Hi Sam, Thanks Sam, I am aware of that problem and for sure it affects to the "absolute" S/N values. There is another issue due to the limited dynamic range and that all S/N values below -24 dB are reported as -24 dB. I have partially tried to avoid the issue by only taking into calculation messages that are received by both receives. I also checked how much those measurements with one of the values reported as -24 dB (less than 1 % of all values) really affect to the calculated average and deviation values and there is a minor shift at one or two higher attenuation point, when I leave those out.
73, Reino OH3mA -----Original Message----- From: Sam W2JDB via wsjt-devel [mailto:wsjt-devel@lists.sourceforge.net] Sent: maanantai 10. huhtikuuta 2023 14.38 To: wsjt-devel@lists.sourceforge.net Cc: Sam W2JDB <w2...@aol.com> Subject: Re: [wsjt-devel] Truncated Waterfall Display "Feature" Hi Reino, The one thing to be aware of when analyzing SNR is that since you can not account for the signals that you did not decode, those that are below the threshold, the mean will always be biased towards the positive side and those missing decodes invariable would have affected the variance and standard deviation values. 73, Sam W2JDB -----Original Message----- From: Reino Talarmo via wsjt-devel <wsjt-devel@lists.sourceforge.net> To: 'WSJT software development' <wsjt-devel@lists.sourceforge.net> Cc: Reino Talarmo <reino.tala...@kolumbus.fi> Sent: Mon, Apr 10, 2023 3:33 am Subject: Re: [wsjt-devel] Truncated Waterfall Display "Feature" Hi, Just as promised, I looked what happens, when I set waterfall from 200 Hz to 1500 Hz (just bit more). The measurement was a side result of a study how a narrower radio filter affects to the S/N values. I have two similar receivers and I feed the same antenna signal via a power divider to those receivers. I calculate using ALL.txt files differences of the S/N values for the same (common) messages. That removes transmit power and propagation from the equation. For the filtering study I calculate in 25 Hz slots the number of messages and an average and standard deviation of S/N differences. By this method the filter passband and out of band attenuation on quite wide range up to 60 dB on longer measurement times. The typical standard deviation is about 3 dB. Quite good for a S/N calculation that is not designed at all for this purpose. Also linearity seems to be better that in some earlier versions. Now the result: there is not a single decoding in 35 361 decoded messages above 1500 Hz. A related observation close to the waterfall edge. The standard deviation increases from 3.8 dB to 7.7 dB in the last 25 Hz slot. The increase starts at 1425 Hz slot. There is also 4.6 dB gain increase over that range. So there is some S/N calculation anomaly close to the band edge and unnatural high S/N values are seen close to band edge. Of course there is also the natural calculated S/N increase over the narrow passband due to missing noise outside the passband. That easily results to misinterpretation that a narrow RX bandwidth will enhance decoding probability, it is just the actual "wrong" noise bandwidth in the S/N calculation! 73, Reino OH3mA >From: Reino Talarmo [mailto:reino.tala...@kolumbus.fi] Sent: lauantai 8. huhtikuuta 2023 10.46 >Hi All, >You may discuss different issues! In FT8 mode Sam is correct as the signals >are reached only on the frequencies displayed on the waterfall. In JT65 and >Q65 the situation is different at there is a spinner called F Tol that defines >the search range independent of the waterfall range. All that is written in >the User Guide. If something else is happening, then it is extra for the >operator benefit! I’ll do a FT8 test in my next 24 h recording. 73, Reino OH3mA _______________________________________________ wsjt-devel mailing list wsjt-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/wsjt-devel _______________________________________________ wsjt-devel mailing list wsjt-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/wsjt-devel