Lawrence Mayhead G3AQC wrote:
As I understand it about 3db improvement is possible by making the
FFT sample time equal to the dot length.
Laurie,
that's not strictly true. Of course, the optimum is to collect data
for the FFT analysis for a period of time coinciding with the actual
transmission, which, for nonoverlapping FFTs, implies two things : 1)
the dot length must be equal to the inverse of the frequency resolution
of
the FFT, 2) the transmission must start synchronized with the data
gathering period. But, as shown graphically by ON7YD on his LF pages, what
is usually done is to compute several FFTs spaced in time just a
fraction of the dot length, so that to have the guarantee that at least
one of them coincides with the actual transmission. This has a
drawback, what our US friends call the "dogboning", caused by a
smearing in frequency when the analysis interval and the actual data
interval are only partially overlapping in time.
IMHO, a real gain could be achieved.if we would forget noncoherent
detection, as that based of the computation of the magnitude of the
FFT, neglecting the phase, to switch instead to a sort of matched
filter, where a continuous correlation is computed between the incoming
signal and a locally generated digital sinusoid of the same frequency
of the expected signal. In case of random noise as input, the
correlation will not show any peak. If instead a valid signal is
present, a marked peak will manifest when there is phase coincidence
between the local and the received signals. This technique implies that
both signals have the same frequency, as the peak sharpness depends on
this. This requirement in turn implies that both the Rx and the Tx are
locked to GPS, with an accuracy in time and frequency much better than
that 1 sec you mention. A GPSDO is mandatory, like the HP Z3801A or a
selfbuilt unit (not so difficult to do).
These are only qualitative considerations, without any computation of
the gain that could be achieved by using that technique, and are meant
only to stimulate further discussions on the subject.
73 Alberto I2PHD

