Hi Steve,
On 10/26/2015 10:27 PM, Steven Franke wrote:
> Hi Joe,
>> In the WSJT slow modes PTT is nominally asserted at t=0 s and audio
>> starts at t=1 s in a UTC minute. SimJT generates wav files in which the
>> noise starts at t=0 and signal starts at t=1. I guess this is
>> essentially what you observed?
>
> Perhaps this is what I’m seeing. I incorrectly assumed that the dtx
> parameter would be an estimate of the delay to onset of the signal
> with dtx=0 corresponding to onset at the first sample of the dd vector.
We have always started Tx audio at t=1 s so as to accommodate computer
clock mis-alignments up to +/- 1 s without losing any signal at the Rx end.
> I looked at what you are doing with the experience-based approach.
> Great idea! I’m going to guess that you’ll have to defend this
> approach to some who might argue that this is “unfair”. I’m not
> one of those - any method for generating candidate codewords to
> be placed on the list and tested against the senseword is fair.
> There should be no penalty for being clever and testing the
> most-likely possibilities first!
Agreed. As you suggest, if we use the same acceptance criterion as in
the "ntrials" loop then all we have done is to speed up the decoding
process (by orders of magnitude) by testing likely codewords first.
> I’m slightly uneasy, though, about using a looser acceptance
> criterion when testing an experience-based codeword vs. a codeword
> produced by the errors-and-erasures decoder. It seems that this
> could raise the probability of seeing
> “highly-believable-because-we’ve-seen-this-codeword-before” false
> decodes.
Indeed, setting a looser acceptance criterion for experience-based
decodes means not only decoding weaker signals but also an increased
probability of false decodes. Just as with ear-and-brain reception of
CW or SSB transmissions, it's easier to recognize and "decode" familiar
callsigns than arbitrary ones -- but when operating very close to the
noise limit, occasionally one can make a mistake.
Setting a looser threshold should be optional, and an operator would
have to recognize and reject occasional false decodes. This is
generally easy to do. I think experience-based decoding will be a
useful feature only VHF and above, and possibly also at LF/MF.
Certainly not on crowded HF bands, where signals are typically much
stronger.
Congratulations on your excellent results with two-pass JT65 decoding!!
We are definitely coming up with some very significant enhancements to
WSJT-X!
-- Joe, K1JT
------------------------------------------------------------------------------
_______________________________________________
wsjt-devel mailing list
wsjt-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/wsjt-devel
