Conceptual Design Report
(Original: February 2004 Paul Fallon)
(Revision: Oct 2005, Sergio Zimmermann)

Conceptual Design Report
for the
Gamma Ray Energy Tracking In-Beam Nuclear Array (GRETINA)
At Ernest Orlando Lawrence Berkeley National Laboratory
February 2004


_________________________________________ Date: _______________
I-Yang Lee
GRETINA Contractor Project Manager

_________________________________________ Date: _______________
Barry Savnik
GRETINA Federal Project Director
DOE Berkeley Site Office


_________________________________________ Date: _______________

James Symons
Director of the Nuclear Science Division
Lawrence Berkeley National Laboratory
1. Introduction................................................................................................................................ 4
2. Background to GRETINA......................................................................................................... 5
3. Functional Requirements and Performance............................................................................... 7
3.1 General characteristics......................................................................................................... 7
4. Technical Design ..................................................................................................................... 10
4.1 Overview............................................................................................................................ 10
4.2 Germanium Detector Modules........................................................................................... 11
4.2.1 Detector Module Design............................................................................................. 11
4.2.2 GRETINA Geometry.................................................................................................. 13
4.2.3 Detector Tests ............................................................................................................. 17
4.3 Electronics.......................................................................................................................... 17
4.3.1 Pre-Amplifier .............................................................................................................. 17
4.3.2 Digital Signal Processing Module .............................................................................. 18
4.3.3 GRETINA Triggering and Timing System................................................................. 22
4.3.4 High Voltage............................................................................................................... 28
4.4 Computing System............................................................................................................. 29
4.4.1 Hardware..................................................................................................................... 29
4.4.2 Software ...................................................................................................................... 32
4.5 Mechanical System............................................................................................................ 39
4.5.1 Support Structure ........................................................................................................ 39
4.5.2 Liquid Nitrogen Filling System.................................................................................. 40
4.5.3 Target Chamber System.............................................................................................. 42
4.6 System Assembly............................................................................................................... 42
Appendix A: Angular positions of the hexagonal crystals in the 120 hexagon geometry........... 44
Appendix B: Proposed detector acceptance and characterization tests ....................................... 45
1. Introduction

GRETINA is a gamma-ray detector array capable of reconstructing the energy and spatial
positions of gamma-ray interactions within the germanium crystals. It will be used to study the
structure and stability of nuclei under various conditions. The new capabilities provided by
gamma-ray tracking will give large gains in sensitivity for a large number of experiments,
particularly those aimed at nuclei far from beta stability.

GRETINA consists of 27
highly segmented coaxial germanium crystals. Each crystal is
segmented into 36 electrically isolated elements and three crystals are combined in a single
cryostat to form a triple-crystal module. There will be 10 modules in total. The modules are
designed to fit a close-packed spherical geometry that will cover one quarter of 4t with an inner
radius of 15 cm and a germanium thickness of 9 cm. The 1110 detector signals are digitized
using custom electronics. A processor farm of ~150 commercially available CPUs will be used
to reconstruct the final energies and spatial locations of the detected gamma-rays. GRETINA is
transportable and can be used at several existing stable- and radioactive-beam facilities in the

A proposal for GRETINA was submitted to DOE in June 2003. It presented the scientific case,
the readiness of technical development, the design, the suggested management organizations,
and a proposed cost and schedule. The GRETINA proposal received its CD0 approval in
August 2003.

The present document describes a technical implementation of GRETINA, including its cost,
schedule, and resources. The design presented here meets the functional requirements outlined
in section 3 and defined through numerous community workshops and in conjunction with the
GRETINA steering committee and Gamma-ray Tracking Coordinating Committee. It has four
major components; mechanical systems, detector modules, electronics, and computing systems.
Each is described separately in section 4. The fifth major item is system assembly also
described in section 4.

Plus one pre-existing 3-crystal module, making 30 crystals (10 modules in total)
2. Background to GRETINA

Studies of nuclei at the limits of spin, excitation energy, or isospin require efficient and
selective detector systems. This is because the states of interest are either populated very weakly
with stable beams, or are reached with low intensity radioactive ion beams (RIBs). Therefore,
these studies benefit greatly from the use of a γ ray detector that simultaneously provides large
photopeak efficiency, high peak-to-total ratio, and excellent energy and position resolution in
the energy range of a few tens of keV to about 20 MeV.

The current generation of 4π gamma-ray detector arrays, for example Gammasphere, are based
on modules of Compton suppressed Ge detectors. They use high-purity Ge crystals, which have
intrinsically good energy resolution of 2-3 keV. Although the largest available crystals are used
most of the gamma rays do not deposit all their energy in a single crystal. Such partial-energy
events contribute to a background, which can be rejected by detecting the gamma rays that
scatter out of the Ge crystal into a “Compton shield” (made with a high density scintillator such
as BGO) surrounding the Ge detector. While this improves the peak-to-total ratio it does not
improve the efficiency. Furthermore, the suppressors occupy about the same solid angle
coverage as the germanium detectors. This limits the full-energy peak efficiency in
Gammasphere, for example, to ~10% (at 1 MeV) and the peak-to-total ratio is 55%. To explore
new scientific regions, as identified in the 2002 Long Range Plan for Nuclear Science, new
technologies and capabilities are required that go beyond those provided by arrays of Compton-
suppressed detectors.

The efficiency limit reached in current arrays can be overcome by eliminating the Compton
shields and by closely packing the Ge crystals. Rather than suppressing events that scatter out of
individual crystals, the gamma rays can be tracked across crystal boundaries by determining the
location of the scattering points (a 1 MeV gamma ray has typically 3 to 4 interactions within the
Ge before depositing its full energy). This can be achieved by using highly segmented Ge
detectors. These detectors have their outer electrical contact divided into a number of individual
segments. By analyzing the direct and induced charges from these segments the interaction
locations of gamma-rays can be determined to better than a few millimeters. The path of
scattered gamma rays through the crystals can then be followed and the gamma-ray energies
reconstructed by using suitable algorithms. This is the new concept of a gamma-ray energy
tracking array, and GRETINA will be the first practical implementation.

To date, considerable progress has been made towards the R&D necessary to design and build
GRETINA. This includes 1) the manufacture of segmented detectors and pre-amplifiers that can
provide high quality signals needed to resolve and locate individual interaction points, 2) the
use of signal processing methods to determine energy, time, and position based on pulse shape
digitization and digital signal processing, 3) the development of a tracking algorithm that uses
the energy and position information to identify interaction points belonging to a particular
gamma ray, 4) the design and packing schemes for a close-packed array of segmented coaxial
germanium detectors, and 5) the design and construction of a pulse shape digitizer board that
has the high sampling rate and good energy resolution needed for GRETINA. This development
started in 1994 and approximately $1.5M of R&D funds and 20 FTE-years of scientific effort
have been invested in this area.
The scientific opportunities that result from building new arrays based on the tracking principle
were already noted in the 1996 Long Range Plan for Nuclear Science, which stated that they
would `…represent a massive leap in resolving power compared to the existing systems and
have a profound impact on the entire nuclear structure field.’ The physics case for gamma-ray
tracking arrays was further developed in a series of community-wide workshops at Lawrence
Berkeley National Laboratory (1998), Michigan State University (2000), and the University of
Massachusetts at Lowell (2001) and 2002 Long Range Plan for Nuclear Science added `The
physics justification for a…tracking array is extremely compelling, spanning a wide range of
fundamental questions in nuclear structure, nuclear astrophysics, and weak interactions.’ In
July 2002 the Gamma-Ray Tracking Coordination Committee endorsed the design of gamma-
ray tracking array using a close packed arrangement of highly segmented co-axial germanium
detectors, and the March 2003 report of the NSAC Facilities Subcommittee gave the highest
ratings possible to the scientific case for, and the technical readiness to proceed with, a gamma-
ray tracking array.
3. Functional Requirements and Performance

In this section we present an overview of the technical requirements that have a direct influence
on the expected performance of the array.

3.1 General characteristics

Large full-energy efficiency (ε), good energy resolution (ΔE
), and high peak-to-total ratio (P/T)
are the most crucial parameters for any gamma-ray detection system. In a gamma coincidence
experiment, where K-gammas are detected, the sensitivity (RP) for detecting the weakest
reaction channel or decay path increases as [(1/ΔE
, while the counting statistics
improve as (ε)
. Therefore, our primary design goals are to maximize the efficiency and P/T of
the array and to preserve the energy resolution close to the intrinsic values, even in experiments
involving fast-moving sources by providing good position resolution for accurate Doppler

Ge Detector: Hyperpure germanium (Ge) detectors with low noise, good energy resolution, and
minimum cross-talk between segments are needed to produce the high quality signals necessary
to isolate individual gamma-ray interaction points and reconstruct the gamma-ray energy.
Several critical performance requirements are given in table 3.1.

Full-Energy Efficiency: The full-energy efficiency is related to the average interaction length of
the gamma ray vs. the depth of the Ge crystals. Practical issues regarding the commercial
available Ge crystals determine a length of ~9 cm.

Segmentation: The segmentation size is determined by the need to observe an image charge and
to minimize the probability of multiple interactions in the same pixel. Multiple interactions
increase the uncertainty in position and energy resolution and reduce the tracking efficiency.
The number of detection elements (segments) must be large compared to the total interactions to
minimize the multi-hit probability. Optimum segment sizes vary with gamma ray energy
(interaction length) and event multiplicity. Experience with the LBNL GRETA prototype II
detector and simulations indicate that six azimuthal segments with a depth segmentation of
approximately 1 - 2 cm can provide 2 mm position resolution.

Peak-To-Total (P/T) Ratio: The sensitivity to detect the weakest gamma ray in a reaction
increases as (P/T)
and a high P/T is essential for all experiments, especially those involving
high-multiplicity. In conventional arrays, a P/T value of ~60% is achieved with the help of an
active BGO shield to suppress the Compton-scattered events. In a tracking detector, tracking
algorithms may be tuned to optimize the P/T or full-energy efficiency depending on the
experimental requirements. When a tracking algorithm is optimized to achieve the highest RP,
GRETINA is calculated to have a P/T ratio of 60% for a 1.33 MeV gamma ray and an
efficiency of 6.2%.

Position Resolution: A position resolution of ≤2 mm (standard deviation) average for E
keV is needed (i) to determine the location of the first gamma-ray interaction in the crystal and
perform accurate Doppler correction for moving source velocities up to v/c ~ 50%, and (ii) to
ensure efficient and accurate gamma-ray energy reconstruction through tracking for high
multiplicity events.

Timing Resolution: Invariably, studies of far-from-stability nuclei or other rare and exotic
phenomena require detection of gamma rays in coincidence with other gamma rays, or with
signals from auxiliary detectors, or both. These experiments greatly benefit from having good
time resolution (e.g., FWHM of s7 ns at 1.33 MeV) for detection of gamma rays.

Geometry: In a tracking array efficiency depends not only on the volume of active Ge, but also
on the ability to track and reconstruct the full-energy events. To obtain an optimal tracking
efficiency the individual elements of the array must be mounted in a close packed geometry that
minimizes the gaps between the crystals. It is highly desirable for the array to have azimuthal
symmetry. The geometry should be designed to be expandable to a 4t array.

Auxiliary detectors: The partial cross section for the production of exotic nuclei decreases
rapidly as we approach extremes of spin and isospin. To cope with this difficulty we need to
combine gamma detection with one or more auxiliary detectors that would help select the
reaction channels of interest cleanly and efficiently. While many of the auxiliary detectors are
deployed in the inner cavity of the array, some devices are mounted outside the cavity and may
have to replace some elements of the array. The requirements posed by auxiliary devices should
be considered in the mechanical design and in the electronics requirements.

Electronics and Data acquisition: The electronics should have a dynamic range of 10 keV to
20 MeV for the incident gamma rays and a trigger threshold of better than 10 keV. The trigger
logic and data acquisition systems should be versatile enough to accept both prompt and
delayed coincidence triggers from a variety of auxiliary detectors and to incorporate an adequate
level of signal processing to select the interesting events with minimum loss of data. The system
should be capable of processing a rate of approximately 20000 gamma-rays/sec at a position
resolution of ~2 mm.

Table 3.1 Summary of expected array characteristics and performance parameters.

Detector module
Number of Ge crystals
≥ 27
Number of segments 6 longitudinal × 6 transverse
Segment Energy resolution ≤ 2.5 keV (FWHM) average, at 1.33 MeV
Noise per segment ≤ 7 keV (standard deviation) average at 35MHz bandwidth
Time resolution ≤ 10 nsec (FWHM) average, at 1.33 MeV
Array peak efficiency ≥ 7.2 % at 1.33 MeV
Array peak-to-total ratio ≥ 40% at 1.33 MeV
Position resolution ≤ 2 mm (standard deviation) average for E
> 300 keV
Digital Signal Processing Module
Digitizer sampling rate ≥ 75 MHz
Digitizer resolution
≥ 12 bits
Final integral nonlinearity

(in E¸)

≤ ± 0 .1% over the top 99% of the dynamic range
Final differential nonlinearity

(in E¸)

≤ ± 1% over the top 99% of the dynamic range
Final energy/gain stability
≤ ± 0.2%/hour gain drift for ≤ ± 5°temperature drift
Trigger and Readout
Readout speed ≥ 10 MB/s/crystal
Additional functionality Accommodate auxiliary detectors in the trigger and the data
Data processing rate ≥ 20,000 gamma/s total
Data storage rate ≥ 10 MB/s
Performance following Signal Decomposition and Tracking
Efficiency ≥ 5.4 % at 1.33 MeV
Peak-to-total ≥ 55 % at 1.33 MeV

[1] Plus one preexisting module with 3 crystals
[2] Resolution refers to the nominal value, not the effective resolution or effective number of bits
[3] As measured in the final energy spectrum

4. Technical Design

4.1 Overview

The four major components of GRETINA, detector modules, electronics, computing systems,
and mechanical systems, as well as the final system assembly, are described in this section.

The detector system will consist of 10 detector modules. A module includes three encapsulated
germanium (Ge) detectors assembled in a common cryostat with a liquid nitrogen dewar. Each
crystal has a tapered irregular hexagonal shape and will be electrically segmented with six
longitudinal segmentation lines along the flat tapered surface and five transverse segmentation
lines (36-fold segmentation). There are 2 different types of module, type A-B-A and B-A-B,
where the labels A and B refer to the two irregular hexagon shapes needed to tile a sphere of
120 hexagons and 12 pentagons. GRETINA will have 30 crystals and will cover 1t solid angle.
The adopted geometry provides an upgrade path to a future 4t array (40 detector modules).

The GRETINA detector module requires one pre-amplifier for each of the 111 segments. The
analog signals from the pre-amplifiers are input to Digital Signal Processing (DSP) modules.
The DSP module has a total of 10 analog inputs and one segmented Ge detector requires a total
of four DSP modules. The DSP board digitizes the analog information from the detector and
provides a first level trigger. The trigger and timing electronics is divided into local and global
subsystems. Local trigger and timing modules combine information from four DSP modules to
process a local level trigger for a segmented Ge crystal. The global subsystem combines
information from all 30 crystals, as well as any auxiliary detectors, and generates a global
trigger decision about the event. Timing information is generated initially at the global
subsystem level and then distributed to every local subsystem for local distribution to the DSP

The architecture for the computing system hardware is logically divided between front-end data
acquisition computers, data processing, and storage computers connected by high-speed
switched Ethernet. The front-end computers are either Compact-PCI or VME processor boards
with bus access to the detector electronics and gigabit Ethernet interfaces. A ‘farm’ of ~150
workstations receives event data from the front-end computers across gigabit Ethernet interfaces
and writes the processed event data to a network-attached mass storage device.

The software for the computing system is divided between event building and event processing
applications. The event building software assembles events from the DSP boards and forwards
the event to a workstation on the processor farm. The event processing software receives this
data and processes it in a queue for subsequent signal decomposition, tracking, and storage of
the resulting data.

The mechanical system consists of a detector support structure, a target chamber, and a system
to supply liquid nitrogen to the detector modules. GRETINA will employ a hemispherical
structure to support the detectors at a target-to-detector distance of 15 cm. A near spherical
chamber will hold multiple targets at its center, and the targets will be positioned by a remote-
control system. The chamber will accommodate various auxiliary detectors. The liquid nitrogen
system supplies and maintains the liquid nitrogen in the detector dewars.

System assembly is the assembly of the mechanical sub-systems, the installation of all Ge
detector modules, and the installation, configuration, and assembly of the electronics and
computer systems in a site at LBNL. It marks the end of GRETINA MIE construction.

4.2 Germanium Detector Modules
4.2.1 Detector Module Design

The basic detector unit for GRETINA is the triple crystal module. It consists of three
encapsulated Ge crystals placed in a single cryostat with a liquid nitrogen Dewar (figure 4.2.1).
The Dewar will have a holding time of at least 16 hours and will be connected to an automated
liquid nitrogen filling system. Each crystal will be segmented with six longitudinal
segmentation lines along the flat tapered surface and five transverse segmentation lines, which
separate the crystal into 6 unequal “layers” with thickness of 1.0, 1.2, 1.6, 1.8, 2.0, and 1.4 cm.
A crystal provides 37 signals (36 segments and one central electrode). These signals pass
through a feed-through on the encapsulation canister and are amplified by cold FET’s mounted
in the cryostat and go through another set of feed-throughs to exit the cryostat (a total of 111
signals per module). Additional feed-throughs are provided for high voltage and pulser input.
General crystal specifications are given in table 3.1.

Each detector is cut from a single cylindrical Ge crystal with a diameter of 8 cm and a length of
9 cm into one of two different tapered irregular hexagon shapes, labeled type A and B. The two
shapes are determined by the symmetry of the 120 hexagon geometry (see below) and a
GRETINA detector module is either a triple cluster of type A-B-A or type B-A-B crystals.
Figure 4.2.2 shows the dimensions and packing of the three Ge crystals in a type A-B-A
detector module.

The crystals have a taper angle of ~10 degrees between the axis and the center of a flat surface,
and the apex of the tapered surface is 15 cm from the front surface of the crystal. The first ~5cm
of the crystal are fully tapered, the next ~3 cm are partially tapered, and the last ~1 cm retains
the original cylindrical shape of the crystal (not tapered). This shape maximizes the distance
from the source to the detector, allows more space for an auxiliary detector in the target
chamber, and optimizes the germanium coverage. GRETINA will have a target to crystal
distance of 15 cm and a Ge thickness of 9 cm. The distance between crystals is 3.5 mm and the
distance between the crystal and the cryostat wall is 4.5 mm. The cryostat walls on the side of
the detector follow closely the tapered shape of the crystals and the taper continues until they
intercept the cylindrical part of the cryostat. This ensures that there is no interference from the
neighboring modules when one is removed from the array. The detector modules are mounted
on the support structure using a precision flange.


Figure 4.2.1. Upper. Drawing of the GRETINA 3-crystal module detector showing dimensions and layout of the
various components. Lower: Photograph of one encapsulated GRETINA prototype germanium crystal. This crystal
has a regular hexagonal geometry and 36-segments (the feed-throughs are visible).

LN Dewar

Figure 4.2.2. Cross–section projection of the three crystals in a type A-B-A detector module. Inter-crystal
dimensions and crystal-to-cryostat distances are shown. Dimensions are given with respect to the nominal modified
dodecahedron geometry that has been developed for the GRETINA crystal array. Crystal surfaces between adjacent
detector modules are parallel and are offset 5.0 mm from the nominal geometry plane as measured perpendicular to
the nominal geometry plane. The detector module outer housing or shell is offset 0.5 mm from the nominal
geometry and the crystal is offset 4.5 mm from the outer surface of the shell for a total crystal offset of 5.0 mm.
Crystal surfaces within a given detector module are parallel and are offset 1.75 mm from the nominal geometry
plane, resulting in a 3.5 mm parallel gap between crystals within a detector module. There is a nominal parallel gap
of 1.0 mm between the outer surfaces of the detector module shell.

4.2.2 GRETINA Geometry

The array geometry is based on a spherical geodesic design comprising 120 hexagons and 12
pentagons that tile the complete sphere. GRETINA detectors will occupy 30 of the possible 120
hexagon locations and have a solid-angle coverage of 1t. The design provides a close-packed
geometry to maximize efficiency and allows future expansion to a 4t coverage using the
present detector design.

The 120-hexagon geometry (figure 4.2.3) has two types of irregular hexagons; 60 of each type
A and B, arranged in 24 rings. Each ring contains 5 hexagons of the same type. Nominal
angles and lengths for GRETINA type A and type B hexagons are shown in tables 4.2.1 through
4.2.4. The labeling scheme for angles and lengths is given in figure 4.2.4. (Angular positions
(u, |) of the centre of the 120 hexagon locations are given in appendix A.)

Figure 4.2.3: The proposed GRETINA configuration based on the 120 hexagon geometry. In the picture the red
hexagons represent hex-A detectors, the green represent hex-B detectors. Each crystal is encapsulated and three
crystals share a common cryostat to for a detector Module. GRETINA will contain 30 crystals (10 Modules) and
will cover a solid angle of 1t.

1 2
4 5

Figure 4.2.4: An illustration of the lengths and angles of the hexagon geometry showing the labeling scheme used
in this document
Table 4.2.1 Angular dimensions for the type-A irregular hexagon.

Table 4.2.2 Angular dimensions for the type-B irregular hexagon.

¢ [Degree] u [Degree] q [Degree] Taper Angle
51.71 64.14 64.14 10.19
60.98 59.52 59.49 9.77
60.52 58.78 60.70 9.70
59.07 61.41 59.51 9.76
59.07 59.51 61.41 9.76
68.63 56.49 54.88 9.28

Table 4.2.3 Lengths of the sides and radii of the front surface of type-A hexagons used in the 120 hexagon

¢ [Degree] u [Degree] q [Degree] Taper Angle
57.37 62.36 60.26 9.84
60.98 59.49 59.52 9.77
68.63 54.88 56.49 9.28
60.52 60.70 58.78 9.70
57.37 60.26 62.36 9.84
55.12 62.44 62.44 9.85
Side Distance
Radius Distance
1-2 2.85 0-1 2.94
2-3 3.04 0-2 3.00
3-4 3.35 0-3 3.00
4-5 2.99 0-4 2.94
5-6 2.85 0-5 3.00
6-1 2.72 0-6 2.94

Table 4.2.4 Lengths of the sides and radii of the front surface of type-B hexagons used in the 120 hexagon

Figure 4.2.5: Two geometric configurations of GRETINA. The first configuration (top) is a closed packed
symmetric section of the sphere clustered tightly around 0 or 180 degrees with respect to the beam, with an angular
coverage of 17 degrees to 55 degrees. The second configuration (bottom) is an asymmetric section of the sphere,
consisting of the 10 neighboring clusters which better cover the angle between 17 and 101 degrees. The planar
projection of the sphere is shown on the right side of the picture.
Side Distance
Radius Distance
1-2 2.61 0-1 3.00
2-3 3.04 0-2 3.00
3-4 2.99 0-3 3.00
4-5 2.93 0-4 2.94
5-6 2.93 0-5 2.99
6-1 3.35 0-6 2.94

The 10 triple-crystal modules can be arranged in a number of ways depending on the physics
requirements. Two proposed configurations for GRETINA are shown in figure 4.2.5.

1. A symmetric arrangement clustered tightly around 0 or 180 degrees with respect to the
beam, with an angular coverage of 17 degrees to 55 degrees.
2. An asymmetric arrangement, which extends the angular coverage from 17 to 101 degrees.

The first configuration offers a more compact solution with the minimum edge effects, while
the second provides the capability to perform angular distribution measurements.

4.2.3 Detector Tests

Detector testing can be divided into two specific tasks. After delivery, acceptance tests will be
carried out to check the detector performance against the manufacture’s specifications.
Following these acceptance tests, a full characterization of the module and detailed comparison
with simulations should be performed. The tracking performance of the detector will be
characterized using gamma-ray sources and in-beam measurements. In particular, the module
should be thoroughly tested in “realistic” in-beam conditions, and the Doppler reconstruction
studied in detail. In addition, results on long-term reliability and detector annealing to recover
from neutron damage also will be obtained.

The proposed acceptance and characterization measurements are listed in appendix B.

4.3 Electronics

This section describes the pre-amplifier, digital signal processing modules, and trigger hardware

GRETINA requires high-bandwidth low-noise preamplifiers capable of preserving the position
information in the transient current signal from each detector segment, fast signal digitizer
modules capable of performing real-time calculations of energy and time and capturing a
portion of the digitized waveform for use in signal decomposition, and a flexible trigger system
to control data readout based on experimental requirements.

The major change in the design of this electronics system compared to previous gamma-ray
detector arrays is the extensive use of digital processing. Substantial effort has already gone into
developing a digital processing board and the proposed system is an extension of this past

4.3.1 Pre-Amplifier

An operational detector with pre-amplifiers installed will be delivered by the manufacturer. The
input transistor and feed back capacitor are also supplied by the manufacturer and mounted
inside the sealed detector. For noise reduction considerations the input transistors are cooled
along with the detector.

An alternative option being considered is to install custom designed pre-amplifiers as done with
a previous GRETA prototype II. The LBNL GRETINA pre-amplifier will consist of a charge
sensitive first stage and a differentiating second stage (figure 4.3.1.)

Figure 4.3.1 Pre-amplifier Block Diagram
This charge sensitive pre-amplifier is suitable for JFETs with noise levels below 0.7nV/\Hz.
The InterFET IF1331 input transistor installed by the manufacturer has approximately 2nV/\Hz
noise. Therefore the noise contribution by the pre-amplifer is small and can be disregarded.

The rise time of the pre-amplifier will be determined by internal wiring and, from experience
with the GRETA prototype II, the rise time from the GRETINA pre-amplifier signals will be
~30 ns, or less. This value is sufficient to satisfy the timing requirements given in section 3.

The pre-amplifiers are mounted in a circular arrangement, similar to our design used on the
GRETA prototype II, and all three crystals will share a common ring for amplifier mounting.
The pre-amplifier height will be reduced for GRETINA and closer board spacing will be used.
In the original design, to improve performance, the wiring to the input of the pre-amplifier was
connected directly to the amplifier without being fed through the motherboard. However, this
strategy may be reviewed in the GRETINA design.

4.3.2 Digital Signal Processing Module

The GRETINA DSP module is a combination digitizer and signal processing board. It accepts
10 inputs directly from the detector module pre-amplifiers and digitizes at a nominal frequency
of >75 MHz with >12 bits ADC precision. Signal-processing algorithms calculate the energy
and constant fraction timing in real-time through the use of an FPGA. Also, the DSP module
communicates through fast serial links with the trigger system and, when directed, transfers the
data from buffers within the FPGA to its own output FIFOs. This data is composed of the
processed data and a portion to the original waveform for use in complex signal decomposition.

The data processing algorithms planned for the DSP module are described below (figure 4.3.2
shows block diagrams of the DSP module and the data processing logic):

a) Leading Edge Discrimination (LED):
- y
= x
– x
- y
= (x
+ x
) + 2x
(repeat 4 times)
- Threshold comparison ÷ LED time
b) Constant Fraction Discrimination (CFD):
- y
= x
– x
- y
= (x
+ x
) + 2x
(repeat twice)
- y
= x
– f x
(constant fraction)
- Zero crossing comparison ÷ CFD time
c) Trapezoidal filter and energy determination
- y
= y
+ ( (x
+ x
) – (x
+ x
) )
- Maximum tracking ÷ energy
d) Pole-zero correction
- y
= x
+ I
- I
n =
n-1 +

f is an attenuation factor and t is the pre-amplifier time constant (in units of the sampling rate).

A successful prototype of this DSP module was designed, assembled, and tested. It has 8 input
ADC channels on the board and a VME readout. Figure 4.3.3 shows a photograph of this
module. Each channel has its own discriminator. For event building purposes, trigger output
primitives for each channel are available on the front panel as well as validate inputs for each
channel. There is also a global validation output and an external trigger input. In addition to the
processed data, a portion to the original waveform is captured and read out for use in complex
signal decomposition. Figure 4.3.4 shows an energy spectrum obtained with the prototype DSP
board and a Ge detector (1 crystal from a “Clover-type” detector) illuminated by a
Co source.
Because signal decomposition is carried out on a crystal-by-crystal basis it is advantageous to
process the information from a single crystal as if it were one unit. A GRETINA germanium
crystal requires 37 channels of digital electronics (36 signals from the outer segments and 1
from the central contact) and four DSP boards will be used to instrument a single crystal. This
arrangement is schematically illustrated in figure 4.3.5. These boards will have a local trigger
and readout control, described in the following section.


FPGA fifo VME/
Clock distribution &
local oscillator
Digital Inputs/Outputs
Gigabit Serial Interface
100 MHz
100 MHz clock
External synch
Control and monitor registers
10 Inputs

Figure 4.3.2: GRETINA Digital Signal Processor block diagrams. Upper: DSP board components. Lower: Data
processing logic.

Delay 1
Delay 2
Delay 3
Delay 4
LED Time
CFD Time
CFD Amplitudes

Figure 4.3.3: DSP Prototype Module with 8 ADC input channels

Figure 4.3.4: Spectrum of
Co obtained from the prototype 8-channel digitizer board. The energy was obtained
from the signal waveform applying a software trapezoidal filter described above and a software pole-zero
correction. The 1332 keV peak has a FWHM of 2.5 keV.


Figure 4.3.5: Schematic diagram showing four DSP boards used to instrument a single germanium crystal together
with the local trigger module and readout CPU (embedded processor)

4.3.3 GRETINA Triggering and Timing System

The main functions of the GRETINA triggering and timing system are to:

a) Determine when a trigger has occurred and notify appropriate DSP Modules.
b) Synchronize the timing information for the GRETINA electronics.

The trigger system has a maximum of 20 microseconds to notify the DSP Modules that a trigger
has occurred. After 20 microseconds the data will be flushed from the digitizer’s buffers and
will no longer be available for readout.

Trigger Module Description and Block Diagram

Since the main logic in the Trigger Module is based on a FPGA, it is a flexible module that can
be reprogrammed to serve multiple functions and use a variety of triggering algorithms. By
reprogramming the Trigger Module it can be configured as a Local Trigger Module, a Fan
In/Out Trigger Module, or as the Global Trigger module. Its auxiliary I/O will be used as
trigger signals to and from external devices.

Size Document Number Rev
Date: Sheet of
<Doc> A
1 1 Tuesday, October 07, 2003
Tr ansei ver
Tr ansei ver
Tr ansei ver
Tr ansei ver
Tr ansei ver
Tr ansei ver
Tr ansei ver
Tr ansei ver
cPCI I nt er f ace
Tr ansei ver
Conf i gur at i on
St at us LEDs
4 I nput s
4 Out put s

Figure 4.3.6; Trigger Module block diagram

As shown in the block diagram (figure 4.3.6) the Trigger Module consists of the following
sections. (We have chosen an example using a cPCI bus.)

- FPGA with its configuration Prom
- cPCI interface
- 9 parallel to serial transceivers
- 4 programmable input signals
- 4 programmable output signals
- A number of programmable status leds

The FPGA is programmed by the onboard configuration PROM when there is a power cycle or
a system reset. The PROM data can be updated over the cPCI bus from a configuration file
transferred over the internet. The cPCI interface provides the glue to interface with the cPCI
bus. It is possible that this logic may be included inside the FPGA.

Eight of the serial transceivers will communicate with either eight DSP Modules or eight other
Trigger Modules. The data coming into these eight serial transceivers will go directly into the
FPGA. Since the trigger commands going out of these eight serial transceivers are the same,
they all get their data from the same data bus. The other serial transceiver will communicate
with an upper level Trigger Module. Both its input and output data bus will go directly into the
FPGA. By modifying the FPGA program, the data going in and out of the transceiver can be

The 4 programmable input and output signals may be used to input or output external signals
from auxiliary devices. These signals may be used as trigger inputs or outputs, and/or can be
used to monitor activity inside the Trigger Module.

Trigger Flow

The trigger information will flow from each of the DSP Modules through two layers of Trigger
Modules and then into a Global Trigger Module. The Global Trigger Module will collect all of
the information and decide if a trigger has occurred. If the Trigger Module detects a trigger
condition it will send a trigger command back through all of the Trigger Modules to the DSP

The first-level trigger is done by the DSP Modules. When the charge in a single segment goes
above a lower level threshold, a first-level trigger occurs. The DSP Module adds up the number
of segments that have a first-level trigger within a programmable window, and sends this trigger
sum to the layer-1 Local Trigger Modules. Layer-1 consists of 15 Local Trigger Modules
connected to 120 DSP Modules and one other Local Trigger Modules connected to external
devices. Each Local Trigger Module collects data from 8 DSP Modules and passes its trigger
information to a layer-2 Fan In/Out Trigger Modules. Layer-2 consists of 2 Fan In/Out Trigger
Modules connected to 16 Local Trigger Modules and the Global Trigger Module. Each layer-2
Fan In/Out Trigger Module collects data from 8 layer-1 Local Trigger Modules and passes the
trigger information to the Global Trigger Module. The Global Trigger Module makes the
second-level trigger decision.

When the Global Trigger Module detects a second-level trigger it sends the trigger command to
the DSP Modules and the auxiliary devices through the Trigger Modules. When the DSP
Modules receive the trigger command they save the data from the relevant segments (containing
either net or transient signals), which is then readout by the crate controller and transferred to
the computer farm. The Trigger Modules may also be programmed to send the trigger
command out one of its auxiliary outputs as a pulse. This pulse may then be used by auxiliary
devices as a trigger signal.

Figure 4.3.7 shows how the DSP Modules, Local Trigger Modules, Auxiliary Modules and
Global Trigger Modules are interconnected. Each double arrowed line represents a serial
transmitter and receiver between modules. Information about first-level triggers flow up to the
Global Trigger Modules, and information about second-level triggers flow through the Trigger
Modules to the DSP Modules.

Serial Link between Trigger Modules

The serial link between Trigger Modules consists of a transceiver with a serial transmitter and a
serial receiver (figure 4.3.8). We are evaluating different transceivers with different data rates
with either an 8 bit or 16 bit data width. One being considered is the low latency transceiver
from Cypress (CYP15G0101DBX). One of the requirements in selecting the transceiver is that
that the latency between when a message is sent out, and when it is received is the same for all
data paths. In order to keep the latency the same all data paths must have the same length and
the proper transceiver must be selected.

Figure 4.3.7. Trigger flow diagram

A 16 bit transmitter consists of a parallel-to-serial conversion performed on the 16 bit input data
and a differential serial driver. The parallel data is clocked in with a reference clock. The 16 bit
parallel data is encoded into 20 bit using 8-bit/10-bit (8B/10B) encoding format. The resulting
20-bit word is then transmitted differentially at 20 times the reference clock rate. The receiver
performs the serial-to-parallel conversion on the input data, synchronizing the resulting 20-bit
wide parallel data to the extracted reference clock. A phase-locked-loop (PLL) is used to
extract the reference clock from the serial data stream. By using this extracted reference clock
and keeping all line lengths the same we can generate a synchronized clock that is common
throughout the GRETINA Data Acquisition System.
Digitizer Module 1

Digitizer Module 2
Digitizer Module 3
Digitizer Module 4
Digitizer Module 5
Digitizer Module 6
Digitizer Module 7
Digitizer Module 8



AUX Inputs
-Trigger In
AUX Outputs
- Trigger Out
- Timing Sync
Digitizer Module 57

Digitizer Module 58
Digitizer Module 59
Digitizer Module 60
Digitizer Module 61
Digitizer Module 62
Digitizer Module 63
Digitizer Module 64
Digitizer Module 65

Digitizer Module 66
Digitizer Module 67
Digitizer Module 68
Digitizer Module 69
Digitizer Module 70
Digitizer Module 71
Digitizer Module 72
Digitizer Module 113

Digitizer Module 114
Digitizer Module 115
Digitizer Module 116
Digitizer Module 117
Digitizer Module 118
Digitizer Module 119
Digitizer Module 120




Trigger Modules
Fan In/Out
Trigger Modules
Trigger Modules

Figure 4.3.8: Serial link between trigger modules

Trigger Logic

The Global Trigger Module makes the second-level trigger decision based on information from
the Local Trigger Modules and the Auxiliary Trigger Module. The trigger hardware is designed
to be sufficiently flexible to allow a wide variety of triggering methods by reprogramming the
Trigger Modules. The first trigger method to be implemented will be similar to the one used
successfully at Gammasphere. This section describes the details of this trigger method.

The triggering method is based on the multiplicity of first-level trigger events that are detected
in the DSP Modules. In this method the first-level triggers are summed in the Global Trigger
Module and are stored in a running trigger total. After a user programmable time period, with a
range of 1 to 20 µsec, the triggers get subtracted out of the running trigger total. When the
running trigger total goes above a user specified multiplicity, the Global Trigger Module sends
out a multiplicity trigger command. This technique creates a moving trigger window equal to
the programmable time period.

Figure 4.3.9 is an example of what the running trigger total in the Global Trigger Module looks
like. In this example the programmable delay is set to 2 µsec. At 1 µsec, a single trigger comes
in and is added to the running total. At 1.2 µsec, two triggers come in and get added to the
Transmitter D16
Receiver D16
Stream at
20 x
Serial Link with 100 MHz Reference Clock
running total so that the running total is now three. At 2.4 µsec, three triggers come in and get
added to the running total which is now six. If the multiplicity is set to five, a trigger happens
and the trigger command is sent out and propagates down to the DSP Modules. What happens
to the running total when a trigger occurs is still being evaluated, but for this example the
running total is cleared. If the multiplicity is set to 7, a trigger does not happen and the running
total goes back down to zero as each trigger expired. At 3 µsecs, the 2 µsec delay for the first
trigger is over, so 1 gets subtracted from the running total. At 3.2 µsec the 2µsec delay is over
for the two triggers, so they are subtracted from the running total and the running total goes to 3.
At 4.3 µsec, the 2 µsec delay is over for the three triggers, so they are subtracted from the
running total and the running total goes back to zero.

The waveform in figure 4.3.10 shows the timing of how the trigger sums and the trigger
command propagate through the trigger modules. This example shows the DSP Module
grouping together triggers that happen within a 200 nsec window and sending out the trigger
sum once every 200 nsec. The transfer of the trigger sum takes about 10 nsec plus some
propagation delay. The Trigger Modules in layer 1 and layer 2 perform the same task, adding
up the trigger sums from their eight inputs and sending out the total number of triggers. Since
the Local Trigger Modules and the Fan In/Out Trigger Modules receive trigger sums every 200
nsec, they will be performing the sum and output every 200 nsec. The Global Trigger Module
will be performing a sum on its eight inputs and using its running trigger sum to keep track of
the multiplicity.

In the example there are trigger sums coming from two DSP Modules which are added up in
Trigger Module 1 in layer 1. The sum is passed to Trigger Module 17 in layer 2. Since the rest
of the Local Trigger Modules in layer 1 have no triggers, Trigger Module 17 simply passes the
sum it received to the Global Trigger Module. When the Trigger Sum in the Global Trigger
Module goes above the multiplicity level it generates a multiplicity trigger command that is
propagated down to the DSP modules.

The time from when a DSP Module sends a trigger sum, to when it receives a trigger command,
is related to the programmable delay in the Global Trigger Module. So when the DSP Module
receives a trigger command it knows how far back it needs to look in its buffer for any first-
level triggers. If it has any first-level triggers the DSP Module save the data so that it can be
readout and sent to the computer farm.

Other triggers will be based on information from the auxiliary inputs and outputs on the Trigger
Modules and spare inputs on the DSP Modules. As stated above the hardware is flexible
enough to allow a number of auxiliary triggering methods by reprogramming how the auxiliary
inputs are used. Some of the options that will be available are:

- An auxiliary input on a Trigger Module can be used to gate the triggers
- An auxiliary input on a Trigger Module can be used to veto triggers
- A logical combination of the auxiliary inputs may be used to generate a trigger
- A spare analog input on a DSP Module may be used to generate a trigger
- An auxiliary output may be used to send out a pulse when a trigger has happened.

In the cases above, the module with an input transition will send a status command up to the
Global Trigger Module, which will use the data to determine if a trigger has occur.

Figure 4.3.9: Example of running trigger multiplicity in the global trigger module.

Figure 4.3.10: Timing of trigger multiplicities and commands
4.3.4 High Voltage

Trigger Timing Example – Multiplicity =
Level-1 Trigger sums
from Digitizer Module 1
Local Trigger 1
Sum sent to layer 2
Global Trigger sends
Trigger command to
Local Trigger layer 2
Local Trigger layer 2
sends trigger command
to Local Trigger layer 1
Local Trigger layer 2
sends trigger command
to Digitizer Modules
2 5 1 0 1
Level-1 Triggers sums
from Digitizer Module 2
1 3 0 2 0
3 8 2 1
Local Trigger 16
Sum from layer 2 to
Global Trigger
(Rest of Layer 1 = 0)
3 1 8
2 1
Global Trigger Sum with
a window of 2 usec
11 3 12 14
200 nsec
200 nsec
10 nsec
10 nsec
610 nsec
200 nsec?
250 nsec?
300 nsec?
Time in usecs
1 2 3 4 5
Multiplicity = 5
7 Multiplicity = 7
If Trigger If no Trigger
Stable, low current, low noise high voltage power supplies capable of providing up to 5000V,
will be required to bias the germanium detectors. Each crystal will have its own HV supply and
will include a bias shutdown circuit. Several commercial systems exist today that can perform
this task.

4.4 Computing System

The primary task of the GRETINA computing system is to reconstruct tracks of gamma-rays
incident on the array from the raw waveforms captured by the 1100 digitizer channels which
instrument the array of 30 segmented Ge crystals. This is done by first decomposing the time
correlated waveforms from the 36 segments of a given crystal into a linear combination of basis
signals to identify the position of each scattering site. Next, the positions and energies of the
scattering sites among the different crystals are time correlated so that a tracking algorithm,
which clusters and orders the calculated scattering sites, can determine the number, energy, and
path of gamma-rays which strike the array.

For this task the GRETINA computing system must perform a number of functions. Real-time
data acquisition software is required to readout and assemble events from the individual
digitizer cards and forward the data to a farm of commodity processors which run the
decomposition algorithm, tracking algorithm, and performs global event building. Following
gamma-ray reconstruction from the computing farm, a system will be implemented to monitor
and archive the processed data from the spectrometer. Slow controls are also required for the
digitizers, trigger system, and reconstruction software running on the farm.

In this section we will first present an overview of the computing hardware of the GRETINA
computing system. This will be followed by a description of the decomposition and tracking
algorithms and their computational requirements as currently implemented. Also included will
be a brief discussion of the online data analysis, data archiving, and slow controls.

4.4.1 Hardware

The computer hardware consists of one single-board ‘read-out’ computer for each digitizer
crate, a processor farm of about 75 dual processor machines, a gigabit Ethernet switch for
connecting the read-out computers to the farm, and a network-attached storage device for
storing the processed data. Figure 4.4.1 gives a schematic overview of the computing hardware.

The computer hardware is, for the most part, available today from multiple vendors. The
exception is the processing farm where we assume a factor of 4 increase in processor
performance to enable signal decomposition to be performed with 150 CPUs. We expect to see
this needed improvement by the time the processors are purchased.


Figure 4 4.1 Schematic of the GRETINA computing hardware. Data rates are based on the assumption of ~20,000
gamma-rays per second in the processing farm.

Computer system software such as the operating systems, the tape backup utility, and cluster
management tools should be chosen with the same objectives as the hardware. Industry standard
operating systems such as Linux and VxWorks are common in a laboratory environment and
familiar to the support staff.

The computer hardware consists of the following items:

Read-out computer:
There will be a standard VME or cPCI crate containing the digitizers, trigger module, and a
single board computer for readout and control. The read-out computer performs the first level of
event building; it reads segment event data from each digitizer over the bus (e.g. a rate of 40
MB/sec can be achieved with VME64), sorts the data in time-stamped order, and sends the
resulting data to a processor in the decomposition farm via a Gigabit Ethernet switch. For these
purposes the computer should be configured with at least 2 processors and an on-board gigabit
Ethernet interface. An example is the SBS Technologies VG5 with dual PowerPC processors
and 2 gigabit Ethernet interfaces.

The read-out computer will be configured with an off-the-shelf real-time OS, such as VxWorks
from Wind River Systems, and with the EPICS software for communicating with the slow

Data Acquisition System Schematic
Aux. Det. Trigger
Processing Farm
30 Crystals
Data Storage
2.2 MB/s
Aux. Det. Data
66 MB/s
75 dual Processors
Workstations, Servers
6.9 MB/s
2.3 MB/s + Aux. Det. Data
Processing farm:
The nature of GRETINA experiments simplifies the problem of the workstation farm
configuration. The event data coming from each crystal can be processed independently of the
others. There is no requirement to assemble a general parallel processing compute farm, only a
loosely coupled network of computers that can read and write shared mass storage is required.

A network connected but densely packaged array of about 150 processors will be required for
signal decomposition and tracking. These processors will need fast CPUs, sufficient memory,
and dual Ethernet interfaces. For ease of maintenance and for added reliability, they should be
configured as diskless to boot from a central server.

There are many strategies for building up the processor farm. The solutions range from building
our own to ordering the farm delivered and configured from a vendor. We plan to utilize the
experience of the LBNL computer support staff for the procurement and assembly of the farm.

The design used for the cost estimate assumes that standard computer motherboards with dual
processors that can be packaged in 1U. About 40 can be installed in one 19" rack requiring two
racks for the 75 computers. We estimate each computer uses about 2 amps requiring
approximately 100 amps per rack. For console interaction with these workstations, 2 Cyclades
TS Series console switches, which are each capable of switching between 48 consoles in a 1u
package, can be used.

A Gigabit Ethernet switch will connect the 10 read-out computers to the 75 processors in the
farm. A Cisco Catalyst 4506 with 2 ea. 48-port 10/100/1000BASE-T, and 64 Gb/s switching
capacity would be suitable for this purpose as well as for connecting the processor farm to the
mass storage.

Data storage:
The maximum data rate from the farm to storage is estimated to be less than 10 MB/sec totaling
about 4.3 TB of data following about 5 days of continuous running.

We propose using a network attached storage device for the primary storage. These devices are
designed for high availability/reliability and high performance for applications in data centers.
Our experience with one vendor, Network Appliance, has been positive in these regards. Lower
cost devices of this kind are becoming available due to the competition in the data storage area.

Data Export:
There will be provisions to export experimental data to either portable hard drives and/or high
density tape medium before erasing it from the primary data store. Portable drives of adequate
density should be available by the end of this project.

Development Server/Storage:
A server will be required for software development. The storage for software development must
be highly reliable. Establishing a backup regimen early is crucial. We propose using a separate
network attached storage device and an attached tape backup unit for software development

Boot server:
A workstation will be dedicated for remote booting the digitizer and computer farm operating

Application server:
This workstation will host development tools such as compilers, CVS, MySQL database, and
Web server software. Since this server will store our source code we want high reliability and a
good backup process.

Operator Console:
This workstation will be used to monitor the progress of the data collection and for slow control

4.4.2 Software Event Building

Event building in GRETINA refers to the assembly of data from a single reaction within the
spectrometer from the 1100 digitizer channels, which instrument the 30 Ge crystals. Events are
assembled according to a common 48-bit timestamp with 10 ns resolution that is distributed to
all digitizers. In the GRETINA system there are three levels of events; segment events, crystal
events, and global events, requiring two software-based event builders.

The lowest level events are segment events that are produced by the 10-channel digitizer cards.
Segment events contain information concerning a hit in a Ge detector segment which includes
the energy deposited in the segment, a timestamp generated by the global trigger, a constant
fraction time referenced to the timestamp and the relevant portion of the signal trace for
decomposition. In order for these traces to be decomposed into charge deposition locations, one
needs to assemble all segment events from a crystal for a given timestamp range. This will be
done by an embedded processor (readout cpu described in 4.4.1) which reads the four digitizer
boards corresponding to a single crystal and time orders the data. This time ordering is required
as the digitizer boards buffer several thousand segment events that are read out as a block by the
embedded processor. The time-ordered data that a single crystal produces is divided into fixed
length packets by timestamp range and dispatched to available nodes in the decomposition farm.
Data within a short predetermined window is considered coincident and is termed a crystal
event. Signal decomposition is a trivially parallel process in that one only requires information
from a given crystal (not the entire array) to properly decompose the signals.

Following signal decomposition, one has the calculated positions and energies of charge
deposition sites within a given Ge crystal as well as a timestamp derived from the crystal event.
In order to carry out the next step, tracking (which groups interaction points from multiple
crystals corresponding to individual gamma-rays and assigning a scattering order), one must
time correlate the data (crystal events) corresponding to individual Ge crystals distributed
among the ~120 nodes of the decomposition farm into global events. Global events also contain
data from auxiliary devices coincident with the timestamps derived from the crystal event. The
construction of global events will be carried out by one or more processors which are fed by the
decomposition farm. These global events are then forwarded to computers that carry out the
tracking algorithm that adds grouping/ordering information to the global event before it is
archived. Signal Decomposition

Each pulse shape can be the result of multiple interactions, at different locations, each with
different energy deposition. In order to identify all the interaction positions every pulse shape
must be decomposed into the individual components resulting from each interaction. This is the
process called `signal decomposition’.

The Current Decomposition Algorithm
A signal decomposition algorithm has been successfully developed and implemented for
determining the positions and energies of multiple interactions based on a least squares
Event Building Data Flow Diagram
Crystal Event Builder
Segment events
Crystal events
Signal Decomposition
Interaction points
Global Event Builder
1-36 segments
1-30 crystals
Global Events
Data from
minimization procedure. An illustration of the signal decomposition process is presented in the
figure 4.4.2.

Figure 4.4.2: Demonstration of the decomposition of signals due to multiple interactions into their individual
components in a coaxial detector with 6 fold azimuthal segmentation. The upper right indicates grid positions
(green squares: A=4 mm) where signals - shown on the left for A=8 mm (blue triangle) are calculated. Assuming
three interactions (at the locations of the red stars on the upper right figure and which result in the red signals in the
lower left figure) the minimization procedure is able to find the closest grid points to the interactions. The resulting
signals for three segments are shown in green and can hardly be distinguished from the input red signals. The lower
right figure shows the amplitudes (proportional to the area of the squares) obtained for all grid points found to
contribute to the signal.

The set of basis pulse shapes (upper left of figure) representing single interactions are calculated
at grid points throughout the detection volume (upper right of figure) so that pulse shapes
involving one or more interactions may be represented as:

) , , , ( ) (
t z y x X E t q k k k k k

where, q(t)
is the calculated charge on segment electrode s at time t. The basis functions X
the calculated pulse shapes, which result from energy deposition at a single location (x, y, z).
Each basis function is weighted in the summation by E
, which represents the fraction of the
total energy deposited during the interaction, k.

For each event an initial estimate of the interaction positions is made based on the energy
distribution and pulse shapes measured in each segment, which are then compared with the
calculation. A figure of merit, χ
, is calculated and this value along with the initial estimates of
adjustable parameters such as the fraction of energy deposited, E
, and the position (x,y,z)
each interaction, and the total number of interactions, M, is passed to a minimization routine.
The χ
is then minimized via the iterative adjustment of these parameters to find the global
minimum. There is a very large parameter space over which to search when multiple
interactions occur in a segment and this is the reason that the signal decomposition procedure is
the most computationally intensive part of the data analysis.

An example of calculated versus measured pulse shapes (from a simulation) after the
minimization procedure is shown in the lower left panel of figure 4.4.2. The deduced positions
and energy at each interaction point is shown in the lower right panel of the figure. Interactions
occurring between basis grid points result in a sharing of energy between points at the level of
the position resolution (~2 mm) and can be taken into account by the tracking algorithm

Computational Requirements
Signal decomposition is the most computationally intensive part of the data analysis for
GRETINA. To achieve ~2 mm position resolution for a gamma ray with multiple interaction
points within a segment, the current least squares minimization algorithm requires 0.016s of
CPU time with a 1.9 GHz P4 processor. Advances in microprocessor technology and
improvements in the algorithms should provide significant gains in performance. Assuming a
conservative factor of four gain over today’s CPU speeds and given a minimum requirement
that we are able to decompose ~20,000 gamma rays per second at maximal position resolution
requires approximately 125 CPU’s. Tracking

Once the energies and positions of all interaction points in an event have been identified
through the signal decomposition process, the gamma rays can be reconstructed by application
of a `tracking’ algorithm. The tracking algorithm must:

i) Identify and sequence interactions that belong to individual gamma rays;
ii) Distinguish gamma rays that deposit their full energy in the detector from those that
deposit only some fraction their energy before escaping;
iii) Determine the first and second interaction points for each gamma ray.

The Current Tracking Algorithm
An algorithm has been successfully developed and implemented to fulfill the above criteria for
gamma rays in the energy range of main interest (between 150 keV and 5 MeV) in most nuclear
structure experiments. It consists of three steps, which are described below and illustrated in the
flow diagram.

Step 1: Interaction points are identified within a given angular separation, as viewed from the
target, and grouped together into a cluster.

Step 2: The Compton scattering formula is used to determine the most likely scattering
sequence from the position and energy of the interaction points. If the cluster contains the
interactions belonging to only one gamma ray then the deviation from the Compton scattering
formula will be small. The finite position and energy resolution will mean some clusters are

Step 3: Candidates for misidentified clusters are re-evaluated after splitting or adding groups of
interaction points.
Position and Energy of Interactions
From Signal Decomposition
Cluster Identification
Based on Angular Separation
Cluster Evaluation
Using Compton and Pair-
production Tracking
Good Add Split Bad
Gamma Rays Reconstructed
Energy, Interaction Points, and Scattering
Flowchart of Tracking Algorithm

The algorithm not only separates and identifies multiple gamma rays from an event but also
gives the time sequence of the interactions for each gamma ray. The first interaction point is
required for accurate Doppler correction while the additional information from the second
interaction point allows the linear polarization of a gamma ray to be determined.

Computational Requirements
To investigate the performance of the tracking algorithm simulations for tracking an event of 25
coincident 1.33 MeV gamma rays have been performed. This represents the most challenging
situation that could occur in experiments with GRETINA and most applications will be
significantly less complex. Requiring a position resolution for the interaction points of 2 mm
and using an angular separation parameter of 8

degrees, it takes 0.01 seconds of CPU time with
a 700 MHz PIII processor to track such events. This is significantly less time than that required
for the signal decomposition process. Assuming the same minimum requirement as that for the
signal decomposition of tracking ~20,000 gamma rays per second (or ~1000 events per second)
requires approximately 10 CPU’s (<10% of the computational resources required by the
decomposition algorithm). Online Data Analysis and Data Archiving

As with earlier gamma-ray spectrometers such as Gammasphere, experimenters will carry out
both on-line and off-line analysis of the data. The on-line analysis system provides rapid
diagnostics to catch and correct problems early in the experiment while complete analysis of the
data is done offline at the Users home institution. Offline analysis requires GRETINA to
provide a data archiving facility that will allow users to extract their data from the system in a
wide variety of media formats. This section will describe both the online-analysis and data
archiving systems for GRETINA.

Online Data Analysis
Online data analysis is required to provide diagnostic information for the spectrometer and basic
physics analysis tools. Traditionally, such systems are divided into two parts; a histogrammer,
where predetermined histograms of interest are defined incremented, and a display that provides
access to these histograms and allows experimenters to carry out simple analysis tasks. For
GRETINA, the histogrammer will examine the data written to the disk array and create a series
of 1D and 2D histograms for a variety of parameters such as detector hit pattern, TAC spectra,
and various energy spectra. Also included will be spectra to monitor external devices. While
there will be a subset of histograms that are fixed, several custom histograms are usually
tailored to meet the specific monitoring requirements of any given experiment. These tailored
histograms can be fixed or dynamically extracted from special purpose databases.

The software for display and analysis of histograms should be such that multiple users may
remotely access the histogrammer though the network. Requirements include the ability to
compare spectra, multiple peak fitting, and the ability to take arbitrary gates in multi-
dimensional histograms. There are several software packages that could serve as the basis for
such a system; e.g., the Root framework, which is being used to construct a new online analysis
system for Gammasphere at ANL. Our current plan is to adapt this work for use with

Data Archiving
GRETINA will write its data directly to a disk array rather then a tape system. For the
conceptual design we assume a data rate from the array of 10 MB/s for an experiment of
duration 5 days. This produces a dataset 4.3 TB in size. We expect this to be an upper limit on
the single experiment storage requirement and defines the size of the disk array.

A key requirement of the disk array is reliability as failure could result in the loss of an
experiment. Excellent hardware reliability can be achieved through use of 2 RAID disk arrays
that have a failover mechanism such as those produced by Network Appliance. However, a
more complex problem is to prevent data loss through User error. One approach would be to
automatically archive all data to a tape robot so such errors are recoverable. The capacity of the
tape robot would minimally be that of two experiments requiring roughly 10 TB. Alternatively
(or additionally) a software/administrative procedure could be developed to ensure data is
written another source before deletion from the disk array.

A key requirement is to provide the data collected during the experiment to the User in a wide
variety of media formats. This will be done though an archiving server which will support a
number of tape and optically based devices. Backups will be done from the disk array though a
set of scripts running on the archive server. Facilities will also be provided to back-up data
directly to disks provided by Users or though the network to their home institutions. Network
access to the Users data could also be carried out through protocols developed for computing
grids that we plan to investigate. GRETINA Slow Controls

A control system will be needed for operating and interacting with the GRETINA spectrometer.
The areas handled by this system include configuration of the electronics, monitoring of status
and alarm conditions, and archiving for fault analysis. The response rate of this system will be
on the order of 100 ms, but will need to support up to 10,000 channels including the 1200
channels of detector ADCs, high voltage power supply status, the status of the LN system, and
any configuration data needed to configure the experiment. In addition, this system needs to
display the status of the ongoing data collection.

These types of control systems have been implemented successfully by many laboratories
(including LBNL for Gammasphere) with the EPICS control system software originally
developed by LANL. This software is actively supported by a community of national
laboratories and is designed as a client-server model with the server running on a real-time OS,
such as VxWorks, and the client applications running on workstations configured with one of
many desktop operating systems (e.g. Linux, Windows, Solaris). For GRETINA, the EPICS
server software will be installed on the 10 embedded processors in the digitizer modules. The
client applications such as the Event Logger and Alarm Handler, and Display Manager will be
installed on the workstation used for operating the experiment. From much experience with this
software, the typical load on a 64000 class CPU scanning 1000 channels at 100 ms is about
30%. On today's gigahertz PowerPC processors the CPU load will be proportionally smaller.
Many of these newer embedded processor boards are configured with multiple on-board 100
BaseT Ethernet interfaces one of which could be dedicated to the slow controls data. Using this
second interface will minimize the contention on the Gigabit Ethernet used to dispatch the
crystal event packets.

4.5 Mechanical System

The GRETINA mechanical system consists of three main sub-systems: the support structure,
the liquid nitrogen distribution system, and the target chamber system.

4.5.1 Support Structure

The principal geometry of the detector array is a dodecahedron. The geodesic sphere consists of
12 perfect pentagons and 120 irregular hexagons. There are 2 unique hexagons designated type
A and type B. There are 60 hexagons of each type and 3 hexagon crystals will be grouped
together in a cluster to form a detector module. There will be 2 different types of module: A-B-
A and B-A-B. Hence the entire 4t array would consist of 40 detector modules. GRETINA will
consist of 10 detector modules or one quarter of the entire possible array.

The primary function of the support structure is to position the gamma-ray detector modules in
a precise spherical array around the target chamber. It provides a means for installing and
removing detector modules with minimal effort and allows access to the target chamber for
changing targets and installing auxiliary detectors. The support structure must be sufficiently
rigid so that any deformation due to its weight combined with the weight of the detector
modules does not cause interference between adjacent detector modules. Looking to the future,
the deformation of the support structure must also be such that adequate clearances are
maintained with detector modules attached to an identical support structure when the two arrays
are mated around the target chamber. Finally, the support structure accommodates the liquid
nitrogen distribution system as well as the electrical cables from the detector modules to the
electronics racks.

The conceptual design of the GRETINA support structure is based upon that used for
Gammasphere. It consists of a monolithic hemisphere capable of accommodating 20 detector
modules (figure 4.5.1). The hemisphere is fabricated from a forged aluminum disk and can be
machined to very accurate tolerances using conventional machining practices. It is on the order
of 2” thick, 48” in diameter, and weighs approximately 1000 lbs. The detector modules weigh
approximately 100 lbs each; hence the weight of one fully loaded hemisphere is on the order of
3000 lbs. The hemisphere is supported by a steel weldment. The Gammasphere support
structure oriented the axis of the hemisphere transverse to the beam and each hemisphere could
be translated orthogonal to the beam a distance of 30” to provide access to the target chamber.
The GRETINA support structure can orient the axis of the hemisphere coincident or orthogonal
to the beam. The support structure is mounted on precision rails to translate the hemisphere to
provide full access to the target chamber. Gammasphere had a system to rotate each hemisphere
+/- 90 degrees to allow the heavy yet fragile detectors to be installed and removed horizontally.
The GRETINA support structure as proposed is also capable of rotating the hemisphere +/- 90
degrees. The GRETINA detector modules are heavy and awkward to handle. They are also
fragile and valued in excess of $500K each, and hence it is advantageous to include a rotation
system in the final design so that detector modules may be installed and removed horizontally.

Figure 4.5.1: Schematic diagram of the GRETINA support structure

A rotation/translation system for GRETINA would utilize various fail-safe limit switches,
position sensors, and hard stops to assure that the hemisphere could not be rotated unless they
had been first translated to a safe distance from the beam-line and target chamber (similar to
those used in Gammasphere). Clearances between the detector modules and the target chamber
are typically on the order of 0.25”, hence the precision and accuracy of this system need not be
excessive. However, if a second hemisphere is added in the future, then the 0.040”clearance
between detector modules from the two mating hemispheres must be taken into consideration.

4.5.2 Liquid Nitrogen Filling System

The Ge detectors that make up GRETINA will generally be maintained at liquid nitrogen (LN)
temperature, except during shipment and repair. GRETINA will comprise ten detector modules,
each with their own LN dewars. This is significantly fewer than Gammasphere so the filling
system can be correspondingly smaller. The LN filling system must be highly reliable and
carefully interlocked with the temperature monitoring and high-voltage control systems. Should
a cluster of detectors warm significantly above liquid nitrogen temperatures while the detectors
are biased, the field-effect transistors of the preamplifiers are likely to be damaged,
necessitating costly repair.

Figure 4.5.2 Schematic layout of the GRETINA LN system

The GRETINA liquid nitrogen filling system includes three subsystems:

1. Liquid nitrogen hardware; this includes LN tanks, cryogenic valves, plumbing, temperature
and pressure sensors, and "bayonets" to fill the detector module dewars.
2. Computer control hardware, including one or more computers, digital output registers and
solenoids to control valves, and analog input devices to digitize the output of temperature and
pressure sensors, etc.
3. Computer control software and user interface to the filling system.

The GRETINA LN system will be similar to that of the CLARION array at ORNL. The
computer-controlled system will automatically initiate fill cycles at preset time intervals.
Resistive temperature devices (RTDs) will be used to detect overflows of liquid nitrogen and
thus determine that detectors have been successfully filled. RTDs incorporated into the detector
modules themselves will also be continually monitored; should a detector module begin to
warm up, the bias voltages for those detectors will immediately be shut off and an alarm
generated. Alarms will also be generated as a result of fill cycles which fail for any reason, and
by off-normal values of other monitored parameters (such as fill line pressure, environmental
temperatures, etc.) A watch-dog timer will be incorporated into the control system to
immediately shut off voltages and disable fills should the monitor/control processor fail for any
reason. Oxygen level sensors will be installed to sound an alarm, and close the master LN
valve, should a LN line rupture and result in depleted oxygen levels.

Several different user interfaces to the LN system will be made available. For local use, a
sophisticated user-friendly graphical interface will facilitate use of the system. An additional
simple command-line interface will however also be available for remote access over the
network, removing any requirements for a particular remote operating or windowing system.

4.5.3 Target Chamber System

The GRETINA target chamber system will combine several features of the Gammasphere and
Microball target chambers with provisions for operation in a stand alone mode (GRETINA
only) and in conjunction with a variety of auxiliary detector systems. For these purposes easy
access to the interior of the chamber will be important. An adequate number of low mass ports
for signal or control cables should be provided in a way that avoids as much as possible placing
mass in front of the Ge detectors.

The target chamber will be suitable for housing a target assembly that will allow experiments to
be done with minimal degradation of the GRETINA performance from absorption and/or

The target chamber system includes three subsystems:

1. The vacuum vessel itself, which consist of a vertical cylindrical section. This couples to the
entrance and exist beam lines. An additional support structure along the pentagonal position in
the vertical plane will provide strength and allow the target latter motion. The chamber will be
closed with two “hemispheres” (bowls) with a concealed O-ring. Two dowel pins and the
vacuum will support the “hemispheres” (no screws). The vacuum requirements for the chamber
are on the order of 10
torr, hence organic o-ring seals are acceptable.

2. The target holding assemblies. One of these will be similar to the Gammasphere target
holder (capable of supporting multiple target frames) and the other will be inserted from a top
entrance port that will permit a dual target frame to be inserted in a 4t charged-particle detector

3. The target remote control actuator. This will be similar to one used in Gammasphere.

4.6 System Assembly

System assembly involves the final assembly of the mechanical sub-systems, the installation of
all Ge detector modules, and the installation, configuration, and integration of the electronics
and computer systems.
System assembly will occur during the last two quarters of GRETINA construction at LBNL.
At this point, all mechanical components will have been designed, fabricated, procured,
inspected, and received. The detectors will have been received, tested and characterized and
ready for installation. The electronics will have been developed, fabricated and tested.
Computer programming will have been completed, and the computer hardware procured,
configured and loaded.

A suitable location will be identified and secured at LBNL for system assembly and testing.
This location will most likely be an experimental cave at the 88” Cyclotron. All of the LBNL
supplied mechanical subsystems will be assembled including the support structure, the
hemisphere, the support structure translation system and the hemisphere rotation system. All of
the associated pneumatics, electro-mechanical devices, wiring harnesses, and motion control
devices shall be completely installed and connected. The liquid nitrogen distribution system will
be mounted to the structure and connected to the facility liquid nitrogen supply and exhaust
lines. The ten Ge detectors will be installed and tested for electrical isolation. Electronics cables
will be run from the electronics racks to the individual detectors, secured to the hemisphere and
connected to the detectors. Individual liquid nitrogen supply and return lines will be run from
the manifolds to the individual detector dewars. The liquid nitrogen supply system will be
activated and will be used to fill the detectors throughout the duration of the system tests. The
target chamber will be mounted to the hemisphere and prepared for testing using radioactive

The electronics will be installed in the crates, mounted in the racks and interconnected to the
detectors and the switch based event builder. The processor farms and storage system will be
configured and interconnected. The subsystems will be tested for full functionality.

A series of performance tests will be conducted to compare actual results with predicted.
Calibration tests will be carried out and the parameters for proper operation measured and
recorded. Following the final commissioning of the system, the last DOE review will be carried
out and the last project milestone will be completed. This will represent the end of GRETINA
Appendix A: Angular positions of the hexagonal crystals in the 120 hexagon

Ring u (degree) Crystal Type | (degree)
1 16.91 A 36.00 108.00 180.00 252.00 324.00
2 30.69 B 0.27 72.27 144.27 216.27 288.27
3 35.46 B 35.43 107.43 179.43 251.43 323.43
4 47.38 A 59.12 131.12 203.12 275.12 347.12
5 48.44 B 13.32 85.32 157.32 229.32 301.32
6 54.85 A 35.61 107.61 179.61 251.61 323.61
7 64.27 A 71.46 143.46 215.46 287.46 359.46
8 68.19 B 19.30 91.30 163.30 235.30 307.30
9 74.33 A 38.74 110.74 182.74 254.74 326.74
10 79.48 A 58.26 130.26 202.26 274.26 346.26
11 82.71 B 5.76 77.76 149.76 221.76 293.76
12 86.94 B 24.91 96.91 168.91 240.91 312.91
13 93.06 B 44.35 116.35 188.35 260.35 332.35
14 97.29 B 63.50 135.50 207.50 279.50 351.50
15 100.52 A 11.00 83.00 155.00 227.00 299.00
16 105.67 A 30.52 102.52 174.52 246.52 318.52
17 111.81 B 49.96 121.96 193.96 265.96 337.96
18 115.73 A 69.80 141.80 213.80 285.80 357.80
19 125.15 A 33.65 105.65 177.65 249.65 321.65
20 131.56 B 55.94 127.94 199.94 271.94 343.94
21 132.62 A 10.14 82.14 154.14 226.14 298.14
22 144.54 B 33.83 105.83 177.83 249.83 321.83
23 149.31 B 68.99 140.99 212.99 284.99 356.99
24 163.09 A 33.26 105.26 177.26 249.26 321.26

Appendix B: Proposed detector acceptance and characterization tests

1. Acceptance test
1. Measure energy resolution (at 4 µsec time constant).
2. Measure absolute efficiency and peak-to-total ratio of the individual crystal and in 3-
crystal summing mode.
3. Measure bandwidth of preamp output signal.
4. Measure noise power spectrum (0 – 35 MHz).
5. Measure time resolution.
6. Measure cross talk between channels.
7. Measure depletion voltage.
8. Measure detector cooling down time, and dewar holding time.
9. Verify that cooling is sufficient to keep detector and FET at design temperatures.
10. Verify that temperature signal and HV shut down signal are functioning.
11. Verify heater is functioning.
12. Verify the mechanical dimensions of the cryostat and mounting flange.

2. Full characterization
a) Pulse shape measurement – scan with collimated sources
1. Singles measurement – 2D scan to confirm detector geometry and efficiency from both
central contact and segment signals.
2. Coincidence measurement – confirm calculated signal shapes which are determined by:
a. Segment geometr
b. Electric field distribution, and impurity concentration
c. Crystal orientation, and drift velocity variation.
d. Possible region of incomplete charge collection.

b) Source measurement with full analysis*
- Determine efficiency and P/T to confirm signal decomposition and tracking

c) In-beam measurement with full analysis*
– Confirm Doppler correction ability.

(* to be performed for the first few modules)