I just did a comparison of different event lengths wrt the
appearance of echo using Sept11 and BaghdadBridge data. Both
previously showed echoes at a length of 4 hours.
EventLength34.jpg shows lengths 3 and 4 hours with real data
on the left and pseudo data on the right. The dataset name
has 3 or 4 appended.
I used the Euclidean distance method to search for echoes in
the LP_Variance array, and I used the exponential filter
with coefficient .001.
Sept11Length.jpg and Sept11LengthPED.jpg show the power
spectrum analysis of the data in the graphs for event
lengths 2, 3, 4, 5, and 6 hrs. Similarly for
BaghdadLength.jpg and BaghdadLengthPED.jpg. The pseudo data
is nicely flat compared to the real data. From these
graphs, it appears that we should use event lengths of 3 or
4 hours. I'm leaning towards 3 hours.
Incidentally, the peak position corresponds to about 6
hrs/cycle for both real data sets.
Maybe we could use the appearance of this kind of ringing as
additional evidence of a real effect.