We've always had the idea that multiplets should consist of "unique" signals - e.g. if the DB has several spikes that are close in time and otherwise similar, they represent the same source, and only one of them - perhaps the most powerful - should be included in a multiplet.

But our logic for doing this was flawed; in particular, it didn't take into account that sometimes the telescope stares at a fixed point for a while. So Eric Korpela cleaned up this thinking and made the following definitions:

Integration time (t_i): This is the time over which one signal was observed.
Spike: Defined by FFT length. 128K fft=13.4 seconds
Gaussian: Defined by workunit length. 107.4 seconds
Triplet, Pulse: min(beam_width in samples, min(40960 samples,107.4 seconds)
(40960 samples is only < 107.37 seconds for 8 and 16 point FFTs)
Autocorr: Defined by FFT length, but currently only done at 128K FFT
(13.4 seconds)
Beam crossing time (t_b): Defined by the motion of the telescope,
it's the length of time for the telescope beam to cross a point on the sky
(variable)
Exclusion time (t_e): The time in which which two adjacent signals
should be considered to be a "the same single signal".
In principle we would want this to be max(t_b, t_i)
with some upper limit for the case where the telescope is stationary
(or actually finding when the scope has moved a beamwidth),

I implemented these; we'll see the results in the next scoring round.

-- D