Wavelet transform

Wavelet transform is a relatively new concept (about 10 more years old)

First of all, why do we need a transform, or what is a transform anyway?

We have answered in Fourier Transform.

TIME-DOMAIN FREQUENCY-DOMAIN

Signal is stationary

A non-stationary signal. Interval 0 to 300 ms: 100 Hz, Interval 300 to 600 ms: 50 Hz, Interval 600 to 800 ms: 25 Hz, Interval 800 to 1000 ms: 10 Hz sinusoid.

At what times (or time intervals), do these frequency components occur?

FT cannot answer it!

FT is not a suitable for non-stationary signal, with one exception:

FT can be used for non-stationary signals, if we are only interested in what spectral components exist in the signal, but not interested where they occur. However, if this information is needed, i.e., if we want to know, what spectral component occur at what time (interval) , then Fourier transform is not the right transform to use.

FT gives what frequency components exist in the signal. Nothing more, nothing less!

THE ULTIMATE SOLUTION:THE WAVELET TRANSFORM

Wavelet transform is capable of providing the time and frequency information simultaneously, hence giving a time-frequency representation of the signal.

Example:

Suppose we have a signal which has frequencies up to 1000 Hz

Using filter, we can break the signal into different bands, say, 0-125 Hz, 125-250 Hz,250-500 Hz, and 500-1000 Hz. We can continue this until we have decomposed the signal to a pre-defined certain level.

We can plot them on a 3-D graph, time in one axis, frequency in the second, and amplitude in the third axis. This will show us which frequencies exist at which time.

Before introducing WT, let’s look at short time Fourier Transform (STFT)

There is only a minor difference between STFT and FT. In STFT, the signal is divided into small enough segments, where these segments of the signal can be assumed to be stationary.

)t-w(tx(t) How to do it? Using window as a filter.

Do FT on series of )t-w(tx(t) The red one shows the window located at t=t1', the blue shows t=t2', and the

green one shows the window located at t=t3'. These will correspond to three different FTs at three different times. Therefore, we will obtain a true time-frequency representation of the signal.

Interval 0-250 s: 300 Hz, 250-600s: 200 Hz,600-800s: 100 Hz,800-1000s: 50 Hz

0

250

600

800 1000

60 30 20 10 5

X10

Isn’t that great?

What if frequency changes continues? we can only know what frequency bands exist at what time intervals.

Resolution ProblemThe uncertainty principle, Heisenberg states that the momentum and the position of a moving particle cannot be known simultaneously. Same thing here, We cannot know what spectral component exists at any given time instant.

The best we can do is to investigate what spectral components exist at any given interval of time. This is a problem of resolution

Higher frequencies are better resolved in time, and Lower frequencies are better resolved in frequency.

Let’s look at the problem using window

If a window is of finite length, it obtains time information, but it covers only a portion of the signal, which causes the frequency resolution to get poorer.

If a window has an infinite length, it gets back to FT, which gives perfect frequency resolution, but loses time information.

Narrow window ===>good time resolution, poor frequency resolution.Wide window ===>good frequency resolution, poor time resolution.

Examples /2)exp(-atw(t) 2window

0

250

600

800 100060 30 20 10 5

X10

Four peaks are well separated from each other in time. But in frequency domain, every peak covers a range of frequencies, instead of a single frequency.

0

250

600

800 1000

60 30 20 10 5

X10

Peaks are not well separated from each other in time, but, in frequency domain the resolution is much better.

This is the main reason why researchers have switched to WT from STFT. STFT gives a fixed resolution at all times, whereas WT gives a variable resolution

0

2

50

600

8

00

1000

60 30 20 10 5

An excellent frequencyresolution, buta terrible time resolution

X10

What kind of a window to use? Narrow windows give good time resolution, but poor frequency resolution.

The main differences between the STFT and the WT

The width of the window is changed as the transform is computed for every single spectral component in the WT, but is fixed in the STFT

Fourier analysis

Wavelet analysis

Continuous Wavelet Transform

dtf(t)e)F( tj-

dtposition) (scale,f(t)position) C(scale,

A wavelet transform is defined as

dt)(x(t)s),WT(s-t*

s1

analyzed be tosignal :x(t)

signal. he through tshifted is window theas , window

theoflocation thedetermone to(position)on translati:

scale :s

The parameter scale s in the wavelet analysis is similar to the scale used in maps. As in the case of maps, large scales correspond to a non-detailed global view, and small scales correspond to a detailed view. Similarly, in terms of frequency, low frequencies (large scales) correspond to a global information of a signal (that usually spans the entire signal), whereas high frequencies (small scales) correspond to a detailed information of a signal (that usually lasts a relatively short time).

Large scale Low frequency Small scale High frequency

eletmother wav called function, ing transform:)(s-t

*: complex conjugate

Fourier analysis

Wavelet analysis

)(s-t

Morlet wavelet2

2

0)(x

eex xi

Small scalecompressedhigh frequency

Large scaledilatedlow frequency

Translation

Commonly used mother wavelet other than Morlet wavelet

Haar

1.0] [0, xif 0,x)(

1.0] [0.5, xif 1,x)(

0.5] [0, xif 1,x)(

Mexican Hat

2/x-24/13

22

e)x1)((x)(

Meyer

]1, [0 )20708435()( where,

], [ xif 0,x)(

x if 1)),x(cos(e)2(x)(

x if 1)),x(sin(e)2(x)(

324

38

34

38

34

23

2ix/21/2-

34

32

23

2ix/21/2-

How to do wavelet transform?

)( series, A time tx

wavelet transform dt)(x(t)s),WT(s-t*

s1

1

0 s-n*

s1 t)()nx(s),WT(N

nnn

Reconstruction

(t) of ansformFourier tr theis )(ˆ where

,,ddx(t)2

2s-t

)(ˆ)(,s)WT(1

dCssC

scalechoosen of maximum theofindex theis where

,x(n)1

0

1

0

)(),sWT(1

2js

-j

J

stJ

j

N

n s

tn

Cj

nn

Wavelet transform is much easier and faster in the Fourier domain

dtetxx ti )()(ˆ

wavelet transform

22

22

k

1

0

*

:

:

where

,)(sˆ)(ˆs),WT(

NtN

NtN

tniN

kk

k

k

ekxn k

1

0

/21 )()(ˆN

n

NkniN

enxkx

dex i )(sˆ)(ˆs),WT( *

Example

6,)( 2

2

oti

o

to eet

wavelet transform

)(ˆ)(ˆ sss o

]3:01.0:0[],)1.0(2sin[)( ttttx

Morlet wavelet

2/)( 22

2

2)(ˆ ot

o edteee titio

Wavelets

)()( 1, s

toss t

dex i )(sˆ)(ˆs),WT( *

Time domain

Fourier domain

Time domain

Fourier domain

How to Connect Scale to Frequency?

Pseudo-frequency corresponding to a scale.

cf

a f

is a scale.is the sampling period.is the center frequency of a wavelet in Hz.is the pseudo-frequency corresponding to the scale a, in Hz.

cf

af