Hi Joe et al, I just did my first svn commit into the wsjtx_exp branch. This revised version of the wsprd decoder is 10-20% faster (I only calculate the sin/cos tables when absolutely necessary) and has a slightly increased probability of a successful decode on average (it yields a couple of % more decodes in my tests, depending on the band) and is significantly better on very weak signals.
I also added a new option (-z) which is the metric table “bias” parameter, currently set at a default value of 0.42. This bias “knob" is one way to control and study the tradeoff between the probability of a failed decode (i.e., a decoder time-out) and the probability that the decoded message contains an error. The “standard” value for the bias is 0.5. Lower values bias the decoder so that it wants to go forward and finish at the risk of making errors, whereas raising the bias slows the decoder down, resulting in more timeouts, while decreasing the probability of errors in the successfully decoded messages. Aside from being fun to play with, I thought that this “knob” might be useful in a QSO mode like JT9 where the operator might be willing to accept some frames with errors in trade for maximizing the probability of getting the QSO. Just FYI, Joe, to add the metric bias I added a file called “metric_tables” that contains the un-biased and un-scaled 2p0/(p0+p1) metric for 4 different Es/No values (0, 3, 6, 9 dB). I experimented with having the decoder switch metric tables and symbol scaling factors based on the estimated SNR, and it just didn’t buy us anything. So the decoder is currently set up to use the 6 dB table with a default bias of 0.42. Here’s a summary of some comparisons between old and new: 100 test files with -30 dB SNR: old “new” wsprd: 66 successful decodes in 30 seconds new “new” wsprd: 75 decodes, 2 bad decodes (73 successful decodes) in 26 seconds. 10% more decodes, 13% reduction in computation time. On my test suite of 214 “off-the-air” 20m files the comparison between the old and new decoders is as follows: old “new” wsprd: 1306 successful decodes (no bad decodes) in 164s new “new” wsprd: 1330 successful decodes (1331 decodes, 1 bad decode) in 138s 2.2% more decodes, 16% reduction in computation time. 73 Steve k9an ------------------------------------------------------------------------------ One dashboard for servers and applications across Physical-Virtual-Cloud Widest out-of-the-box monitoring support with 50+ applications Performance metrics, stats and reports that give you Actionable Insights Deep dive visibility with transaction tracing using APM Insight. http://ad.doubleclick.net/ddm/clk/290420510;117567292;y _______________________________________________ wsjt-devel mailing list wsjt-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/wsjt-devel
