1) Byte Limit in the WFD readout
2) PMT fanout coupling capacitor value for the WFD outputs (only)
3) Capacitor value for the discriminator RC circuit on the WFD d-cards
4) Extending capacitor value after the negative discriminator
5) Choice of charge threshold for the WFD fast (<164usec) stop
It is HARD, ALMOST IMPOSSIBLE, to isolate the contribution of each of
the above components to the final fix. The above listing follows an
order that tries to suggest a hierarchy in the way they should be
considered/investigated. However, it remains that all of the them are
heavily inteconnected.
Here, we are proposing a fix that involves:
1) Increasing the WFD readout byte limit from 40,000bytes to MINIMUM
50,000bytes.
2) Increasing the PMT fanout coupling capacitor from 2.2uF to 33uF (tantalum).
3) Decreasing the capacitor on the discriminator RC circuit on the WFD
d-cards from (1.00+0.01)uF to (0.022+0.01)uF.
4) Not changing the extending capacitor on the WFD d-card.
5) Defining a fast (~120usec) WFD stop for PMT charges greater than ~20Vusec.
The fix has been tested in representative counters in the detector.
What we have seen is that the channel-to-channel variation in PMT afterpulsing,
s.p.e. discrimination efficiency by the WFD d-cards and general fanout
noise is more than a factor of two, something that makes any further fine
tunning of the above values not having much sense. The fix needs to be applied
to at least a SM worth of channels as soon as possible and judging from
the SM-wide LED tests we may adjust the fix IF necessary. The values of
the components have been chosen in order to accomodate the
worst-case-scenarios with channels we have encountered. Item #5 (fast stop
charge threshold) will hopefully be the only thing to adjust.
The fix guarantees full digitization of PMT signals up to their saturation
and for most anticipated (up to 10usec) durations. For signals less than
~20Vusec, the 1msec stopping scheme is used while for signals of bigger
charge the faster stop is utilized. It is only for signals at the few mV
level lasting on the order of 10usec (and greater) where we have observed few
short "dents" in digitization (lasting 30-60ns). This is most likely a result
of inefficient discrimination of the s.p.e.'s from the WFD discrimination
circuitry.
In testing the fix, we performed several special calibration runs in these
representative counters, all of which are available in the standard MACRO
data distribution tapes (and currently on VAXGS disks).
Byte Limit in the WFD readout
=============================
The WFD hardware comes with 64KB (=65,536bytes) that is organized in two
"pages" of 32KB each. One "page" is allocated to ADC values and the other
one for TDC/discriminator bits. The way bits are assigned, every 5ns
sample requires 2bytes of memory, or 400bytes for every microsecond.
Thus, at any given time, 164 microseconds of PMT history resides in the
WFD memory. This is the hardware limit of the WFD readout.
Since the WFD turn-on (August 1995) and until 17 July 1996 (RUN12513)
we have been operating with the WFD readout limited at 20,000 bytes
(=50microsecs of PMT history). Since 17 July 1996 (RUN12514) and until
09 April 1997 (RUN13790) we have been reading 40,000 bytes (=100microsecs of
PMT history). On 09 April 1997 (RUN13791) we have attempted to raise the WFD
readout byte limit closer to its hardware limit (60,000bytes). This resulted
in 5 times more ID=2 buffers with respect to the previous running (40,000bytes)
BUT with no increase of the actual data volume. This was simply a result
of how acquisition writes data on disk. We DO NOT believe this to be
a problem for the WFD system per-se. According to the acquisition experts
the only danger that this presents is for Gravitational Collapse detection.
The whole argument surrounds the fact that more possible ID=2 buffers resulting
from a GC will increase the "effective" event rate that the "write-data-to-disk"
process will have to handle, thus limiting the maximum event rate that
our acquisition can handle in a real GC. Nobody could give us an answer
as to how this maximum event rate relates to the WFD readout limit. We plan
to do a test on that (using the pulser, next Wednesday); this will define
the ultimate WFD readout byte-limit. As of 10 April 1997 (RUN13796) this
limit is at 50,000bytes (=125microsecs of PMT history) and should be considered
the MINIMUM for the final solution.
Let us return to the question of what a higher byte limit for the WFD
readout can buy you. The amount of WFD data as a function of the PMT
charge can be generally described by two lines of different slopes.
The "knee" in this response is determined by the clipping diode voltage
which is roughly 0.5V and corresponds to an approximate charge of 0.5Vusec.
Above this, the amount of WFD data as a function of the PMT charge increases
much slower with respect to the region below 0.5V. The limit of 40,000bytes
translates to a maximum charge of ~0.7Vusec while for 60,000bytes this
would had been more than ~2Vusec. The above numbers reflect the present
fanout/WFD hardware BUT it buys us EVEN MORE in the solution that we'll
discuss below.
PMT fanout capacitor value for the WFD outputs (only)
=====================================================
The second item that was studied had to do with the coupling capacitor
on the PMT fanout output that is used from the WFD system (all the
other 5 fanout outputs used by the rest of the systems remain intact).
The current value is 2.2uF. All the fixes that have been investigated
until now were looking for bigger capacitor values that will provide:
less distortion to the PMT signal, smaller positive overshoot and
as a result of that smaller negative undershoot (created by the second
RC at the discr. circuitry on the WFD d-card.)
The 1mF electrolytic cap is the value that really guarantees all of the
above in an almost asbolute way. The drawback in this is the fact that
it doesn't filter out the 50/100/150/... Hz noise coming from the
PMT fanout power supply/ground loops. Since the capacitor on the fanout does
not present a solution by itself (and an accompaning fix on the WFD d-card
has to be considered), the value that can do the job may actually be less
than 1mF. Yes, but how smaller? More than the positive overshoot alone, it is
the negative undershoot created by the WFD RC circuit that hurts us the
most. We need to choose a combination of values for the PMT fanout cap
and the WFD RC cap such that for the grid of all PMT pulse height/widths
this negative undershoot will never exceed the negative discriminator
threshold of 2.7mV. Analytical calculations, SPICE simulations and work
bench tests have shown that the combination of 33uF/0.2uF caps for fanout/WFD
doesn't give threshold-exceeding negative undershoot. Obviously, any
capacitor bigger than 33uF (for the fanout) and/or smaller than 0.2uF
(for the WFD) share the same feature. Moreover, the 33uF cap on the fanout
has a cutoff frequency (at 3dB) of ~97Hz, something that cuts the 50Hz
fanout noise. Availability/price/lifetime/tolerances are no longer concerns
for the 33uF (tantalum) capacitor.
Capacitor value for the discriminator RC circuit on the WFD d-cards
===================================================================
Even driven by the ideal (no noise, DC coupled) PMT fanout, a WFD d-card
would still suffer from problems of the same nature due to the RC circuit
that drives the discriminator signal. As is, it has a time constant
of (1.0+0.01)uF*125Ohm=126usec, almost the same with the 2.2uF*50Ohm=110usec
RC circuit on the PMT fanout. So we would either have to go to a higher
cap value that would guarantee that the positive overshoot amplitude remains
below threshold (~3mV) or go with a lower cap value so that the positive
overshoot drops below threshold fast enough. Due to noise problems,
choosing a bigger cap does not solve the problem. Thus, we've looked into
lower (<1uF) cap values. Very small capacitor values would still not
do the job since they would shut the discriminator off fast and thus lose
fraction of the PMT signal. Photoelectron statistics are always present
in a PMT signal and they actually work in our favor. The RC circuit,
being a high-pass filter would always allow them to propagate to the
discriminators. There are two factors that limit this from being fully
efficient. One comes from the clipping diode that attenuates (up to
killing) p.e. statistics. The second factor comes from the non-100%
efficiency of the WFD discriminator circuit to spe's.
For high (>2V) pulse amplitudes, we have to make sure that the capacitor
decays slower than the PMT returns to ground on big, long pulses. This is
mostly determined by the PMT base saturation. This requirement imposes the
lower bound on the capacitor value.
Our tests IN the detector (3B01-0 thru 3B04-0, two LEDs per tankend firing at
a time) have shown that the (0.00+0.01)uF option leaves gaps for big, long
pulses while options greater than (0.01+0.01)uF do the job. This is based on
certain assumptions of p.e. statistics present.
Extending capacitor value after the negative discriminator
==========================================================
In the present WFD hardware there is a 12pF capacitor that extends
negative discrimination by ~40ns. Thus, one s.p.e lasting 15ns will
extend for roughly 40ns. For long, low amplitude PMT signals, the
above choice of (0.022+0.01)uF decays with a time constant of 4usec.
For these signals, after 2-3 time constants (~10usec) digitization
relies completely on s.p.e. statistics. Given the non-100% efficiency
of the WFD discriminator circuit to spe's, we observe small (50-100ns)
dents in the digitization of the PMT pulse. THESE DENTS WILL BE PRESENT
IN THE PROPOSED SOLUTION for PMT signal of low (10-20mV) amplitude,
large duration (>=8usec). One option in order to "bridge" these dents
is doubling this capacitor values. We have tested this option and indeed
it bridges these dents for most of the cases. The drawback in
this is that it effectively increases the volume of WFD data.
Since this will affect all WFD readout, we think it is hiding
risks that we wouldn't like to take; for that we think we should
live with the few dents for some class of events.
Choice of charge threshold for the WFD fast (<164usec) stop
===========================================================
Even with the ideal (no overshoots/undershoots) fanout and WFD hardware,
we would need several hundred microseconds worth of memory depth in order
to store the ENTIRE 1 millisecond PMT activity following a "big" charge
event. This is a result of the serious PMT afterpulsing following such a
big event. Given the limited memory available (up to 164usec), we need to
stop the WFDs earlier than 1 millisecond (our current WFD stopping scheme)
in order to guarantee that the PMT signal is recorded.
We have performed several special calibration runs (using two LEDs
in a horizontal counter) in order to study the PMT afterpulsing and
PMT base saturation. For example, a 15usec, 6V PMT pulse would had
been lost under any circumstances due to afterpulsing.
This behavior has raised the need for a fast stop of the WFD for PMT
signals that exceed some charge threshold.
With a WFD byte limit of 40,000bytes, we have been losing ~10 evnts/SM/6-hr
due to the byte limit. In SM2 (since 15 April 1997 RUN 13820) we have introduced
a stand-alone module that "triggers" for PMT charges >=0.5Vusec and stops
the WFDs in SM2 at ~120usec, the current maximum memory depth of the WFD
readout (50,000bytes). Sophia K's analysis for RUN 13834 has shown that
we are no longer losing any event in SM2. This module uses the CSPAM fanout
outputs (20 signals/SM), integrates the PMT signals and provides a voltage
proportional to the charge. When this exceeds a user-defined threshold,
this earlier WFD stop takes place. The threshold that is set at the moment
corresponds to the approximate maximum that the WFD can record WITHOUT any
fanout/WFD modification. The module triggers at about ~60 evnts/SM/6-hr
mostly at events that involve multimuons/showers. There are several
details that have to be worked out with regards to such a module.
The major consideration is the fact that it relies on PMT signals that
are highly multiplexed and moreover, it multiplexes them even more.
Given the time of flight for all muon-related events PMT signals coming
from different detector faces will practically overlap and thus intergated
together by such circuit.
Although a more sophisticated configuration can be worked out later,
the module in its present state does the job. Notice that the fast
stop provided by this module (either now or following any fix), does
pose limits to the acceptance for particles that have a time of flight
greater than the fast stop (i.e. ~120usec in SM2 now) and release an
amount of light greater that the fast stop threshold (i.e. >=0.5Vusec).
In the above conditions, ONLY the ENTRY PMT waveform will be recorded. At
the same time, any catalysis analysis will be limited to the fast stop
window only.
Erik/Ioannis K.