|
| |
Example 2The second example we consider is a bat sonar signal, recorded with a sampling frequency of 230.4kHz and an effective bandwidth of [8kHz, 80kHz] (this recording was part of the research program RCP 445 supported by CNRS (Centre National de la Recherche Scientifique, France) [Fla86]). First, load the signal from the MAT-file bat.mat (see fig. 1.7):
>> load bat
>> t0=linspace(0,2500/2304,2500);
>> plot(t0,bat); xlabel('Time [ms]');
From this plot, we can not say precisely what is the frequency content at
each time instant
>> dsp=fftshift(abs(fft(bat)).^2);
>> f0=(-1250:1249)*230.4/2500;
>> plot(f0,dsp); xlabel('Frequency [kHz]');
we can not say at what time the signal is located around 38kHz, and at
what time around 40kHz (you can use the zoom function to see more
precisely what is happening around these frequencies ; see the Matlab
Reference Guide). Let us now consider a representation called the pseudo
Wigner-Ville distribution, applied on the most interesting part of this
signal (this distribution was obtained with the M-file tfrpwv.m, stored in the matrix tfr and saved
with the signal in the MAT-file bat.mat ; the corresponding
time- and frequency- samples t and f where also
saved on bat.mat) (see fig. 1.9):
>> contour(t,f,tfr,5); axis('xy');
>> xlabel('Time [ms]'); ylabel('Frequency [kHz]');
>> title('TFRPWV of a bat signal');
We then have a nice description of its spectral content varying with time :
it is a narrow-band signal, whose frequency content is decreasing from
around 55kHz to 38kHz, with a non-linear frequency modulation
(approximately of hyperbolic shape).
Eric Chassande-Mottin 2005-10-26 | |