Hacker News new | ask | show | jobs
by FredFS456 1935 days ago
Why not just use a low-pass linear-phase FIR filter instead of FFT/iFFT? Same result but achieved in time-domain, somewhat more performant.
2 comments
hctaw 1935 days ago
That depends on the filter order. Evaluating a FIR is O(N M^2) where N is the number of data points and M is the size of the FIR. Doing the same with an FFT is O(N Mlog(M)).

Neither is particularly "performant." FFTs and FIRs are big guns!

link
hntrader 1935 days ago
Would FFT even be appropriate for nonstationary time series? Wavelets seem more appropriate for such time series (not sure about how peformant they would be though, the R package runs pretty slow even for small datasets)
link