User name sawatzky
Log entry time 11:00:42 on March 17, 2006
Entry number 167025
Followups:
- 167197 03/18/06 02:10 R. Feuerbach De-bouncing removed from T1
- 167259 03/18/06 07:25 sawatzky Copies of T1,T2,T3 plugged into v775 as of Friday morning (run 3058 on)
Ok, I tracked down a couple of problems tonight -- we'll see how much
things improve.
Here's what I did:
1) The T7 'fake physics' pulses were plugged back into the BB and NA.
These should now generate a physics-timed peak which is both a T3 and a
T7. Make sure you EXclude T7 events in your analysis.
2) I noticed that the airflow in the upper trigger NIM bin was almost
non-existant and the modules were quite warm (the NIM-ECL module in
particular). The filter was completely clogged so I removed it and set
it on the HV crate in the same rack. The modules cooled enough that a
100 ns gate actually shortened by 5 ns after the temps stabilized... I
also pulled the filter on the NIM bin below the trigger NIM bin. That
filter wasn't quite so clogged, but at least two of the fans are dead and
the others don't seem to be working so well. There isn't anything
important in that NIM bin, but the fan tray should probably be replaced
anyway...
I know the ECL-NIM translators have a history of going crazy when they
get hot -- this could have been a contributing factor to some of the
weird TDC spectra we saw with the TS triggers (which were fed by that
ECL-NIM module).
3) This one is important. The TS L1A timing follows the TRAILING edge of
the input triggers, NOT the leading edge. I first noticed this under
some orchestrated conditions on the scope. I later tracked down the TS
manual and found a comment on p. 33 where it says the "rising edge" sets
the timing). This caused a problem where double pulsing on the NA logic
signal arriving at our trigger rack would produce pulses of various
lengths since our modules are all updating. This could easily throw the
L1A pulse off by 40-80 ns or more depending on how many after-pulses
existed. Even worse is that there is a resolving time on the order of
10-15ns on the TS inputs. Pulses separated by less than that amount
would have the L1A timed off the trailing edge of the LAST pulse instead.
Not good.
I ended up implementing a 'debounce' circuit for the NA trigger by
feeding a delayed copy of the initial NA pulse back into a veto in the
same module (the veto just operates on the NA sub-unit, not the whole NIM
module). This forces the NA trigger that is propogated to the TS to be
dead for ~100ns after the first pulse arrives. I was careful to make
sure that there were no changes in the relative timing of the NA pulse
after juggling the cables.
This cleared up a variety of timing problems and oddities in the
D.bit3[0], D.ctimeL1A spectra. See Figure 1 (Run 3046, before the
change) and Figure 2 (Run 3055, after the change) for an example. It
also seemed to clear up a strange 'divot' in the L1A histogram shown in
Figure 3 (Run 3050, before the change). Note the strange gap in the
randoms between channels 1900--2100.
A copy of this log entry has been emailed to: rom@jlab.org
Figure 1
Figure 2
Figure 3
