[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: n-point FFT

Paul Woodford wrote:
> In article <onzoit8fai.fsf@cow.physics.wisc.edu>,
> craigmnet@cow.physics.wisc.edu wrote:
> > It's very easy to
> > do any needed zero-padding yourself (ie, fft([x, fltarr(nzeros)]) ).
> Easy, but inconvenient.  I also missed the Matlab notation when I
> switched to IDL, and finally hacked a small "efft" function to recreate
> it.

Different strokes, but I would never recommend this sort of thing mostly because too many
times I've seen (both others and myself) get in trouble with FFT's because something I or
someone else assumed about the particular FFT implementation (fortran, idl, matlab, C,
various flavours - take your pick) was wrong (or non-portable). Particularly when it comes
to specifying exactly where the Nyquist point is.

Thus, I would always hope that my colleagues are painstakingly clear about how many points
they expect to have in, and with how many zeros they padded, their spectra(um).
> I keep forgetting to send this feature request to RSI...

I wouldn't consider it a feature, but a user bug waiting to happen...... +/- 1 point
errors in FFTs can be capriciously subtle  :o(

BTW, I'm not bitter...no, not at all (snuckin' fessin' rotten rasterdly FFTs.....)



Paul van Delst           Ph:  (301) 763-8000 x7274
CIMSS @ NOAA/NCEP        Fax: (301) 763-8545
Rm.207, 5200 Auth Rd.    Email: pvandelst@ncep.noaa.gov
Camp Springs MD 20746