Wavelet Transform, Wavelets.
Fourier analysis provides a description of signals in terms of
sinusoids and is ideal for signals whose frequency content does not
change with time. One way of applying Fourier analysis to signals
with time varying spectra is to use the STDT (Short Time Fourier
Transform) where the signal is broken up into short segments and
Fourier Analysis applied to each section. This will be done in
DSP using the FFT.
Here a different approach is used. In the frequency domain, a sinusoid
occupies 1 point on the frequency axis and in the time domain it spreads
out from -¥ to + ¥. One way to do analysis of
signals is to use a function which extends
over a finite range of frequency and time. The continuous wavelet
transform is such an approach. Unfortunately some lack of mathematical clarity
exists here and the result has been a vast literature addressing
questions like :- what is an acceptable wavelet, types of wavelets, how to construct wavelets,
orthogonality, relation of wavelets to other concepts, reconstruction of signal from wavelet etc. Here we
introduce 2 wavelets, the Mexican Hat(MH) and the Gabor(GB). The
MH wavelet is suited for analysis of fine scale transients in
signals while the GB wavelet is suited for signals with sections
looking "sinusoidal" like audio. After introducing the wavelets,
the GB wavelets is used to produce Scalograms of some basic
The equation describing the MH wavelet is shown below:-
This wavelet has 2 variables t and w, the
width of the wavelet.
The equation for the GB wavelet is shown below. It has 3 variables
t, w(width) and n(frequency).
This wavelet is complex and is one of a class of non orthogonal
wavelets. Since the Fourier Transform of a gaussian is also a
gaussian, then it is reasonable to use a gaussian for a function
required for a time frequency representation of a signal. The GB
wavelet in the frequency domain is a gaussian translated along
the frequency axis.
From the basic wavelet, scaled wavelets can be produced using the
The applet below show the MH and GB wavelets and indicates how the
parameters s, w and n
affect both the time and frequency domain description of the wavelet.
Both s and w have a similar effect.
Increasing both of them widens the wavelet in the time domain
and narrows it in the frequency domain. In the frequency domain,
the center frequency is also reduced. The opposite happens
when s and w are reduced. The parameter
n applies to the GB wavelet and changes the period of the oscillations in the time domain while translating the
spectrum along the frequency axis without affecting its width. The
spectrum of the GB wavelet is also a gaussian.
When enabled the following gif file show how the applet should
Return to main page
Return to page index
COPYRIGHT © 2007 Cuthbert Nyack.