Fourier Transform: part 1

SCC0251/5830 – Image Processing

Prof. Moacir Ponti

Instituto de Ciências Matemáticas e de Computação – USP

2021/1

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 1 / 50

 Fundamental concepts

Agenda

1 Fundamental concepts
Representation of functions using points and coefficients
Fourier Series and the complex exponential

2 Fourier Transform
Motivation, algorithm, examples

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 2 / 50

 Fundamental concepts

Introduction

Mathematical transformations are used to obtain information not available

(or not visible) directly in the original data.

Can be seen as a map between different domains. Although the values in

different domains are different, they represent the same data.

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 3 / 50

 Fundamental concepts

Introduction: same information, different value

⇔ (-22.00257, -47.89855) ⇔

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 4 / 50

 Fundamental concepts

Introduction: same information, different value

Av.Trabalhador
Saocarlense, 400

⇔ (-22.00257, -47.89855) ⇔

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 4 / 50

 Fundamental concepts

Introduction: same information, different value

Av.Trabalhador
Saocarlense, 400

⇔ (-22.00257, -47.89855) ⇔

USP São Carlos main entrance

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 4 / 50

 Fundamental concepts

Introduction

Mathematically a signal/image can be seen as a function

There are important (and often non-obvious) information about the

function that is not trivial to grasp in their original domains.

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 5 / 50

 Fundamental concepts

Introduction

A 1 − d signal is often represented in the time domain in its original

form
plots are often in terms of time-amplitude

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 6 / 50

 Fundamental concepts

Introduction

An image (2 − d signal) is represented in the space domain

display is in terms of space-amplitude (or space-intensity)

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 7 / 50

 Fundamental concepts Representation of functions using points and coefficients

Agenda

1 Fundamental concepts
Representation of functions using points and coefficients
Fourier Series and the complex exponential

2 Fourier Transform
Motivation, algorithm, examples

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 8 / 50

 Fundamental concepts Representation of functions using points and coefficients

Representations of a function
Given n = 10 unique
points:

x f(x)
-1.0 , 4.0
-0.79 , 4.04
-0.58 , 4.18
-0.37 , 4.4
-0.16 , 4.71
0.05 , 5.1
0.26 , 5.59
0.47 , 6.16
0.68 , 6.82
Can I represent it using a different set of values?
0.89 , 7.57

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 9 / 50

 Fundamental concepts Representation of functions using points and coefficients

Representations of a function

Let us define that

it is a polynomial of degree 2

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 10 / 50

 Fundamental concepts Representation of functions using points and coefficients

Representations of a function

Let us define that

it is a polynomial of degree 2
3 values represent this function
(since a polynomial of degree n − 1 has n coefficients!)

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 10 / 50

 Fundamental concepts Representation of functions using points and coefficients

Representations of a function

Let us define that

it is a polynomial of degree 2
3 values represent this function
(since a polynomial of degree n − 1 has n coefficients!)

But how to obtain/compute such representation?

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 10 / 50

 Fundamental concepts Representation of functions using points and coefficients

Representations of a function

Build and solve a linear system with the following matrices:

x1N−1 . . . 1
 N   
x1 f (x1 )
x N
2 x2N−1 . . . 1 Y = f (x2 )
 
A= . . .   ... 
xnN xnN−1 . . . 1 f (xn )

With the coefficients given by:

C = (AT A)−1 (AT Y )

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 11 / 50

 Fundamental concepts Representation of functions using points and coefficients

Representations of a function
In our example
1.0 −1.0 1.0 4.0
   
0.62
 −0.79 1.0 
4.04
 
0.34
 −0.58 1.0 
4.18
 
0.14
 −0.37 1.0 
 4.4 
 
0.03 −0.16 1.0  Y = 4.71
 
A=
 0.0
 0.05 1.0 
 5.1 
 
0.07 0.26 1.0 5.59
  

0.22 0.47 1.0 6.16
  

0.46 0.68 1.0 6.82
0.79 0.89 1.0 7.57
 
1.0
C = 2.0
5.0
Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 12 / 50
 Fundamental concepts Representation of functions using points and coefficients

Representations of a function

f (x) = 4.0, 4.04, 4.18, 4.4, 4.71

f (x) = 5 + 2x + x 2
5.1, 5.59, 6.16, 6.82, 7.57

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 13 / 50

 Fundamental concepts Representation of functions using points and coefficients

Representations

Representation using coefficients

f (x) = c0 + c1 x + c2 x 2 + · · · + cn−1 x n−1

Representation using points

f (x) = f (x1 ), f (x2 ), · · · f (xn )

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 14 / 50

 Fundamental concepts Representation of functions using points and coefficients

the Fourier Transform will take as

input points or sampled intervals of a function

(i.e. in the way they are acquired),

output the coefficients that define the function

(its fundamental components).

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 15 / 50

 Fundamental concepts Representation of functions using points and coefficients

Synthesis and Analysis

Two aspects of Fourier Transform:

Analysis: divide the signal (or function) by defining it via simpler
parts.

Synthesis: reconstruct the signal (or function) from its parts.

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 16 / 50

 Fundamental concepts Representation of functions using points and coefficients

Synthesis and Analysis

Two aspects of Fourier Transform:

Analysis: divide the signal (or function) by defining it via simpler
parts.

Synthesis: reconstruct the signal (or function) from its parts.

Both can be achieved via linear operations,

i.e. series and integrals.

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 16 / 50

 Fundamental concepts Fourier Series and the complex exponential

Agenda

1 Fundamental concepts
Representation of functions using points and coefficients
Fourier Series and the complex exponential

2 Fourier Transform
Motivation, algorithm, examples

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 17 / 50

 Fundamental concepts Fourier Series and the complex exponential

Fourier Series

Jean-Baptiste Fourier, 1822, studying heat transfer,

claimed that a function of a single variable could be
expanded in terms of a series of sinusoids of multiples
of the variable.
After Lagrange and Dirichlet studies using this
expansion, it was refered to as Fourier Series.

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 18 / 50

 Fundamental concepts Fourier Series and the complex exponential

Fourier Series and Periodicity

Fourier Series are associated to the mathematical analysis of periodic

patterns.

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 19 / 50

 Fundamental concepts Fourier Series and the complex exponential

Fourier Series and Periodicity

Fourier Series are associated to the mathematical analysis of periodic

patterns.

Periodicity
Time: harmonic movement (e.g. of a string)
Space: some physical measure distributed to a certain region in a
symmetric way (periodicity from symmetry, repetition of a pattern).

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 19 / 50

 Fundamental concepts Fourier Series and the complex exponential

Fourier Series and Periodicity

Fourier Series are associated to the mathematical analysis of periodic

patterns.

Periodicity
Time: harmonic movement (e.g. of a string)
Space: some physical measure distributed to a certain region in a
symmetric way (periodicity from symmetry, repetition of a pattern).
e.g. heat distribution in a circular object: the temperature repeat itself
in cycles.

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 19 / 50

 Fundamental concepts Fourier Series and the complex exponential

Fourier Series and Periodicity

Fourier Series are associated to the mathematical analysis of periodic

patterns.

Periodicity
Time: harmonic movement (e.g. of a string)
Space: some physical measure distributed to a certain region in a
symmetric way (periodicity from symmetry, repetition of a pattern).
e.g. heat distribution in a circular object: the temperature repeat itself
in cycles.
that is why Fourier Analysis is often associated with symmetry.

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 19 / 50

 Fundamental concepts Fourier Series and the complex exponential

Fourier Series and Periodicity

Mathematical descriptors of periodicity

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 20 / 50

 Fundamental concepts Fourier Series and the complex exponential

Fourier Series and Periodicity

Mathematical descriptors of periodicity

Ideas
Time: frequency — number of pattern repetitions along the time
(e.g. 1 second)
Space: period (wavelength) — size of the repeating pattern

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 20 / 50

 Fundamental concepts Fourier Series and the complex exponential

Fourier Series and Periodicity

Mathematical descriptors of periodicity

Ideas
Time: frequency — number of pattern repetitions along the time
(e.g. 1 second)
Space: period (wavelength) — size of the repeating pattern

In some cases time and space are involved at the same time — e.g.
wave movement
If we fix the position (in space), we can measure frequency
(distribution of the pattern in time)
By fixing an instant (in time) we can measure the size (distribution of
the pattern in space).

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 20 / 50

 Fundamental concepts Fourier Series and the complex exponential

Fourier Series and Periodicity

Relationship space (wavelength) and time (frequency)

v is the velocity (rate) of the wave and F its frequency then:
λ = v · F1 , considering one complete wave in 1
F
or F · λ = v
There is a reciprocal relationship between wavelength and frequency

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 21 / 50

 Fundamental concepts Fourier Series and the complex exponential

Wavelength vs Frequency

Let a sequence yellow-blue define the wavelength, then:

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 22 / 50

 Fundamental concepts Fourier Series and the complex exponential

Fourier Series
There are mathematical functions for which

f (t + T ) =f (t) (1)
f (t + nT ) =f (t), n = 0, ±1, ±2, · · · (2)
(3)
some can be used to model periodic behaviour, in particular sinusoids

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 23 / 50

 Fundamental concepts Fourier Series and the complex exponential

Fourier Series
There are mathematical functions for which

f (t + T ) =f (t) (1)
f (t + nT ) =f (t), n = 0, ±1, ±2, · · · (2)
(3)
some can be used to model periodic behaviour, in particular sinusoids
Why not others: square wave, other periodic signals?

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 23 / 50

 Fundamental concepts Fourier Series and the complex exponential

Fourier Series
There are mathematical functions for which

f (t + T ) =f (t) (1)
f (t + nT ) =f (t), n = 0, ±1, ±2, · · · (2)
(3)
some can be used to model periodic behaviour, in particular sinusoids
Why not others: square wave, other periodic signals?

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 23 / 50

 Fundamental concepts Fourier Series and the complex exponential

Fourier Series

Sine and cosine are periodic with period 2π

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 24 / 50

 Fundamental concepts Fourier Series and the complex exponential

Fourier Series

Sine and cosine are periodic with period 2π

associated with space periodicity, in particular the “simpler periodic
object”

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 24 / 50

 Fundamental concepts Fourier Series and the complex exponential

Fourier Series

Sine and cosine are periodic with period 2π

associated with space periodicity, in particular the “simpler periodic
object”
circle: cos t is coordinate x and sin is coordinate y of a unitary circle.

cos(t + 2πn) = cos(t) (4)

sin(t + 2πn) = sin(t) (5)

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 24 / 50

 Fundamental concepts Fourier Series and the complex exponential

Fourier Series and Periodicity

Can we write an arbitrary function in terms of sinusoids?

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 25 / 50

 Fundamental concepts Fourier Series and the complex exponential

Fourier Series and Periodicity

Can we write an arbitrary function in terms of sinusoids?

Must this function we want to write be periodic?

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 25 / 50

 Fundamental concepts Fourier Series and the complex exponential

Fourier Series and Periodicity

Important remarks:
Functions with the simplest periodic behaviour: sines and cosines;

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 26 / 50

 Fundamental concepts Fourier Series and the complex exponential

Fourier Series and Periodicity

Important remarks:
Functions with the simplest periodic behaviour: sines and cosines;
circle: cos t is coordinate x and sin t is coordinate y of a unitary circle.

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 26 / 50

 Fundamental concepts Fourier Series and the complex exponential

Fourier Series

Given a periodic function f (t) of a continuous variable t with period T :

∞ ∞ ∞
X X X 2πn
f (t) = an cos 2πnt + bn sin 2πnt = cn e j T
t
(6)
n=0 n=1 n=−∞

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 27 / 50

 Fundamental concepts Fourier Series and the complex exponential

[Complex Numbers and Euler’s formula]

A complex number C is defined by

c = R + jI , (7)
√
R and I are real numbers and j is the imaginary j = −1
Geometric interpretation: a complex Cartesian plane with real axis R,
and imaginary axis I .

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 28 / 50

 Fundamental concepts Fourier Series and the complex exponential

[Complex Numbers and Euler’s formula]

In polar coordinates, we have:

c = |c|(cos ω + j sin ω) = a cos(ω) + jb sin(ω), (8)

|c| is the vector size extending from the origin of the complex plane to
the point (R, I ); and ω is the angle between the vector and the real
axis.

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 29 / 50

 Fundamental concepts Fourier Series and the complex exponential

[Complex Numbers and Euler’s formula]

Euler’s formula relates the complex sum of sine and cosine using a
complex exponential:

e jω = cos ω + j sin ω, (9)

we can substitute so that

X = |c|(cos ω + j sin ω), (10)

and obtain
c = |c|e jω . (11)

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 30 / 50

 Fundamental concepts Fourier Series and the complex exponential

[Complex Numbers and Euler’s formula]

Euler’s formula relates the complex sum of sine and cosine using a
complex exponential:

e jω = cos ω + j sin ω, (9)

we can substitute so that

X = |c|(cos ω + j sin ω), (10)

and obtain
c = |c|e jω . (11)
Example: x = 1 + j2
√
in polar coordinates: 5e jω , with ω = 64, 4

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 30 / 50

 Fundamental concepts Fourier Series and the complex exponential

[Complex Exponential]

e jx cos x sin x
Thanks to Jim Clay

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 31 / 50

 Fourier Transform

Agenda

1 Fundamental concepts
Representation of functions using points and coefficients
Fourier Series and the complex exponential

2 Fourier Transform
Motivation, algorithm, examples

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 32 / 50

 Fourier Transform

Fourier Transform – interpretation

A signal can be represented by the independent sum of each number

in each point in time: f (t) = f (t1 ) + f (t2 ), ...
instead of summing points, we are going to sum functions cosine and
sine with different coefficients.

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 33 / 50

 Fourier Transform

Fourier Transform

The Fourier series allows writing a function by a discrete sum of

complex exponentials with different frequencies.

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 34 / 50

 Fourier Transform

Fourier Transform

The Fourier series allows writing a function by a discrete sum of

complex exponentials with different frequencies.

Fourier Transform is the evaluation, for each frequency ω,

of its coefficient cω
∞
X
F (ω) = f (t)e −jωt
t=−∞

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 34 / 50

 Fourier Transform

Fourier Transform

the functions cover all the input axis:

cω e jωt = aω cos(ωt) + jbω sin(ωt)

When summing all possible sinusoids with different frequencies, we

have a series of values:

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 35 / 50

 Fourier Transform

Fourier Transform

the functions cover all the input axis:

cω e jωt = aω cos(ωt) + jbω sin(ωt)

When summing all possible sinusoids with different frequencies, we

have a series of values:
ω1 and its coefficients a1 , b1

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 35 / 50

 Fourier Transform

Fourier Transform

the functions cover all the input axis:

cω e jωt = aω cos(ωt) + jbω sin(ωt)

When summing all possible sinusoids with different frequencies, we

have a series of values:
ω1 and its coefficients a1 , b1
ω2 and its coefficients a2 , b2
ω3 and its coefficients a3 , b3
...

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 35 / 50

 Fourier Transform

Fourier Transform

Fourier Transform takes (a given signal) from time/space domain to

the frequency domain (per seconds / per measure).
signals (time): f (t) to F (ω)
images (space): f (x, y ) to F (u, v )

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 36 / 50

 Fourier Transform

Fourier Transform

When plotting the function in the Fourier domain, we use, for each
frequency a complex exponential with:

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 37 / 50

 Fourier Transform

Fourier Transform

When plotting the function in the Fourier domain, we use, for each
frequency a complex exponential with:
the relative amplitude of the cosine (real part) and of the sine
(imaginary part) as a function of ω,
the representation of the signal in the frequency domain:
an (ω) = Re(F (ω))
bn (ω) = Im(F (ω))

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 37 / 50

 Fourier Transform

Discrete Fourier Transform

N−1
X
F (ω) = f (t)e −jωt
t=0

evaluating F (ω) for different frequencies, we obtain the amplitudes

of cosines (real part) and sines (imaginary part) so that we can
reconstruct f (t) if needed.

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 38 / 50

 Fourier Transform Motivation, algorithm, examples

Motivation

The universe has a lot of periodic phenomena

Humans often observe time and space phenomena

the energy propagation of the electromagnetic spectrum is described

in waves, including the light that generate images.

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 39 / 50

 Fourier Transform Motivation, algorithm, examples

Motivation

The universe has a lot of periodic phenomena

Humans often observe time and space phenomena

the energy propagation of the electromagnetic spectrum is described

in waves, including the light that generate images.

Also, differential equations are key to many applications in science and

engineering. Taking signals to the frequency domain makes it easier to
solve many problems.

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 39 / 50

 Fourier Transform Motivation, algorithm, examples

Information in Frequency

There is relevant (and often

non-obvious) information about
the signal in its frequency
content.
It indicates how the amplitude of
the signal changes along time or
space
e.g. is it dominated by abrupt
or smooth changes?
N = 500 # sample points
Fs = 1.0/1000.0 # frequency of sampling
x = np.linspace(0.0, N*Fs, N) # sampling vector

# signal with frequency Fr

Fr = 10
y = np.sin(Fr*2*np.pi*x)

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 40 / 50

 Fourier Transform Motivation, algorithm, examples

Fourier Transform: translated

n−1
X
F (ω) = f (t)e −jωt dt
t=0

1: for i = 0 to n − 1 do
2: multiply: f (t) × e −jωi t ,

5: end for

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 41 / 50

 Fourier Transform Motivation, algorithm, examples

Fourier Transform: translated

n−1
X
F (ω) = f (t)e −jωt dt
t=0

1: for i = 0 to n − 1 do
2: multiply: f (t) × e −jωi t , or f (t) × [cos(ωi t) + j sin(ωi t)].

5: end for

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 41 / 50

 Fourier Transform Motivation, algorithm, examples

Fourier Transform: translated

n−1
X n−1
X n−1
X
−jωt
F (ω) = f (t)e dt = f (t) cos(ωt)dt + f (t)j sin(ωt)dt
t=0 t=0 t=0

1: for i = 0 to n − 1 do
2: multiply: f (t) × e −jωi t , or f (t) × [cos(ωi t) + j sin(ωi t)].

5: end for

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 41 / 50

 Fourier Transform Motivation, algorithm, examples

Fourier Transform: translated

n−1
X n−1
X n−1
X
−jωt
F (ω) = f (t)e dt = f (t) cos(ωt)dt + f (t)j sin(ωt)dt
t=0 t=0 t=0

1: for i = 0 to n − 1 do
2: multiply: f (t) × e −jωi t , or f (t) × [cos(ωi t) + j sin(ωi t)].
3: sum (integrate) for all t getting coefficients a (real) / b (imag)

5: end for

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 41 / 50

 Fourier Transform Motivation, algorithm, examples

Fourier Transform: translated

n−1
X n−1
X n−1
X
−jωt
F (ω) = f (t)e dt = f (t) cos(ωt)dt + f (t)j sin(ωt)dt
t=0 t=0 t=0

1: for i = 0 to n − 1 do
2: multiply: f (t) × e −jωi t , or f (t) × [cos(ωi t) + j sin(ωi t)].
3: sum (integrate) for all t getting coefficients a (real) / b (imag)
4: F (ωi ) = aωi + jbωi
5: end for

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 41 / 50

 Fourier Transform Motivation, algorithm, examples

Frequency analysis

Signal obtained by summing a sine with amplitude 0.6 and frequency 3Hz
and a cosine with amplitude 0.8 frequency 8Hz:
f = 0.6*sin((2*pi) * 3 * t) + 0.8*cos((2*pi) * 8 * t)

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 42 / 50

 Fourier Transform Motivation, algorithm, examples

How the sum behaves in each frequency

function overlayed with the real part (cosine) in frequency 3Hz

function overlayed with the imaginary part (sine) in frequency 3Hz

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 43 / 50

 Fourier Transform Motivation, algorithm, examples

How the sum behaves in each frequency

Function of the product between the input function and the cosine and sine
terms:

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 44 / 50

 Fourier Transform Motivation, algorithm, examples

How the sum behaves in each frequency

After multiply using 3Hz cosine the sum is near zero, since this component
is not part of the signal (see positive and negatives cancel each other)

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 45 / 50

 Fourier Transform Motivation, algorithm, examples

How the sum behaves in each frequency

On the other hand, for a 3Hz sine, most values are positive because this
wave is part of the signal.

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 46 / 50

 Fourier Transform Motivation, algorithm, examples

Frequency analysis

yf = scipy.fftpack.fft(y) # computes Fourier Transform

N2 = N//2 # gets half of the samples

# reescaled/normalised spectrum
yw = ( 2.0/N*np.abs(yf(:N2)) )
Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 47 / 50
 Fourier Transform Motivation, algorithm, examples

Applications

ECG (electrocardiogram diagnosis)

Thanks to Murray Bourne http://www.intmath.com/blog/mathematics/math-of-ecgs-fourier-series-4281

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 48 / 50

 Fourier Transform Motivation, algorithm, examples

Frequency analysis in stationary signals

Fourier analysis suits better stationary signals, e.g. with frequencies 3

and 10 at any point

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 49 / 50

 Fourier Transform Motivation, algorithm, examples

Frequency analysis in non-stationary signals

Signals in which a part (∼ 75%) has frequency 5 Hz and the
remaining has frequency 13 Hz, makes it hard to analyse.
Frequency analysis allow us to see what are the frequencies present in
the signal, but not in which position they occur.

Moacir Ponti (ICMC–USP) Fourier Transform 2021/1 50 / 50