Hi Bill and all,
Thanks for the CRUNCHed file. I took a close look at it, to see
how well a machine might decode it.
First, I found the spectral peak at about 799.875 Hz. This is
amazing Tx and Rx frequency accuracy. Dividing by the 40x speedup
factor, the error at LF is only three millihertz!!!
Demodulating at that frequency, I then searched for the bit timing.
I assumed that elements were precisely three seconds long (0.075
after CRUNCH), and just looked for the phase. A CW signal has
a discrete spectral line at one half the element rate, because the
key is down far more at odd element times than at even. For
example, "LEK" is key down for 14 elements; 11 of those are in the
odd intervals. So I just shifted a window of alternate elements
until the signal was maximum.
Using the phase obtained above, a rectangular matched filter was
then applied, obtaining this result:
Amplitude is arbitrary linear "voltage" units; phase is in degrees
(and not meaningful when the key is up).
You can see that "LE" could be easily decoded, but noise towards
the end of "K" would result in a bit error, regardless of the
threshold chosen. Coherent detection did not help in this case -
I suspect that the path caused too much phase jitter. However, a
sophisticated program which knew the rules of Morse might be able
to correct the error.
So, IMO, it will be a while before robot QRSS decoders can compete
with the human brain. But, if we all use the same precise element
timing, it should be possible to have an automated *detector* of
QRSS signals, which could trigger a screen capture or audible
Bill, if you have the audio file before CRUNCHing, could you please
post it to your site or email it? I have been playing with some
noise reduction schemes, and would like to attempt an S/N
improvement on this signal.
----- Original Message -----
From: "Bill de Carle" <[email protected]>
To: <[email protected]>; <[email protected]>
Sent: Wednesday, December 06, 2000 10:23 PM
Subject: [Lowfer] LEK's QRSS-CW signal copied in Canada using CRUNCH
If you go to my web page at:
down at the bottom of the page you'll find a link you can use to
download a file called QRSS-LEK.ZIP - it has a few seconds of WAV
file snipped from last night's overnight CRUNCH run trying to hear
Although I've copied Lyle before using BPSK, I believe this is the first
time I copied him on CW; the signal came through in the wee hours of the
morning, Wednesday Dec. 6th, 2000. I have Lyle's co-ordinates in
Aitkin MN as 46.53013N, 93.71042W and my QTH in St-Adolphe d'Howard, PQ
Canada is at 45.9351N, 74.2993W. Just re-calculated the distance using
Clarke's ellipsoid: it works out to 1495 Km or 929 miles.
Antenna was a 30 x 20 foot rectangular tuned loop. Rx was a Kenwood
TS850S with external freq standard. Computer was a Compaq laptop
with a 100 Mhz Pentium processor running CRUNCH V3.0. An S16 PCMCIA
soundcard was used for 16-bit audio input. CRUNCH settings were:
40:1 time compression, 800-Hz CW tone out, 300-Hz bandpass filter on
the output. If you have CoolEdit handy you can do a frequency analysis
of this file (select the entire snippet and press the SCAN button) -
it shows Lyle's signal on 800-Hz jutting up just a little over the
smoothed background noise envelope.
If you listen carefully you can just hear Lyle's CW ident in there -
starting about halfway through the audio snippet, you'll hear "LEK"
very distinctly. The signal is weak so you may have to play it over
a few times in LOOP mode before you get it, but it's definitely there.
This is a good example of a signal that seems to come up out of the
noise when you listen to it over and over. It's the equivalent of
the "GRAB" process but done in your head. Pretty neat!
To unsubscribe, send to [email protected] "unsubscribe lowfer" (Do not
send to list!!) Send on list submissions to [email protected]