The reading I have done indicates there are three main natural issues in dealing with HF propagation: noise, fading, and multipath. There are other factors like impulse from either man-made or natural causes but the one I mentioned are clearly critical.
It may even be that noise is less of an issue that fading and multipath. Or, from another perspective, based on Shannon-Hartley we know the theory behind the limitations due to noise. Fading is a temporary decline in the signal strength. It may even occur selectively across the bandwidth of a signal. But fading is time limited. Multipath propagates a signal over different length paths. The result is that symbols overlap as one is one path delays part of the signal sufficiently to overlap the symbol propagated on a shorter path. A tricky aspect of multipath is that NVIS propagation is more susceptible than DX. An optimal waveform for NVIS may not be optimal for NVIS. Making some guesses from existing protocols here are some thoughts: Pactor 3 uses 100 baud as its fixed symbol rate, regardless of bit rate, to minimize fading. A 10 msec signal appears to be long enough to outlast typical fading which my reading indicates is 2 msec or so. The 100 baud may also be long enough to avoid significant problems from multipath. MT63 addresses fading by recognizing the frequency its selective nature. The bandwidth of MT63 is sufficient that fading often occurs on only part of its bandwidth. That leaves sufficient numbers of its multiple tones unfaded as to reconstruct the symbol. DominoEx takes the approach of limiting bandwidth by only sending a nybble (4 bits) of data per baud keeping the symbol rate low with that fading / multipath advantage. It also uses an interesting technique I have not read about elsewhere that I would call differential FSK, in analogy with DPSK. The absolute tone is not important. What is important is the difference between the previous tone and current tone. The allows for a lot of drift and Doppler since those are not significant over a symbol period. Please take potshots at those observations, suggest similar analysis based on other modes, and provide similar observations. One of the issues I see with many of the existing waveforms is they extrapolate that since PSK31 works well we can use PSK64 or PSK128 to do better. The pitfall is that, for these examples, they are moving the baud into area that make them more susceptible to fading and multipath and therefore the extrapolation is invalid. The development of a good waveform needs to avoid that type of blind extrapolation. Rud Merriam K5RUD ARES AEC Montgomery County, TX http://TheHamNetwork.net
