eHam

eHam Forums => Computers And Software => Topic started by: AG1LE on January 05, 2013, 12:55:04 PM



Title: Towards Bayesian CW Decoder
Post by: AG1LE on January 05, 2013, 12:55:04 PM
I spent few sleepless nights to implement my first version of Bayesian CW decoder. It is still in alpha level quality but it does decode Morse code to -10dB SNR (3Khz) using 32Hz Matched filter almost perfectly.

I used FLDIGI v.3.21.64 as the basis and I documented my learnings in here:

http://ag1le.blogspot.com/2013/01/towards-bayesian-morse-decoder.html (http://ag1le.blogspot.com/2013/01/towards-bayesian-morse-decoder.html)


It is not ready for prime time yet but if anybody is interested developing this further please let me know.



73
Mauri AG1LE


Title: RE: Towards Bayesian CW Decoder
Post by: NK2F on January 27, 2013, 06:49:39 AM
How about using the database of known call signs to improve the ability to decode call signs?


Title: RE: Towards Bayesian CW Decoder
Post by: N9RO on January 28, 2013, 11:31:29 AM
Mauri,

The Bayesian page is great, not often you visit a ham site where you can learn something.

Tim  N9RO


Title: RE: Towards Bayesian CW Decoder
Post by: KB4QAA on January 28, 2013, 01:25:39 PM
Bayesian CW?  No thanks, I use Tylenol.  :)

Way over my head, but I enjoyed reading about this fascinating area of statistic/mathematics.  Thanks.  bill


Title: RE: Towards Bayesian CW Decoder
Post by: AG1LE on February 04, 2013, 09:55:23 PM
How about using the database of known call signs to improve the ability to decode call signs?

I think this is very good idea and in fact should also be doable. 
There are well known algorithms for matching received word against a database of known words.   
I believe there is also a database of known contest stations that some of the logging programs are using.

I am still working on the lower level of the stack. Right now my focus has been to improve the classifier part.
This part of software looks at received "dit" "dah"  timing, translates them into symbols and runs them through a classfier. 
See my latest post in here http://ag1le.blogspot.com/2013/02/probabilistic-neural-network-classifier.html (http://ag1le.blogspot.com/2013/02/probabilistic-neural-network-classifier.html)

73
Mauri AG1LE


Title: RE: Towards Bayesian CW Decoder
Post by: AG1LE on February 04, 2013, 10:07:49 PM
Mauri,

The Bayesian page is great, not often you visit a ham site where you can learn something.

Tim  N9RO


Hi Tim

I appreciate the feedback. I found that writing is a good way to refresh my memories and learn new topics.
Writing software actually forces you to build better insight in these topics.   

I did rewrite this particular Bayesian decoder section three times before I understood what is really going on.
And now I scrapped the whole section to use  Probabilistic Neural Network algorithm http://ag1le.blogspot.com/2013/02/probabilistic-neural-network-classifier.html (http://ag1le.blogspot.com/2013/02/probabilistic-neural-network-classifier.html) instead.
That was introduced by D.F. Specht in the early 1990s and has some great advantages compared to my initial attempt.

73
Mauri AG1LE



Title: RE: Towards Bayesian CW Decoder
Post by: AG1LE on February 04, 2013, 10:10:30 PM
Bayesian CW?  No thanks, I use Tylenol.  :)

Way over my head, but I enjoyed reading about this fascinating area of statistic/mathematics.  Thanks.  bill

Hi Bill
Way over my head too,  but I keep trying.  Sometimes with Tylenol :)
Eventually something will stick and I can make good use of it.

73
Mauri AG1LE



Title: RE: Towards Bayesian CW Decoder
Post by: ZENKI on February 06, 2013, 03:09:03 AM
Brainsian is best :)