My inquiry is regarding the so called cepstrum analysis.

By conducting some experiments, i have real time data at my disposal. The acquisition rate is 1600Hz.

I wanted to try cepstrum analysis just out of curiosity to see how it performs in comparison with trivial spectral analysis.

So what i do with the acquired data:

  1. Use a lowpass filter with a cut-off frequency of 340 Hz for noise attenutation.
  2. Fourier transform the filtered data, removing the DC component by subtracting the mean value from the filtered data.
  3. Take the log of the magnitude of the FFT.
  4. Apply IFFT on the resulting vector to acquire the real cepstrum.

I notice that, despite the filtering that has taken place at step 1, the cepstrum plot displays activity beyond the cut-off frequencies. So my question is this: Is this supposed to happen? To me it seems that this is not normal. Does anyone have any experience with that?

Edit: Plot image added for clarification. Frequencies of interest lie on the right of the red line displayed

Cepstrum vs Quefrency plot


Presence of frequencies above cutoff in Cepstrum could be because of -

  • Is lowpass filter used a perfect lowpass filter(does it removes all the frequencies above cutoff or just attenuates them) ?
  • If lowpass filter is not perfect, what are the relative magnitudes of high frequencies(above cutoff) with respect to low frequencies. if high frequencies are relatively dominant then they will be visible in cepstrum.

also "Fourier transform the filtered data, removing the DC component by subtracting the mean value from the filtered data."

After fourier transform subtracting mean should not be the way to remove DC.

| improve this answer | |
  • $\begingroup$ Hello, this is indeed the case as the lowpass filter does not completely remove them but just attenuates them. I might try manually setting them to 0 to see what kind of different results i take out of them. Regarding the magnitude of the cut off frequencies in terms of cepstrum plot, they do appear to be higher than the rest, this is why i was wondering what is going on. P.S: Regarding DC removal, just to clarify, i remove the mean component of the data before Fourier Transform takes place, not afterwards. $\endgroup$ – Paraskevas Dimitris Apr 5 '17 at 7:30

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.