FFT pad or no pad
Started by ●August 21, 2009
I have 300sec data collected at 1Hz. I want to look at power spectrum of this
time series, so first I need to do FFT(). But I am not sure if I need to pad it
with zeros to make it 512. Matlab FFT() lets me do it either way, but I
don't really understand which way I should do. Please advise. Thanks.
Reply by ●August 22, 20092009-08-22
KC-
> I have 300sec data collected at 1Hz. I want to look at power
> spectrum of this time series, so first I need to do
> FFT(). But I am not sure if I need to pad it with zeros to
> make it 512. Matlab FFT() lets me do it either way, but I
> don't really understand which way I should do. Please advise. Thanks.
A power spectrum implies more than one frame; i.e. averaging of some type in the frequency domain. But you have only
one 300 pt. frame.
I think at this stage in your analysis, either DFT or zero-padded FFT would be fine. The main reason to use FFT is in
a real-time or product implementation where there are constraints on your available processing resources.
Also, if you do use a zero-padded FFT, be sure to take into account "edge noise" artifacts, due the cut-off at the end
of your 300 pts. You can mitigate this by applying a window prior to the FFT.
-Jeff
> I have 300sec data collected at 1Hz. I want to look at power
> spectrum of this time series, so first I need to do
> FFT(). But I am not sure if I need to pad it with zeros to
> make it 512. Matlab FFT() lets me do it either way, but I
> don't really understand which way I should do. Please advise. Thanks.
A power spectrum implies more than one frame; i.e. averaging of some type in the frequency domain. But you have only
one 300 pt. frame.
I think at this stage in your analysis, either DFT or zero-padded FFT would be fine. The main reason to use FFT is in
a real-time or product implementation where there are constraints on your available processing resources.
Also, if you do use a zero-padded FFT, be sure to take into account "edge noise" artifacts, due the cut-off at the end
of your 300 pts. You can mitigate this by applying a window prior to the FFT.
-Jeff