*From*: Stefan Westerfeld <stefan space twc de>*To*: Beast Liste <beast gnome org>*Subject*: MERGE: bug-491577 - Gsl FFT test improvement*Date*: Thu, 16 Sep 2010 11:36:54 +0200

Hi! I've spent some time trying to figure out the difference between Gsl FFT and FFTW. The issue already had a bug report: http://bugzilla.gnome.org/buglist.cgi?bug_id=491577 491577 - Gsl FFT code can produce incorrect results ======= I think I've found the reason for the difference between Gsl FFT and other FFTs. While Wikipedia (and FFTW) use out[k] = SUM{n=0..N-1} (in[n] * exp (-2 * pi * j / N * k * n)) as definition for the FFT, Numerical Recipies (and the Gsl FFT) use (one minus sign less) out[k] = SUM{n=0..N-1} (in[n] * exp (2 * pi * j / N * k * n)) so there seems to be no bug here, just a different definition of what we want to compute. The two versions are relatively easy to convert into each other (O(n)), and I do it in SpectMorph. In any case I suggest adding the DFT test, I modified it to match the "Gsl/Numerical Recipies" formula. Will send a merge request (stwbeast -> bug-491577). ======= repo: http://space.twc.de/public/git/stwbeast.git branch: bug-491577 Cu... Stefan -- Stefan Westerfeld, Hamburg/Germany, http://space.twc.de/~stefan

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