[CQ-Contest] Skimmer accuracy...

Wed Feb 20 14:04:59 EST 2013

Sure, glad to help.   It would be interesting to see what can be done with
today's computer power.

The project I mentioned was done in a university lab environment.  We
(staff) created the hardware/software platform, and then a stream of
students could build various new modules that plugged in to add new
"opinions".  It was good for term projects, theses, etc.

One class of modules was "identity" functions - software that tried to
identify (i.e., "spot") the operator and/or station  that was sending each
particular Morse signal.  Various modules used all sorts of measurements to
"fingerprint" a signal - things like dot/dash ratio, shape of keying
envelope, hum/chirp characteristics, spectrum shape, presence of artifacts
like aurora flutter, etc.  Anything we could think of that a human op might
use to identify an op on the band was fair game to capture in software
form.   Once an identity (like a callsign) was determined, it was  saved
with the fingerprint for future use.

With such modules in place, the software could identify a station/operator
even though they never sent a callsign at all.   It was fun to tell the
machine to "go find K3FIV on 15 meters" and watch it methodically scan the
band, pausing briefly to "listen" to each signal in turn and compare in its
fingerprint database to the K3FIV fingerprint, and then report where K3FIV
was, and how confident it was of that opinion.

A lot of this just might also be applicable to voice now, since computers
are so much more capable.   SSB Skimmer anyone...?

If Alex is interested in this ancient history, we should probably take it
offline and I can describe what I remember.

73,
/Jack
On Feb 20, 2013 8:37 AM, "Pete Smith N4ZR" <n4zr at contesting.com> wrote:

>  Jack, I think you could, if you're willing, have a very fruitful
> conversation with Alex, VE3NEA.  I know a little bit about the techniques
> he's using to assess the plausibility of a given prefix and suffix, but
> you have "been there and done that" , and I think it would be a great
> contribution.
>
> 73, Pete N4ZR
> Check out the Reverse Beacon Network at http://reversebeacon.net,
> blog at reversebeacon.blogspot.com.
> For spots, please go to your favorite
> ARC V6 or VE7CC DX cluster node.
>
> On 2/19/2013 2:52 PM, Jack Haverty. wrote:
>
> In the spirit of improving accuracy....a suggestion
>
> In the late 1970s, I was involved in a project to use computers to
> understand Morse code.   Google "haverty understanding Morse code" if
> you're curious.
>
> We struggled with the same kind of inaccuracies that Skimmers are now
> experiencing.   We used many independent techniques to generate plausible
> decodings from what came off the antenna plus lots of ancillary
> information, like dictionaries, and "common sense" like expected
> propagation behavior and current band conditions.
>
> The key technique in making the final decoding very accurate was that each
> module that generated an "opinion" about what was being decoded also
> attached a "confidence" to that opinion.   So, a suggested decoding from an
> S9+30 signal with little QRM/QRN/QSB would have a higher confidence than a
> S2 signal in the middle of a crowd with summer static conditions.   A
> decoder might look at a few seconds of signal and suggest that the sender
> said "E5EEE", but if it was a strong signal it would have a very high
> confidence level and if a weak one very low.
>
> That "confidence" factor enabled other modules further down the line to
> make more intelligent decisions about the most likely decoding.   For
> example, a subsequent "callsign module" might look at the decodings
> suggested by the earlier "dot/dash decoders", and conclude that "E5EEE" was
> not very likely, especially if "E5EEE" had been decoded with low
> confidence.  It would offer its own opinion about what the signal contained
> -- for example that it was just noise.
>
> If the Skimmers could produce spots that contained not only a callsign but
> also a "confidence", consumers further down the line (e.g., contest
> programs displaying spots) could filter those spots based on confidence -
> e.g., only display spots that are high confidence.
>
> Perhaps a spotting network convention like <callsign>.<confidence> -- so
> spots would come out across the spotting network looking like "E5EEE.2" or
> "EK3LR.4", or "K3LR.9" or something similar.    Logging software receiving
> such spots could apply their own heuristics, e.g., by consulting a list of
> known contest callsigns, and decide whether or not to display the spot to
> the operator.
>
> 73,
> /Jack de K3FIV   (not VK3FI as I am sometimes labelled....)
>  On Feb 19, 2013 9:46 AM, "Shane Mattson-->K1ZR" <k1zr at comcast.net> <k1zr at comcast.net>
> wrote:
>
>
>  I'm starting a new thread for increased visibility to this issue.
>
>
>
> This is by no means a skimmer bashing post, so please don't go there..
>
>
>
> I'm just curious if others have observed the following:
>
>
>
> Just for kicks I performed a RBN spot search for my call and the following
> spots came up during a period of time that I was QRT.  I did not operate at
> all on Sunday.
>
>
>
> de    dx    freq  cq/dx snr   speed time
>
> K3LR  K1ZR 28005.3     CW CQ 13 dB 32 wpm      2042z 17 Feb
>
> K3LR  K1ZR 14046.6     CW CQ 24 dB 31 wpm      2014z 17 Feb
>
> KM3T  K1ZR 14046.6     CW CQ 15 dB 33 wpm      2011z 17 Feb
>
> VE2WU K1ZR 14046.6     CW CQ 20 dB 31 wpm      1927z 17 Feb
>
> K8ND  K1ZR 21006.9     CW CQ 8 dB  32 wpm      1743z 17 Feb
>
> NY3A  K1ZR 28062.4     CW CQ 28 dB 31 wpm      1654z 17 Feb
>
> K2DB  K1ZR 7020.5      CW CQ 10 dB 29 wpm      0855z 17 Feb
>
> K2DB  K1ZR 7020.4      CW CQ 11 dB 30 wpm      0753z 17 Feb
>
> K3LR  K1ZR 3510.3      CW CQ 26 dB 29 wpm      0207z 17 Feb
>
>
>
> Perhaps the skimmers misinterpreted a like-call sign?  I'm not excited
> about
> seeing my call detected and spotted during a period for which I was QRT.
>  An
> issue like this emphasizes the importance of double checking the call sign
> for accuracy before logging a point and shoot qso.
>
>
>
> Could someone within the RBN domain provide some insight as to how to
> address an issue like this?
>
>
>
> -Shane
>
>
>
> _______________________________________________
> CQ-Contest mailing listCQ-Contest at contesting.comhttp://lists.contesting.com/mailman/listinfo/cq-contest
>
>  _______________________________________________
> CQ-Contest mailing listCQ-Contest at contesting.comhttp://lists.contesting.com/mailman/listinfo/cq-contest
>
>
>