A continuación te dejo un documento escrito por Joe Taylor, K1JT, y Steve Franke, K9AN, donde comparan las sensibilidades medidas de FT8 SuperFox y FT8 estándar (incluida la transmisión multi-stream).
The experimental SuperFox operating mode in WSJT-X 2.7.0 has now been thoroughly
exercised by DXpeditions N5J (Jarvis Island) and CY9S (St. Paul Island). According to
Club Log statistics, each of these major efforts made over a hundred thousand QSOs,
of which about half (46% at N5J, 50% at CY9C) used FT8. When propagation was
adequate to yield SNRs of –14 dB or better, each group used the SuperFox protocol
most of the time and often achieved QSO rates more than 200 per hour. After pileups
had thinned out, and in order to work weaker stations, operators sometimes switched to
standard Fox-and-Hound mode with one or two FT8 streams.
Chatter on several WSJT-related forums shows that many Hound users are confused
about how to compare the weak-signal performance of SuperFox and multi-stream FT8.
As part of our development procedure we made exhaustive measurements of decoding
probability for both protocols, using simulations that cover a wide range of signal
strengths and propagation types. The following plot summarizes performance of
SuperFox and single-stream FT8 on two simulated channels: additive white Gaussian
noise (AWGN) and the ITU standard “Mid-latitude Moderate” (MM) channel, which
assumes Doppler spread 0.5 Hz and differential delay 1 ms. In practice, on both
channels standard FT8 can decode signals 3 to 4 dB weaker than SuperFox.
Transmitting FT8 with N=2, 3, 4, or 5 simultaneous streams necessarily attenuates each
stream by respective amounts 6, 9.5, 12, or 14 dB. Thus, for example, circumstances
that yield a SuperFox signal with SNR = –14 dB would yield two FT8 streams at –20 dB
each. The graph shows that decoding probability on the MM channel is greater than
90% for SuperFox and less than 15% for each of the two FT8 streams. With everything
else equal, a Fox is more likely to be decoded with FT8 than SuperFox only for the
single-stream case, N=1. Otherwise, SuperFox always wins on decoding probability,
and of course even more so for achievable QSO rate, since SuperFox messages can
always include information for as many as 9 Hounds.
We have not yet seen statistical summaries that show the hourly QSO rates achieved at
N5J and CY9C separately for SuperFox and standard Fox mode. Based on what we
already know, we think the operators generally made wise decisions about when to
switch from SuperFox to Fox using one or two streams. We note that for paths with the
weakest signals, standard FT8 with N=1 would probably produce QSO rates at least as
high as those for N=2, and often higher.
Curves in the plot show that SuperFox decoding thresholds are degraded less on more
difficult propagation paths than those for FT8. Our simulations show that this trend
continues over progressively worse paths, such as those over high-latitude regions. The
SuperFox tone spacing is nearly twice that of FT8, so this behavior is just as expected.
Moreover, the FT8 decoder tries to make use of available inter-symbol coherence,
which difficult propagation paths tend to destroy. The SuperFox decoder uses only
noncoherent demodulation of transmitted symbols, so again it is expected to degrade
less on difficult propagation paths.
One further point deserves mention here. The plot shows that probability of decoding
an FT8 signal with SNR = –23 dB is nearly zero, yet we sometimes see a decode with
reported SNR = –24 dB. So what gives? For simulations we know the signal strength
accurately, because we generate the signal! For received signals we obtain an
estimate for SNR by averaging signal power over the full transmission, dividing by an
estimate of baseline noise power in the same interval, scaled to a 2500 Hz reference
bandwidth, and finally converting the ratio to decibels. Especially at the lowest
decodable signal levels, the total SNR error budget can be as large as several dB,
especially on the low side.
