Hi Joe, Some quick comments in-line. 73, Reino OH3mA
> Sent: Sunday, December 10, 2023 4:56 AM > Subject: [wsjt-devel] WSJT noise estimates > Been wondering how WSJT-X generates the noise power estimate it uses to calculate SNR for each FT8 signal. Does it simply collect all the signals and noise over the bandwidth selected on the waterfall and call that the noise power level or does it take a quick snapshot of the background noise level during the brief quiet period at the end of each 15 second FT8 sequence? Or is it more complicated than that? 3mA: The S/N calculation process is pretty complicated. It estimates noise over the whole receiver bandwidth and uses 2500 Hz as the reference bandwidth. Some other implementations try to calculate S/N over the individual signal bandwidth, 50Hz, and the values are not directly comparable. > I am plagued with a S2 -S3 noise level on 6 meters nearly all the time that if not AWGN is pretty close to it. 10 meters is even worse. The DSP noise blanker in my TS590 will reduce it slightly. I estimate this is degrading my ability to decode FT8 signals on 6 by nearly 20 dB compared to the noise level generated by a 50 ohm resistor. I dont use an LNA ahead of the radio would be pointless. I dont use the noise reduction feature in the radio either as it tends to lose very weak signals completely. 3mA: Noise blanker may help only, if the noise have high and short peaks and the noise is cancelled e.g. limited at a wide bandwidth point of the receiver chain. On AWGN it does not help. Then the resulting distortion noise is distributed over a wide frequency range. Your assumption about the degradation may be a bit pessimistic, depending on the noise type, S2 - S3 is quite low compared what is typical on lower HF. > Wondering if I can use the DSP in my TS590 to narrow the receiver bandwidth to perhaps 300 -500 Hz around a known offset to help pick weak signals out of the noise? I realize that the WSJT program filters the audio into much narrower BW bins so all the receiver filtering can do is reduce the receiver gain reduction caused by the noise pumping up the AGC but that might be beneficial. 3mA: The additional receiver bandwidth reduction improves the calculated S/N values as there is less noise. But the decoding sensitivity does not usually improve as long as the linearity of the receiver chain is kept. There is some S/N calculation oddities at the edges of a narrow bandwidth. I don't know the actual reason, but I have recorded up to 40 dB additional variations of S/N values in a comparison of the same messages between narrow and wide bandwidths. > Likewise, would using the DSP notch to suppress a single strong local signal or birdie help since strong signals also reduce receiver gain? 3mA: That seems to help. Especially, when the disturbing signal is not very close to the wanted signal. > Should I deselect the flatness option if I use these tools? 3mA: To my knowledge the flatness option is only for waterfall display, not for the decoding process. Would narrowing the waterfall span help any since the program ignores anything outside that span? 3mA: That is about equivalent as usage of a narrow filter. If signals outside of the waterfall are not causing distortion noise (overdriving receiver), then there will be an apparent S/N improvement, but not a real decoding improvement. There is another possible decoding improvement as there are less signal candidates to process, Decoding is only attempted on frequencies inside the waterfall. Well at least in principle, I have seem rear instances, where also signal outside the waterfall has been decoded. > Would appreciate any insight you can share. Joe W0FY PS. 3mA i.e. is 3 milliamps, my local nickname. _______________________________________________ wsjt-devel mailing list wsjt-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/wsjt-devel
