Design and engineering aspects of a high resolution positron ...

1446
Abstract
IEEE TRANSACTIONS ON NUCLEAR SCIENCE, VOL. 41, NO. 4, AUGUST 1994
Design and Engineering Aspects of a High Resolution
Positron Tomograph for Small Animal Imaging*
R. Lecomte, J. Cadorette, P. Richard+, S. Rodrigue and D. Rouleau
Department of Nuclear Medicine and Radiobiology,
Universid de Sherbrooke, Sherbrooke, Quebec, Canada J1H 5N4
We describe the Sherbrooke positron emission
tomograph, a very high resolution device dedicated to
dynamic imaging of small laboratory animals. Its distinctive
features are: small discrete scintillation detectors based on
avalanche photodiodes (APD) to achieve uniform, isotropic,
very high spatial resolution; parallel processing for low
deadtime and high count rate capability; multispectral data
acquisition hardware to improve sensitivity and scatter
correction; modularity to allow design flexibility and
upgradability. The system implements the "clam-shell"
sampling scheme and a rotating rod transmission source. All
acquisition parameters can be adjusted under computer
control. Temperature stability at the detector site is ensured
by the use of thermoelectric modules. The initial system
consists of one layer of 256 modules (two rings of detectors)
defining 3 image slices in a 118 mm diameter by 10.5 mm
thick field. The axial field can be extended to 50 mm using
4 layers of modules (8 rings of detectors). The design
constraints and engineenng aspects of an APD-based PET
scanner are reviewed and preliminary results are reported.
I, INTRODUCTION
Small discrete detectors with individual readout and
parallel signal processing are required in order to reach the
theoretical resolution limit in PET [ 11. At the same time, 2-
D detector arrays with a high packing density are necessary
to achieve good sensitivity and isotropic resolution in
volumetric imaging. These requirements can be satisfied
using solid state photodetectors as a replacement of
photomultiplier tubes for reading out the scintillation of
individual crystals. Avalanche photodiodes (APD) coupled
to BGO scintillators have been shown to achieve both the
spectroscopic and timing performance required for PET [2-
51. However, implementing APD detectors in PET raises a
number of specific problems in addition to the well-known
tradeoffs between sensitivity, resolution and cost per
channel.
* This work was supported in part by the Medical Research
Council of Canada under Grant MA-8549. The first author is a
senior scholar of Le Fonds de la Recherche en S ad did Qrribec.
+ Now with Alliance Medical Inc., St-Laurent, Quebec, Canada
H4S 1x8.
0018-9499/94$04.00 0 1994 IEEE
The construction of the first PET camera based on APD
detectors at University of Sherbrooke is nearing completion.
In this paper, we discuss the design constraints and
engineering aspects of an APD-based PET scanner dedicated
to high resolution dynamic imaging of small animals.
11. DESIGN CONSTRAINTS
The Sherbrooke animal PET camera is based on the
EG&G C30994 detector module consisting of two BGO
scintillators, each coupled individually to a silicon "reachthrough"
APD [6]. The properties and performance of these
detectors were described elsewhere [7] and have been fully
investigated using the Sherbrooke PET simulator [S-lo].
The detectors are enclosed in an hermetic package of
dimensions 3.8 mm x 13.2 mm x 30 mm which determines
the channel packing density (4 channels/cm2). The following
were the main design constraints imposed by the use of the
BGO-APD detectors:
- the timing performance of the detector is critically
dependent on the preamplifier series noise and source
capacitance; therefore, a wideband, low noise-voltage
preaniplifier must be used [5,11];
- the preamplifier must be located in proximity oE the
detector to decrease stray capacitance and minimize
electromagnetic interference;
- fast and slow signals must be derived for optimum timing
and energy selection of the events;
Fig. 1 - Photograph of front-end cassettes showing their tangential
arrangement on a ring and axial assembly for a multi-layer system.
EG&G C30994 APD detector modules are also shown.

- APD detectors have a narrow optimum operating range
and, therefore, must be biased independently:
- access to the detector rim noise and energy spectrum is
required for implementing automated detector tuning
procedures [ 101;
- a steady temperature (within less than l°C) must be.
maintained at the detector site because of the temperature
dependence of the APD gain.
Modularity was another major requirement to facilitate
servicing and allow design flexibility and upgradability.
Finally, strong emphasis was placed on computer control of
calibration, setup and fault detection in system design.
111. SYSTEM DESCRIPTION
A. Tomograph Configuration
The detector modu1.e~ are physically and logically
grouped into cassefres which also incorporate the front-end
electronics. Cassettes include 8 modulesflayer forming a 2 x
8 detector array and they can be assembled into "blocks" of
several layers (Fig. 1). The dimensions and shape of the
cassettes are such that they can be used in the construction of
various diameter rings without modification. The animal
tomograph utilizes 32 cassettes (256 detectordring) on a 3 10
mm diameter cylinder. 'The cassettes are further organized
into groups of four which are made coincident with three
opposing groups over the ring, thus defining 12 coincident
pairs over a 118 mm diameter UFOV (Fig. 2). The port
diameter is 135 mm, which is suitable for small laboratory
animals such as rats, rabbits or the brain of small monkeys.
Since each layer of modules consists of two adjacent rings of
detectors, three image planes (two direct, one cross) are
defined within a 10.5 mm thick transaxial slice. The design
characteristics of the animal tomograph are summarized in
Table I.
Fig. 2 - Geometry of the animal tomograph. The ring is divided
into 8 groups, each made of 4 cassettes with 8 modules/layer. Each
group is in coincidence with 3 opposing groups. Only LORs
crossing UFOV are useful to the image.
1447
Table I
Physical description of the Sherbrooke animal tomograph
Detector type
Number of detectors
BGO crystal size
Module dimensions
BGO crystal spacing
Number of detector rings
Ring diameter
Port diameter
Useful field-of-view
Shielding gap width
Reconstruction planes
Sampling
Number of LORs - total
- useful
EG&G C30994, low-k
256 1 ring
5 12 1 layer of modules
3 x 5 x 20 mm (beveled)
3.8 x 13.2 x 33 mm
3.8 mm in-plane
5.5 & 7.7 mm axially
2 to 8
(1 to 4 layers of modules)
310 mm
135 mm
118 mm
10.5 - 50 mm
3 (2 direct, 1 cross) per layer
"clam-shell''
24,576 / plane 98,304 1 layer
16,384 / plane 65,536 1 layer
B. Electronics
The tomograph electronics is distributed between the
front-end cassettes installed in the gantry, the processing and
control functions housed in FASTBUS crates and the
acquisition system located in an industrial FT computer.
Fig. 3 illustrates the system architecture.
The front-end cassettes (for one layer) include 16
channels shared between two 8-layer dual-side surface-
mount printed circuit cards. Each channel consists of a
MOSFET-based charge sensitive preamplifier and fast/slow
shaping amplifiers [ 11,121. The cassettes also include
analog summing circuits to provide a single total energy
signal per module, individual APD rms noise monitors with
associated readout logic, a common test input and a
temperature sensor. The power consumption is about 600
mW per channel or 10 W for the cassette.
The shaped analog signals from the front-end are brought
through miniature coaxial ribbon cables to three external
FASTBUS crates housing a total of 67 boards of eight
different types:
- 16 discriminator boards, each containing 32 constant
fraction discriminators (CFD). The CFD timing pulse
(DTP) is used to trigger the energy processing cycle and is
routed to the coincidence units for time validation.
- 32 energy processor boards, each containing 8 channels
of a gated integrator, ADC and digital single channel
analyzer (SCA). The boards contain encoder/multiplexer
circuitry to generate a common timing pulse for the
cassette (DTP-c), a 5-bit crystal address (ADD.C) and 8
bits of energy data (ENR-c). Programmable delay lines
were also inserted to compensate for propagation time
differences of the DTP signal between channels.
- 4 encoder boards, each concentrating the data from two
groups of detectors.

1448
FRONT-END
____-_----
FASTBUS CRATES COMPUTER
___--___-_
DlSCRlMINATOR
HIGH VOLTAGE
Fig. 3 - Block diagram of the APD PET scanner electronics. The number of boards are indicated for a one-layer animal tomograph (256
modules, 512 detectors). CSP: charge sensitive preamplifier; E analog sum of energy signals; DTP: timing signal; ADD: detector address;
ENR: energy data. The extensions -M, -C and -G refer to the module, cassette and group data, respectively.
- Two multiplexer boards routing the data from the 8
groups of detectors into 12 coincidence units as required
by the 12 coincidence pairs.
- 3 coincidence processor boards, each containing 4
coincidence units. Each unit incorporates an on-time
window for true coincidence detection and a delayed time
window for random coincidences. A random coincidence
flag and up to 5 bits of physiological information are
inserted into the data stream on the board to form 36-bit
data words which are stored into 512-word deep FIFOs.
The sampling position flag and a 2-bit code identifying
the coincidence unit are added on emptying the FIFO into
-
-
the acquisition system located in the PC computer.
8 high-voltage control boards providing individual bias to
the 512 APD detectors with a precision of about 2 volts.
A common stable voltage reference and safety protection
circuits were implemented on a separate board.
One control board implementing the control signals and
I/O bus to read/write parameters into DACs and LUTs
distributed on the other FASTBUS boards.
A total of 2824 parameters per layer of modules must be
loaded to set-up the AF'D bias, the CFD noise threshold, the
ADC reference conversion gain, the energy lower and upper
thresholds, the delay compensation and the time coincidence
window width. In order to facilitate calibration and
troubleshooting, it is possible to enable or inhibit individual
detectors, modules, cassettes or groups.
The acquisition system was designed to allow on-line
histogramming of the dah with maximum throughput and
high flexibility. In addition to the standard data acquisition
modes (static, dynamic/gated, sampling,...), provision was
made for:
- simultaneous acquisition of the direct or coincidence
IX, -Am
energy spectra from all detectors for calibration, set-up
and diagnostic purposes;
- multispectral data acquisition whereby the coincident
events are recorded in up to 16x16 energy windows [13].
This was achieved by using three parallel fully
programrnable data selectors directly interfaced to three
RISC processors (AMD 29K), each with a 16 MB memory
histogramming capacity. The data selectors are based on
field programmable gate arrays (Xilinx, XC3030) and their
function is to select and rearrange up to 24 out of the 39 bits
of information supplied by the coincidence units. The
processed data are then sorted and histogrammed in real time
by the RISC processors. Note that only six address bits are
required to code the detector position in groups with the
present single-layer implementation of the system. The two
unused bits are available to expand group size in a four-layer
animal scanner or larger ring diameter systems.
C. Gantry
Fig. 4 is a photograph showing the APD animal
tomograph during assembly. A cross section of the
mechanical design of the scanner with one and four layers of
detector modules is displayed in Fig. 5.
Temperature stability at the detector site was the most
stringent requirement to satisfy in designing the gantry for
the APD tomograph due to the temperature dependence of
the APD gain [2]. Temperature control in the gantry is
complicated by the heat load from the front-end electronics (
-300 W per layer of modules) and by the necessity to
maintain proper electromagnetic isolation of the front-end.
The heat from the electronics is evacuated by circulating
forced air between the circuit boards. A method previously
developed for the Sherbrooke PET simulator was used to

Fig. 4 - Photograph of the APD animal tomograph during assembly
with one of the front half-:ring frame removed to show the
cassettes.
cu
Pb
Fig. 5 - Cross section showing the mechanical design of the animal
PET scanner with a single (top) and four (bottom) layers of
detector modules.
1449
provide a stable detector temperature [ 141: twelve
thermoelectric cooling modules (Melcor CP1.4- 127, Trenton
NJ) fixed to the lead shields on either side of the ring are
used to slightly cool the detectors below ambient
temperature (typically 18°C). Water-cooled heat sinks
remove the heat generated by the thermoelectric modules.
Temperature sensors on each half of the annular lead shields
(4 in total) are connected in closed loop to simple custom-
made digital remote control systems which are capable of
keeping the temperature constant within HI. 1°C over
extended periods of time. The lead shields were copper-
plated on both surfaces to ensure good thermal contact and
uniform temperature all over the detector ring. The shield
mass provides sufficient inertia to withstand rapid ambient
temperature changes without any significant effect on the
detectors. Nylon screws and rubber spacers were used to
electrically isolate the shields (and detectors) from the ring
frame.
Since the intrinsic detector resolution is 1.9 mm FWHM
and the linear sampling distance in stationary mode is also
1.9 mm, the sampling density must be increased by at least a
factor of two to reach the resolution limit of the system. The
"clam-shell'' motion scheme was retained [15]. The ring
frame supporting the detectors and electronics is separated in
two halves which are suspended to a clam pivot aligned with
the detectors on the upper side of the ring. The clam action
mechanism is based on a worm screw driving toggle arms
attached to the bottom of the two ring halves (see Fig. 6). A
smooth, continuous motion, free of mechanical shocks and
vibrations is achieved using a stepper motor. The
positioning accuracy is very high (

1450
Provision was made for expanding the axial field of the
tomograph to 50 mm by inserting additional layers of
modules (see Fig. 5). Therefore, the supporting frame, the
cable ducts and the heat management capabilities were
designed to accommodate up to four layers of modules (8
rings of detectors). The front shield collimator is adjustable
and can easily be removed to insert or remove inter-slice
septa. In the current single layer system, no slice septa has
been installed between the two rings of detectors.
The whole ring assembly can be tilted +90°/-300 relative
to the vertical to facilitate selection of imaging planes
through the subject and to permit imaging of phantoms
immersed in a jar in the upright position.
IV. PRELIMINARY RESULTS
The design performance characteristics of the APD
animal tomograph are summarized in Table 11. Preliminary
results were obtained using the Sherbrooke PET simulator
[14], set up with the same geometry and equipped with the
same front-end and processing electronics as the animal PET
camera. All measurements reported here were taken with the
lower energy threshold set at 350 keV on each detector and a
coincidence timing window of 40 ns.
A. Spatial Resolution
Spatial resolution was measured using a 0.85 mm
diameter 22Na line source. Examples of the in-plane and
axial coincidence responses obtainable with the scanner are
shown in Figs. 7 and 8, respectively. These data were not
corrected for randoms and detection efficiency. As can be
observed from these distributions, individually coupled
detectors yield sharp, undistorted responses which follow
rather closely the expected geometric response functions of
the crystals. The overlap between response functions of
adjacent parallel LORs is less than 20% of the maximum
height in the central region of the field. The slight
asymmetry of the base of the axial response for the direct
planes in Fig. 8 is due to the absence of septa between the
Table I1
Design performance characteristics of the
Sherbrooke animal tomograph
Spatial resolution (center)
Intrinsic: Transaxial 1.9 mm FWHM, 3.5 mm FWTM
Axial 3.1 mm FWHM, 5.4 mm FWTM
Reconstructed 2.1 mm FWHM, 3.9 mm FWTM
Absolute efficiency 0.4 Yo
Sensitivity 3.3 kcps/pCi/ml
(1 lcm0 flood, 350 keV)
Energy resolution 125% FWHM (511 keV)
Timing resolution 15-20ns FWHM
Timing window 20 to 40 ns
W
2
a
I-
z
9
8
W
E
4
W
U
1000
800
600
400
200
0
5 10 15 20
SOURCE POSITION (mm)
Fig. 7 - Transaxial response functions for three parallel LORs at a
distance of 10-20 mm from the center measured by sweeping the
line source radially in the field.
I-
z
3
8 1000
W
2
I-
4 500
W
U
0
-10 % -6 -4 -2 0 2 4 6 8 1
SOURCE POSITION (mm)
Fig. 8 - Axial response functions of the two direct planes and of the
cross slice defined by one layer of detector modules. The profiles
were measured by moving the line source axially midway between
diametrically opposite detectors.
two crystals in the same module and to the energy validation
performed on the summed signal from the two crystals.
Spatial resolution data measured in the reconstructed
images and in the axial direction are summarized in Figs. 9
and 10. The resolution in-slice (Fig. 9) was measured from
images acquired by a double sampling method described
previously [ 161 and reconstructed by filtered backprojection
with a ramp filter of cut-off frequency 5.3 cm-'.
B. Sensitivity
An event rate of 70 cps/pCi/cm, corresponding to an
absolute efficiency of about 0.4%, was measured with the
22Na line source placed axially in the center of the field.
The sensitivity measured with a 11 cm diameter flood is 0.8,
1.7 and 3.3 kcps/pCi/ml for the direct planes, cross plane and
full layer of dual detector modules, respectively. These data
were corrected for randoms, but not for scatter radiation.

e
'1
6
2 4
23
8
w2
a
Fig. 9 - Radial and tangential resolution in the reconstructed image
measured using a 0.85 mm diameter line source of 22Na.
7
't
, , . .
0 1 2 3 4 5 6
DISTANCEFROMCPCIEA (an)
Fig. 10 - Axial resolution as a. function of distance from the center.
C. Scatter
In a high resolution tomograph based on small discrete
detectors, a scatter component originating from the detection
system can be identified in addition to the object scatter
component [ 171. The scatter distributions were estimated
from the projections measured with the 22Na line source in
air and in a 11 cm diameter by 2.54 cm thick Plexiglas
cylinder. The scatter within the peak was estimated by
extrapolating the scatter distributions fitted with
monoexponentials under the peak. With the 350 keV
threshold, the detector and object scatter fractions
(scatter/total) at the center are 24% and 17%, respectively.
Since the detector scatter component is confined to a narrow
region around the peak (-2 cm), these events would
conventionally be considered as true events. Note that the
object scatter fraction measured at the center of a 11 cm
diameter cylinder is a worst case for this scanner. The
scatter fraction determined following the SNM/NEMA
recommendation [18] is estimated to be less than 7%.
145 1
D. Count rate
The data processing and acquisition circuits have not
been tested as yet with the full detection system. Thus, we
can only make general comments about the count rate
performance at this time. The overall count rate capability is
the result of limitations in the detector front-end, address
encoding circuits and data acquisition system [ 191. Thus far,
only the detector front-end and part of the encoding circuits
were investigated.
The front-end maximum throughput for a dual detector
module is determined by the 2.2 p energy processing cycle.
However, the effective event rate is affected by the noise
threshold of the CFD which is used to trigger the energy
cycle: if the noise threshold is too high, useful events are
lost: if it is too low, the circuit deadtime becomes important
and the rate of random events due to electronic noise
increases. In addition, accidental pulses due to electronic
noise may invalidate useful events which will be rejected by
the coincidence timing window. Initial experiments with the
PET simulator have shown that a CFD noise rate of the order
of 2-5 kcps per detector is optimum [lo]. Thus, deadtime
due to noise in the front-end is typically on the order of 2%
or less.
Deadtime also arises from the address encoding circuits
which concentrate the timing pulses, reject multiple events,
compensate for propagation differences and generate the
crystal address. This is performed in two steps for the
crystal address in the cassette (0.55 p) and cassette address
in the group (0.135 ps). Assuming an energy-validated
event rate of 5 kcps per detector, which is considerably
higher than most planned investigations, the deadtime due to
encoding is less than 10%.
Each coincidence unit accepts the timing pulses from two
groups and determines the true (on-time) and random (off-
time) coincidences. This is performed in about 105 ns.
Since the timing pulses are generated at a minimum interval
of 135 ns by the group encoders, no deadtime is introduced
by the coincidence units. This is the last real time operation
of the data acquisition process. The validated coincident
event addresses are stored in FIFOs which are subsequently
unloaded at a total maximum speed of about 1.2 MHz.
For a singles rate of 2 kcps per detector, the fraction of
accidental coincidences (accidentals/trues) was measured as
10% with the 40 ns coincidence timing window. Using the
design timing window of 30 ns, it can be extrapolated that a
50% accidental fraction will be reached for a system trues
rate of the order of 500 kcps or a total singles rate in excess
of 5 Mcps.
V. CONCLUSION
The first PET scanner using solid state photodetectors as
a replacement of the photomultiplier tubes has been
developed and its major design characteristics have been
described. In spite of several specific design constraints, the
implementation has been demonstrated to be technically

1452
feasible and economically within reach for instruments of
this class. The APD scanner achieves a nearly isotropic
volumetric resolution of the order of 0.015 cc ", a
factor of at least 2 better than all existing tomographs. In
providing functional images of unprecedented definition, this
new device will be an important research tool for small
animal studies.
VI. ACKNOWLEDGEMENTS
We thank S. Setian and I. Stark for contributions to the
mechanical design of the gantry and ISIS Inc. (Lachine, QC)
for financial support during its construction. The authors
would like to express their appreciation to all those who
contributed, over the years, to the design and development of
this tomograph.
VII. REFERENCES
W.W. Moses and S.E. Derenzo, "Empirical observation of
resolution degradation in positron emission tomographs
utilizing block detectors", J. Nucl. Med., vol. 34, p. lOlP,
1993.
R. Lecomte, D. Schmitt. R.J. McIntyre and A. Lightstone,
"Performance characteristics of BGO-silicon avalanche
photodiode detectors for PET', IEEE Trans. Nucl. Sci.. vol.
NS-32, pp. 482-486, 1985.
C. Carrier and R. Lecomte, "Recent results in scintillation
detection with silicon avalanche photodiodes", IEEE Trans.
Nucl. Sci., VO~. 37, pp. 209-214, 1990.
C. Carrier and R. Lecomte, "Timing performance of
scintillators read out by silicon avalanche photodiodes", Nucl.
Instrum. Meth. Phys. Res., vol. A299, pp. 115-1 18, 1990.
D.M. Binkley, J.M. Rochelle, M.J. Paulus and M.E. Casey,
"A low-noise, wideband, integrated CMOS transimpedance
preamplifier for photodiode applications", IEEE Trans. Nucl.
Sci., vol. 39, pp.747-752, 1992.
A.W. Lightstone, R.J. McIntyre, R. Lecomte and D. Schmitt,
"A bismuth germanate-avalanche photodiode module
designed for use in high resolution positron emission
tomography", IEEE Trans. Nucl. Sci., vol. NS-33, pp. 456-
459, 1986.
R. Lecomte, C. Martel and C. Carrier, "Status of BGO-avalanche
photodiode detectors for spectroscopic and timing
measurements", Nucl Instrum. Meth. Phys. Res., vol. A278,
pp. 585-597, 1989.
R. Lecomte, J. Cadorette, M. HOon, D. Rouleau and G. Gauthier,
"High resolution positron emission tomography with a
prototype camera based on solid state scintillation detectors",
IEEE Trans. Nucl. Sa., vol. 37, pp. 805-81 1,1990.
R. Lecomte, C. Martel and J. Cadorette, "Study of the resolution
performance of an array of discrete detectors with independent
readouts for positron emission tomography", IEEE
Trans. Med. Imag.,vol. 10,pp. 347-357.1991.
J. Cadorette, S. Rodrigue and R. Lecomte, "Tuning of
avalanche photodiode PET camera", IEEE Trans. Nucl. Sci.,
VO~. 40, pp. 1062-1066. 1993.
D. Schmitt, R. Lecomte, M. Lapointe et al., "Ultra-low noise
charge sensitive preamplifier for scintillation detection with
avalanche photodiodes in PET applications", IEEE Trans.
Nucl. Sci., vol. NS-34. pp. 91-96, 1987.
1121 P. Richad, D. Rouleau, S. Rodrigue et al., "Design and
performance of the front-end processing electronics for a
PET tomograph based on BGO-avalanche photodiode
detectors", Nucl. Instrum. Meth. Phys. Res, A (to be
published).
[13] R. Lecomte, M. Bentourkia, P. Msaki et al., "Potentials of
multispectral acquisition in positron emission tomography",
Con$ Rec. 1992 IEEE Nuclear Science Symposium and
Medical Imaging Conference, IEEE Catalog No. 92CH3232-
6, VO~. 2, pp. 856-858.
141 R. Lecomte, J. Cadorette, S. Rodrigue et al., "A PET camera
simulator with multispectral data acquisition capabilities",
IEEE Trans. Nucl. Sci., vol. 40, pp.1067-1074, 1993.
151 R.H. Huesman, S.E. Derenzo and T.F. Budinger, "A twoposition
sampling scheme for positron emission
tomography", in Proc. 3rd World Congress of Nuclear
Medicine and Biology, ed. C. Raynaud, NY:Pergamon, 1982,
pp. 542-545.
[16] M. HCon, C. Carrier, J. Cadorette et al., "A stationary
sampling scheme for multilayer positron tomographs", IEEE
Trans. Med. Imag.. vol. 12, pp. 293-298,1993.
[17] M. Bentourkia, P. Msaki, J. Cadorette et al., "Assessment of
scatter components in a very high resolution PET scanner", J.
Nucl. Med., vol. 34, p. 136P, 1993.
[18] J.S.Karp, M.E. Daube-Witherspoon, E.J. Hoffman et al.,
"Performance standards in positron emission tomography",
J. Nucl. Med., vol. 32, pp. 2342-2350, 1991.
[19] C.J. Thompson and E. Meyer, "The effect of live time in
components of a positron tomograph on image
quantification", IEEE Trans. Nucl. Sci., vol. NS-34, pp. 337-
343, 1987.