In the old hitfinder, deconvolution is done seperately in the time and pad
directions. The switches that control the time deconvolution are
TRFF and DECON.
- If TRFF=0, then there is no deconvolution attempted.
- If TRFF=1, then a symmetric Gaussian response is used.
- If TRFF=2, then a Gaussian response is used, with the width (sigma)
varying linearly with time (asymmetric gaussian).
- For TRFF=2, if DECON=1, then the width parameters of the
asymmetric gaussian are fixed.
- For TRFF=2, if DECON=2, then the width parameters of the
asymmetric gaussian can vary. If the fit fails, then they are fixed
and we try again.

## Comparison 1

Take a look at some of the spectrum fragments shown
here, deconvoluted
with the three switch settings listed above.
What I see is that TRFF=2 and DECON=2 either (1) fits a little better
than TRFF=1, or (2) fits a lot worse. Case (2) is fairly rare, but you will
find some examples below. When TRFF=2 fails, it does so spectacularly.
However, this happens when the pulse is wide, and the TRFF=1 peaks are
rejected for being too wide. Based on this comparison, it is
hard to say conclusively which is better.

## Comparison 2

However, it is clear that TRFF=2/DECON=1 is the worst situation, so
we exclude that.
Looking at the spectra from the comparison above, we wondered if the
constraints on the two sigma parameters are messing up the TRFF=2/DECON=2
fits.

By default, and in the comparison above, the constraints are
sig1_hi = 3, sig2_hi=5, as provided in the HIT_SW table.

In this, comparison, we
look only at TRFF=1/DECON=1 vs TRFF=2/DECON=2, but we have loosened
the cuts to sig1_hi = 6, sig2_hi = 10.

From this comparison, it is pretty clear that TRFF=2/DECON=2 is better.

** Also in this comparison, I have "enriched" the spectrum sample
so we can look at some spectra with many peaks.**
Unfortunately, many of
our "multiple-peak" spectra result from poor pedestal subtraction, to which
the two deconvoluters react differently (TRFF=2/DECON=2 seems more robust
to this).