STFT Spectrogram Recognize Linear Regression

Hi Community,

this problem might be considered a low level pattern recognition. The starting dataset is an STFT spectrogram.

As you can see from the plot one signal is constant in FFT domain, the other in STFT domain - I believe this is called sparsity.

The two approximately orthogonal signals need to be seperated and converted back to time continuous signals (constant frequency and FMCW chirp).

One idea was to look for max values across frequency bins and look for concatenated regions across a certain time.

This way chirp slope could be determined.

Regards, Robert

[Robert Worm]

To give an update, I started using Empirical Mode Decomposition which for my case seems to be more promising since I can easily replace data.

[imf,residual,info] = emd(data,'Interpolation','pchip');

As you can see IMF1 data has a varying frequency and constant frequency component. Is there a way to make adjustments to the first sifting stage?

The problem is the main part of the faulty signal is contained in IMF1 but is needed in IMF2 - seperated from the constant signal.