Joe VO1NA wrote:
> Well done Markus and Paul.
Thanks. I am enjoying this foray into FEC and learning
something new. It is fun to push some boundaries.
When I first picked up a VLF signal from Markus (Oct 2010)
I would not have thought that a few years later we would be
able to convey quite reliably a message of 100 bits in less
than half the time that it took to average the spectrum in
http://abelian.org/vlf/df6nm/2010_10_09a.gif
(DF6NM is the signal at 8970 - 0.002 Hz)
A note from Markus suggested the bit clock on our last test
was about a second late. I shifted the recording and got a
better decode: Eb/N0 = -0.4 and rank 0 in the Viterbi list so
list decoding once again not necessary. Seems we can afford
to widen the channel further and try for more bits per hour.
Turned attention last week to making stronger codes. Found a
better way to construct low rate codes 1/(2n) from pairs of
higher rate 1/n. I measure the distance spectra of high rate
codes to identify weak (low weight) codewords then pair these
with polynomials which compensate those with higher weight.
I build 1/16 from 1/8 which in turn come from combining 1/4,
etc. Much faster. Previously about 800 CPU hours produced
one good 1/16 K=25 code with min distance 237. Now the same
hours turns out nine at distance 237 and one at 238 which is
the maximum possible for this rate and K.
Hoping for an on-air VLF test of 1/16 K=25 soon. First have
more trials to do. Use the best (d_min=238) code? Or use
another which has not so good d_min but possibly performs
better at very low S/N when using list decoding?
Perhaps other stations will be encouraged to try sending
coherent BPSK? At VLF this is easy to do. The challenge is
not to make new distances but to see how much information you
can get across per hour.
--
Paul Nicholson
--
|
