Biomedical Engineering Reference
In-Depth Information
EEG (2048x1 real, Fs=250)
6000
4000
2000
0
−
2000
−
4000
−
6000
0
1
2
3
4
5
6
7
8
(a)
×
10
4
FFT spectrum estimate
14
spect1:FFT:Nfft = 1024
12
10
8
6
4
2
0
20
40
60
80
100
120
Frequency
(b)
Figure 3.7
(a) EEG sequence recorded from the left frontoparietal regions of a rat's brain sampled at
250 samples per second. (b) The FFT of the EEG signal.
sequence recorded from the left frontoparietal regions of a rat's brain sampled at
250 samples per second. Figure 3.7(b) shows the FFT of the EEG signal, which, as
expected, reflects the statistical variations of the signal.
Figure 3.8 shows the estimated average PSD or periodogram of the same EEG
signal using the Welch technique with a Hamming window and a 1,024-point FFT;
64 data points overlap and 125 windows. The estimated PSDs using an AR model of
order 10 and 20 of the same EEG signal are shown in Figure 3.9(a, b), respectively.
The preceding example emphasizes the importance of model order selection. An
AR model of a relatively low order of 10 produced a “smoothed” spectrum and was
not adequate to show the details of the EEG spectrum estimate. We needed to raise
the order to at least 20 before seeing any resemblance between classical and modern
spectral estimates. If we keep raising the order, less accurate estimates of the signal
spectrum with spurious peaks will result. Akaike criterion can be used to determine
an optimum order model [45].
Search WWH ::
Custom Search