Paul and Markus, thanks for this excellent work and welcome contribution to state of art. Great test design, analysis and hardware implementation, from FEC to auto-tuner. Are you also approaching the Chu limit? I’m guessing the maximum effective dimension of Marcus’ antenna as (5m -20m?), and with auto-tuning considered, estimating instantaneous antenna bandwidth in the neighborhood of Paul’s bin width, indirectly suggesting performance in the general neighborhood of Chu limit. Do you remember what value was used for the antenna length in the ERP estimate? If the guesstimate that your instantaneous bandwidth (bandwidth for each symbol) is near the Chu limit is correct, even more credit to the auto-tuner. Also noting indications of impressive calibration and accuracy; for example: comparing the parameters from the 1-hour CW transmission yesterday (including SNR by inspection from yesterday’s spectrogram, 20-minute and 1-hour CW intervals) to parameters from the December 13 DK7FC – DF6NM 20uW ERP, 175km 4-hour (64uHz) 21.7dB link, and normalizing for bandwidth, ERP and distance (~2-3 dB/Mm) shows an interesting correlation. D-layer variation is a substantial unknown in this comparison, but the fact of an interesting numerical correlation across equipment and months, using small antennas and long integration times, noise cancellation and other attributes that challenge repeatability and accuracy, is impressive. Well done! 73, Jim AA5BW Thanks Paul, for devising this experiment in the first place, and encouraging me to take part in it. It looks like we are now getting real close to the ultimate Shannon limit, making best use of your superb receive capabilities. The 8270 Hz carrier comes from a Rubidium-referenced decadic synthesizer, with frequency calibrated to Loran-C to within about 1e-10. For the test, an encoded bit sequence for the five-character message was obtained beforehand from Paul's script on his website. SpecLab's digimode terminal (configured similar as previously for Opera, and initiated by "scheduled actions") is used to generate a keyed sidetone, with bit timing controlled by NTP to within a second. The tone is being rectified at the soundcard output and drives an ancient and sensitive telephone relay, commuting the carrier between two complimentary outputs of a small pot core transformer. The phase-keyed signal is then amplified by the usual "Nitro" audio PA to about 60 Watts (28 V rms). A large ferrite transformer (in the brown chocolate box) bought the voltage up to 200 V rms, feeding 0.3 A antenna current to the venerable seven-bucket coil which is sitting in the blue rubber bins outside the window. To keep the antenna on resonance, the phase detector from my LF automatic tuner (basically an XOR comparing the phase of voltage and current, with sensors modified for the lower frequency) was inserted between the amp and the transformer. It drives a little DC motor, pulling on a nylon string, which then rotates a 10x10x0.5 cm^3 ferrite plate inside the coil. This effectively mitigates current reductions and phase variations caused by wind, temperature and humidity. During operation, one can hear the whistle of the coil and transformers, occasionally interrupted by a short "krk" sound when the relay is bouncing during phase changes at full or half minute. Sent: Sunday, May 11, 2014 8:36 AM Subject: VLF: Coherent BPSK at 8270 from DF6NM I am pleased to report reception of two test transmissions from
DF6NM on 8270.000 Hz which took place on Saturday 2014-05-10
morning.
Markus was sending coherent BPSK with UT synchronous symbols
using a symbol period of 30 seconds. The FEC is a terminated rate
1/4 convolutional code with constraint length 21, cascaded
with an outer error detecting code using a 16 bit CRC.
Two transmissions of 46 bits were made, each lasting 132
minutes with a 20 minute carrier test in between. ERP was
probably around 5 or 10 uW and the range is 1028 km.
Eb/N0 was about -0.5dB in the first test and about -1.5 dB
in the second, which is below -7dB in the symbol bandwidth
of 33.3 mHz.
Both messages were decoded with some margin to spare.
The decoder is a list Viterbi decoder using the tree trellis
algorithm with a list length of 2000 and stack size 20000.
Both messages decoded at the top of the list so the list
decoding wasn't actually necessary for this strength of signal.
The transmitter uses a rubidium source and the receiver is
GPS timed. A reference phase at the receiver is obtained by
averaging the phase of the squared signal but at such low signal
strengths the resulting reference is unreliable and the decoder
makes a search for the correct phase and phase drift rate.
The signal is completely invisible in any spectrogram at the
receiver. In a spectrogram running at the symbol bandwidth
the signal is too far below noise and the bandwidth of the
transmission is such that, in a resolution capable of seeing
the signal above noise, the signal is too wide.
For example, this spectrogram uses the symbol bandwidth,
http://abelian.org/vlf/tmp/df6nm_140510a.png
Maybe someone can see some very faint line at 8270.000 ?
The first test ran 07:02 to 09:12 and the second
from 09:32 to 11:44. The 20 minute carrier in between is
also below noise at this resolution.
Following the second test, an hour of carrier is visible in
the 278 uHz spectrogram at
http://abelian.org/vlf/fbins.shtml#p=1399780800&b=110&s=sp
More information on the FEC codes, trials, and search for
good polynomials can be found at
http://abelian.org/fec
Stronger codes with constraint length up to 25 are available
and hopefully further tests will be made soon.
--
Paul Nicholson
-- |
|
