NONSTATIONARY SIGNAL

ANALYSIS

WEEK 1
 Outline
• This course presents the application of the nonstationary signal processing tools to the
Objective analysis of biomedical signals. It shows how clinically relevant information can be extracted
from the signals.

• Clifford, G.D., Azuaje, F., and McSharry, P.E. (2006), Advanced Method and Tools for ECG Data Analysis,
Artech House, Inc.
• Barbieri, R., Scilingo, E.P., and Valenza, G. (2017), Complexity and Nonlinearity in Cardiovascular Signals,
References Springer International Publishing.
• W.J. Tompkins, Biomedical Digital Signal Processing, Prentice Hall, 2000.
• R. Polikar, The Wavelet Tutorial : Fundamental Concepts & An Overview of The Wavelet Theory, 1994.
• Selected papers.

• Stationary vs Nonstationary Signal

Course •
•
EEG signal analysis
HRV signal analysis
outline • Short Time Fourier Transform
• Wavelet Transform

2
 Course Mark
Final
Exam
20%
Mid
Exam
20% Assignments*
40%

Reports
20%

*Assignments : Signal processing practice, application design analysis software, Final project
3
HRV in different stages of cancer
Early detected Chemotherapy Hospice cancer patient
cancer patient cancer patient (last months of life)
EEG Characteristic
 NONSTATIONARY TIME SERIES

Introduction :
Key Questions :
• What is nonstationary?
• Why is it important?
• How do we determine whether a time series is nonstationary?
IDENTIFICATION IN TIME SERIES
 STATIONARY SIGNAL IN TIME SERIES [1]

• A deterministic signal (nonrandom function of time) is said to be stationary if it can be

written as a discrete sum of cosine waves or exponentials:

• As a sum of elements which have constant instantaneous amplitude and instantaneous

frequency
 STATIONARY SIGNAL IN TIME SERIES [2]

• In the random case (varies over time), a signal {x(n)} is said to be wide-sense
stationary (or stationary up to the second order) if its variance is independent
of time.
• Constant mean
• Constant variance
 STATIONARY SIGNAL IN TIME SERIES [3]

• The autocorrelation function for a discrete process of length N {x(n)} with known
mean μ and variance σ, depends only on the time difference m. ➔ depend only on
the “distance” or “lag” between them.

Nonstationary Stationary
AUTOCORRELATION FUNCTION
 NONSTATIONARY SIGNAL

What is nonstationary?
• There is no long-run mean to which the series return (the mean change).
• The variance is time-dependent.
• Theorical autocorrelations do not decay, sample autocorrelations do so very
slowly.
Stochastic/random signal
Deterministic signal with k= 0.45
IDENTIFICATION IN TIME SERIES
IDENTIFICATION NONSTATIONARY IN FREQUENCY DOMAIN

Fourier
Transfo
rm
 Non-stationary signals
3 600

2 Hz + 10 Hz + 20Hz 2 500

Magnitude

Magnitude
1 400

Stationary 0 300

-1 200

-2 100

-3 0
0 0.2 0.4 0.6 0.8 1 0 5 10 15 20 25
Time Frequency (Hz)
1 250
0.0-0.4: 2 Hz + 0.8

0.4-0.7: 10 Hz + Magnitude
0.6 200

Magnitude
0.7-1.0: 20Hz 0.4

0.2 150

Non- 0

-0.2 100

Stationary -0.4

-0.6 50

-0.8

-1 0
0 0.5 1 0 5 10 15 20 25
Time Frequency (Hz)
Chirp signal
◼ Frequency: 2 Hz to 20 Hz ◼ Frequency: 20 Hz to 2 Hz
Different in Time Domain

1 150 1 150

0.8 0.8

0.6 0.6

0.4 0.4
Magnitude

Magnitude
Magnitude 100 100

Magnitude
0.2 0.2

0 0

-0.2 -0.2
50 50
-0.4 -0.4

-0.6 -0.6

-0.8 -0.8

-1 0 -1 0
0 0.5 1 0 5 10 15 20 25 0 0.5 1 0 5 10 15 20 25
Time Frequency (Hz) Time Frequency (Hz)

Same in Frequency Domain

Fourier transform drawbacks
◼ Global behaviour: we don’t know what frequencies happens at a
particular time
◼ Time and frequency are not seen together

◼ We need time and frequency at the same time: time-frequency

representation
◼ Biological or medical signals (ECG, EEG, EMG) are always non-
stationary
 Signal processing

Time domain methods Freq domain methods

TF domain methods
Fourier
Filters Transform
Non/Stationary signals Stationary signals
STFT

WAVELET
TRANSFORMS

CWT DWT MRA 2D DWT SWT Applications:

Denoising
Compression
Detection
Analysis, etc..
 SHORT-TIME FOURIER TRANSFORM (STFT)

• Insert time information in

frequency plot

STFTx f (t ' , f ) =  x(t )  w(t − t ') e − j 2ft dt

t
 MULTI-RESOLUTION ANALYSIS

• MRA is designed to give good time resolution and poor frequency

resolution at high frequencies and good frequency resolution and poor
time resolution at low frequencies.
• This approach makes sense especially when the signal at hand has high
frequency components for short durations and low frequency
components for long durations.
• The signals that are encountered in practical applications are often of
this type.
CONTINUOUS WAVELET TRANSFORM

¥
X ( f ) = ò x(t )exp(- j 2pft )dt
-¥
¥
STFT (t , f ) = ò x(t ).w(t - t ¢)exp(- j 2pft )dt
x
w
-¥

¥ æ t -t ö
CWTx (t , s ) = ò-¥ x(t ).y çè s
y 1
÷dt
*

s ø

y(t): mother wavelet

DISCRETE WAVELET TRANSFORM