2

Is there a rule of thumb in deciding the cut-off value when performing the low time liftering process in Cepstral analysis? Or is it just trial and error analysis?

I am trying to calculate the spectral envelope of the frequency response of data obtained from a vibration sensor. Sampling frequency is 5000 Hz.

sifferman
  • 2,955
  • 2
  • 27
  • 37
Kanmani
  • 479
  • 7
  • 21
  • 1
    For those who might downvote this question on the basis of apparent dyslexia of the OP, think again. – sifferman Mar 03 '17 at 18:10
  • Mr.Sifferman, would you like to elaborate on that 'dyslexia' comment ? Why would this question make anyone think I have dyslexia ? I don't get it. – Kanmani Mar 04 '17 at 18:51
  • Because "liftering" looks like "filtering" misspelled. You asked a legitimate question, a good question, but somebody downvoted it, I can only imagine on the assumption of lousy spelling. I was about to edit your post for spelling until I googled "Cepstral Analysis" and learned about low-time liftering. – sifferman Mar 05 '17 at 13:13
  • I understand. I guess not many know about cepstral analysis. I couldn't add it as a tag to the question since it wasn't available in the suggestions. – Kanmani Mar 05 '17 at 23:35
  • This is really, really cool to find out, but I think this also belongs on EE or DSP stack exchange, or maybe Math.SE. – bright-star Mar 06 '17 at 00:22
  • Thank you. I have added the question on DSP SE too. Haven't found an answer yet. I now understand that this question is inappropriate here. So I will remove it. – Kanmani Mar 09 '17 at 02:08

1 Answers1

1

I had done a project on cepstral analysis and found that 15 or 20 is the low-time liftering cut-off.

To my application 15 seemed to be pretty good.

Both the values are acceptable and works good.

But think for your application.

And I don't think there exist any rule or formula to calculate it. Because it works well for all kinds of values and signals.

So it must be trail and error by someone who ended up in this value.