Quantitative analysis of EEG signals: Time-frequency methods and ...
Quantitative analysis of EEG signals: Time-frequency methods and ...
Quantitative analysis of EEG signals: Time-frequency methods and ...
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
4 Wavelet Transform<br />
4.1 Introduction<br />
Fourier Transform consists in making a correlation between the signal to be analyzed<br />
<strong>and</strong> complex sinusoids <strong>of</strong> dierent frequencies (see upper part <strong>of</strong> g. 13). As stated in<br />
chapter x2, Fourier Transform gives no information about time <strong>and</strong> requires stationarity<br />
<strong>of</strong> the signal. By \windowing" the complex sinusoidal mother functions <strong>of</strong> the Fourier<br />
Transform, a time evolution <strong>of</strong> the frequencies can be obtained just sliding the windows<br />
throughout the signal. This procedure, called the Gabor Transform, consists in correlating<br />
the original signal with modulated complex sinusoids as showed in the middle part <strong>of</strong><br />
g. 13 (for details see section x3.2). Gabor Transform gives an optimal time-<strong>frequency</strong><br />
representation, but one critical limitation appears when windowing the data due to the<br />
Uncertainty Principle (see x3.2 <strong>and</strong> xA Cohen, 1995 Strang <strong>and</strong> Nguyen, 1996 Chui,<br />
1992). If the window is too narrow, the <strong>frequency</strong> resolution will be poor, <strong>and</strong> if the<br />
window is too wide, the time localization will be not so precise. Data involving slow<br />
processes will require wide windows <strong>and</strong> on the other h<strong>and</strong>, for data with fast transients<br />
(high <strong>frequency</strong> components) a narrow window will be more suitable. Then, owing to<br />
its xed window size, Gabor Transform is not suitable for analyzing <strong>signals</strong> involving<br />
dierent range <strong>of</strong> frequencies.<br />
Grossmann <strong>and</strong> Morlet (1984) introduced the Wavelet Transform in order to overcome<br />
this problem. The main advantage <strong>of</strong> wavelets is that they have avarying window<br />
size, being wide for slow frequencies <strong>and</strong> narrow for the fast ones (see lower part <strong>of</strong><br />
g. 13), thus leading to an optimal time-<strong>frequency</strong> resolution in all the <strong>frequency</strong> ranges<br />
(Chui, 1992 Strang <strong>and</strong> Nguyen, 1996 Mallat, 1989). Furthermore, owing to the fact<br />
that windows are adapted to the transients <strong>of</strong> each scale, wavelets lack <strong>of</strong> the requirement<br />
<strong>of</strong> stationarity.<br />
This chapter starts with a brief theoretical background <strong>of</strong> the Wavelet Transform.<br />
Details <strong>of</strong> a straightforward implementation called the multiresolution <strong>analysis</strong> <strong>and</strong> an<br />
alternative decomposition <strong>of</strong> time <strong>signals</strong> called the Wavelet Packets <strong>analysis</strong> will be<br />
given. In section x4.4, I will show the results obtained by applying trigonometric wavelet<br />
packets (Serrano, 1996) to the study <strong>of</strong> tonic-clonic seizures (Blanco et al., 1998a). In<br />
sections x4.5 <strong>and</strong> x4.6, I will analyze two dierent types <strong>of</strong> event related potentials. By<br />
using the multiresolution decomposition method, in the rst case I will study the alpha<br />
b<strong>and</strong> responses (Quian Quiroga <strong>and</strong> Schurmann, 1998, 1999) <strong>and</strong> in the second case the<br />
ones corresponding to the gamma b<strong>and</strong> (Sakowicz et al., 1999). Finally, I will discuss<br />
advantages <strong>of</strong> Wavelets in the study <strong>of</strong> brain <strong>signals</strong>.<br />
36