Маркус тут нам ответил:
In principle, for constant TX power and noise level, SNR reports (dB in 2.5 kHz) should be the same for FW and WSPR, and at all speeds, only the decoding threshold will be lower for slower modes. Someone mentioned that the SNR calibraton had been reviewed in the FST4 development process, and that SNR reports are now 0.5 to 1 dB lower than they were with WSPR.
However here are other differences. Frequency stability and timing accuracy are much more critical for FW. While WSPR is able to track (and report) several Hz of frequency drift, FW cannot cope with that. Also the clock deviation should be within one or two seconds for FW, versus more than 5 seconds for WSPR. I speculate that the tolerances have been sacrificed to save some of the increased CPU load for the new decoding algorithm.
Another topic is the effect of audio or IF bandfilters: The noise level in WSPR is apparently not calculated by simply summing the received power, but by evaluating the quartile from a histogram. This method is also used in Argo and SpecLab. It has the advantage of being robust against "outliers" from strong in-band carriers or short static bursts. When the bandwidth is cut below 210 Hz, more than a quarter of the bins have zero noise, and the SNR will be greatly exaggerated. In Jan 2022, I experimented with different SpecLab bandfilter settings and observed changes of the reported SNR (see screenshots). This effect also happens with FW2, but the transition seems to be softer and the "enhancement" is less dramatic. Thus a bandfilter could produce "better" (but fake) reports for WSPR than FW.
I read that the new modes employ a partly coherent decoding process, where the phase is expected to be constant over some part of the transmission. So FW might perhaps more easily be affected by soundcard dropouts on the TX or RX side. I do not know whether noise blanker settings might have a different effect.
As far as I know, FW modes have been designed by the same team around K1JT, who designed and improved WSPR over the years. If you want, you could probably examine the source codes on Github, but I have never looked into that.
Best 73,
Markus
Вообщем, вопросы к дядьке Джо и его команде надо задавать по поводу алгоритма вычисления того самого с/ш.
Все остальное - ровно, что мы уже обсудили, да, есть разница между ФСТ и ВСПР в 1- пол-дБ в рапортах "просто так", да возможно ЩСБ в эфире.
Я все же думаю, что есть ЩСБ сопоставимые с 5 минутным циклом и они влияют.
Это можно доказать наблюдая реальные сигналы, видимые в Арго или СПЛ и замеряя в течении цикла передачи их уровень.
У меня тут такое не возможно, но методу я описывал уже выше.
Такой эксперимент сразу расставит все точки над и.
Ну, а дядьке Джо надо писать собравшись с духом и сформулировав всё четко, включая ваши уже проведенные эксперименты.
