I have exchanged a couple of emails with Joe K1JT about the QRPp experiment and the -24 dB limit for the FT8 decoder. The following are Joe's remarks.
###########################################################################
By convention all SNRs reported by the software package WSJT-X are
measured in a standard reference noise bandwidth of 2500 Hz.
Detailed measurements of FT8 sensitivity in various simulated
propagation conditions are summarized in a graph posted on the WSJT web site here:
http://physics.princeton.edu/p…T8_Decode_Probability.png
This plot was made by generating 1000 simulated FT8 signals at each
specified SNR, subjecting them to simulated propagation by nine standard
ITU propagation models, and counting the number of resulting decodes for
each case.
Propagation to the geostationary satellite QO-100 ahould closely
approximate the AWGN (additive white Gaussian noise) path, for which the
leftmost curve in the plot shows a 50% probability of decoding success
at SNR = -20.8 dB and a very small (less than 1%) probability at SNR =
-23 dB. The probability of decoding an FT8 signal with true SNR less
than -22 dB is extremely small.
Why is it, then, that the decoder sometimes reports SNRs as low -24 dB?
And why do we never see reported SNR -25 dB or lower?
All physical measurements have associated uncertainties. In principle
we could estimate and report an uncertainty along with each measured
value -- for example, something like SNR = -18 ± 2 dB. If WSJT-X were
intended as a measurement tool, we would do it this way; but for its
intended purpose such reports would be cumbersome, distracting, and
irrelevant.
Reported SNRs lower than -22 dB are nearly always underestimates of the
true value. To avoid unwarranted focus of attention on very small
reported SNRs, we arbitrarily clamp reported FT8 SNRs at -24 dB.
###########################################################################
