Some of the supposed benefits of JT65a are real, but some are illusory. I noticed that I could plainly heard tones and quite easily see them on the waterfall at about the same level of visibility as a nearby Olivia QSO, yet WSJT was reporting -20dB or son S/N.
Reading the paper [1], I found that the S/N number reported isn't what we thought it was. The -24dB claim in the paper for copying arbitrary messages is relative to the 2.5KHz bandwidth of an SSB phone signal, so if we consider noise on a 50Hz window such as for PSK31, that would be 17dB less noise, so an S/N of -7. In its own bandwidth of 177.6Hz, that corresponds to an S./N of -12.5dB. That isn't bad, but it certainly isn't -24dB below the noise, and fits well with the claim that JT65a outperforms human CW by 10dB. OOO, RO, and 73 are QSL acknowledgements of the initial reports. They aren't sent in the MFSK modulation scheme...they are alternated tones. They claim to have a -32dB S/N requirement, but readjusting again we get -20.5dB for recognition of 3 of the required exchange parts of a QSO. So about 10 log(2500/177.6)=11.5dB of this S/N is illusory. The callsign and grid square exchanges are sent in the normal modulation scheme, which claims -12.5dB S/N in its own bandwidth, but if you "assist" the decoder with a list of callsigns and grid squares to choose from, the claimed coding gain is 4dB. I suppose it isn't much different from being able to copy that DX op's callsign that he never seems to send after a quick check of the spot database, but I believe this 4dB, while it may be nice for the EME folks, is an illusion, as those using "deep search" on HF have seen, generating phantom contacts. (Illusory may be too kind a word for this particular feature...) They do claim a 5dB gain using "averaging", though, and this seems real. I believe this uses the proprietary soft RS decoder and multiple minutes of transmission, so this gain does seem reasonable. A final source of "gain" is the placebo ionospheric heating effect often seen on 15m and 10m during contests, and on 20m any time PSK31 signals are booming in but the rest of the band is dead. The ability of this software to search in a 600Hz window and the willingness of operators to devote time to TX and RX testing on specific frequencies is probably one of its biggest advantages over other modes for HF weak signal work. My conclusion is that this mode is about 10-15 dB worse than it appears to be, and we should start doing more careful tests alongside Olivia and plain old MFSK as controls in side-by-side propagation conditions. [1] http://physics.princeton.edu/pulsar/K1JT/K1JT_eme2006.pdf 73, Leigh/WA5ZNU