[Tftb-help] tfrsp doubt (fwd)

[Eric Chassande-Mottin]

hi,

your observation is related to finite size effects. we analyze signals of 
finite duration. to compute a TFR, you need to decide what to do at the 
boundaries. most (if not all) routines of the TFTB (including tfrsp) 
assume that the signal is zero beyond the boundaries.

consequently, from the definition of a short-term Fourier analysis, 
http://gdr-isis.org/tftb/tutorial/node34.html the signal selected by the 
analysis window when it is centered around a point near a boundary is 
partly zero. at t=0, half of the right hand part of the windowed signal is 
zero. we see less signal, thus the spectrogram amplitude decreases.

if you are not satisfied by this, you may like to consider TFRs which 
conserves the support of the signal (i.e., which are zero when the signal 
is zero). see Flandrin, Time-frequency/time-scale analysis, Academic Press 
for more details on this point.

bye, e.

---------- Forwarded message ----------
Date: Tue, 13 Jun 2006 08:27:52 +0800
Subject: tfrsp doubt

Hi,Eric,
I found one probelm in tfrsp.m.
If I enter:"sig=fmlin(128,0.1,0.4); tfrsp(sig);" in Matlab commend,I get one 
plot as per attached file.

Based on this signal, tfr should be at time 0 ->normalised frequency is 0.1, 
whereas, time 128->freq 0.4.

But how come appear weak energy value at the time 0 and time 128 in the plot? 
Is it dealed properly?How to see the energy value exactly cut at location(0, 
0.1) and (128, 0.4)?

Thanks,
Joc

prob.doc
Description: Binary data