Project

General

Profile

Forums » Open discussion »

spectrum operator

Added by Ivan Guettler about 9 years ago

Dear everyone,

Do some of you use cdo operator spectrum (undocumented)? I find it quite elegant but have one short question concerning its results.

Usually, the integral of the power spectral density (this is what operator spectrum produces) should equal the variance of the input time-series.
I've tested several timeseries, and usually I get area integral that is ~95% of the variance. Is there any way one can perform better? 99% would be nice :)

Any hints are welcome.
Ivan


Replies (1)

RE: spectrum operator - Added by Ivan Guettler about 9 years ago

Sorry everyone. I had to remind myself that impact of the segmenting and windowing of the input timeseries will change the final integral of the power spectral density (PSD).
If no windowing and no segmenting of the input timeseries is performed, the PSD is not smooth but the PSD integral is ~99.5% of the variance of the input timeseries.

Cheers, Ivan

    (1-1/1)