Improve STFT Plot Clarity

Hi,

I'm analyzing a time-domain signal using STFT and want to improve the clarity of the spectrogram. Here's my code:

load('D.mat')
nfft_custom = 1024*5;
wlen = nfft_custom;
win = hann(wlen, 'periodic');

[S, f, t_stft] = stft(AAA, 100000, ...
    'Window', win, ...
    'OverlapLength', round(0.5 * wlen), ...
    'FFTLength', nfft_custom);

stft_dB = 20*log10(abs(S) + eps);
imagesc(t_stft, f, stft_dB); axis xy; ylim([0 800]); caxis([-70 0]);
xlabel('Time (s)'); ylabel('Frequency (Hz)');

Answer 1

[Mathieu NOE]

hello @Abo

and welcome back !

fist idea with STFT is to maximize the overlap , but anyway the STFT has not the best performance when you look for time AND frequency accuracy (cwt would be a better choice : look at the example in : CWT-Based Time-Frequency Analysis

to reduce the amount of fft computation I decided to downsample the data first with decimate (otherwise you are just doing a lot of computations and throw 90% at the garbage bin

to further improve the rendering with imagesc you can use the option : 'Interpolation', 'bilinear'

load('D.mat')
Fs = 100000;
% decimate the data 
r = 8;
AAA =decimate(AAA,r);
size(AAA)

ans = 1×2

4096 1

<mw-icon class=""></mw-icon>

<mw-icon class=""></mw-icon>

Fs = Fs/r

Fs = 12500

%% spectrogram
nfft_custom = 1024;
wlen = nfft_custom;
win = hann(wlen, 'periodic');
[S, f, t_stft] = stft(AAA, Fs,'Window', win,'OverlapLength', round(0.95 * wlen), ...
    'FFTLength', nfft_custom,'FrequencyRange',"onesided");
stft_dB = 20*log10(abs(S) + eps);
imagesc(t_stft, f, stft_dB, 'Interpolation', 'bilinear');
axis xy; 
colormap('jet');
ylim([0 800]);
caxis([-80 0]);
xlabel('Time (s)'); ylabel('Frequency (Hz)');
colorbar('vert');

●

[Abo]

Thanks, Mathieu. I’ll check out the CWT method. Your input already helped improve the STFT plot. Much appreciated.

[Mathieu NOE]

as always, my pleasure !