I essentially have two signals X and Y, recorded with PPG devices. These have been filtered already. I want to standardize(z-score) or min-max scale them but I don't know if I should do this on each signal individually, or combine them (This is my main question). I am then calculating the STFT to calculate the power spectral densities in segments for my ML model.

  • $\begingroup$ why would you normalize them? That would mean whatever you do doesn't work on your original signal but a modified one, where you removed a lot of the important info – especially since you're interested in power spectral densities, I'd say this is counter-productive. $\endgroup$ Commented Aug 3, 2021 at 15:49
  • $\begingroup$ I observed that the spectrogram does not change after standardization. Yet, the power spectral density values are smaller. $\endgroup$ Commented Aug 3, 2021 at 17:38
  • $\begingroup$ enzokro.dev/spectrogram_normalizations/2020/09/10/… iopscience.iop.org/article/10.1088/1361-6579/abf9f3/pdf Some resources that I used. $\endgroup$ Commented Aug 3, 2021 at 17:55
  • $\begingroup$ yes, but for a reason, not "just because", so why do you want to normalize, if you're actually interested in the PSD, and not something else. $\endgroup$ Commented Aug 3, 2021 at 19:28
  • $\begingroup$ For the exact same reason that those resources did it. For my particular ML application, I need to have inputs that are distributed between 0 and 1 or have 0 mean and std of 1. Spectrum normalisation is even more challenging and thus I am going for this approach. My question is what method of normalisation is more appropriate i.e. how to apply it on the data. $\endgroup$ Commented Aug 3, 2021 at 22:59


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.