I have a lot of EEG signal data at sampling frequencies of 200Hz, 256Hz and 512Hz. I resampled all the signals and brought them to 256Hz. When I computed the PSD using pwelch function of MATLAB there seems to be a shift in frequency . How should the PSD ideally change or should there be a change??
"pwelch" function code comes by a default sampling frequency but you can adjust that so it gives the best results considering the actual signal sampling frequency. Matlab will not let you change the built in "pwelch" function but what you can do is to save this function under a different name then you make the required change. Now you need to call the new modified function into your main program.
Hello,
Thank you for responding. I will try modifying the sampling frequency in thr pwelch function.
Could you please tell me how upsampling should affect the PSD i.e the power spectral density of the signal theoretically??
It should not !
the reality of your signal (its "analog representation") does not change because you prefer to acquire it at different sample rates, so the frequencies where spread the power over the frequencies is still the same, excepting that you won't see the power above the half sampling rate (shannon).
So if you're finding two differents PSD on the same signal at different sampling rates, my guess is you're making something wrong:
- if you're downsampling the signal, don't forget to low-pass filtering it according to the new sampling rate, otherwise chance are some high frequencies alias into the low-freq part (shannon again)
- an efficient way to upsampling a signal is to make zero-padding in the frequency domain,
- by changing the sampling rate, you're changing the freq value of your rank bin since bin n is for the freq band from n*Fsr/N Hz until (n+1)*Fsr/N Hz where N is the nb of samples used for the FFT and Fsr the sampling rate. So maybe you're wrongly interpret a good result ?
- Badges
- Science topic
- Similar topics
- Engineering
- Computer Engineering