Note: groups.io will be down for maintenance on Wednesday, October 5th, starting at 9AM Pacific Time (4PM Wednesday October 5, 2022 UTC), for approximately one hour.
On 12/07/2021 10:18, SA5NTK wrote:
Hello Everybody!Hi Notko,
the main variable in assessing SNR estimates is themeasured noise level, that is particularly challenging with modes that decode multiple signals from busy sub-bands. The noise estimation algorithms in WSJT-X reflect both the expected band occupancy and the evolution of noise estimating in the software. While SNR numbers are all intended to use the same signal power versus the noise power in a 2500 Hz bandwidth, the results are not going to be perfect. Other factors might include restricted receiver bandwidth, which should be avoided, many overlapping signals, and the fact that as the decoding sensitivity threshold is approached the error band of the SNR number approaches infinity. This last point is worth noting, often users make a big deal of a particularly low SNR number, whereas in fact it is most likely only reflecting this potentially large error band. Each mode enforces a clamp on the lowest SNR number reported, otherwise very large negative numbers that are theoretically impossible would be seen. The clamp value is chosen to be just below the theoretical maximum sensitivity for the mode.