Markus wrote:
> Running the decoder with 61 rather than 60 s offset happened to
> reduce the number of symbol errors from 388 (0.4 dB) to 369
> (1.3 dB Eb/N0).
Thanks Markus, perhaps there's a sample offset error
in the pipeline somewhere. I reproduced the same result with
the Linux decoder too: needs T offset +1 to get 1.2 dB when
using the 1s/s signal.
> when using the SpecLab FFT-export method with very slow
> samplerates, I noticed that I had to add about four samples
> worth to the time offset for best decodes
An out-by-one error is an easy mistake for a novice programmer
to make. An out-by-four error requires much more skill and
generally needs an experienced programmer.
I'll investigate.
If anyone's interested, following are the commands I use to
receive Uwe's signal.
The first pipeline extracts the signal from a raw store which
has been filled by vtwrite, and mixes it down to baseband I/Q
centered at the signal frequency,
vtread -T2016-03-14_03:00:00,+11900 /raw | # Extract 11900 seconds
vtmix -c 0.866,0.500,0.45/-105.71 | # Steer loops and add E-field
vtfilter -h bp,f=8270,w=3000 | # Filter 3kHz bandwidth
vtblank -a1.2 -d0 -t1 | # Remove sferics
vtmult -f 8270 | # Mix with LO at 8270.00000 Hz
vtresample -r240 > dj8wx.160314a.vt # Resample to a low rate
The magic numbers given to vtmix are the loop coefficients for
steering, followed by the coefficient/phase_shift of the E-field.
Then another pipeline does the decode
vtraw -oa dj8wx.160314a.vt | # Convert to ASCII data
ebnaut -dvp8K19A -N19 -S10 -r240 -PS -c4 -L20000 # Decode
--
Paul Nicholson
--
|