"The SNR in CW Skimmer and in SkimServ is computed the same way. The signal is extracted from the I/Q stream using a filter with a bandwidth of 50 Hz. First the key-on/key-off decision is made for every sample in the signal, as part of CW detection. The samples corresponding to the key-off state and on/off transitions are discarded, the key-on samples are analyzed. The signal strength is computed assuming the Rayleigh fading model (in other words, signal strength variations due to QSB are expected and taken into account). The noise density is estimated from the flat part of the power spectrum. The noise power in 500 Hz isthen computed by multiplying the power density by the bandwidth. Finally, theratio of the signal power to the 500-Hz noise power is computed."
One issue people have been wondering about is whether stations operating in the crowded lower part of the band might be "penalized" with lower SNR reports due to QRM, as compared with stations CQing higher up in the band. Alex disagrees. He says:
"I have performed a series of SNR accuracy tests on the simulated CW signals and verified that the SNR is determined correctly even at the station densities as high as 10 stations per kHz, and the average SNR of those stations of at least 60 dB." He goes on, "The SNR is computed using the history of the measurements. The signal strength is estimated in the 50-Hz filter passband, while the noise is estimated globally in the whole receiver bandwidth of 48/96/192 kHz."
I had earlier understood that SNR was calculated in a relatively short interval just before the spot was sent to the Telnet server. Turns out that the calculation is based on data collected in the 50-Hz decoder channel for about 45 seconds up to the moment when the spot is validated and forwarded to the Telnet server.
So, I hope this helps. The SNR numbers are better than I had thought, though still relative and dependent on each RBN contributor's antenna, receiver(s) and noise envirnment.
73, Pete N4ZR