VO1NA 2015-12-04/05 4K19A 0.2S 131C QRO
start Eb/N0 BER rank phase xcor
22:00 7.2 5.7% 0 87.0 0.44
22:30 7.0 6.2% 0 75.6 0.38
23:00 4.6 12.1% 0 56.3 0.28
23:30 no decode 78.7 0.22
00:00 9.0 2.6% 0 69.8 0.46
00:30 no decode 51.3 0.12
01:00 no decode 37.1 0.13
01:30 no decode 0.03
02:00 no decode 0.05
02:30 no decode 128.9 0.09
03:00 1.4 20.8% 6 124.2 0.20
03:30 5.5 9.8% 0 73.0 0.24
04:00 4.9 11.4% 0 35.6 0.27
04:30 3.5 15.2% 0 80.6 0.18
05:00 5.3 10.2% 0 139.5 0.23
05:30 no further decodes 36.8 0.08
06:00 55.9 0.11
06:30 143.7 0.10
07:00 51.4 0.19
07:30 0.04
08:00 71.4 0.10
08:30 0.03
09:00 0.06
09:30 0.05
10:00 0.03
10:30 0.04
11:00 0.03
xcor is the correlation coefficient of the entire message.
phase is measured in bandwidth 1.524mHz (= 1/message_duration).
I have a suspicion from the above data that the decoder is
not working properly on these long messages. It is possible
that the accumulated path metrics are running out of numeric
precision after traversing a trellis of 821 stages. This I
should investigate before any further over-the-air trials
with such long codewords.
Correlation of the received signal with the encoded message
produces nice correlation spikes. For example the correlation
spectrum of the 03:00 message looks like
http://abelian.org/vlf/tmp/151205a.gif
and zooming in on the spike
http://abelian.org/vlf/tmp/151205b.gif
reveals the correct triangular shape as the signal is brought
into alignment with the template. The offset here is about
+20mS and roughly half of that is propagation delay.
With such a large number of symbols (3280), even a message
which doesn't decode produces a clear correlation and very
low auto-correlation sidebands.
With sufficiently accurate tx bit clock, correlation provides
a good way to measure the path length. With a strong enough
signal and a much shorter symbol period it might be possible
to distinguish multiple paths. Unfortunately the width of
the LF band limits what can be done there.
--
Paul Nicholson
--
|
