Uwe wrote:
> message follows at
> 2016-02-1700 02:30:00
> same settings
No problem, decoded with Eb/N0 = +6.7, a bit weaker
thanks to the nighttime noise but would still do 10
second symbols.
Stefan wrote:
> Some decodes appeared but with 2 characters it could
> be wrong.
These are false decodes, just random noise which happens
to have a correct CRC. The default list length of 20000
is too long for 2 chars, probably 1000 is enough.
I'm still working on the difficult problem of finding the
optimum list size and CRC strength for each message length.
As you use longer list length, the chance of a decode increases
(with rapidly diminishing benefit) but the chance of a false
decode increases (roughly linearly). A false decode is
more serious than a failed decode because there's no way to
know if 'GT' or 'NI' was the correct message! At least with
no decode, you know where you stand.
So to make use of longer lists, it is necessary to send with
stronger CRC, but the extra CRC bits is more inner code overhead
which reduces the chance of a decode (for a fixed signal
strength and duration). If CRC and list size are both
increased by the right amounts, the decode chance overall
increases, but only up to a limit corresponding to around
3dB coding gain from the outer code.
The best settings could be determined by a lot of brute-force
trials but the permutations to be tested are many and the
computing time would be huge, so I try to reduce the number
of trials by some performance calculations.
For now, stay with 16 bit CRC and use list length 1000
or less for 2 chars.
--
Paul Nicholson
--
|
