Jim wrote:
> I’m not sure if Paul’s stacking process is more
> akin to coherent integration of the time-domain data
It is. I'm simply adding the daily signals in the
time domain. A box car averaging with a repetition
period of 24 hours. The signal (hopefully) adds
coherently to sum to 9 times the average amplitude
of the received signal. For the plots, I divide
by nine (and by the average normalisation factor)
to give a y axis scale representing the average
received signal strength.
The signal doesn't integrate so well if I just do
a single 9 day integration (actually 817200 seconds
which is 2017-03-11_13:00 to 2017-03-21_00:00, 9.45
days) in 1.22 uHz bandwidth. For this I extract
13:00 to 00:00 each day, padding the gaps with zeros.
The peak is there but S/N not so good, only about
3 sigma.
I haven't been able yet to get anything out of the
four 'unused' days (17th and 21st to 23rd).
> Hard to imagine a tougher frequency range than the
> neighborhood of 2970 Hz.
The average sferic energy is less here, so the sferic
blanker is having to do less work - typically about
5% blanking factor. The threshold doesn't seem
very critical at all. The signal is amidst a forest
of mains harmonics. That's not actually a problem
so long as the rx is impeccably linear and the hum
doesn't have too many sidebands.
Maybe I'm just at an awkward distance where for
example ground wave and sky wave are often
cancelling. In one respect the propagation ought to
be simpler - we're below the waveguide cut-off frequency
of the higher modes.
One problem I have with the very narrow bandwidths
(ie less than say 10uHz) is numerical noise due to
limited floating point precision. This certainly
happens with the Goertzel algorithm of vtnspec and
I have to use vtwspec for the long integrations.
--
Paul Nicholson
--
|
