Review

Fourier series expansion of periodic signal

(with period of T)

f (t )  a0   (ak cos kt bk sin kt ) TRIGNOMETRIC
k 1 2
(  )
T

f (t )  A0   Ak cos(kt  k ) COMPACT
k 1


f (t )  k
c e jkt

k  
EXPONENTIAL

(k  1,2,3,......)
 Review
Fourier series expansion of periodic signal
1
a0 
T  f (t )dt
T

f (t )  a0   (ak cos kt bk sin kt ) ak 
2
 f (t ) cos(kt )dt
T
k 1 T

2 2
T

bk 
T  f (t ) sin(kt )dt
T


Ak  ak  bk
2 2

f (t )  A0   Ak cos(kt  k ) A0  a0
 bk
k 1 k  tan 1 ( )
ak

 ck e jkt 1
f (t )  ck   f (t )e  jkt dt
k   T T

( k  1,2,3,......)
Fourier series expansion of periodic signal

Step 1: Observe symmetry of the function f(t) (even or odd),

determine the period and fundamental frequency, and select
a complete cycle (period) for analysis. Express function f(t)
for the selected period

Step 2: Calculate Fourier coefficients a0 , ak , bk (k  1,2,3,......)

Step 3: Find A0 , Ak (k  1,2,...) for Fourier series expansion

in compact form

Use Matlab to help

plot Amplitude spectrum ( Ak vs. k ) and
phase spectrum ( k vs. k )
 Review
• Exponential Fourier series expansion of a periodic signal

1
f (t )   ck e
k  
jkt
ck 
T 
T
f (t )e  jkt dt

 A non-periodic function f(t) (period of infinity) can be

represented as a sum of sin’s and cos’s of (possibly)
all frequencies:

1  
f (t )   F ( )e jt
d F ( )   f (t )e  jt dt
2  

Inverse Fourier transform Fourier transform (integral)

of f(t) of f(t)
 Amplitude
Fourier series Amplitude Spectrum
expansion of A1
a periodic signal A2
A3 A
4 A
5

1 2 3 4 5 Harmonics
Magnitude
F ( )
Fourier transform
of a non-periodic
signal ( T0   )
T Spectrum
Fourier

f (t)
t Frequency
1
T0 f  (Hz)
T0
 Terms:
- Fourier Transform (FT)

- Discrete Fourier Transform (DFT)

- Fast Fourier Transform (FFT)

• A fast algorithm for calculating DFT
 Processing of Biosignals Using FFT
F(nf)
f[n]=f(nT)
DFT
n  n
fs
T 2
T0 = (N-1)T (N1)f  fs

1
Sampling rate (frequency): f s  ( Hz) Frequency resolution: f 1/ T0
T
Sampling interval: T (sec) Max. Frequency : fmax (Hz)

Number of samples: N (=2M) fs

f max 
Sample length: T0 (sec) = (N-1)T
2
 Biosignals and Processing

Lecture 11 FFT and applications

Using Matlab command fft

 Using Matlab fft command
Example 1: find the DFT of the following function:

x(t )  sin( 20t )  0.5 cos(40t )

Use a sampling frequency of 100 Hz and 256 samples

for the analysis.
Example 1 Matlab script
N=256; %%% sampling points
fs=100; %%% sampling rate (freq.)= 100 Hz
dt=1/fs; %%% sampling interval=0.01;

n=1:1:N;
xn=sin(20*pi*(n-1)*dt)+0.5*cos(40*(n-1)*dt);

figure %%%%%%%%%% plot sampled x(t) sequence x[n]

plot(n,xn,'.-')

xf=fft(xn,N); %%%%%%%% Find fft of x(t), expressed as xf

mag=abs(xf); %%% calculate the magnitude of xf

figure
plot(n,mag,'.-') %%%%%%% plot magnitude
 Example 1 figures from Matlab
x(nt) 1.5

0.5

-0.5

-1

-1.5
0 50 100 150 200 250 300

X(n) 100

80

60

40

20

0
0 50 100 150 200 250 300
Example 1 Matlab script (add time and frequency units)
N=256; %%% sampling points
fs=100; %%% sampling rate (freq.)= 100 Hz
dt=1/fs; %%% sampling interval=0.01;
t=0:dt:(N-1)*dt; %%% length: (N-1)*dt
xt=sin(20*pi*t)+0.5*cos(40*t);
figure %%%%%%%%%% plot x(t)
plot(t,xt,'.-')
xlabel('Time (sec)')
ylabel('x(t)')
xf=fft(xt,N); %%%%%%%%Find fft of x(t)
Mag=abs(xf); %%% calculate the magnitude of X(f)
df=1/(N-1)/dt %%% df: frequency resolution
freq=0:df:fs;
figure,
plot(freq(1:N/2),Mag(1:N/2),'.-') %%% plot magnitude vs. frequency
grid
xlabel('Frequency (Hz)')
ylabel('Magnitude')
figure, %%%% plot magnitude in dB
plot(freq(1:N/2),20*log10(Mag(1:N/2)),'.-')
grid
xlabel('Frequency (Hz)')
ylabel('Magnitude (dB)')
title(‘Fourier Spectrum’)
 Example 1 figures from Matlab
1.5 100
Fourier spectrum
90
1
80

Magnitude
70
0.5
60

Magnitude
x(t)

0 50

40
-0.5
30

20
-1
10

-1.5 0
0 5 10 15 20 25 30 35 40 45 50
0 0.5 1 1.5 2 2.5 3
Frequency (Hz)
Time (sec)
40
Fourier spectrum
Fourier Spectrum

30
Magnitude (dB)

20
Magnitude (dB)

10

-10

-20
0 5 10 15 20 25 30 35 40 45 50
Frequency (Hz)
Sampling, DFT/FFT, and STFT

• Sampling Theorem
-Discrete Fourier Transform (DFT)
/Fast Fourier Transform (FFT)

• Short-Time-FT (STFT) or Spectrogram

 Sampling Theorem
 If a signal contains the highest frequency
fmax, the sampling frequency (rate)
(fs) >= 2 fmax

 Nyquist frequency (rate): fNyquist = 2 fmax

 The sampling theory insures that the

digitized signal truly represents the original
analog signal (No loss of information)
 Insufficient sampling rate time results in
aliasing
 Aliasing
• A characteristic of the A/D process (converting
from an analogue signal to a digital signal)

• Caused by insufficient sampling frequency (rate) -

a higher frequency ‘folds over’ into the lower
frequencies

• This can be illustrated by taking a simple signal of

1000Hz and steadily reducing the sampling rate:

3000 Hz, … 2000Hz, …. 1000 Hz

 Aliasing
Original: (1000Hz)

Sampled at 3000Hz:

Sampled at 1800Hz:

Sampled at 1500Hz:

Sampled at 1200Hz:

Estimate the frequency in each case…

 Aliasing
Original: (1000Hz)

Sampled at 3000Hz: Estimated: 1000Hz

Sampled at 1800Hz: Estimated: 850Hz

Sampled at 1500Hz: Estimated: 500Hz

Sampled at 1200Hz: Estimated: 250Hz

Estimate the frequency in each case…

 Aliasing

Estimated frequencies are plotted against sampling

frequency:
normal

1000Hz
Estimated
frequency

500Hz aliased

Sampling
frequency

1000Hz 1200 1500 1800 2000Hz

20
 Aliasing

What is the affect on speech?

Original (max: 1000 Hz):

Sampled at 2000Hz:
Alias:
Sampled at 1000Hz: Frequencies greater
than 500Hz sound
Folded over as lower
frequencies, creating
distortion

21
 Sampling Theorem
 If a signal contains the highest frequency fmax,
the sampling frequency (rate)
(fs) >= 2 fmax

 Nyquist frequency: fNyquist = 2 fmax

 The sampling theory insures that the digitized

signal truly represents the original analog signal
(No loss of information)
Analog-to-Digital (A/D) Converter
 Analog input

 Anti-alias Filter

 Sampling

 Quantization

 Digital output
 A sampling rate of 200 Hz is not adequate
to sample a 360 Hz signal
Original signal
360 Hz signal, T=0.00278 sec

Frequency:
f=360 Hz;

Period:
T=1/360 (sec)

Sampling rate: 200Hz

40 Hz signal, T=0.025 sec

Signal reconstructed
after sampling
Conventional Clinical Tool: Videostroboscopy
Vocal Fold Cycles
Open
Apparent
Closed
Motion
Strobe flash

 Stroboscope shows an apparent motion,

or a slowed down motion !!
 High-speed Imaging of the larynx
Sampling rate: 2000 Hz

Sustained phonation of /i/ at normal pitch:

f0 ~ 200 Hz

Glottal Area Waveform (GAW) Normal voice

Open

Closed
0.005 sec
1 vibratory cycle

50 image frames (2000 f/sec) 0.025 sec of recording)

Vocal fold vibrations captured at different
recording rate

Video-stroboscopy High-speed imaging

sampled at 25 Hz sampled at 2000 Hz
Vocal fold Vibrations in normal and
pathological voicing
Leukoplakia

Normal voice
Example 1: Signal processing by DFT/FFT
Noisy Sine wave
Sine wave
Original signal

FFT spectrum FFT spectrum

 Example 2: Signal processing bypage
Textbook DFT
Originalofsignal
Upstroke: contraction of the ventricles the
heart; Downward: ventricular relaxation

Systolic
pressure

Diastolic
pressure
Example 2: Signal processing by DFT
FFT spectrum
Example 3: Vocal signal processing by DFT
Sound Pressure

Sound radiation
from lips
Vocal tract: Larynx
filter and
amplifier Sound
propagation in
the vocal tract

Vocal fold:
Source of Sound

Source filter model of speech production (Fant, 1960)

 One-mass model 3-mass model (Titze,1995)

Pi

Ps

mx  bx  kx  P
 Signal from the sound source: (vocal fold vibration)
1

0.5
GAW (Normalized)

-0.5

-1
10 20 30 40 50 60 70 80 90 100
FFT
Time (ms) Power Spectrum
5
10

FFT
4
10 Spectrum
Normal Voice Condition 3
10

2
10

0 200 400 600 800 1000

Frequency (Hz)

280 Hz 560Hz 840 Hz

 Sound output (after travelling through the vocal tract)
acquired by a microphone
1
Acoustic data (Normalized)

0.5

-0.5

-1
10 20 30 40 50 60 70 80 90 100
Time (ms)

FFT Power Spectrum

FFT
4 2
10

3 Spectrum
Normal Voice 10
3

Condition 10
2

1
10

0 500 1000 1500 2000

Frequency (Hz)
Comparison: vocal-fold vibration vs. sound waves
FFT Power Spectrum
5
10

4
10

10
3 2 DFT of the Vocal fold Vibration
10
2 3
0 200 400 600 800 1000
Frequency (Hz)
FFT Power Spectrum

4
10
2
3
3
10 DFT of the
2
10
sound output
1
10

0 500 1000 1500 2000

Frequency (Hz)

 Knowledge of the source-tract transfer function helps understand the

mechanism of voice production;
 Speech synthesis requires understanding of the vocal tract dynamic properties
Voice pathology: Muscle Tension Dysphonia (MTD)
Vocal fold source signal

Period
(ms)
Sound wave signal
Spectral analysis of MTD voice
FFT Power Spectrum

f
3f/2
4
10 2f 5f/2
f/2
3
10 3f
GAW

2
10

1
10

0 100 200 300 400 500 600 700 800 900 1000
Frequency (Hz)

Sub-harmonics
Short-time Fourier Transform (STFT)

• STFT: FFT analysis of short, consecutive

segments of a signal;

• A type of time-frequency analysis ;

effective for analysis of time-varying signals

• Provide information on both time and frequency

• STFT is also called Spectrogram

• Can be plotted in 3-D or displayed as image

STFT: Time-frequency Analysis

(Time)
STFT: Time-frequency Analysis
 STFT of EEG displayed in 3D plot

Amplitude

6 Hz

EEG signals recorded on stimulations (6, 10, 15 and 20Hz)

 EEG examination of Epilepsy
Epilepsy: genetic disorder caused by mutation in
single ion channel

 Patients with suspected epilepsy usually undergo EEG

examination upon a photo stimulation;

 The technique is used to trigger epileptic potentials by

repeated light flashes (at a rate of 10 ~ 30 Hz)

 Patient suffering from epilepsy possibly exhibits a repetitive

spike-wave pattern which occur at ~ 3/second.
Spectrogram of EEG at the onset of an
epileptic seizure displayed in grey-scale image
Original signal

Time (s)
Spectrogram

Time (s)

Amplitude is expressed in brightness

 Spectrogram of a speech signal displayed in color image
Decibel (dB): logarithmic unit
5000 20
4500
1200
dB  20 log10 ( A)
4000 0
3500
1000
Frequency (Hz)

3000 -20
800
Frequency

2500
600
2000 -40
1500 400
1000
-60
200
500

0
0 0.2
00.4 0.6
0 0.10.8 10.2 1.2 1.4
0.3 1.6 1.8
0.4 2 0.5
Time
Time (sec)