Hi Tim,
You have done an interesting experiment with your Bayesian callsign correlator. I think your experiment failed only because your
system missed one important element, a Bayesian squelch. I hope it will not be too difficult to modify your statistical model so
that it includes silence in addition to the callsigns.
GRITTY has a small bug due to a last minute code change, this bug prevents squelch from working properly - that's why you can
see so many CQ's and 599's decoded from the noise. This error is already fixed, a new build will be released shortly.
Another error I have fixed is unnecessary blank lines in the text, TNX to Salvo
IW1AYD for bringing this to my attention.
73 Alex VE3NEA
On 2015-04-13 09:41, Tim Shoppa wrote:
I played around just a little with GRITTY as a third decoder in EA RTTY
contest weekend before last. (Running on an entirely different PC than my
usual 2Tone and MMTTY instances).
Watching it do its Bayesian magic and "back-correcting" sorta like CW
Skimmer is really cool.
The Bayesian smarts in it is really sore thumb obvious - it loves to look
at band noise and print out "599" and "DE" and "QRZ" and "CQ" (which of
course are exceedingly common in real contesting so of course a
straightforward Bayesian approach will see all of them.)
I tried to do something similar with Bayesian decoding last year, where I
built an auto-correlator that looked for several thousand common RTTY
contest callsigns culled from my RTTY contest logs. It of course ended up
printing out well-formed commonly heard callsigns when fed band noise :-).
I'm not sure I'm ready to add it to my "decoder lineup" yet but will be
trying it out some more.
Tim N3QE
_______________________________________________
RTTY mailing list
RTTY@contesting.com
http://lists.contesting.com/mailman/listinfo/rtty
_______________________________________________
RTTY mailing list
RTTY@contesting.com
http://lists.contesting.com/mailman/listinfo/rtty
|