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 ...
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
where k g D k= R 1<br />
;1 jg D(t 0 )j 2 dt 0 . Eq. 18 is the inverse Gabor Transform <strong>and</strong> implies that<br />
the original signal can be completely reconstructed from the coecients G D (f t).<br />
Gabor Transform is highly redundant because it gives a time-<strong>frequency</strong> map from<br />
every time value <strong>of</strong> the original signal. In order to decrease redundancy, asampled Gabor<br />
Transform can be dened by taking discrete values <strong>of</strong> time <strong>and</strong> <strong>frequency</strong>, i.e.:<br />
G D (f t) ! G D (mF nT ) (19)<br />
where F <strong>and</strong> T represent the <strong>frequency</strong> <strong>and</strong> time sampling steps. Small F steps are<br />
obtained by using large windows, <strong>and</strong> small T steps are obtained by using high overlapping<br />
between successive windows. Depending on the resolution required, a proper<br />
choice <strong>of</strong> F <strong>and</strong> T will decrease the redundancy <strong>and</strong> save computational time, but the<br />
price to pay is that the reconstruction will be no longer straightforward as with eq. 18<br />
(Qian <strong>and</strong> Chen, 1996).<br />
In the following, for convenience I will keep the notation <strong>of</strong> the continuous Gabor<br />
Transform dened in eq. 14. The spectrum can be dened as<br />
I(f t) =jG D (f t)j 2 = G D(f t) G D (f t) (20)<br />
<strong>and</strong> a time-<strong>frequency</strong> representation <strong>of</strong> the signal can be obtained by using a recursive<br />
algorithm that slides the time window <strong>and</strong> plots the energy (I) as a function <strong>of</strong> the<br />
<strong>frequency</strong> <strong>and</strong> time. These plots, called spectrograms (see g. 8), give an elegant visual<br />
description <strong>of</strong> the time evolution <strong>of</strong> the dierent frequencies, but it is dicult to extract<br />
from them any quantitative measure.<br />
In order to quantify this information, I will dene the b<strong>and</strong> power spectral intensity<br />
for each one <strong>of</strong> the <strong>EEG</strong> traditional <strong>frequency</strong> b<strong>and</strong>s i (i = :::) as:<br />
I (i) (t) =<br />
Z f<br />
(i) max<br />
f (i) min<br />
I(f t) df i = ::: (21)<br />
where ( f (i) min f (i) max ) are the <strong>frequency</strong> limits for the b<strong>and</strong> i. The division <strong>and</strong> grouping<br />
<strong>of</strong> the spectrum in <strong>frequency</strong> b<strong>and</strong>s is not arbitrary, since as showed in section x1.1.1<br />
<strong>and</strong> section x2.3.3, <strong>EEG</strong> b<strong>and</strong>s are correlated with dierent sources <strong>and</strong> functions <strong>of</strong> the<br />
brain.<br />
Obviously the total power spectral intensity will be<br />
I T (t) = X i<br />
I (i) (t) i = ::: (22)<br />
<strong>and</strong> we can dene the b<strong>and</strong> relative intensity ratio (RIR) foreach <strong>frequency</strong> b<strong>and</strong> i as<br />
23