Biomedical Engineering Reference
In-Depth Information
12
×
10
4
Yule AR power spectral density estimate
spect1:Yule AR:Nfft = 1024
10
8
6
4
2
0
0
20
40
60
80
100
120
Frequency
(a)
Yule AR power spectral density estimate
×
10
4
14
spect1:Yule AR:Nfft = 1024
12
10
8
6
4
2
0
0
20
40
60
80
100
120
Frequency
(b)
Figure 3.9
The estimated PSDs using an AR model of order (a) 10 and (b) 20 for the same EEG signal
used in Figure 3.7.
One solution to this problem is to divide the long-term signal into blocks or win-
dows of short time duration. The Fourier transform is then computed for each of
these “short” signal blocks. One problem that may arise is that a short window will
lead to a poor spectral resolution. If the window width is increased, the frequency
resolution will improve while the time resolution will deteriorate. The time-fre-
quency trade-off is associated with the Heisenberg uncertainty principle, which
Search WWH ::
Custom Search