I have accelerometer signal, which is preprocessed by the actigraph on-device. Original sampling rate is 32 Hz, but activity count is summed for every minute, so I have a signal with 1 measurement per minute.

For denoising and to analyze long-term dependencies (my data spans several days), especially related to circadian rhythm, I aggregate this data for each hour, taking the mean. After this I have a signal with 1 measurement per hour.

Libraries for analysis of frequency features, e.g. Scipy or TSFEL require sampling rate as a parameter, e.g. for calculating periodogram.

What should be my sampling rate after all this processing?

  • 32 Hz, since original measurements were gathered with this frequency?
  • 1 Hz, since as a part of very basic preprocessing (and since I don't care about minor movements and analysis of short term effects) a total activity per minute was calculated?
  • 1/3600 Hz, since after aggregating by hour I have one measurement per hour?
  • it depends on purpose and results and no single answer is totally right or wrong?

1 Answer 1


It depends on which signal you want to do frequency analysis with. If you are doing frequency analysis on the 1/3600 Hz signal, then that is the sample rate the function probably expects. If you are doing frequency analysis on the 1 Hz signal, then use 1 Hz as the sample rate parameter in the function you are using.

  • $\begingroup$ So just to make sure: if I just performed downsampling, i.e. took 1 measurement from each hour, that would be 1/3600 Hz, and if I take mean of measurements for each hour it is also 1/3600 Hz, right? $\endgroup$
    – qalis
    Commented Aug 20, 2021 at 9:22
  • 1
    $\begingroup$ Yes that is correct $\endgroup$
    – Ryan
    Commented Aug 20, 2021 at 9:39

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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