Hi Paul,

 

thanks for the explanation! So the decoder does not only consider the symbol phase but also the magnitude? In case of time-variable noise (e.g. changing QRN), you will then want to flatten out the noise level before decoding (i.e. reduce the gain during high noise background). To do that you may need to look not only at the symbol filter outputs but also evaluate out-of-band energy.

 

Strange that we are not getting the same result from Stefan's transmission. In case you'd like to take a look at my wav, I have uploaded it to
http://df6nm.bplaced.net/VLF/fec_tests/join_160427_2200.wav (128 kB)

 

All the best,
Markus

 

 

-----Ursprüngliche Mitteilung-----
Von: Paul Nicholson <[email protected]>
An: rsgb_lf_group <[email protected]>
Verschickt: Do, 28 Apr 2016 3:09 pm
Betreff: Re: VLF: EbNaut transmission on 8270.1 Hz again - 73


> I noticed that in ebnaut-rx v0.4d the output with
> best rank seems to persist, rather than the one
> with fewest symbol errors

It actually keeps the decode with the best path
metric. The other options (best rank, best BER)
don't perform so well on average. The log file
doesn't show the path metric though, I'll fix that
for the next revision.

With the Viterbi decoder using soft decisions with
Euclidean distance metric, the path metric is equivalent
to the cross-correlation of the signal and the message
symbols. The decoder is effectively doing a 'deep search'
which tests all possible messages, not just a short list.

I can't convince my decoder to produce anything other
than 'V3' on this one. Hamming distance between 73 and
V3 is 180. The differences are pretty evenly spread
through the messages.
--
Paul Nicholson
--