How can I compute spectrogram for data vector containing NaN values?

Dear community,

I am currently working on seismic recordings and I need to compute the spectrogram. The dataset has been cleaned in advance, and thus the original vector contains small portions of NaN values (about a few seconds each with a sample rate of 100 Hz).

However, spectrogram function cannot be applied to NaN values without compromising the entire result, and applying it to not NaN values (applied to data(~isnan(data)) would provide incorrect results (spectrogram function assumes that the sample rate is homogeneous which would no longer be the case).

I do not really know how to manage this issue, and to compute a correct power spectral density for my data, if anyone has a clue...

Thank you

Answer 1

[KSSV]

Multiple options:

1. Discard NaN's
2. Fill the NaN's using interpolation (interp1, spline)
3. Fill the NaN's using fillmissing
4. Fill the NaN's using fillgaps

My favourite is 4 th option.

[Suzanne CARON--SAUTEJEAU]

Thank you, that should do it! :)