Talbot Andrew wrote:

 I think Alberto is using a
2^17 long FFT on an 11025 Hz sampling rate, so each FFT bin occupies
0.084Hz.  Windowing of the FFT data is usually done which widens the
effective bandwidth - a Hamming window gives a factor of 1.85 - so the Rx
bandwidth is now 0.16 Hz.

Correct, only I use a Bartlett window, to reduce the width of the main lobe.
The Gibbs phenomenon is of no much annoyance here.

Hope that explains the process of how to compare these modes.  All rely on
the same fundamental decoding mechanism, incoherent detection of signal
power in a fixed bandwidth.  Now, if instead of power, we operate with known
phase and look for signal voltage we immediately gain another 6dB, but
that's another story.

Yes, a version of Jason locked to a GPS could be done, but the problem is,
how many have the needed hardware to use it ? Besides non-coherent detection,
another factor contributes to reduce the S/N performance : character sync.
I don't know when a tone switching occurs or should occur. So I compute the
statistical 'mode' of the FFT peaks to detect this event. It is my feeling that 
an
integration done on the signal exactly on the interval where you 'know' that
a single tone is to be found, would give much better results.

 Suffice to say,  G3PLX at 393km distance can get 100%
copy of my 30s BPSK transmissions when I am radiating less than 1W of RF,
equating to probably only 100-200 microwatts ERP

In the States, the W4DEX beacon with 1 W (RF power, not ERP) has been
received at 1000 km distance, though not 100% error free. This is not to say
that m-ary FSK, non-coherently detected, is better than BPSK (it isn't), but the
diagrams on the Tom McDermott's book
"Wireless Digital Communications: Design and Theory"  ISBN: 0-9644707-2-1
giving the BER at very low S/N ratios look promising.

73  Alberto  I2PHD