated by
Z Tp ∞
1
Parseval’s identity: |xa (t)|2 dt = ∑ |ck |2 ⇒ |ck |2 power at kF0 .
Tp 0 k=−∞

1 0.15

0.8

0.6

0.4
0.1
0.2

-0.2
0.05
-0.4

-0.6

-0.8

-1 0
0 1 2 3 4 5 6 7 -6 -4 -2 0 2 4 6
F (Hz)

March 3, 2025
0.
Signal decomposition
n  o
The set of pairs kF0 , |ck |2 | k ∈ Z is called the power spectrum of xa (t),
motivated by
Z Tp ∞
1
Parseval’s identity: |xa (t)|2 dt = ∑ |ck |2 ⇒ |ck |2 power at kF0 .
Tp 0 k=−∞

1 0.15

0.8

0.6

0.4
0.1
0.2

-0.2
0.05
-0.4

-0.6

-0.8

-1 0
0 1 2 3 4 5 6 7 -6 -4 -2 0 2 4 6
F (Hz)

March 3, 2025
0.
Signal decomposition
n  o
The set of pairs kF0 , |ck |2 | k ∈ Z is called the power spectrum of xa (t),
motivated by
Z Tp ∞
1
Parseval’s identity: |xa (t)|2 dt = ∑ |ck |2 ⇒ |ck |2 power at kF0 .
Tp 0 k=−∞

1 0.15

0.8

0.6

0.4
0.1
0.2

-0.2
0.05
-0.4

-0.6

-0.8

-1 0
0 1 2 3 4 5 6 7 -6 -4 -2 0 2 4 6
F (Hz)

March 3, 2025
0.
Summary continuous-time periodic signals
Z Tp
1
Analysis equation: ck = xa (t)e− j2πkF0 t dt
Tp 0
∞
Synthesis equation: xa (t) = ∑ ck e j2πkF0 t
k=−∞
Z Tp ∞
1
Parseval’s identity: |xa (t)|2 dt = ∑ |ck |2
Tp 0 k=−∞
1 0.025
| ck | 2
0.8

0.6 0.02

0.4

0.2 0.015

-0.2 0.01

-0.4

-0.6 0.005

-0.8

-1 0
0 2 4 6 8 10 -6 -4 -2 0 2 4 6
t F (Hz)

Cont.-time periodic signal. Power spectrum.

March 3, 2025
0.
Electrocardiogram
• The location of the different peaks are kF0 , k = 1, . . . , K ≈ 30.
• F0 = 1/Tp , where Tp is the period (heart rate).
• The height of each peak is |ck |2 .
• The relative ”strength” of the different harmonics determines the shape
of the ECG waveform.
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)
March 3, 2025
0.
Four types of signals
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 2 4 6 8 10 0 2 4 6 8 10
t t
Continuous and periodic. Continuous and aperiodic.
5 5

4 4

3 3

2 2

1 1

0 0

-1 -1

-2 -2

-3 -3

-4 -4

-5 -5
0 5 10 15 20 0 5 10 15 20
n n
Discrete and periodic. Discrete and aperiodic.
March 3, 2025
0.
Can aperiodic signals have frequency?
Let’s say that that our aperiodic signal xa (t) is zero outside some interval
[−Tp /2, Tp /2]. Then, we can create a periodic extension x p (t) by periodically
repeating xa (t):

x p (t) = xa (t mod Tp ).

By construction x p (t) is periodic and thus has a Fourier series:

∞
x p (t) = ∑ ck e j2πkF0 t , F0 = 1/Tp .
k=−∞
|{z}
??

0.5

-0.5

-1

March 3, 2025
0.
Periodic extension
Note that, as x p (t) ≡ xa (t) for t ∈ [−Tp /2, Tp /2] and xa (t) ≡ 0 for
t∈
/ [−Tp /2, Tp /2]
Z Tp /2 Z Tp /2
analysis 1 1
ck = x p (t)e− j2πkF0 t dt= xa (t)e− j2πkF0 t dt
Tp −Tp /2 Tp −Tp /2
1
Z ∞
= xa (t)e− j2πkF0 t dt.
Tp −∞

Now comes the hand-wavy part: let’s define a function Xa (F), for any F, as
Z ∞
Xa (F) = xa (t)e− j2πFt dt.
−∞
1
Then, we have ck = Tp Xa (kF0 ) ⇒ Tp ck = Xa (kF0 ) = Xa (k/Tp ),
∞ ∞
synthesis 1
x p (t) = ∑ ck e j2πkF0 t = ∑ Xa (k/Tp )e j2πk/Tp t
k=−∞ T
k=−∞ p
∞
= ∑ Xa (k∆F )e j2πk∆F t ∆F , ∆F = 1/Tp .
k=−∞

Where does this lead us?

March 3, 2025
0.
Hand-wavy motivation (1/2)
− j2πkFt dt):
R∞
We have the following (with Xa (F) = −∞ xa (t)e
1. x p (t) ≡ xa (t) for t ∈ [−Tp /2, Tp /2].
2. x p (t) = ∑∞ j2πk∆F t ∆ , ∆ = 1/T .
k=−∞ Xa (k∆F )e F F p
By 2) we have a frequency representation for x p (t), but not for xa (t).

Here comes the (slightly hand-wavy) observation:

x p (t) → xa (t) for all t when Tp → ∞.

March 3, 2025
0.
Hand-wavy motivation (1/2)
− j2πkFt dt):
R∞
We have the following (with Xa (F) = −∞ xa (t)e
1. x p (t) ≡ xa (t) for t ∈ [−Tp /2, Tp /2].
2. x p (t) = ∑∞ j2πk∆F t ∆ , ∆ = 1/T .
k=−∞ Xa (k∆F )e F F p
By 2) we have a frequency representation for x p (t), but not for xa (t).

Here comes the (slightly hand-wavy) observation:

x p (t) → xa (t) for all t when Tp → ∞.

Can we then say something about xa (t)?

March 3, 2025
0.
Hand-wavy motivation (2/2)

1. x p (t) → xa (t) as Tp → ∞, (or equivalently ∆F → 0)

2. x p (t) = ∑∞ j2πk∆F t ∆ , ∆ = 1/T .
k=−∞ Xa (k∆F )e F F p
Then it should follow that
∞
∑ Xa (k∆F )e j2πk∆F t ∆F → xa (t) , ∆F → 0.
k=−∞

But what happens with the left-hand side when ∆F → 0? In fact, this is
a Riemann sum! As ∆F → 0, this converges to the Riemann integral
∞ Z ∞
j2πk∆F t
lim ∑ Xa (k∆F )e ∆F = Xa (F)e j2πFt dF.
∆F →0 −∞
k=−∞

We have found the Fourier transform and the inverse Fourier

transform for the aperiodic signal xa (t)!

March 3, 2025
0.
Continuous-time aperiodic signals
For aperiodic functions, the Fourier series is replaced by the Fourier
transform:
Z ∞
Analysis equation: Xa (F) = xa (t)e− j2πFt dt (Fourier transform)
−∞
Z ∞
Synthesis equation: xa (t) = Xa (F)e j2πFt dF (inverse Fourier transform)
−∞
Z ∞ Z ∞
Parseval’s identity: |xa (t)|2 dt = |X(F)|2 dF
−∞ −∞

• We need intervals of frequencies in order to describe the signal.

• For the existence of the FT, we need more from xa (t): integrability
conditions. Basically, xa (t) has to decay fast enough.
• In fact, we can use these tools can also for periodic signals, but then we
need the concept of distributions (see your Fourier analysis course).

March 3, 2025
0.
Differences between periodic and aperiodic signals
Periodic
Z Tp
1
Analysis equation: ck = xa (t)e− j2πkF0 t dt
Tp 0
∞
Synthesis equation: xa (t) = ∑ ck e j2πkF0 t
k=−∞
Z Tp ∞
1
Parseval’s identity: |xa (t)|2 dt = ∑ |ck |2
Tp 0 k=−∞

Aperiodic
Z ∞
Analysis equation: Xa (F) = xa (t)e− j2πFt dt (Fourier transform)
−∞
Z ∞
Synthesis equation: xa (t) = Xa (F)e j2πFt dF (inverse Fourier transform)
−∞
Z ∞ Z ∞
Parseval’s identity: |xa (t)|2 dt = |X(F)|2 dF
−∞ −∞

March 3, 2025
0.
Energy spectrum
• In contrast to the periodic case, we now have an energy spectrum.
• The spectrum |Xa (F)|2 typically has support in intervals (not only in
points).
• Similar to the periodic case, we can interpret |X(F)|2 as describing how
energy is distributed over frequency.

1 0.14
|X(F)|2
0.8
0.12
0.6

0.4 0.1

0.2
0.08
0
0.06
-0.2

-0.4 0.04

-0.6
0.02
-0.8

-1 0
0 2 4 6 8 10 -8 -6 -4 -2 0 2 4 6 8
t F
Cont.-time aperiodic signal. Energy spectrum.

March 3, 2025
0.
Energy spectrum
• In contrast to the periodic case, we now have an energy spectrum.
• The spectrum |Xa (F)|2 typically has support in intervals (not only in
points).
• Similar to the periodic case, we can interpret |X(F)|2 as describing how
energy is distributed over frequency.

1 0.4
|X(F)|
0.8
0.35
0.6
0.3
0.4
0.25
0.2

0 0.2

-0.2
0.15
-0.4
0.1
-0.6

-0.8 0.05

-1 0
0 0.5 1 1.5 2 2.5 -8 -6 -4 -2 0 2 4 6 8
t F

March 3, 2025
0.
Energy spectrum
• In contrast to the periodic case, we now have an energy spectrum.
• The spectrum |Xa (F)|2 typically has support in intervals (not only in
points).
• Similar to the periodic case, we can interpret |X(F)|2 as describing how
energy is distributed over frequency.

1 0.4
|X(F)|
0.8
0.35
0.6
0.3
0.4
0.25
0.2

0 0.2

-0.2
0.15
-0.4
0.1
-0.6

-0.8 0.05

-1 0
0 0.5 1 1.5 2 2.5 -8 -6 -4 -2 0 2 4 6 8
t F

March 3, 2025
0.
Energy spectrum
• In contrast to the periodic case, we now have an energy spectrum.
• The spectrum |Xa (F)|2 typically has support in intervals (not only in
points).
• Similar to the periodic case, we can interpret |X(F)|2 as describing how
energy is distributed over frequency.

1 0.4
|X(F)|
0.8
0.35
0.6
0.3
0.4
0.25
0.2

0 0.2

-0.2
0.15
-0.4
0.1
-0.6

-0.8 0.05

-1 0
0 0.5 1 1.5 2 2.5 -8 -6 -4 -2 0 2 4 6 8
t F

March 3, 2025
0.
Energy spectrum
• In contrast to the periodic case, we now have an energy spectrum.
• The spectrum |Xa (F)|2 typically has support in intervals (not only in
points).
• Similar to the periodic case, we can interpret |X(F)|2 as describing how
energy is distributed over frequency.

1 0.4
|X(F)|
0.8
0.35
0.6
0.3
0.4
0.25
0.2

0 0.2

-0.2
0.15
-0.4
0.1
-0.6

-0.8 0.05

-1 0
0 0.5 1 1.5 2 2.5 -8 -6 -4 -2 0 2 4 6 8
t F

March 3, 2025
0.
Electroencephalography (EEG)
• An EEG signal is not periodic (it’s actually random, but that’s for a later
course).
• However, Fourier analysis is still useful for understanding the frequency
content of the signal.
• The spectrum allows for identifying different frequency components
(so-called brainwaves).
• This is used for diagnosis of deceases as well as for brain-computer
interfaces.
8 70

6 60

4
50

2
40
0
30
-2

20
-4

-6 10

-8 0
0 0.5 1 1.5 2 0 10 20 30 40 50 60

March 3, 2025
0.
Summary

• LTI systems are easy to understand if the input is sinusoidal.

• If signals can be represented by mixtures of sinusoids, then our life is
very pleasant.
• And we are in luck: with tools from Fourier analysis, we can decompose
signals in sinusoidal components.
• If the signal is periodic, we can use Fourier series.
• If the signal is aperiodic, we can use the Fourier transform.
• Today we have looked at what this means for continuous-time signals.
• Next lecture, we will see what happens for discrete-time signals.

March 3, 2025