Bill, Just an idea: Wouldn't it make sense to discard all the /R ?a2 decodes unless a special operation mode is set?. My observation also with the prior WSJT-X versions was, that if there are false decodes, most often they are /R decodes. Couldn't that significantly reduce rate of false decodes without reducing the sensitivity? And when only the AP decodes were discarded, a true /R call would still be decoded. Couldn't that be a solution?
73 de Uwe, DG2YCB ____________________________________ German Amateur Radio Station DG2YCB Dr. Uwe Risse eMail: [email protected] Info: www.qrz.com/db/DG2YCB -----Ursprüngliche Nachricht----- Von: Bill Somerville [mailto:[email protected]] Gesendet: Montag, 25. Mai 2020 13:43 An: [email protected] Betreff: Re: [wsjt-devel] WSJT-X 2.2.0-rc2 False Decode On 25/05/2020 12:04, Stephen VK3SIR wrote: > Folks, > > I just picked up a false decode that can be replicated when played back through WSJT-X 2.2.0 rc2: > > 104945 -23 -0.5 1623 ~ VK3VM FH9ZZV/R ND49 ? a2 > > Research suggests this is not a genuine call and is definitely a false decode (i.e. ND49 = middle of Southern Ocean). I have noted that many false decodes in the previous version are reported with /R ! > > As one cannot send attachments via email I'll reforward this email plus the .wav file directly to Bill and Joe for further analysis. > > 73 > > Steve I > VK3VM / VK3SIR Steve, AP decodes flagged as low confidence ('?' marker) should always be considered dubious, and unless there is other evidence that the decode is genuine it should be ignored. Without using knowledge not obtained on air it is virtually impossible for the decoder to eliminate such false decodes without damaging the capabilities of the AP decoding mechanism. AP decodes of the 'a2' category are only detected shortly after a CQ call IIRC. The FT8, FT4, and MSK144 decoders give virtually every possible message equal weight. The message types that allow a '/P' or '/R' (grid rover station) prefix to a standard callsign have a 50% probability per possible callsign that random data will unpack as one of those prefixes, so they will be far more common in false decodes than in genuine messages where the expected likelihood is far lower. 73 Bill G4WJS. _______________________________________________ wsjt-devel mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/wsjt-devel _______________________________________________ wsjt-devel mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/wsjt-devel
