Hi Joe, I’m glad to see that you were able to confirm the improved performance of the two-pass decoder. I’m guessing that your dataset includes a more representative mixture of bands and conditions than the group of 20m files that I used. Hence the smaller, but still significant, increase in the number of decodes over the default wsprd.
I am surprised by your observation that the two-pass decoder is faster than the default one. That’s not what I see here. Are you using your wspr_timer.out times? Or some other measure of execution time? The numbers that I reported were the “Total” times from wspr_timer.out. I just ran the analysis of my 20m files using the current default wsprd. It was slightly slower than running wsprd_exp with the -s (single-pass option): wsprd: 1315 decodes in 163s wsprd_exp -s: 1315 decodes in 159s (taken from Table 1 in the .pdf writeup) But in my case, both of these single-pass times are much shorter than the two-pass execution time, which was 270s. I wonder if there is some compiler optimization setting that was different between your wsprd and wsprd_exp instances? Steve k9an > On Jun 26, 2015, at 11:00 AM, Joe Taylor <j...@princeton.edu> wrote: > > Hi all, > > I have committed a "Makefile.win32" for building wsprd_exp.exe during > testing. > > I have now run compartson tests of wsprd.exe and wsprd_exp.exe using a > group of 410 *.wav files, with the following results: > > --------------------------------------------------------- > Decoder Decodes Time > (s) > --------------------------------------------------------- > 1. wsprd 2291 524 > 2. wsprd_exp 2551 418 > 3. wsprd_exp 2653 434 > --------------------------------------------------------- > > Run #1 used the default wsprd.exe now built along with WSJT-X v1.6.0. > Run #2 uses Steve's wsprd_exp with incoherent (symbol-by-symbol) signal > subtraction; #3 used his fully coherent subtraction routine. > > Important note: to make the coherent subtraction work in Windows I had > to increase the stack size -- see LDFALGS in Makefile.win32. > > So, with these example files I got 11% more decodes with > symbol-by-symbol subtraction and 16% more decodes with coherent > subtraction. > The two-pass decoder appears to be faster than the single-pass one. (I > don't yet know why, but in a couple of runs this seems to be repeatable.) > > I think these results are very impressive! We should certainly make > wsprd_exp (renamed to wsprd) the default WSPR decoder. > > -- 73, Joe, K1JT > > ------------------------------------------------------------------------------ > Monitor 25 network devices or servers for free with OpManager! > OpManager is web-based network management software that monitors > network devices and physical & virtual servers, alerts via email & sms > for fault. Monitor 25 devices for free with no restriction. Download now > http://ad.doubleclick.net/ddm/clk/292181274;119417398;o > _______________________________________________ > wsjt-devel mailing list > wsjt-devel@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/wsjt-devel ------------------------------------------------------------------------------ Monitor 25 network devices or servers for free with OpManager! OpManager is web-based network management software that monitors network devices and physical & virtual servers, alerts via email & sms for fault. Monitor 25 devices for free with no restriction. Download now http://ad.doubleclick.net/ddm/clk/292181274;119417398;o _______________________________________________ wsjt-devel mailing list wsjt-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/wsjt-devel
