The Chicagoland Observatory Underground for Particle Physics cosmic ray veto system
ABSTRACT A photomultiplier (PMT) readout system has been designed for use by the cosmic ray veto systems of two warm liquid bubble chambers built at Fermilab by the Chicagoland Observatory Underground for Particle Physics (COUPP) collaboration. The systems are designed to minimize the infrastructure necessary for installation. Up to five PMTs can be daisy-chained on a single data link using standard Category 5 network cable. The cables is also serve distribute to low voltage power. High voltage is generated locally on each PMT base. Analog and digital signal processing is also performed locally. The PMT base and system controller design and performance measurements are presented.
The Chicagoland Observatory Underground for
Particle Physics Cosmic Ray Veto System
M. Crisler, J. Hall, E Ramberg and T. Kiper
Fermi National Accelerator Laboratory Batavia, Illinois*
Abstract–A photomultiplier (PMT) readout system has been
designed for use by the cosmic ray veto systems of two warm
liquid bubble chambers built at Fermilab by the Chicagoland
Observatory Underground for Particle Physics (COUPP)
collaboration. The systems are designed to minimize the
infrastructure necessary for installation. Up to five PMTs can be
daisy-chained on a single data link using standard Category 5
network cable. The cables is also serve distribute to low voltage
power. High voltage is generated locally on each PMT base.
Analog and digital signal processing is also performed locally.
The PMT base and system controller design and performance
measurements are presented.
he COUPP dark-matter search collaboration has
constructed two bubble chambers at Fermilab. These
bubble chambers are surrounded with liquid filled tanks which
form part of their cosmic ray veto systems. One bubble
chamber with an active volume of 2 liters uses a tank filled
with scintillator oil and the other with an active volume of 60
liters a water tank. Light generated by transiting cosmic rays is
detected by an array of 19 five inch photomultiplier tubes
(PMTs) mounted on the oil tank and 12 eight inch PMTs
mounted on the water tank. Cameras running at 100 frames
per second are used to detect bubble formation within the
chambers and supply a trigger to the data acquisition system
(DAQ) which in turn triggers the PMT systems. These
systems must operate at a low threshold and, given the bubble
formation and detection timing, store about 100 milliseconds
of data. To achieve this in a mechanically simple, inexpensive
and compact way, we are providing the analog and digital
functions required along with a high voltage power supply all
mounted directly on each PMT base.
The analog processing section includes a charge integrator
for use by the ADC which has a range of 12 bits with a least
count of 25 femto-Coulombs (fC) and a sampling rate of 40
Msps and an anode discriminator for use by the on-board
TDC. The digital section includes a small FPGA which is used
to implement a variety of control functions and a 1.6 ns per
bin TDC, a 32 Mbyte RAM, a microcontroller and a serial
data and control link. The high voltage supply is a low noise
resonant mode Cockroft-Walton type whose interference level
is a fraction of an ADC count. Up to five bases can be daisy-
chained on one data link. The base consumes approximately
A controller module housed in a 2" high 19" rack mount
box is used for timing, synchronization, control and readout of
up to four data links. The controller delivers power to the
Manuscript received November 12, 2010.
Operated by Fermi Research Alliance, LLC under Contract No. DE-AC02-
07CH11359 with the United States Department of Energy.
bases over the links and has outputs for driving an LED
flasher system. The data path from the controller to the DAQ
is 100 Megabit Ethernet.
A cosmic ray veto is commonly used in low background
searches to aid the rejection of background events. Typically
scintillator either solid or liquid is used. PMTs are the still the
sensor of choice for these systems because of their large
sensitive are. What is unusual in the case of COUPP is the
time scale over which PMT signals must be stored pending a
trigger decision from the cameras observing the active volume
of the bubble chambers up to several 10s of milliseconds. A
conventional PMT system assembled from
components requires an in-time trigger to record the PMT
signals. The long look-back requirement presents the
opportunity to design a simple to use, minimal infrastructure
system which would have general applicability to a variety of
PMT based systems.
The system consists of two parts: a controller and a base.
The controller is the interface between the bases and Ethernet,
supplies power, timing reference and triggers to the bases over
standard category 5 cables.
The bases contain a supply to deliver the high voltage
required by the PMT, an analog section for integrating the
anode charge and discrimination the time of arrival of the
signal, an ADC, a large data buffer, a microcontroller for
control and status functions, an FPGA for the TDC and
interfacing the various digital sections, and a synchronous data
link for communication to the controller. Twenty-four volts
from the cable is connected to a series of secondary power
converters for generating several voltages required. A key
advantage of providing all the functions on the base is the
reduction of the cable plant. A typical PMT system would
require an ADC cable, a TDC cable and a high voltage cable
for each tube. With this system, 15 patch cables connecting 19
PMTs into strings and four cables connecting the strings to the
controller are all that are needed for the two liter chamber.
A. Cable Signal Assignment
The cables are standard category 5e. The voltage levels are
LVDM, with FM encoding running at 25Mhz. Four cables
operating at 25Mhz exceed the bandwidth available with a
100Mbit Ethernet link. In practice the Ethernet link achieves
Fig. 1. Cable signal assignments.
about four Mbytes/sec transfer rate. Of the four pairs in the
cables, two are bussed; two are daisy-chained. The bussed
pairs are used for commands from the controller; the daisy
chained pairs are the responses from the PMTs. With this
scheme, there is no need for any drivers to change direction.
The use of FM makes it simple to phase lock the PMT
system clocks to that of the controller. Since FM has constant
activity on a pair, the first module in the chain can sense that it
is at the end opposite the controller if it sees no activity on its
input port. Addresses can be automatically assigned based on
a PMTs position in the chain, so no address switches are
needed. Once a geographic address is assigned, any logical
address can be assigned afterwards. One bussed pair is used as
the clock frequency reference and for synchronous messages
such as triggers and timing initialization. One daisy chained
pair is used for event data which is sent in binary format. The
second bussed pair is used for control commands. Since
control bandwidth is not large, ascii bytes are used to make the
data readable with a terminal emulator. The second daisy-
chained pair returns status information, also in ascii form.
Fig. 2. Controller block diagram.
The controller is housed in a 2 inch high rack mount box.
Mounted in the same box is a 50W 24 volt power supply,
which is enough to power the controller and 20 PMTs. The
links use standard power over Ethernet transformers for the
power distribution. Included on the controller card are four
channels of LED flasher outputs, which are AC coupled TTL
pulsed outputs offset by a variable DC voltage. Small “postage
stamp” cards at the far end of the flasher cables are fitted with
a simple transistor switch which discharges a capacitor biased
with this DC voltage across an LED. There is also a DDS
frequency synthesizer chip configured as a numerically
controlled divider within the VXO phase lock loop which
allows the controller to phase lock to an arbitrary external
reference frequency. Finally, there is an input compatible with
a Fermilab built GPS receiver card which can be used as a
reference for the VXO if absolute timing is desired.
The controller extracts event data from the PMTs upon
receipt of an external trigger and builds a single block of data
with a global header and presents this block to a Labview
program running on the DAQ host. Status data is gathered
periodically and pooled into single a block in the controller,
simplifying the access to this information by the DAQ host.
Fig. 3. Photograph of the controller installed for four liter test run at
Fermilab. The green cables are the PMT links, the black is the Ethernet cable.
The LEMO inputs are visible to the left of the PMT links.
C. PMT Base
Each base is equipped with a resonant Cockroft-Walton
voltage multiplier, an anode charge integrator, a discriminator
for timing information, a 40Msps 12 bit ADC with 25fC per
count sensitivity, a DRAM buffer memory, a 1.6ns per bin
TDC a, microcontroller for status and control, and serial links
for communication. The buffering of the digital data allows
for both pre and post trigger data readout.
Having the charge integrator right at the anode of the PMT
affords a degree of insensitivity to interference from the high
voltage supply. Careful arrangement of the multiplier
components is still necessary, but the bases have a fraction of
an ADC count of HV power supply interference.
Fig. 4. Block Diagram of the PMT base
1) Anode signal processing
The since the PMTs are large (5 inch and 8 inch), the bases are
positive high voltage, so the anode signal is AC coupled to the
integrator input. Having the charge integrator adjacent to the
anode means that there are no input impedance restrictions on
the integrator as it is being driven by nearly an ideal current
source. The negative input signal is inverted by the integrator.
This positive integrator signal is differentiated and sent to a
discriminator. By using the integrator output, a single supply
comparator can be used, and a wide range of pulse widths,
wider or narrower than the original PMT pulse that can be
implemented by choosing the appropriate R-C network. A
decay time constant (300ns) was chosen to allow the ADC to
deliver multiple samples of the waveform and at the same time
have a low probability of pile-up. In principle overlapping
pulses can be resolved using the TDC information, but is not
implemented here because the rates (KHz in the worst case)
are so low.
Fig. 5. An oscilloscope trace of a PMT anode signal and the integrator output.
A test pulse injector is included. A DAC channel on the
microcontroller is used to set the magnitude and a pulse
pattern memory is included in the FPGA logic. The pattern
memory can be loaded with a list of times. When a counter
clocked at 80MHz matches the time of a particular list entry, a
test pulse is issued. The span if the counter is about 30 seconds
and the table has 256 entries.
2) Data formatting
The discriminator output is fed to the FPGA which has a
1.6ns per bin TDC implemented in its logic. The minimum
binning for this FPGA is 1.2ns, but 1.6ns conveniently
subdivides one ADC sample interval by 16. TDC hits are used
to zero-suppress the ADC data. For each hit, a programmable
number of pre-hit and post-hit samples are written into the
data buffer. Each event consists of a leading word count, the
TDC time followed by the ADC data.
3) Buffering Scheme
In the COUPP case, the PMT system is externally triggered.
The desired data is a list of hits that occurred from a specified
interval before the trigger up to a specified interval after the
trigger. To do this, three time counters are used. On receipt of
a synchronous initialization message (INIT) one is set to
negative pre-store value and the other to a (presumably)
positive post-store value and the third, the counter used to
from the upper order bits of the TDC word, is set to zero. Also
in response to INIT, the event begin pointer is set to the
beginning of the DRAM. During running, the value of the pre-
store time counter is compared to the TDC data field of the
oldest event in the buffer. When the counter equals the TDC
value, the event read pointer is incremented to the next oldest
event in the buffer and so on. This way the oldest relevant
event is always ready to be read out. Upon receipt of a trigger
events are copied from the DRAM to a single event buffer in
the FPGA until the TDC value of a given event equals or
exceeds that of the post-store counter. The reason for the
second buffer is the desire to send data to the controller with a
leading word count. This count isn’t known until all the
relevant hit data have been collected. When an event has been
copied to the staging buffer, the on card microcontroller is
interrupted, status information is appended and a total word
count calculated. At this point the PMT microcontroller waits
for a data request from the controller. When a readout request
message with the correct address is received, the
microcontroller starts a DMA controller implemented in the
FPGA and data is transmitted over the link to the controller.
There is nothing in the hardware that precludes a self-
triggered system, which might be desirable in other
4) High Voltage supply
The resonant mode Cockroft-Walton topology offers
advantages of high efficiency and low noise. The power
consumption of the high voltage supply is less than 80mW.
The operating frequency is around 400 kHz which allows the
use of small magnetic components and filter section
capacitors. The leakage inductance associated with separated
transformer windings is not a problem for this topology, so it
is straight forward to make a transformer that can isolate the
PMT voltage, a desirable feature for a positive base. By
isolating the high voltage at the transformer, the match
between an N stage multiplier chain output current capacity
(1/N) and a N stage PMT dynode chain current consumption
(1/N~4) is as it would be for a negative base. The control
variable is frequency, in this case generated by a digitally
controlled oscillator implemented in the FPGA. The supply is
operated above resonance. Lowering the frequency increases
the output voltage, raising the frequency lowers the voltage.
One problem with this is that care must be taken not to go
below resonance, which results in a positive feedback
condition A resistive voltage divider provides the voltage
feedback to an ADC input on the microcontroller. A PID servo
controller updating at one kHz is implemented in the
microcontroller. The frequency output of the FPGA is fed to a
bridge driver which delivers a 11V pk-pk square wave voltage
to the transformer primary circuit. A series inductor (L in
figure 6) forms a resonant tank with the capacitance of the
reversed biased diodes in the multiplier chain reflected back
through the transformer.
Fig. 6. The resonant Cockroft-Walton multiplier.
The Q of this resonance results in a quasi-sinusoidal primary
current which in turn delivers a quasi-sinusoidal secondary
voltage. The Q of the tank also gives a large voltage ratio
across the transformer using a relatively small turns ratio.
Typically the secondary voltage is 100V pk-pk with an 11V
drive and a 1:3 turns ratio.
Fig. 7. Trace taken from a negative base of transformer primary current
and secondary voltage which is very nearly sinusoidal. A positive base has the
secondary voltage offset by the full multiplier chain voltage.
In spite of the smooth high voltage waveforms, care must be
taken with the filtering and layout to keep power supply
interference to a minimum. The reverse biased capacitance of
all the diodes in the multiplier chain is effectively in parallel.
As a result, there is a capacitive divider between the diodes
and any dynode filter capacitors. The diodes have a reverse
capacitance of order one pF and the filter capacitors are 10nF.
Twenty-fours diodes in parallel in a divider with 10nF results
in a ripple of about 300mV at the extreme ends of the chain
where the capacitors are directly connect to ground. In the
middle of the chain, the ripple is greater since the capacitance
to ground is divided by the number of capacitors in series.
This ripple must be filtered before connecting with the
dynodes, so there are two stages of filtering per dynode. A
particular feature of positive bases is that the anode must also
be biased. This critical connection calls for a three stage filter.
Unfortunately, the filter capacitors associated with the anode
must withstand the full high voltage and are thus rather large.
Since positive high voltage is usually associated with large
tubes, this is not a problem. With the COUPP bases the ADC
pedestal width is measured to be 1.3 bins RMS with the high
voltage supply turned off and 1.7 bins RMS with the supply
operating at nominal voltage, but with no PMT attached.
Voltage multiplier chain
Fig. 8. A photograph showing the two cards that are stacked to form a base
for an eight inch PMT.
Fig. 9. A photograph of the top of the veto tank surrounding the four kg
chamber at Fermilab. Nineteen five inch phototubes are read out over four
cable chains. The green cables in the foreground can be seen daisy-chaining
from base to base.
IV. TEST RESULTS
The tanks were set up on the surface and fitted with a
scintillator telescopes. The measured efficiency of the vetos
was about 98%, the expected value given the geometry of the
setup. Tests with the LED flasher indicate the TDCs are
synchronized at the level of 100ns which is consistent with the
internal reflection times of the tanks.
Fig. 10. A screen capture of the Labview display during the surface test of the
60 liter chamber showing a single scintillator trigger. The vertical axis is ADC
counts, the horizontal time scale is milliseconds.
The 2 liter chamber was operated 150m below grade in the
NuMi experimental at Fermilab hall for a few months. The
veto system functioned properly with no failures. The chamber
is as of this writing beginning a data collection run at the
Sudbury Neutrino Observatory (SNO) in Ontario Canada. The
site is sufficiently deep such that for a sensitive mass of 4kg,
there is no need for a veto.
Fig. 11. A screen capture of the Labview display during the NuMi hall test of
the two liter chamber. This shows all the camera triggers over a three day
period. The peak visible at time zero is due to muons generated by the NuMi
neutrino beam. The cluster of dots at time 31ms and 3000 ADC counts comes
from the LED flasher system which is fired for each trigger to verify time
The 60 liter chamber is now in a test run in the NuMi hall
and its veto is functioning normally. The expectation is that it
too will be moved to SNO, and since it is larger, will be fitted
with its veto even at the deep site.
 Spin Dependent WIMP Limits from a Bubble Chamber. Science 319:
 Henry W. Ott, Noise Reduction Techniques in Electronics Systems, 2nd
ed., John Wiley & Sons 1988, pp. 79-95.
 C. Rivetta, G.W. Foster, S. Hansen, “HV Resonant Converter for
Photomultipliers Tube Bases”, Nuclear Science Symposium conference
record, 1992, vol.1, pp. 438-440.