Discrete Fourier Transform and Spectral analysis using

Matlab
Introduction to Fourier Transform and Spectral Analysis – Part 2.
Alexander Taratorin
1
 Course Outline
• Discrete Fourier Transform
• Definition,Frequency resolution
• Spectrum leakage and Windows
• FFT
• Introduction to Matlab
• Matlab command window, editor and figures
• Variables and workspace, scalars, vectors, matrices, For loops and If operator
• Basic plot functions
• Complex numbers
• Sample Scripts
• Generation of signal – sine wave, sine wave conversion to square wave
• Angle, phase and phase unwrapping
• Signal delay and phase shift calculation
• Square wave and harmonics
• FFT, spectral leakage and spectral windows
• Up-conversion and down-conversion: time and spectrum view
• Product of signals and DFT output
• Convolution operation, examples of convolution with symmetrical and asymmetrical pulses
• Cross-correlation – signal detection in noise
• More complicated example: Measurement of unknown frequency using time, spectrum and phase
Introduction to Fourier Transform and Spectral Analysis – Part 1 2
Discrete Fourier Transform and Spectral analysis using
Matlab
Introduction to Fourier Transform and Spectral Analysis – Part 2.
Alexander Taratorin
3
 Discrete Fourier Transform
Introduction to Fourier Transform and Spectral Analysis – Part 2.

4
 Discrete spectrum of Periodic Function

∞ ∞
𝟏 𝒌
𝓕 𝒔(𝒙 − 𝒌𝑻 = 𝑺(𝒇 𝜹(𝒇 −
𝑻 𝑻
𝒌=−∞ 𝒌=−∞
Spectrum of periodic function consists of discrete harmonics at frequencies k/T, amplitude is proportional to 1/T

1
Repetition frequency: 𝑓0 = , spectrum values 𝑓0 𝑆(𝑘𝑓0 ,
𝑇

Introduction to Fourier Transform and Spectral Analysis – Part 1 5

 Discrete Fourier Transform (DFT)

Continuous Fourier Transform:

N signal samples 𝑠𝑛 = {𝑠0 , 𝑠1 , … , 𝑠𝑁−1 }. ∞

1 𝑆 𝑓 = 𝑠(𝑥 𝑒 −𝑖2𝜋𝑓𝑥 𝑑𝑥
Sampling frequency 𝑓𝑠 = 𝑇𝑠 −∞
𝑘 𝑘𝑓𝑠
N frequencies of spectrum given by 𝑓𝑘 = 𝑁𝑇 = ,𝑘 = 0,1, … . , 𝑁 − 1
𝑠 𝑁

DFT assumes that signal is periodic with period N

Discrete Fourier Transform:
𝑁−1
2𝜋
−𝑖 𝑘𝑛
𝑆𝑘 = 𝑠𝑛 𝑒 𝑁
𝑛=0

Introduction to Fourier Transform and Spectral Analysis – Part 2 6

 Discrete Fourier Transform

N=100 samples
Ts=0.01 s
fs=1/0.01=100 Hz

DFT results in 100 frequency points

𝑘𝑓𝑠
𝑓𝑘 = , 𝑘 = 0,1, … . , 𝑁 − 1
𝑁

𝑓𝑠
Frequency step Δf = =100/100=1 Hz
𝑁

Introduction to Fourier Transform and Spectral Analysis – Part 2 7

 Discrete Fourier Transform

Introduction to Fourier Transform and Spectral Analysis – Part 2 8

 Discrete Fourier Transform

N=200 samples
Ts=0.01 s
fs=1/0.01=100 Hz

DFT results in 200 frequency points

𝑘𝑓𝑠
𝑓𝑘 = , 𝑘 = 0,1, … . , 𝑁 − 1
𝑁

Note that frequency step is twice smaller

Δf=100/200=0.5 Hz

𝒇𝒔
FREQUENCY RESOLUTION OF DFT ∆𝒇 =
𝑵

Introduction to Fourier Transform and Spectral Analysis – Part 2 9

 Discrete Fourier Transform – Spectrum Leakage

DFT OF 10.6 Hz signal

Introduction to Fourier Transform and Spectral Analysis – Part 2 10

Discrete Fourier Transform – Spectral Leakage and
WIndows
Introduction to Fourier Transform and Spectral Analysis – Part 2.

11
 Discrete Fourier Transform – Spectrum Leakage

DFT OF 10.6 Hz signal

Sampling Frequency = 100 Hz Frequency = 10.6 Hz
0.4

0.35
Multiple frequency samples
Max at 11 Hz
0.3

0.25
Where these distortions come from?
Magnitude

0.2

0.15

0.1

0.05

0
0 5 10 15 20 25 30 35 40 45 50
Frequency, Hz

Introduction to Fourier Transform and Spectral Analysis – Part 2 12

 Discrete Fourier Transform – Spectrum Leakage

DFT OF 10.6 Hz signal

• DFT assumes periodic signal

• 10.6 Hz signal using 100 points results in
artificial “step” - signal not periodic

Introduction to Fourier Transform and Spectral Analysis – Part 2 13

 Discrete Fourier Transform – Spectrum Leakage

DFT OF 10.6 Hz signal

• Use 94 samples instead of 100 samples

(signal period is 0.09434s)
• Distortion is smaller since frequency
samples are
𝑘𝑓𝑠 100
𝑓𝑘 = =𝑘 = 1.0638k
𝑁 94

For k=10 we are close to 10.6 Hz

Introduction to Fourier Transform and Spectral Analysis – Part 2 14

 Discrete Fourier Transform – Spectrum Leakage

DFT OF 10.6 Hz signal

• Increase number of samples

𝑘𝑓𝑠
𝑓𝑘 =
𝑁
Larger number of samples results in higher
DFT resolution:

For N=100 Δf=1 Hz

For N=400 Δf=0.25 Hz
For N=800 Δf=0.125 Hz

Introduction to Fourier Transform and Spectral Analysis – Part 2 15

 Discrete Fourier Transform – Spectrum Leakage as “rectangular window”

1.5 100 Samples – Rectanglular window

Amplitude 0.5

-0.5

-1

-1.5
0 50 100 150 200 250 300
SAMPLES

DFT is equivalent to using “rectangular window” rect(N) having length of N samples

Using Fourier transform properties – product of signal and rectangular window results in spectrum convolution with
sinc function
sin 𝜋𝑓𝑘 𝑁 sin 𝜋𝑓𝑠 𝑘
𝑊 𝑓𝑘 = =
𝜋𝑓𝑘 𝑁 𝜋𝑓𝑠 𝑘

Introduction to Fourier Transform and Spectral Analysis – Part 2 16

 Discrete Fourier Transform – Spectrum Leakage as “rectangular window”
Sampling Frequency = 100 Hz Frequency = 10.6 Hz
0.6

0.5

0.4

Magnitude
0.3

0.2

0.1

0
6 8 10 12 14 16
Frequency, Hz

When frequency coincides with DFT bin When frequency is between bins, sinc
value, sinc equals 1 and zero for all other function is sampled “between” DFT bins
frequency bins resulting in distortion of spectral values

Introduction to Fourier Transform and Spectral Analysis – Part 2 17

 Discrete Fourier Transform – DFT windows

• Main idea of using window functions

– multiply signal by some “smooth”
function and try to minimize DFT
leakage distortions

• However this approach comes with a

cost:
• Spectrum amplitude is reduced
• Frequency resolution is increased

Introduction to Fourier Transform and Spectral Analysis – Part 2 18

 Discrete Fourier Transform – DFT windows

• Blackman Harris Window: lowest

resolution about 3 bins of DFT,
however no “leakage” above 3 DFT
bins

2𝜋𝑛 4𝜋𝑛 6𝜋𝑛

𝑊 𝑛 = 𝑎0 − 𝑎1 cos +𝑎2 cos − 𝑎3 cos
𝑁 𝑁 𝑁
𝑎0 = 0.35875
𝑎1 = 0.48829
𝑎2 = 0.14128
𝑎3 = 0.01168

Introduction to Fourier Transform and Spectral Analysis – Part 2 19

 Discrete Fourier Transform – DFT windows

Introduction to Fourier Transform and Spectral Analysis – Part 2 20

 Discrete Fourier Transform – DFT windows

Spectral resolution – mix of 10.6 Hz and 10% signal at 17 Hz

While Blackman-Harris window have low spectral resolution, spectral leakage is low and low 17 Hz signal is detected

Introduction to Fourier Transform and Spectral Analysis – Part 2 21

 Fast Fourier Transform (FFT)

Fast Fourier Transform is DFT, using smart method of DFT calculation

Direct implementation of DFT for N samples requires N2 multiplications

FFT of N samples results in 𝑁𝑙𝑜𝑔2 (𝑁 multiplications

For 1000 points direct DFT calculation requires 1 million of multiplications

FFT results in 10000 multiplications – 100 times less operations!

Most common radix-2 FFT algorithms require N=2M (32,64,128,256,512 etc)

More complicated methods use prime-factor decomposition and operate for arbitrary N (e.g. Matlab)

Introduction to Fourier Transform and Spectral Analysis – Part 2 22

Introduction to Matlab - workspace

23
 MATLAB Workspace
size(…) returns size of object
No need to define type, except for strings
A=1 constant, size y=1 size x=1
String example B=‘Hello’ – size y=1, size x =5
Array of numbers -“horizontal” or “line” vector A=[1,2,3], size y=1, size x=3
1
Column vector B=2 size y=3, size x =1
3
Product of A*B=A(1)*B(1)+A(2)*B(2)+A(3)*B(3)=14
𝐵 1 𝐴(1 𝐵 1 𝐴(2 𝐵 1 𝐴(3
Product of B*A= matrix , size y=3, size x=3 B ∗ A = 𝐵 2 𝐴(1 𝐵 2 𝐴(2 𝐵 2 𝐴(3
𝐵 3 𝐴(1 𝐵 3 𝐴(2 𝐵 3 𝐴(3

Dot Product for vectors of same size A.*C=A(1)*C(1), A(2)*C(2)+A(3)*C(3)

Power for vector A.^2=A(1)*A(1),A(2)*A(2),A(3)*A(3)
Arithmetic operations: *,/, ^,sqrt,log,log10,sin,cos,exp… Built in constant pi and 1i(imaginary one)
Comparison operators: >,<,>=,<=,== (equal),~= (not equal)
help – will show help to the command
clear all – clears all data from workspace
who – shows variables currently in workspace
Semicolon ; disables numerical output to command window
Introduction to Fourier Transform and Spectral Analysis – Part 2 24
 MATLAB basic language constructs
For loop:
Assign array values to index

for k=1:10
A(k)=k;
end;

If operator

If A>B,
‘A is bigger’
elseif A<B,
‘A is smaller’
else
‘A equals B’
end;

Introduction to Fourier Transform and Spectral Analysis – Part 2 25

Introduction to Matlab – Scripts, Editor and
plot functions

26
 MATLAB script

Script name can be called from workspace and executes sequence of matlab commands
Example: testscript.m

%
% Simplest example of executing script
%

N=input('Number of points: ') % ask for number

A=[1:N] % array of N numbers 1,2,….,N
B=A.*A % square of N points

figure(1) %select figure number 1

clf %clear figure (not required if not in use before)

%plot square function

plot(A,B)

Introduction to Fourier Transform and Spectral Analysis – Part 2 27

 MATLAB plot functions
figure(n) – open graphic window with number n
clf – clear figure, erase all previous plots
plot(x,y) – simple plot of y vs x

plot(x,y,’r’) – selects red for plot

plot(x,y,’b—’) – blue dashed
plot(x,y,’r-o’) – use circle marker
plot(x,y,’mo’) – use magenta circles, lines not connected
plot(x,y,’b’,’LineWidth’,3) – use thicker line (default is 1)

axis([xmin,xmax,ymin,ymax]) – sets plot limits in x and y

Label axes:
xlabel(‘this is x axis’)
ylabel(‘this is y axis’)

grid on – set grid visible

hold on – hold previous plots, do not erase when another plot is performed
legend(‘plot 1’,’plot 2’) – sets legend – number of legends should be equal to number of plots

subplot(n,m,k) where k could be from 1 to n*m – splits figure into n*m subplots and selects subplot k
Introduction to Fourier Transform and Spectral Analysis – Part 2 28
Introduction to Matlab – Complex Numbers

29
 MATLAB Complex numbers
i – imaginary 1, better use 1i to avoid conflict if i is redefined

A=2+i*3 – complex number

real(A) – real part
imag(A) – imaginary part
abs(A) – absolute value
angle(A) – angle in radians

Complex arithmetic is supported, for example

teta=0:2*pi/360:2*pi; % define angle from 0 to 2*pi
E=exp(i*teta) – complex exponential of angle

Introduction to Fourier Transform and Spectral Analysis – Part 2 30

Introduction to Matlab – Sine Wave Example

31
Introduction to Matlab – Phase Unwrapping

32
 MATLAB Phase Unwrapping
1

Cosine Amplitude
0.5

-0.5

-1
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time

15 Phase
Phase, Radians

Unwrapped Phase
10

-5
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time,s

Angle Function in matlab returns complex number/vector angle within – to ,

PH=angle(s);
Unwrapping procedure adds 2 whenever phase transition occurs
PH_unwrapped=unwrap(angle(s));

Introduction to Fourier Transform and Spectral Analysis – Part 2 33

Introduction to Matlab – Square Wave
Approximation

34
 Square Wave
Square wave is represented by sum of odd harmoncis of sine wave
Sum of 484 harmonics
1.5

0.5
Amplitude

-0.5

-1

-1.5
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time, Seconds

𝟒 𝟏 𝟏 𝟏
𝒔 𝒙 = 𝒔𝒊𝒏 𝟐𝝅𝒇𝟎 + 𝒔𝒊𝒏 𝟐𝝅𝟑𝒇𝟎 + 𝒔𝒊𝒏 𝟐𝝅𝟓𝒇𝟎 + 𝒔𝒊𝒏 𝟐𝝅𝟕𝒇𝟎 …
𝝅 𝟑 𝟓 𝟕

Introduction to Fourier Transform and Spectral Analysis – Part 2 35

Introduction to Matlab – DFT, Spectrum
Leakage and Windows

36
Introduction to Matlab – Time and Phase
Shift Example

37
 Square Wave

Shift ℱ 𝑠 𝑥−𝑎 = 𝑒 −𝑖2𝜋𝑓𝑎 𝑆 𝑓

∞ ∞

ℱ 𝑠 𝑥−𝑎 = 𝑠(𝑥 − 𝑎 𝑒 −𝑖2𝜋𝑓𝑥 𝑑𝑥 = 𝑠(𝑢 𝑒 −𝑖2𝜋𝑓(𝑢+𝑎 𝑑𝑢 = 𝑒 −𝑖2𝜋𝑓𝑎 𝑆 𝑓

−∞ −∞

Introduction to Fourier Transform and Spectral Analysis – Part 2 38

Introduction to Matlab – Up and Down
Conversion

39
 Fourier Transform Properties

Frequency shift (modulation) ℱ𝑠 𝑥 = 𝑆(𝑓 ℱ 𝑠 𝑥 𝑒 𝑖2𝜋𝑓0 𝑥 = 𝑆(𝑓 − 𝑓0 Spectrum shifts “up”

ℱ 𝑠 𝑥 𝑒 −𝑖2𝜋𝑓0𝑥 = 𝑆 𝑓 + 𝑓0 Spectrum shifts “down”

∞
ℱ 𝑠 𝑥 𝑒 𝑖2𝜋𝑓0 𝑡 = −∞
𝑠 𝑥 𝑒 −𝑖2𝜋(𝑓−𝑓0 𝑥 𝑑𝑥 = 𝑆(𝑓 − 𝑓0 )

Fourier transform of derivative upconversion

𝑑𝑠 𝑥
ℱ = 𝑖2𝜋𝑓𝑆(𝑓)
𝑑𝑥

downconversion

Introduction to Fourier Transform and Spectral Analysis – Part 1 40

Introduction to Matlab – Product of signals

41
 Fourier Transform Properties
∞

Product of two functions example ℱ 𝑠(𝑥 ℎ(𝑥 =𝑆 𝑓 ∗𝐻 𝑓 = 𝑆 𝑢 𝐻 𝑓 − 𝑢 𝑑𝑢

−∞
𝑺 𝒇 = 𝒔𝒊𝒏𝒄(𝝅𝒇𝑻

𝜹 𝒇 − 𝒇𝟎 + 𝜹 𝒇 + 𝒇𝟎
𝒔 𝒙 𝒉 𝒙 = 𝒓𝒆𝒄𝒕 𝑻 𝐜𝐨𝐬(𝟐𝝅𝒇𝟎 𝒕 𝑯 𝒇 =
𝟐

-T/2 0 T/2
−𝒇𝟎 0 𝒇𝟎
𝒔𝒊𝒏𝒄 𝝅 𝒇 − 𝒇𝟎 𝑻 + 𝒔𝒊𝒏𝒄 𝝅 𝒇 + 𝒇𝟎 𝑻
𝓕 𝒔(𝒙 𝒉(𝒙 =
𝟐
Introduction to Fourier Transform and Spectral Analysis – Part 1 42
Introduction to Matlab – Convolution

43
 Convolution Property of Fourier Transform
∞

Convolution 𝑔 𝑦 =𝑠 𝑥 ∗ℎ 𝑥 = 𝑠 𝑥 ℎ 𝑦 − 𝑥 𝑑𝑥
−∞

Fourier transform of convolution equals product of spectra ℱ 𝑠(𝑥 ∗ ℎ(𝑥 = 𝑆(𝑓 𝐻(𝑓

Discrete Convolution 𝑔 𝑘 = 𝑠 𝑚 ℎ(𝑘 − 𝑚 ℎ = ℎ 1 ,…,ℎ 𝐿 s = 𝑠 1 ,…,𝑆 𝑁

𝑚
𝑔 = 𝑔 1 ,…,𝑔 𝑁 + 𝐿 − 1

MATLAB: g=conv(s,h)

Introduction to Fourier Transform and Spectral Analysis – Part 2 44

 Discrete Convolution example and ‘same’ option
Discrete Convolution
ℎ = ℎ 1 ,ℎ 2 ,ℎ 3 𝑠 = 𝑠 1 , 𝑠 2 , 𝑠 3 , 𝑠(4
h(1) h(2) h(3)
𝑔 1 =𝑠 1 ℎ 1 4

𝑔 2 =𝑠 1 ℎ 2 +𝑠 2 ℎ 1 2

𝑔 3 =𝑠 1 ℎ 3 +𝑠 2 ℎ 2 +𝑠 3 ℎ 1 0
0 1 2 3 4 5 6 7 8 9 10

𝑔 4 =𝑠 2 ℎ 3 +𝑠 3 ℎ 2 +𝑠 4 ℎ 1 4 s(1) s(2) s(3) s(4)

𝑔 5 =𝑠 3 ℎ 3 +𝑠 4 ℎ 2 2

𝑔 6 =𝑠 4 ℎ 3 0
0 1 2 3 4 5 6 7 8 9 10

20

Output length equals length(s)+length(h)-1 10

0
0 1 2 3 4 5 6 7 8 9 10

g(1) g(2) g(3) g(4) g(5) g(6)

g=conv(s,h,’same’)

Introduction to Fourier Transform and Spectral Analysis – Part 2 45

 Discrete Convolution

1
Signal {s(1),…s(51)}
0.5

0
10 20 30 40 50 60 70 80 90 100

0.5
Response {h(1),…h(51)}

0
10 20 30 40 50 60 70 80 90 100
10

5 Full Convolution {g(1),….g(101)}

0
10 20 30 40 50 60 70 80 90 100
10

5 Center part of convolution {g(26),….g(76)}

0
10 20 30 40 50 60 70 80 90 100

Introduction to Fourier Transform and Spectral Analysis – Part 2 46

Introduction to Matlab – Cross-Correlation

47
 Cross-Correlation

∞
Cross-Correlation 𝑅 𝑦 =𝑠 𝑥 ⋆ℎ 𝑥 = 𝑠 ∗ 𝑥 ℎ 𝑦 + 𝑥 𝑑𝑥
−∞
ℱ 𝑠(𝑥 ⋆ ℎ(𝑥 = 𝑆 ∗ (𝑓 𝐻(𝑓

Discrete Cross-Correlation r=xcorr(s,h)

In Matlab if lengths of s and h are different, smaller length vector is zero padded

ℎ = ℎ 1 ,….ℎ 𝑁 s = 𝑠 1 ,….𝑆 𝑁

𝑟 = 𝑟 1 , … . 𝑟 2𝑁 − 1

g=xcorr(s,h,’coeff’)

Introduction to Fourier Transform and Spectral Analysis – Part 2 48

 Cross-Correlation: Auto Correlation example xcorr(s,s)
Signal {s(1),…s(51)}
1

0.5

0
0 10 20 30 40 50 60 70 80 90 100
4
Peak at 51 =s(1)^2+…+s(51)^2
2 Cross correlation {r(1),…r(101)}

0
0 10 20 30 40 50 60 70 80 90 100
4

0
-50 -40 -30 -20 -10 0 10 20 30 40 50

Cross-Correlation x axis: (1:101)-51 extends from -50 to +50

Introduction to Fourier Transform and Spectral Analysis – Part 2 49

How to Find Sine Wave Period?

50
 Find Sine Wave Period Time Domain Method

Transmission 𝑓1 = 𝑓0 + ∆𝑓
Sampler
(ADC)

Frequency 𝑓0
e.g. 31 MHz Clock provided by
based on local oscillator another local oscillator

What is exact frequency?

Introduction to Fourier Transform and Spectral Analysis – Part 2 51

 Find Sine Wave Period Time Domain Method

Detect Zero crossings

Linear Fit of 2 adjacent samples

y=ax+b
Zero crossing
x=-b/a

Introduction to Fourier Transform and Spectral Analysis – Part 2 52

 Find Sine Wave Period : DFT bins interpolation

• Use symmetrical window with minimum leakage – e.g. Blackman Harris

• While DFT resolution is low - 0.2 MHz we can artificially increase it by interpolation
• Use smooth “spline” interpolation and locate maximum frequency between DFT frequency bins

Introduction to Fourier Transform and Spectral Analysis – Part 2 53

 Find Sine Wave Period : Linear Phase Detection
1000

800

Magnitude
600

400

200

0
0 100 200 300 400 500 600 700 800 900 1000
Frequency, MHz

f0=31.1234 measured f=31.1234 MHz

20

15
Phase, Radian

10

-5
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time, microseconds

• “Down-Convert” signal by multiplying with complex exponential 𝑒 −𝑖2𝜋𝑓𝑎 𝑡 using 𝑓𝑎 - approximate estimate of 𝑓0
• Reject negative frequency by averaging (low-pass filtering)
• Phase of signal should be linear and equal to −2𝜋(𝑓𝑎 −𝑓0 𝑡
• Use linear fit of unwrapped phase to estimate frequency difference
Introduction to Fourier Transform and Spectral Analysis – Part 2 54