----- Original Message -----
Sent: Monday, August 27, 2012 10:22
PM
Subject: LF: WD2XES - Opera detected and
identified by correlation
Recently I have been pondering wether it is
possible to detect QRSS or Opera transmissions by signal correlation against a
known waveform. This should far be more sensitive than the conventional
incoherent decoding process. John's stable and phase-coherent Opera
transmission last night's provided a welcome opportunity to test this
scheme.
Using SndInput from DL4YHF, I recorded a long IQ
file at 2 samples / second, ie. 2 Hz wide centered on 137561 Hz. The audio was
taken straight from the Rubidium-locked receiver, with no noise blanking
inserted. The data was then postprocessed using a MathCad spreadsheet. Some
results can be viewed in
http://df6nm.bplaced.net/opera/xes/
,
along with TA grabber screenshots showing both XGJ and a weak trace from
XES in 21 mHz FFT (testTA.jpg).
First a high resolution spectrogram was
generated, at 1.9 mHz per bin (xes_spectrogram.png). The central carrier
component of the transmission can clearly be seen. Best SNR occured between
3:30 and 4:30 UT (as indicated by marker ticks), when the peak was 9.0 dB
above the noise (xes_spectrum.png). Scaling noise bandwidth from 2.9 mHz to
2.5 kHz (-49.5 dB), and adding 6 dB for 50% duty cycle, we get a peak-power
SNR of -44.5 dB. This corresponds to -48.5 dB on the Opera SNR scale, about 9
dB below the current decoding threshold for Op-32.
The peak appeared about 0.2 Hz off-center because
the 12 kHz samplerate had not been not calibrated. Once the peak frequency is
accurately identified, the received signal can be correlated against a
"prototype" waveform, which contains the Opera sequence for WD2XES, 16-fold
oversampled. The correlation is efficiently implemented as a multiplication in
Fourier space.
The result (xes_correlation_wd2xes.png) shows
four distinct peaks in time domain at 2:15, 2:48, 3:21 and 3:54 UT, which
should correspond to the a-priori unknown start times of John's Opera
sequences. The repetition period was apparently 32.92 minutes. As the DC
component in the reference waveform had not been removed, the peaks are riding
on a pedestal caused by the self-correlation of the carrier
component.
To check the ability to identify an unknown
station, the correlation to a different callsign was also plotted (WD2XGJ just
as an arbitrary example, see xes_correlation_wrongcode). In spite of the weak
cross-correlation peaks, we find that a correct selection from a list of
potential candidates would certainly be feasible.
It would not be too difficult to automate this
process and create an "Opera deep search" software, which should be able to
detect and identify signals reliably down to about 12 dB below the threshold
of the current Opera decoder. This means at same sensitivity we could go
16 times faster!
To reap this benefit, the following prerequisits
need to be fulfilled:
- As we need a carrier component, transmit keying has
to be phase coherent. Thus simple keying schemes which interrupt the
oscillator or divider would not work.
- as we look at phase-sensitive
integration over the whole sequence rather than a single symbol duration, the
frequency stability has to be much tighter.
- As there is no bit-wise
decoding involved, we will need to supply a list of potential candidate
callsigns (similar to deep-search in K1JT EME modes).
Thanks again to John and Warren for the
signals!
Best 73,
Markus
(DF6NM in
JN59NJ)
