Hi Joe and all, Thanks Joe for implementing the fft-based convolution. It makes a big difference in execution time.
It might save even more time if we calculated the averaged complex amplitude at, say, 1/100’th or even 1/1000’th as many points enabling us to use a much smaller fft for the return to the time-domain. We could interpolate or even use a piece-wise constant approximation to the complex amplitude when calculating the reconstructed signal. Do you think that this is worth pursuing? Note that this idea, taken to the extreme of one sample of the complex amplitude per symbol, reduces to the original symbol-by-symbol approach that I tried first when we implemented two-pass decoding for wspr. Recall that the symbol-by-symbol approach worked pretty well - so it might take only a few samples per symbol to essentially get all of the available benefit... FYI - when I ran the latest jt65 on my data just now I got significantly fewer decodes than I got last night before the fft-based-filter implementation. I traced the problem to a mis-print in the scaling factor “fac”. The incorrect scaling meant that only 78% of the signal was subtracted - giving a maximum suppression of about 13 dB. I’ve committed the fix. If you get a chance to run this latest version on your files, I expect that you should see a bigger improvement with the second pass. Steve k9an > Program Decodes Time > -------------------------------------------- > WSJT-X v1.6.0 v5636 (kvasd) 2270 230 s > jt65 (1 pass, sfrsd2) 2296 767 > jt65 (2 passes, sfrsd2) 2819 2144 > > Note that although the 2-pass decoder is significantly slower than the > others, it still takes less than 5 s per .wav file, on average. This is > perfectly acceptable performance -- especially since the all-important > decode at the QSO frequency always happens first, and will generally be > done in less than a second. > > -- Joe, K1JT > > ------------------------------------------------------------------------------ > _______________________________________________ > wsjt-devel mailing list > wsjt-devel@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/wsjt-devel ------------------------------------------------------------------------------ _______________________________________________ wsjt-devel mailing list wsjt-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/wsjt-devel
