Randomized sample rate

[liquidFuzz]

Hi

How do I find the frequencies in a signal obtained by samplings with a random sample rate? Normally I use a fft analysis, but then I have a fixed sample rate.

Thanks!

 

[f95toli]

I don;t understand the questions
Do you mean that you don't know the sample rate?
Or that the signal was sampled at random intervals?
Or that it was sampled at random intervals but you know WHEN the measurement was done (i,e., each sample has a time tag)?

Regardless, you obviously need SOME form of information about the timing of the measurement to estimate the frequency.

 

[tech99]

If you use a random sampling method then it must always be faster than the Nyquist rate. If some samples are spaced longer than the Nyquist interval then there will be random errors on the recovered signal. This amounts to a source of noise which accompanies the signal, and amounts to quantisation noise. Any sampling below the Nyquist rate will produce gross distortion.

 

[Twigg]

If you're talking about data sampled with random intervals (as opposed to constant intervals where you'd use the FFT), then instead of FFT, you want to use LSSA. There's an efficient implementation in Numerical Recipes in C (I don't have it on me or I would give you the chapter number).

Other implementations:
Scipy
Matlab

LSSA is recommended when you expect a periodic signal. I've looked in the literature before but never found answers about how it performs for spectral analysis of non-periodic signals (noise power spectra, for example).

 

[Baluncore]

Time is of the essence. What do you know about the timing of your sampling?

 

[liquidFuzz]

Thanks for all the input! I'll check the numerical approach suggested.

So far I only really tried to attack the issue by contacting the manufacturer of the measuring device I'm tinkering with to see if I can change the sample rate.

 

[liquidFuzz]

Oh, i forgot mentioning. The timing is periodic atm. 0,335 and 0,408 between samples. The 0,335 is used 5 times for each 0,408 timing.