1

My instrument performs measurements at non-equal time intervals, i.e. non-uniform sampling. Apart from this, I usually oversample (by a factor of $n$) for reasons related to filter design, and afterwards downsample by taking every $n^\text{th}$ sample. However, with non-uniform sampling, things seem to be non-obvious. I cannot simply take every $n^\text{th}$ sample since the samples themselves are non-uniform.

I can argue that, since they are non-uniform in the first place, then my downsampling will itself by non-uniform, yet at the end it does not really matter since I still have a non-uniformly sampled signal but now down-sampled. Am I correct?

Gilles
  • 3,386
  • 3
  • 21
  • 28
student1
  • 384
  • 6
  • 16
  • 1
    In general, more data is better. If you just want to plot the data, it doesn't really matter what you do with it. However, if you want to do something else (like filter it) you're better off interpolating it to be uniformly sampled before either decimating it (taking every $n^{\rm th}$ sample) or filtering it. – Peter K. Sep 02 '15 at 13:13
  • You should first resample at a fixed rate. This could help http://dsp.stackexchange.com/questions/8488/what-is-an-algorithm-to-re-sample-from-a-variable-rate-to-a-fixed-rate/20087#20087 – Rhei Sep 02 '15 at 13:38
  • Well, my intent is to ultimately do Lomb-Scargle periodogram: link which is one solution to find spectral density of non-uniformaly sampled signals. However, if I am doing an interpolation and making the signal uniformly sampled, then I can just do an (regular) FFT...right? – student1 Sep 02 '15 at 13:44
  • I guess that you can follow point 2 in that answer – Rhei Sep 02 '15 at 14:41

0 Answers0