Progress report on the SuperBigbite Project

Response to the recommendations from the secondTechnical Review

with contributions from Evaristo Cisbani, Ole Hansen, Mark Jones,Mahbub Khandaker, Nilanga Liyanage, Vahe Mamyan,

Vladimir Nelyubin and Igor Rachek

edited by Gordon Cates, Kees de Jager, John LeRoseand Bogdan Wojtsekhowski

for the SuperBigbite collaboration

July 13, 2011

Contents

1 GEM quality, GEM spacers, Manpower and Cost of the gas system 61.1 GEM foils Quality Assurance and acceptance criteria . . . . . . . . . . . . . 61.2 GEM foils spacer strips . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81.3 Technical personnel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91.4 GEM gas-handling system cost . . . . . . . . . . . . . . . . . . . . . . . . . 11

2 GEM UV readout and Number of samples 122.1 The UV strip configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122.2 The 3-sample readout . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

3 The low-energy photon response of the GEM chambers 16

4 SBS magnet and optics 184.1 Acquisition of 48D48 magnet and associated equipment . . . . . . . . . . . . 184.2 SBS Magnet Powering Options . . . . . . . . . . . . . . . . . . . . . . . . . 194.3 The field calculation and spectrometer optics . . . . . . . . . . . . . . . . . . 19

5 The Lead-glass Calorimeter 255.1 Calculations of the BigCal energy and position resolution for a 20 cm Al absorber 255.2 Effect of radiation damage on the BigCal energy and position resolution . . . 265.3 Effect of energy resolution on the BigCal trigger rate . . . . . . . . . . . . . 275.4 Test of UV Curing rate in BigCal . . . . . . . . . . . . . . . . . . . . . . . . 28

6 GEM module design 296.1 Bandwidth considerations between HCAL FADCs and Trigger Processor . . 296.2 Multi-mode readout of the APV25S1 chip . . . . . . . . . . . . . . . . . . . 306.3 Mechanical details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

6.3.1 Routing of the cables . . . . . . . . . . . . . . . . . . . . . . . . . . . 326.4 Signal quality with long cables . . . . . . . . . . . . . . . . . . . . . . . . . . 36

7 The Coordinate Detector 427.1 Coordinate Detector vertical position resolution . . . . . . . . . . . . . . . . 427.2 Determining the track search region in SBS . . . . . . . . . . . . . . . . . . . 43

1

8 MC generation of the GEM data 478.1 Front Tracker simulation framework . . . . . . . . . . . . . . . . . . . . . . . 478.2 Monte Carlo simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

8.2.1 Digitization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 498.2.2 GEM multiplication . . . . . . . . . . . . . . . . . . . . . . . . . . . 528.2.3 Charge Collection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 538.2.4 Pulse Formation and Timing . . . . . . . . . . . . . . . . . . . . . . . 568.2.5 Digitization Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . 58

8.3 Charge Collection Implementation . . . . . . . . . . . . . . . . . . . . . . . . 618.4 Data structure implementation . . . . . . . . . . . . . . . . . . . . . . . . . 62

9 Tracking Efficiency 649.1 The GEM tracker concept . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

9.1.1 The COMPASS GEM tracker . . . . . . . . . . . . . . . . . . . . . . 649.1.2 The SBS GEM tracker . . . . . . . . . . . . . . . . . . . . . . . . . . 65

9.2 Intensity of the background particles . . . . . . . . . . . . . . . . . . . . . . 659.2.1 The experimental layout . . . . . . . . . . . . . . . . . . . . . . . . . 669.2.2 Results of Monte Carlo simulation . . . . . . . . . . . . . . . . . . . . 67

9.3 Event generation and data digitization . . . . . . . . . . . . . . . . . . . . . 699.4 Track Reconstruction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

9.4.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 709.4.2 Simulation Decoder . . . . . . . . . . . . . . . . . . . . . . . . . . . . 719.4.3 GEM Cluster Finding . . . . . . . . . . . . . . . . . . . . . . . . . . 719.4.4 Hit pattern Construction . . . . . . . . . . . . . . . . . . . . . . . . . 729.4.5 TreeSearch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 729.4.6 2D De-Cloning & Road Construction . . . . . . . . . . . . . . . . . . 739.4.7 2D Projection Track Fitting . . . . . . . . . . . . . . . . . . . . . . . 739.4.8 Projection Track Matching . . . . . . . . . . . . . . . . . . . . . . . . 739.4.9 3D Track Fitting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

9.5 Data analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 749.5.1 Input track distributions . . . . . . . . . . . . . . . . . . . . . . . . . 749.5.2 Signal pulse shape . . . . . . . . . . . . . . . . . . . . . . . . . . . . 759.5.3 Background filtering . . . . . . . . . . . . . . . . . . . . . . . . . . . 769.5.4 Tracking efficiency . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77

9.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

10 Cost Estimate 82

11 Experimental Tests 8411.1 Experimental tests of GEM detector . . . . . . . . . . . . . . . . . . . . . . 84

11.1.1 GEM tests during the PREX experiment . . . . . . . . . . . . . . . . 8411.1.2 GEM tests in Hall A during Spring 2011 . . . . . . . . . . . . . . . . 8411.1.3 Future plans for GEM tests . . . . . . . . . . . . . . . . . . . . . . . 85

2

Executive Summary

Executive Summary

This document is a response to the Second Technical Review which was held on January 22,2010. We have tried to write it such that it will be readable both to the Second TechnicalReview Committee as well as others that may be asked to review the project.

Since the Second Technical Review the project has gained significant strength in the formof a growing collaboration, strong physics proposals, and funding prospects. Extensive stud-ies have been completed, including both experimental measurements as well as simulationsbeginning with the generation of data as will be seen in the GEMS and following throughthe full process of track reconstruction. The simulations were benchmarked against bothdedicated measurements as well as published data and experiments that have run at JLabsubsequent to the Review. Progress has been made in the development of key instrumenta-tion, e.g. in FY12 a contract will be signed with CMU for the construction of the hadroncalorimeter. Pre R&D is now proceeding at UVa, and engineering support will begin shortlyat JLab.

We reproduce here (in the executive summary) both the summary from the report ofSecond Technical Review as well as a synopsis of our responses to the recommendations.

Summary of the Second Technical Review

The summary section of the report from the Second Technical Review was:

The SuperBigBite project in Hall A has been reviewed. The Technical Re-view Committee is impressed by the broad scope of the physics program andthe anticipated performance of the spectrometer.

The SBS project is aiming at the combination of large solid angle coverageat forward angles with the highest luminosity achievable with the upgraded 12GeV CEBAF. The SBS consists of a transverse field dipole magnet equippedwith high-rate GEM tracking detectors. Calorimeters and Cherenkov counterswill be used for triggering and particle identification.

3

The initial experimental program consists of three nucleon form factor experi-ments that have been approved by the JLab PAC. It is obvious to the Committeethat the SBS will become the instrument of choice for a large variety of otherimportant physics problems requiring small-angle coverage, high luminosity, andmodest resolution.

The Committee finds that the SBS experimental design has a very high prob-ability of meeting the experimental requirements. The high rate and high res-olution capability of the GEM detectors make them the ideal solution for thisapplication. The SBS Collaboration has the required expertise to carry out theproject within the time schedule presented.

Remaining uncertainties in background rates and electronics performance canbe reduced by performing experimental tests under similar conditions. Thisshould be done as soon as possible since the results could lead to modificationsof the segmentation scheme and the readout rates which need to be known beforethe start of mass production.

Response to the committee’s recommendations

We summarize below the actions we have taken in response to the recommendations. In eachcase we indicate the section of the report from which the recommendations came, as well asthe section of the present report (in parentheses) in which a detailed discussion is presented.

• Recommendations from 3.1 (Section 1): We have established a list of specifications andQA acceptance criteria. We are still in discussion with the CERN workshop regardingthe post-assembly criteria, and will formalize them when the details of the productionsetup at UVa have been settled. On the GEM chamber design, we have reconfigured thespacer strips, resulting in a significantly smaller dead area. Finally, we have evaluatedthe level of technical support for the GEM construction at UVa. In addition to thefull-time Ph.D. level individual described in the SBS budget, we have made plans forthat individual to go to CERN and Rome for training, and we have established a planfor two post-docs in the groups of Liyanage and Cates to each work half-time withthe full-time individual, including helping in the coordination of the efforts of fourgraduate students.

• Recommendations from 3.2 (Section 2): We have presented a careful analysis of theexpected noise based on experimental data and show that it is well within our needs.On the subject of the second and third trackers (ST and TT), we show that theincreased noise is acceptable because of relaxed requirements on spatial resolution. Wehave adopted the recommendation of using a three-sample readout with our APV25s,and this has been incorporated into our Monte Carlo simulation. Finally, we note that

4

we have dropped all GEM chambers with a UV configuration, because our simulationshowed them to be unnecessary, and hence an unneeded expense.

• Recommendations from 3.3 (Section 3): The photon-detection efficiency was measuredusing a prototype GEM detector and a 137Cs source, and was found to be in goodagreement with our GEANT4 MC simulation. A small five-plane GEM prototype hasbeen constructed, and will be tested in beam this fall.

• Recommendations from 4.1 (Section 4): Written assurances have been received fromBNL regarding the availability of the 48D48 magnet. A new power supply for themagnet has been added to the SuperBigbite budget, so that the magnet will be operatedwith all coils in series. TOSCA calculations are continually being refined.

• Recommendations from 4.2 (Section 5): Detailed simulations were performed and haveshown that the energy and position resolution of the electromagnetic calorimeter re-main sufficient under the operating conditions of the GEp(5) experiment. The calcu-lations were benchmarked against actual data from the GEp(3) experiment. The UVcuring of the lead glass will be tested this summer at high light intensity.

• Recommendations from 4.4 (Section 6): All recommendations have been implementedand described in the text, including a plan for routing the signals, an analysis of theconsequences of using the multi-mode output of the APV25s, mechanical issues andan analysis of signal quality while using long cables.

• Recommendations from 5 (Sections 7, 8 and 9): A detailed Monte Carlo simulationof the coordinate detector (described in Section 7) has shown that our analysis willyield a spatial resolution that is sufficient for the track selection in the GEMs of thefront tracker (FT) of the hadron arm. A full Monte Carlo of the experiment has beendeveloped with the data generated within the framework of GEANT4 (described inSection 8). These data were then used as input for the full chain of track reconstructionin the first SBS tracker. The precise spectrum of the background was studied usingGEANT3, and we benchmarked the code using the recent Transversity experiment inHall A. The pulse shapes generated using GEANT4 were cross-checked against actualexperimental data. The track reconstruction (described in Section 9) achieved 90%efficiency, with contamination by false tracks shown to be smaller than 5%. Furtheroptimization using the programmable APV25s appears feasible.

• Recommendations from 6 (Section 10): A new and detailed funding profile has beenworked out as part of an MIE proposal to DOE.

• Recommendations from 7 (Section 11). We have tested a small prototype GEM cham-ber during the recent PREx run in Hall A in June of 2010. As mentioned earlier,we have also constructed a five-plane GEM and will be testing it with APV25-basedelectronics during an upcoming run in the fall.

5

Chapter 1

GEM quality, GEM spacers,Manpower and Cost of the gas system

1.1 GEM foils Quality Assurance and acceptance cri-

teria

Recommendation #3.1a: In view of the very large production of GEM foils andreadout PCBs by the CERN workshop, the Committee strongly recommends toset up a list of specifications and QA acceptance criteria (e.g. max. leakage cur-rents, inner/outer hole diameter range, max. mask misalignment, etc.) beforethe start of mass production this year. Strict QA rules and documentation of ac-ceptance tests at the institutes receiving the GEM foils and PCB boards shouldbe set up and followed. Depending on the details of the QA requirements, thelevel of required contingency should be discussed with the CERN workshop. Thequality of GEM foils and PCB components during the mass production processhas to be closely monitored, and rapid feedback to the CERN workshop shouldbe ensured.

Response

The initial quality acceptance criteria and tests have been defined based on the followingsources of information:

• the COMPASS production guide [1],

• the information gathered during the RD51 GEM and Micromegas detector design andassembly training Feb. 17-20, 2009 (e.g., lecture on TOTEM Quality Control [2]),

• discussions with our colleagues (G. Bencivenni and F. Murtas), experts on GEM tech-nology, at the Laboratori Nazionali di Frascati,

6

• the experimental experience we are gaining with the full-scale prototype test and as-sembly that started in the Fall of 2010.

Two clean rooms have been set up, one in Rome/ISS (class 1000) and one in Catania(class 100). The Catania room is equipped with:

• two large granite tables,

• three gas lines with associateds regulator and mixers,

• an extractor fan adequate for any vapors produced during gluing and assembling,

• a large foil stretcher with 5+4 load cells (the first version has been used in Rome forthe first full-scale prototype assembly).

The Rome room includes:

• one large granite table,

• one stainless steel table,

• one ultrasound bath,

• one large transparent box equipped for HV testing (terminal for HV and gas tight),

• one white board for overall visual inspection,

• one microscope (with recordable video system) for detailed inspection and measure-ments of the GEM holes and readout strips.

These detailed acceptance criteria, quality controls and assembly procedures were opti-mized in Rome during the assembly and tests of the first full-scale (40×50 cm2) prototypes.

In the production phase, we expect to proceed as follows (details still to be finalized):

• the first quality checks on the GEM foils, readout planes and PERMAGLAS frameswill be performed in Rome,

• the items that pass these checks will be sent to Catania for final assembly,

• the GEM foil resistance, up to HV= 650 V, in nitrogen atmosphere, will be checkedduring different phases of the assembly at Catania,

• the foils that do not pass the checks in Rome will be tested in more detail, possiblycured, and perhaps sent back to CERN (details to be worked out with the CERNworkshop),

• the assembled modules will be tested and characterized in the lab and some of themin a test beam.

7

1.2 GEM foils spacer strips

Recommendation #3.1b: The Committee recommends investigating the possi-bility to reduce the number of spacer strips between GEM foils which will resultboth in a smaller dead area and a smaller amount of material.

Response

The original spacer layout of the GEM frames was driven by the idea of using a smalltension on the GEM foil, and therefore providing an open sector with an area of ∼10×10cm2. In the first version, the frame spacers had exactly the same segmentation as the upperGEM foil HV distribution (20 identical sectors), and therefore the spacers coincided with theHV sector separation, resulting in an increased dead area of the GEM. In the second designof the spacers we minimized the overlap of the HV sector borders and the spacers and, atthe same time, the spacer material. We ended up with a reduction of the spacer materialby 30% with respect to the first design and a negligible overlap of the HV sector bordersand the spacers. The first 40 × 50 cm2 full-scale GEM prototype has been built accordingto the second design with 3× 6 sectors (∼8.3×13.3 cm2 area each), with a very conservativeGEM foil stretch of about 3 kg/cm; a picture of the prototype during assembly is shown inFig. 1.1.

Figure 1.1: First prototype during assembly; the spacers and a stretched GEM foil are visible.

8

Taking into account the suggestions of the committee, the advice of the CERN workshop,an improved analysis of the gas flow, and a revised version of the GEM stretcher resulted ina final (pre-production) design with a significant reduction of spacers (about 50%) comparedto the original version; a drawing of the GEM frame and spacer is presented in Fig. 1.2. Thespacers form open sectors of about 12.5× 13.3 cm2; we intend to apply a conservative GEMfoil tension of 2 kg/cm which should keep the maximum deformation of the foil well below1% (20 µm) assuming a pressure on the foil of 10 N/m2.

Figure 1.2: Design of the GEM frame with the new spacer layout. Note that the spacershave small rectangular slots for the gas flow (not shown in drawing).

1.3 Technical personnel

Recommendation #3.1c: The Committee also proposes to investigate the possi-bility to increase the technical personnel for the construction and system integra-tion aspects of GEM chambers and electronics for ST and TT at the Universityof Virginia.

Response

9

The UVa GEM construction will be led by Nilanga Liyanage, an expert in trackingdetectors, who previously led the successful project to develop and build the large MWDCtracker for the Bigbite spectrometer. As the manager of that three-year 443 k$ project,he supervised a senior post-doctoral associate, 3 graduate students and 10 undergraduatestudents. His group has already constructed and tested a GEM telescope consisting of five10 × 10 cm2 prototype GEM chambers. These prototype chambers were constructed usinga procedure similar to the one expected for the construction of the final SBS chambers:

• The GEM foils and 2D readout boards were purchased from CERN.

• The GEM foils and the 2D readout boards were checked visually using magnifiers uponarrival at UVa.

• The GEM foils were stretched using a foil stretcher and were glued on to the supportframes.

• The framed GEM foils were high-voltage tested up to 550 V in a dry N2 box.

• The readout boards were glued to the base frames, HV connections were fabricatedand were incorporated with the locally made HV dividers.

• The drift windows, the spacer frames, and the gas windows were fabricated at UVa.

• The GEM chambers were fully assembled and tested in the UVa clean room.

A full-time PhD level technical expert will be hired by the UVa group for the durationof the project (four years). This technical expert will be responsible for the developmentof the GEM chamber units, for the construction and assembly of the GEM chambers andfor the supervision and training of the graduate and undergraduate students working on theproject. He/she will also travel to Rome and stay there for several weeks to participate inthe INFN GEM construction to gain expertise. During this visit he/she will participate inall major steps of the fabrication of the GEM chamber, from the inspection of the foils onarrival from CERN to the final testing of a completed chamber.

Two post-docs, one from the group of Gordon Cates, the other from that of NilangaLiyanage, will each work half-time on the GEM chamber R&D, construction, testing andinstallation. In addition, four graduate students and two undergraduate students will workon the construction and testing of the GEM chambers. Three of the four graduate studentshave already been recruited and are currently gaining GEM chamber construction experience.Each of the four graduate students will work full time on the GEM chamber project overa period of two to three years. Nilanga Liyanage, the post-docs and some of the graduatestudents will also travel to Rome to gain expertise in large GEM chamber construction.

In addition to the above, personnel from the UVa Physics department electronics shopand UVa Physics department machine shop will be employed in the fabrication of readoutelectronics and the machining of chamber components as needed. The electronics shop hasa full-time electronics engineer and a full-time electronics technician skilled in electronics

10

development, prototyping, fabrication and optimization services. The machine shop has astaff of two full-time machinists.

Several members of the UVa group are part of the CERN Micro Pattern Gas Detectorworking group (RD-51) which is spearheading the development of large GEM detectorsaround the world.

1.4 GEM gas-handling system cost

Recommendation #3.1d: The Committee recommends to make sure that thecost estimates for the GEM foils and readout circuits include the contingencyfor the final yield, based on the QA procedures, and to review the cost estimatefor the gas system which seems to be on the low side.

Response

In the current funding proposal for the SBS [3], the GEM gas system is listed as a 6.75 k$item plus 20% contingency. The plan is for the GEMs’ gas supply to use Ar/CO2 premixedin a 70/30 ratio. The existing Hall A gas system will be used to premix the two gases fordelivery to the chambers. The listed cost for the GEM gas system is only for the additionalpiping and flow meters that will be necessary for integration into the Hall A system.

The Hall A gas delivery system currently has enough capacity to supply the four VDCsand the large area straw chambers of the FPP detectors in the HRS spectrometers. The gasvolume of all our GEM chambers together (a total of ∼150 l for all eighty-two 40× 50 cm2

modules of the FT, ST, TT, and CD trackers) is less than the gas volume of the FPP strawchambers. Studies of aging measurements on the triple-GEM detectors of the COMPASSexperiment [4] have shown that a flow rate of 80 cm3/min, into a single detector volume of∼810 cm3, with a Ar/CO2 mixture of 70/30 prevents any noticeable etching of the GEMfoils, while no loss of gain or energy resolution was observed. Providing a flow rate of 80-100cm3/min for the SBS GEMs is well within the capacity of the existing Hall A gas system(∼1450 cm3/min for Ar).

For the performance requirements of the SBS GEMs, the proposed budget is deemedadequate for the integration of the SBS gas supply into the existing Hall A gas system.

11

Chapter 2

GEM UV readout and Number ofsamples

2.1 The UV strip configuration

Recommendation #3.2a: Noise performance studies of the chamber with UVstrip orientation, and therefore varying strip lengths, and an analysis of its im-pact on resolution and efficiency are of a great importance before the start ofmass production. Special tests to estimate S/B performance should be also fore-seen for the ST and TT chambers, where four strips are connected into a singlereadout channel (longer effective strip length mean higher capacitance, i.e., morenoise).

Response

The technical committee expressed concern over noise issues arising from varying lengthsof the U and V orientated strips. We would like to point out that readout strips of varyinglengths have been used in already constructed GEM chambers of circular disk geometry: theGEM chambers of the TOTEM T2 telescope [5] and the STAR forward GEM tracker [6]. Inthese chambers, the azimuthal readout strips change from a very short length at the insideedge of the disk segment to a much larger length at the outside edge. In the case of theTOTEM chambers two chamber sections make a disk with an inner radius ∼4.25 cm andan outer radius ∼14.5 cm. This results in strip-length variations between 13 cm and 45 cm.The variation of strip capacitance between these two cases, as measured by the TOTEMcollaboration [7], is from ∼3 pF to ∼13 pF. No noise problems associated with the varyinglengths of the strips have been reported.

The pulse-shaping time for the APV25 amplifier is 50 ns. In this short time range, theamplified signal noise level is dominated by the voltage noise sources in the detector. Thus,the Equivalent Noise Charge (ENC) is roughly proportional to the detector capacitance.This was verified in the testing of the APV25 chip up to a load capacitance of 22 pF and theresults were reported in Ref. [8]. For the peak mode of APV25 the noise at 0 pF was 246 e−

12

Figure 2.1: Equivalent noise charge for the APV25 readout as a function of the input loadcapacitance in peak (plus symbols) and deconvolution (open circles) mode (from Ref. [8]).

rms, with the noise increasing linearly with capacitance with a slope of 36 e− rms/pF (seeFig. 2.1). It was also determined that the pulse shape does not change significantly withthe increasing load capacitance, at least up to 22 pF. We note here that the maximum stripcapacitance we expect in the front tracker of SBS, where we need high resolution, is about20 pF. We are using the experimentally measured noise levels from Ref. [8] to estimate theexpected noise levels for our setup.

The main contribution to the strip capacitance arises from the capacitance between thetwo layers of strips in a 2-D readout plane. Smaller contributions come from the capacitancebetween the strips in the same plane, and the capacitance between the strips and the lastGEM foil. We calculated the strip capacitance for different readout strip lengths that we areconsidering for SBS chambers using the formalism developed in Ref. [9]. This capacitanceagreed well with the measured values reported for the TOTEM GEM chamber readoutboards.

Table 2.1 shows the expected capacitance and the corresponding equivalent noise chargelevels for different strip length and strip combinations considered for the SBS GEM chambers.

13

Tracker Strip length Capacitance [pF] ENC [e−]

Front tracker (shortest strips) 1 cm ∼ 1 pF ∼ 300 e−

Front tracker (longest strips) 70 cm ∼ 20 pF ∼ 1000 e−

Back trackers (4 strips combined) 280 cm ∼ 80 pF ∼ 3100 e−

Coordinate detector (1-D readout,4 strips combined) 4 m ∼ 36 pF ∼ 1500 e−

Table 2.1: Estimated capacitance values and equivalent noise charge levels for different striplengths of the SBS GEM detectors. The capacitance per unit area is less for the coordinatedetector because there is no contribution from the capacitance between two readout layers.

Minimum Ionizing Particles are expected to create about 20 electron-ion pairs in theionization region of the GEM chamber. With a typical gain of 5 × 103, we can expect asignal strength of roughly 1 × 105 e−. For an average cluster size of 3-strip hits, the totalcluster noise levels in the front tracker will vary between 900 e− and 3000 e−. Since theworst case signal-to-noise ratio will be about 32:1, we do not expect any complications dueto varying length of front tracker readout boards.

In the case of the back trackers, we expect the cluster size to be at most two readoutchannels. In this case, the worst-case SN ratio will be about 17:1. Since we need only acoarse track resolution (∼1 mm) from the back trackers, this SN ratio is more than sufficient.

However, we stress that we have dropped GEM chambers with a UV configuration,because our simulations have shown them to be unnecessary, and hence an unneeded expense.

2.2 The 3-sample readout

Recommendation #3.2b: In view of the high background levels (∼ 500 kHz/cm2)in the GEp(5) spectrometer, the Committee recommends that the 3-samplereadout method of the APV25 be adopted as the default solution for all trackers(FT, ST, TT). This will increase the bandwidth requirement and data rates fromtracking stations to the DAQ which, however, seems to be consistent with theplans for the Hall A DAQ upgrade.

Response

Our Monte Carlo studies (see chapter 9) confirm that at least three samples should be readfrom the APV25 to obtain sufficient rejection of noise and pileup. Hence, we concur with thecommittee’s recommendation to adopt the 3-sample readout method for all trackers in theproton arm of GEp(5). Implementing this readout method in the VME digitizer modules(MPDs) downstream of the APV25s is a simple matter of firmware programming in theMPDs.

14

As pointed out by the committee, the 3-sample readout will increase the data rate fromthe trackers and hence the bandwidth requirements of the DAQ. Our response to recommen-dation 4.4b is presented in section 6.2.

15

Chapter 3

The low-energy photon response ofthe GEM chambers

Recommendation #3.3: The Committee strongly recommends that the responseof a GEM detector to low-energy photons should be measured using a proto-type detector and electronics. The results should be compared to the GEANTmodeling to confirm that the background levels in the Monte Carlo simulationare realistic. The expected level of occupancy in the GEM detectors, using anAPV time window of 250 ns and an average number of strips in cluster per MIPparticle ∼ 3.5, seems to be exceedingly high.

Response

A GEANT4 simulation was used to study the photon response of GEM chambers, asoutlined in Sec. 5.11 of the SBS CDR [10]. This simulation indicated that the detectionprobability is around 4 × 10−3 for photon energies between 0.5 and 1 MeV. Above 1 MeVthe photon detection efficiency was shown to increase but not to exceed 1%.

We conducted a test to verify the low-energy photon detection rate using a 10× 10 cm2

prototype GEM chamber and a high-power (0.42 mCi) 137Cs gamma source. The activity ofthe source was determined to 5% using the decay time since the original certified measure-ment. The source was placed in a lead collimator, as shown in Fig. 3.1, so that the solidangle for the photon exposure of the GEM chamber could be accurately calculated. 137Csemits electrons up to 1.7 MeV and 0.662 MeV photons. A 6 mm thick sheet of plastic wasplaced close to the source between the source and the chamber to block the electrons fromthe source so that the chamber was only exposed to photons. This thickness of plastic issufficient to stop almost all electrons from the source, while it removes only about 3% of thephotons. This loss was factored into the calculation.

The DAQ was triggered by a pulser with an acquisition window of 1 µs. Since the triggeris uncorrelated with the photon hits on the GEM chamber, the count rate for good chamberhits depends on the photon detection efficiency.

16

lead cylinder

137Cs source

0.6 cm plastic cover

GEM chamber

aluminum plate

Figure 3.1: The experimental setup used for the 137Cs source test. The source was collimatedusing a lead cylinder with an inner diameter of 6.4 cm and a wall thickness of 2 cm. Thesource was located so that it is 5 cm above the bottom of the cylinder. The opening of thecylinder defines the solid angle exposure for the GEM chamber. All distances were measuredwith an accuracy of ∼ ±2 mm, and the location of the source inside the sealed enclosurewas known to within 5 mm. These measurements were used to calculate the solid angle tobe 0.96 ± 0.1 sr

The results from this test yielded a photon detection efficiency of 0.46% ± 0.06% for662 keV photons, in good agreement with the GEANT4 simulation result. The main uncer-tainty of this measurement comes from the location of the source. We have also assigneda 6% uncertainty to cover unaccounted factors such as scattering through the edges of thecollimator.

The UVa group has constructed a GEM telescope consisting of five 10×10 cm2 prototypeGEM chambers. This telescope with a GASSIPLEX [11] readout has been tested under beamconditions in Hall A in the fall of 2010 and in the spring of 2011. A full set of APV25-S1electronics has been ordered. It will allow us to implement a realistic readout of this tracker.The GEM telescope will be tested during the g2p-GEp experiment run in Hall A in November2011 with the APV25 electronics. These beam data will be compared to a full GEANT4simulation created for the exact conditions of these runs.

17

Chapter 4

SBS magnet and optics

4.1 Acquisition of 48D48 magnet and associated equip-

ment

Recommendation #4.1a: Execute a letter property transfer of the 48D48 mag-net, spare coils, and power supply with BNL to secure the ownership of themagnet for JLab.Recommendation #4.1b: Transfer of the SBS magnet, spare coils, and powersupply to JLab should occur as soon as funds can be obtained for shippingand storage. The 48D48 magnet (and its power supply) is an excellent generalpurpose device and will be a valuable asset for JLab for many uses including itsobvious value to the SBS project.Recommendation #4.1c: A JLab representative should be dispatched to BNL toinspect the 48D48 DC power supply(s) to ascertain the suitability for continueduse at JLab as part of the SBS Spectrometer. DC Power Supply documentation,spare parts inventory, and overall power supply condition should be determinedby on site first hand inspection.

Response

Since the review we have had further contact with Phil Pile and Al Pendzick, the BNLDivision chief engineer. Al Pendzick said on 4/12/2010, “The power supplies are old, re-builtin the 80s with new SCR bridges and control circuits. They are operational (we use somefor RHIC) but you should assume you will need to interface them with your control system.We have many supplies with different outputs and will need to know what your operatingcurrent will be. One of our common choices is 2 ea. 150 V, 3600 A supplies in series for amagnet full out without booster coils. SCR spares are available, controls are custom, butwe can give you spares, prints specs. for all. We will disassemble the magnets into 18 tonpieces, 5-6 trucks per magnet, no special permits needed, about 100 m-h per magnet to ship≈ 10 k$. Knowing the uncertainty of DOE funding, I would wait for late summer to come

18

up. This equipment is going nowhere — and we will tell you if there is other interest. Themagnets are straightforward — I would bring your power supply engineer.”

As of now, there are no funds to ship the magnets, but we have been assured that themagnet(s) will remain available. Regarding the power supplies, after some consideration ithas been decided to purchase a new one adequate for our needs. The money for that powersupply is in the MIE budget (161 k$ escalated dollars) for FY2014.

4.2 SBS Magnet Powering Options

Recommendation #4.1d: The SBS magnet should be operated with all coils inseries for best spectrometer stability and accuracy.

Response

Of course powering the coils in series is a much preferred option. The power them inparallel option was presented as a what if you had to do it tomorrow with power suppliesalready available at JLab option. As noted above, we have budgeted money to purchase anew power supply that will, of course, allow us to power the magnet coils in series.

4.3 The field calculation and spectrometer optics

Recommendation #4.1e: The magnetic modeling of the 48D48 magnet as mod-ified for the SBS that has been performed using “Mermaid” should be cross-checked against a fresh TOSCA model. The field plots shown during the pre-sentations show some evidence of a grainy mesh especially in the angled beampass through the channel. According to the presenters this was a result of lim-itations of the Mermaid installation. A TOSCA model with higher resolutionwould provide confirmation of the efficacy of the beam pass through shieldingand the field in the gap including fringe fields.Recommendation #4.1f: The SBS optics that was presented was based on modelfields to represent the SBS dipole. The results presented indicate that the SBSmagnet optics, resolution, and acceptance are well matched to the experimen-tal requirements. The optics calculations and evaluation of the resolution andacceptance should be repeated with calculated 3D fields from a high resolutionTOSCA model or at least the present Mermaid model. Using actual calculatedSBS magnet fields will necessarily include the effects of the field gradient along”Z”, a complete fringe field description, and information about actual fields alongthe beam pass through.

Response

19

Extensive TOSCA modeling of the SBS magnet is underway. A TOSCA analysis indicatesthat the SBS requirements can be met with the proposed magnet. In Figs. 4.1 and 4.2 theresults of those calculations are presented.

Tasks remaining for the magnet design are: Continuing TOSCA iterations to reduce thefield gradient near the target, i.e. modify the field clamp dimensions to reduce the field,modifying the model of the beam line shielding to reduce the field along the beam line,developing yoke modification drawings, and designing the support structure.

The final optics test will be completed when the magnet design has been finalized. How-ever, in the interest of making sure we are on the right track, an interim optical analysis,using a map with all the above modifications, the field clamp and a solenoid (that requiresfurther optimization), was performed. This was a “standard” optics simulation, in which aset of approximately 2000 random trajectories spanning the expected acceptance was tracedthrough the spectrometer using the SNAKE program. Those trajectories were then used asa database for determining the reconstructed target parameters, δ, θ, y, and φ. With thethus developed reconstruction tensors the resolution capabilities were tested by running eachtrajectory through a simulator 100 times. Each time through, the trajectory parameters thatwould be measured at the detectors are smeared using the expected detector resolution anda Gaussian random number generator. The difference between the reconstructed values andthe actual values are shown in Fig. 4.3. In Fig. 4.4 the effect of a 500 µm beam spot isexplored. As can be readily seen from a comparison with Fig. 4.3 the addition of a 500 µmbeam spot makes no significant contribution to any of the resolutions. With σδ = 4.6 · 10−3,σθ = 1.9 ·10−4, σφ = 3.2 ·10−4, and σy = 1.1 ·10−3 m the achieved resolutions meet or exceedthe design specifications.

20

Figure 4.1: TOSCA layout of the SBS magnet.

21

Figure 4.2: Field in the horizontal plane along the magnet center line from the target throughthe exit of the magnet. Red is plotted against the right vertical axis, blue against the leftone.

22

Figure 4.3: Results of a simulation test of the SBS using a TOSCA generated map. Theupper left panel shows the distribution of the difference between the actual value of δ and theone determined by the reconstruction tensor with positional and angular errors of σ = 70 µmand ∼ 0.3 mrad, respectively, folded in. The upper right is the same but here the differencebetween the actual θ and the measured θ is shown. The middle two plots are for φ and y.The bottom left plot shows the distribution of θ final as used in the reconstruction. Thebottom right shows the distribution of the δ values used in the simulation.

23

Figure 4.4: The same as in Fig. 4.3 except that here the sample distribution of trajectoriesincludes a 500 µm full width vertical distribution at the source point to simulate the expectedcontribution from a 500 µm beam spot size.

24

Chapter 5

The Lead-glass Calorimeter

5.1 Calculations of the BigCal energy and position res-

olution for a 20 cm Al absorber

Recommendation #4.2: Provide calculations of the energy and spatial resolutionwith the 20 cm Al absorber taking into account the average radiation damage.Evaluate the impact of the resolutions on the general performance includingtracking and trigger rate. Clarify the impact of the expected energy resolutionnot meeting the requirement on page 107. Provide evidence or arguments thata 5 fold increase in the UV light intensity will increase the rate of curing by afactor of about 5.

Response

The energy and position resolution of BigCal has been studied with GEANT calculations.In Fig. 5.1, the energy resolution (σE/Epeak) is plotted versus the initial electron energy. Thered open circles are GEANT calculations for the BigCal setup during the GEp(3) experiment,when a 10 cm aluminum block, a lucite plate and a 2.5 cm aluminum plate were placed infront of the BigCal lead glass. The filled red circle at an incident electron energy of 1.0 GeV isthe energy resolution measured during GEp(3). The agreement with the GEANT simulationis good. The black diamonds are GEANT calculations for a 20 cm aluminum block in frontof BigCal, the configuration proposed for GEp(5). At 1 GeV, the simulation predicts aresolution of 18% which is close to the estimate of 15-16% cited on pg. 113 of the CDR.For the GEp(5) experiment the scattered electron energy is between 3.3 and 3.5 GeV for allkinematics, so the energy resolution would be ∼9%.

In the CDR, a position resolution of the calorimeter of 0.4 cm was assumed. The sameGEANT simulation with a 20 cm aluminum absorber in front of BigCal yielded a positionresolution 6 mm, independent of the incident energy. This value (compared to that of 4 mmquoted in the CDR) will increase the predicted number of pseudo-hits in the CoordinateDetector from 0.028 to 0.042, still an acceptably small number.

25

0 1 2 3 4Incident Electron Energy [GeV]

0

0.05

0.1

0.15

0.2

0.25

0.3

σ E/Epe

ak

10 cm Al , 1.9Al and Luc20 cm Al GEp(3) datum (10 cm, 1.9Al and Luc)

Figure 5.1: The BigCal energy resolution (σE/Epeak) is plotted versus the initial electronenergy. The red open circles (labeled “10cm Al , 1.9Al and luc”) represent GEANT calcu-lations for the BigCal setup during the GEp(3) experiment, when a 10 cm aluminum block,a lucite plate and a 2.5 cm aluminum plate were placed in front of the BigCal lead glass.The filled red circle at an incident electron energy of 1.0 GeV is the energy resolution mea-sured during GEp(3). The black diamonds represent the GEANT calculations with a 20 cmaluminum block in front of BigCal.

5.2 Effect of radiation damage on the BigCal energy

and position resolution

The deterioration of the BigCal energy gain and resolution due to radiation damage wasmonitored during the GEp(3) and GEp2γ experiments, when BigCal was placed at a differentdistance and angle for each kinematic point. In Fig. 5.2 the energy resolution (multiplied by√Epeak) is plotted versus the total charge for all GEp(3) and GEp2γ kinematics. Multiplying

by√Epeak matches the end of one data set to the beginning of the next set to first order. One

clearly sees a different slope in the degradation of the energy resolution for each kinematicpoint, the rate of degradation being faster when BigCal is at more forward angles . Thehighest soft-photon flux occurred during the RCS experiment, during which a radiator wasplaced in front of the hydrogen target with BigCal at its most forward angle. The BigCalenergy resolution could not be monitored during RCS, but elastic data taken at the endof RCS yielded the energy resolution shown as the purple triangle in Fig. 5.2. The slopebetween the last GEp(3) point and the RCS datum is the steepest of the graph. The BigCallead glass was UV cured after the RCS run, which resulted in a dramatic improvement in

26

the resolution. For an estimate of the radiation damage during GEp(5), one should use thekinematics where BigCal was at θe = 32.5◦ and a distance of 11 m from the target. At thiskinematic point the energy resolution increased by 0.15%/C. On pg. 110-111 of the CDR thesoft-photon flux for the highest Q2 point of GEp(5) is predicted to be a factor of 13 higherthan for the GEp2γ kinematics. This translates in a slope of 1.95%/C for GEp(5), so that,with a 75 µA beam delivering 1.1 C in 8 hours (assuming 50% accelerator efficiency), theresolution will increase by 2% every hour.

0 100 200 300 400Integrated Charge [C]

0.1

0.2

0.3

σ E/Epe

ak✳

sqrt(

E peak

)

ΘE = 60.3o d=9.6m

ΘE = 105o d=4.9m

ΘE = 44.9o d=12m

ΘE = 32.5o d=11m

ΘE = 30.8o d=11m

ΘE = 105o d=4.9m

0 100 200 300 4000.1

0.2

0.3

σ E/Epe

ak✳

sqrt(

E peak

)

ΘE = 26o d=8m (RCS)

ΘE = 69o d=4.3m

ΘE = 44.2o d=6.1m

GEp(3) GEp2γRCS

Figure 5.2: The measured energy resolution (σE/Epeak) (multiplied by√Epeak) versus charge

throughout the GEp(3), GEp2γ and RCS experiments.

5.3 Effect of energy resolution on the BigCal trigger

rate

Now, the BigCal trigger rate can be estimated based on the predicted energy resolution.On page 107 of the CDR, it is stated: we need an energy resolution better than 10%/

√E.

This is dictated by the trigger rate considerations. As the Technical Review points out thisis contradicted on page 113 in the CDR where an energy resolution of 16% at 1 GeV wasextrapolated from previous GEANT simulations using a 10 cm aluminum block in frontof BigCal. In the CDR, the trigger rate was estimated to be 60 kHz with a threshold at85% of the elastic scattered electron energy which corresponded to a 2.5σ cut for an energyresolution of 10%/

√E. For the updated energy-resolution estimate this would would lead to

a 1.5σ cut. If the threshold was lowered to 73% (about 2.5σ) the trigger rate would increase

27

to 203 kHz. Clearly, the ultimate rate capability of the DAQ will determine the thresholdin a compromise between the trigger rate and a loss of efficiency in detecting elastic events.

5.4 Test of UV Curing rate in BigCal

The measurements of gain loss were fitted with aebC resulting in b= 0.53%/C. Using theabove mentioned increase in soft-photon flux of 13, the gain loss during GEp(5) can beexpected to increase to b=6.7%/C and the lead glass would suffer a gain loss of 0.87% perhour. In the CDR, a plan is presented to UV cure the lead glass for one hour after every sevenhours of beam which assumes a curing rate of 6%/hr to make up for the 6% gain loss duringthe preceding seven hours. During a down time in the GEp(3) and GEp2γ experiments, thelead glass was cured with a UV light source and the curing rate was found to be 1.24%/hr.Steps are outlined in the CDR to increase the UV light intensity by a factor of five, to reacha curing rate of 6% per hour.

The Technical Review report expresses skepticism that the needed curing rate can beachieved by increasing the UV light intensity. The curing during GEp(3)/GEp2γ was doneduring a one month down so part of the increase in gain could be due to a natural cure. Todetermine what UV light intensity is needed to reach the curing rate desired by GEp(5), wewill perform tests, using lead-glass blocks that were in Hall A during the last run period.This summer, we will measure the curing rate with the UV light intensity proposed in theCDR. The effort in the EEL building will be led by Mark Jones with help from the otherGEp(5) spokespersons.

28

Chapter 6

GEM module design

6.1 Bandwidth considerations between HCAL FADCs

and Trigger Processor

Recommendation #4.4a: Develop a plan for routing the signals from all boardsin the backplane to the FPGAs that is consistent with the expected data rates.

Response

At issue is the bandwidth over the backplane of the VME crate holding the electronicsfor digital summing of the hadron calorimeter (HCAL) signals. Our proposal calls for 13JLab FADCs, each digitizing the signals from 16 calorimeter blocks, which will be integratedover a 16 ns window. The 8 most significant bits of each channel’s data will be transmittedto a custom-built Crate Trigger Processor (“Smart Trigger” or “Smart HCAL”). FPGAsinside this module then compute various sums of amplitudes of 16 adjacent blocks. Eachsum represents a “cluster amplitude”, which will be used for trigger decisions in downstreamcoincidence FPGAs.

Transmission of data from 16 channels at 8 bits/channel every 16 ns requires a bandwidthof 8 Gb/s between each FADC and the Smart Trigger module. The present design of CTP-style modules by the JLab Fast Electronics Group specifies only one-half of this bandwidth,viz. 4 Gb/s, consistent with the requirements of Hall D. Several solutions are conceivable toachieve a higher bandwidth:

1. Operate the backplane lanes at twice the frequency. In fact, the FPGAs on the latestrevisions of the FADC and CTP boards already have a theoretical maximum aggre-gate bandwidth of 8 Gb/s (plus 20% signaling overhead). However, the higher speedcapability has not yet been tested. Such testing is foreseen once pre-production unitsof the FADC boards are received at JLab in FY12.

2. Double the number of backplane lanes from currently two to four, retaining the currentsignaling speed of each lane. The VXE backplane already supports four serial lanes

29

from each module to the trigger processor, however the current-generation CTP onlyuses two lanes. Since we are proposing to design a new CTP-type module anyway, wecould easily accommodate the higher bandwidth requirement at the design stage bytaking advantage of all backplane lanes.

3. Further reduce the number of bits transmitted per channel. For example, if 6 bitsof resolution can be shown to be sufficient, the bandwidth requirement will drop to6 Gb/s. The effect of lower resolution would need to be studied in detailed simulationsof the calorimeter response.

4. Digitize only 8 channels in each FADC. Since each crate can hold a maximum of 16FADC boards, the digital summing electronics would then have to be split over twocrates. Because there is significant overlap between the calorimeter regions that eachcrate would have to process, these two crates would need to be instrumented with 14and 15 FADCs, respectively, for a total of 29 FADCs.

Given the fact that the electronics is, in principle, already capable of delivering the requiredbandwidth, we are presently hopeful that this problem is already solved, pending successfulbench testing by the Fast Electronics group. If testing should not be successful, we willexplore the other options. If no reliable solution for increasing the bandwidth can be found,option 4 offers a safe fallback. The additional cost of implementing this last option would beapproximately 85 k$ (16 additional FADCs @ $3800, one additional crate & VME controller@ $16400, and one additional Smart Trigger Module @ $7500).

6.2 Multi-mode readout of the APV25S1 chip

Recommendation #4.4b: Clarify the consequences of the multi-mode readout ofthe APV25S1 chip for event size and data acquisition rates.

Response

As already discussed in section 2.2, the APV25 chips of the GEM tracking systems willbe operated in the so-called multi-mode where, following a trigger, three samples are readout. This entails the following bandwidth and data rate considerations:

1. The analog data from the APV25s on the front-end cards will be transmitted to customVME digitizer modules (multi-purpose digitizer, MPD) via HDMI signal cables. EachAPV25 chip provides 128 channels of analog data, all of which must be read out for eachtrigger due to the design of the APV25. Each channel will generate three amplitudesamples. Thus there are 384 analog samples to be read and digitized for each trigger.These analog sample data are transmitted to the MPDs sequentially at a clock rateof 40 MHz. Each APV25 chip will be connected to a dedicated flash ADC in VME,i.e. the APV25s will be read out in parallel. Thus, it will take 384/(40 MHz) ≈ 10 µs

30

to transfer the analog data from all trackers to the MPDs, including the expectedhandshake overhead. The links between the APV25 front-ends and VME thereforesupport a maximum trigger rate of 100 kHz. This number is independent of thedetector occupancy since all channels are always read out. In GEp(5), we plan to readthe APV25s at an expected L2 trigger rate of approximately 1 kHz. This is well belowthe rate limit of the analog links discussed above.

There is a small probability that another L2 trigger will be generated within the 10 µsreadout time. To avoid deadtime in this situation, the MPDs will need to retrigger theAPV25s during the readout in order to latch additional data in the analog pipelines ofthe APVs, which can then be read in a subsequent 10 µs period. We currently believethat this mode of operation is technically feasible, although not yet implemented ortested.

2. The flash ADCs within the VME digitizer modules operate at 200 MHz. Since this issignificantly faster than the signal transfer clock of 40 MHz, the digitization time isnegligible for bandwidth considerations.

3. The digitized data will be pedestal-subtracted and zero-suppressed in FPGA firmware.Also, an average of the pedestals of all inactive channels will be calculated for eachevent and used for common-mode suppression. Additionally, a simple ratio of theamplitude values of the three samples per channel can be calculated and used to rejectnoise originating from the tail of pre-trigger pulses (see section 9.5.3 for details on thepulse-shape analysis). Each MPD will need to process 16 chips × 384 samples/chip =6144 analog samples. The exact processing time for the operations listed above withthis amount of data has not yet been determined. However, it is certain that, giventhe L2 trigger rate of 1 kHz, no bandwidth limitation will occur at this stage if theprocessing time remains below 500 µs, which should be very easy to accomplish. Indesigning the MPD firmware, our goal is to achieve a processing time of 10 µs or lessin order to support the maximum incoming data rate of 100 kHz.

4. Now considering the event size and data rate from the MPDs to the downstream DAQsystem, we note that, after sparsification, we expect from Monte Carlo a maximumoccupancy of 15% in the hadron arm GEM trackers for GEp(5) (see section 9.5 on thesimulation analysis). The total number of GEM channels is approximately 100,000.Each channel will provide three ADC values with 12 bit resolution. These data mustbe prefixed with the channel number address within the corresponding APV25 chip(7 bits). Thus, each active channel carries 43 bits of information. For simplicity, thiscould be rounded to 48 bits (6 bytes) per channel, which would also allow us to addcheck and status bits. Data sets from each APV25 chip must also be identified withthe chip number (10 bits), which occurs on average once for every 15% × 128 ≈ 20active channels, i.e. 0.5 bits overhead per channel. This results in a maximum eventsize of

15%× 105 chan× 48.5 bits/chan ≈ 90 kB

31

At the expected physics (L2) trigger rate for GEp(5) of 1 kHz, the corresponding datarate from the trackers would be 90 MB/s. This is well within the 200 MB/s design rateof the planned 12 GeV-era Hall A DAQ. If necessary, for example if the trigger rate isactually higher, the event size can be decreased by reducing the resolution of the ADCdata (10 or even as few as 8 bits per sample may be sufficient for the analysis). With 8bits/sample, we would have 4 bytes/channel and therefore an event size of 60 kB. Thiswould allow a trigger rate of up to 3 kHz within the 200 MB/s design bandwidth.

6.3 Mechanical details

Recommendation #4.4c: Specify the mechanical details of the FEC electronics,in particular the support for the electronics and the routing of the cables.

Response

The GEM electronics consists of three main components:

• Front-end card: hosts the APV25 chip (128 channels) and provides the amplification,shaping and analog (pipelined) sampling of the strip signals.

• Backplane: a rigid PCB that transmits the front-end cards signals under controlledimpedance, provides the mechanical support to the front-end cards and electromagneticshielding. Each backplane can connect up to 5 cards.

• VME multi-purpose digitizer (MPD) module: controls the APV25 and digitizes theanalog samples; provides a fast processing of the data (e.g., zero suppression andcommon noise subtraction); transmits the data over the VME64x bus to the VMEmaster and then to the DAQ node. Each MPD can control and acquire up to 16front-end cards.

The front-end card, hosting one APV25 chip, is 75× 49.5× 6 mm3 including componentsand connectors (see Figs. 6.1 and 6.2). The card is connected to a rigid PCB backplane, withcontrolled impedance bus and mechanical support to the front-end cards (see Figs. 6.3, 6.4and 6.5). All cards and backplanes are identical; they are positioned in-plane and out-of-plane with respect to the GEM modules.

The backplane will be anchored to the outer mechanical frame that will run all aroundthe chamber (as presented in the TR meeting). This outer frame will in addition supportthe GEM modules, the gas pipes and the cabling.

6.3.1 Routing of the cables

The single front-end card requires the following signal and power levels:

• Inputs:

32

Figure 6.1: A front view of the front-end card prototype 1 with the kapton ZIF interfaceused on the 10×10 GEM prototype.

Figure 6.2: A back view of the front-end card prototype 1 with the kapton ZIF interfaceused on the 10x10 GEM prototype.

1. Reset (digital)

2. Trigger (digital)

3. Clock (digital)

4. I2C lines (2 digital channels)

• Analog outputs: differential line, with positive and negative channels

• Power: one power line and one reference

These signals will be transmitted over the backplane buses and from there on HDMIcables to the MPD modules. Note that all cables contain shielded and twisted-pair channels.The preliminary scheme of the routing of the cabling is shown in Figs. 6.6 and 6.7. Withthis design, the total path length (MPD – cards – MPD) is identical for all cards and onlyrequires two different lengths of cables, The cables will run on the outer chamber frame from

33

Figure 6.3: Bottom view of the readout electronics layout: the supporting mechanical frameof the GEM modules, electronics and services (gas pipes and cabling) below the cards is notshown. The in-plane front-end cards (dark-green) are fixed to their respective backplanes(light-green) and are shielded by the backplanes on one side and a conductive layer on theback of the outer frame.

patch panels (that passively collect the different signals and regroup them properly) to thebackplanes (digital signals), and from the backplanes to the patch panels (analog signals).

This scheme presents an evident drawback: the presence of potential loops. All con-nections are made via shielded twisted pairs, (expected to not be influenced by noise) andtherefore only the shielding may represent a sink for noise. However, we intend to break theshielding loop on the patch panel.

HDMI Type A cables will be used for the digital signals (5 twisted and shielded pairstype cable is available) and HDMI Type B will be used for the analog signals (8 twisted andshielded pairs type cable is available). In order to preserve the modularity of the chambers,some of the channels of each MPD are unused (spare channels); this will also improve per-formance of the system in terms of speed. Four MPDs serve all the chamber electronics: 2

Figure 6.4: Top view of the readout electronics layout (for details see caption of Fig. 6.4).

34

Table 6.1: Components of the electronics and relative cabling for one GEM chamber.MPD HDMI-10m Digital HDMI-3m Analog

module (A+B) Patch Panel (A+B) Backplane Patch Panel Cards1 1+2 1/2 3+3 3 1/2 141 1+2 1/2 3+3 3 1/2 131 1+2 1/2 3+3 3 1/2 141 1+2 1/2 3+3 3 1/2 13

digital patch panels, 12 backplanes, 54 front-end cards, 2 analog patch panels according toTable 6.1.

Figure 6.5: Detail of the out-of-plane cards and backplane of the readout electronics. Cardsare in the sandwich of two (identical) backplanes that provide mechanical support and elec-tromagnetic shielding (see also Figs. 6.3 and 6.4)

35

Figure 6.6: A preliminary detailed design of the cable routing of a single chamber

6.4 Signal quality with long cables

Recommendation #4.4d: Examine the signal quality of the GEM chambers asa function of cable length to make sure that the GEM performance is not com-promised by the 7 – 10 meter cable run.

Response

The maximum length from the front-end cards (actually the backplanes) to the VMEmodules is expected to be about 10 m. Tests have been made to study the degradation

36

Figure 6.7: A preliminary logic design of the overall cable routing of a single chamber.

of the analog differential signals running on shielded twisted cables. None of the tests useanalog buffers on the APV25 front-end cards1.

A very preliminary test consisted of a visual inspection of the APV25 analog frame onthe oscilloscope for very short and relatively long cables; screen-shots of the oscilloscope areshown in Figs. 6.8 and 6.9. Basically, no noticeable degradation was observed.

In a more detailed test we sampled the shape of the APV25 calibration pulse (at differentcalibration currents, ICAL = 25, 45 and 85 of the APV25 register) for analog cable lengths of3, 6, 10, 13, 16, 26, and 36 m. The sampled shapes are presented in Fig. 6.10 for ICAL = 45.The amplitude reduces by about 15% from 3 to 15 m, while the shape of the signal remainsstable. At the greater lengths the fluctuations of the shape tend to increase slightly.

We tested the front-end cards on a 10 × 10 cm2 prototype. The very first results showbasically no effects, within noise, on the pedestal width using 3, 10 and 13 m (see Fig. 6.11).

Recently we started to test the cards connected to the first 40× 50 cm2 GEM prototype(which uses only an XY configuration) with the above described backplanes (see Fig. 6.12).This work is in progress, but the first data show improvement with respect to the previousversion of the cards, and the presence of the backplane seems to act quite well as electro-magnetic shielding (and noise suppressor). As expected, when all strips are connected tothe electronics, the overall noise decreases. Moreover, good grounding of the cards (by theconnector to the backplane and ground pads on the PCB card) also acts in favor of noisereduction.

1The front-end card includes space and connections to host an analog buffer, to improve transmissionperformance if needed.

37

Figure 6.8: Oscilloscope image of the APV frame with the analog levels over 50 cm cable;without analog buffer.

Figure 6.9: Oscilloscope image of the APV frame with the analog levels over 7 m cable;without analog buffer.

38

sampling time [ns]0 200 400 600 800 1000 1200 1400-0.005

0

0.005

0.01

0.015

0.02

0.025

0.03

sampling time [ns]0 200 400 600 800 1000 1200 1400

-0.005

0

0.005

0.01

0.015

0.02

0.025

0.03

sampling time [ns]0 200 400 600 800 1000 1200 1400

-0.005

0

0.005

0.01

0.015

0.02

0.025

sampling time [ns]0 200 400 600 800 1000 1200 1400

-0.005

0

0.005

0.01

0.015

0.02

0.025

sampling time [ns]0 200 400 600 800 1000 1200 1400

-0.005

0

0.005

0.01

0.015

0.02

0.025

sampling time [ns]0 200 400 600 800 1000 1200 1400

0

0.005

0.01

0.015

0.02

sampling time [ns]0 200 400 600 800 1000 1200 1400

0

0.005

0.01

0.015

0.02

Cable length [m]0 5 10 15 20 25 30 35

0.016

0.018

0.02

0.022

0.024

0.026

0.028

0.03

0.032

3 m 6 m

10 m 13 m

16 m 26 m

36 m Amplitude

Figure 6.10: APV25 calibration pulse shapes with different cable lengths and the amplitudeof these pulses as a function of the length.

39

ft.adc-100 -50 0 50 1000

10

20

30

40

50

60

ft.adc-100 -50 0 50 100 150 200 2500

20

40

60

80

100

120

ft.adc-100 0 100 200 300 400 5000

20

40

60

80

100

120

140

160

180

ft.adc-100 -50 0 50 100 150 200 2500

10

20

30

40

50

60

70

80

90

ft.adc-100 -50 0 50 1000

5

10

15

20

25

ft.adc-100 -80 -60 -40 -20 0 20 40 60 800

5

10

15

20

25

ft.adc-100 0 100 200 300 400 500 6000

10

20

30

40

50

60

ft.adc-100 0 100 200 300 400 5000

10

20

30

40

50

ft.adc-100 -50 0 50 100

0

10

20

30

40

50

ft.adc-80 -60 -40 -20 0 20 40 60 80 100 120

0

10

20

30

40

50

60

70

ft.adc-100 0 100 200 300 400 500 600

0

20

40

60

80

100

120

140

160

ft.adc-100 0 100 200 300 400 500

0

20

40

60

80

100

120

140

160

3 m

10 m

13 m

Figure 6.11: Pedestals from cards connected to the 10x10 GEM chamber prototype. Columnscorrespond to different APV25 channels (on different cards), rows to different cable lengths(3 m upper, 10 m middle, 13 m lower). The RMS value slightly increases with cable length.The pedestal is subtracted from the common noise on an event-by-event basis.

40

Figure 6.12: The first 40×50 cm2 prototype fully equipped with front-end electronics duringthe beam test in December 2010. Also visible is the small 10× 10 cm2 GEM chamber.

41

Chapter 7

The Coordinate Detector

7.1 Coordinate Detector vertical position resolution

Recommendations #5.5a: Clarify the pattern recognition in the CD, taking intoaccount showers in the 20 cm thick Al absorber in front and the backsplash fromthe calorimeter.

Response

The coordinate detector (CD), preceded by a 20 cm thick aluminum plate to absorb lowenergy photons, will be placed in front of the BigCal electron detector. The CD consists oftwo planes with horizontal strips at a pitch of 1 mm, separated by 4 cm. The function of theCD is to measure the electron vertical coordinate in order to reduce the search region fortracking the proton using the constraints of elastic kinematics. Almost every electron willstart a shower in the aluminum absorber.

A GEANT3 simulation was run for 3.4 GeV electrons into a 20 cm aluminum plate, a gapof 10 cm and the lead-glass bars. The particle type and its position and energy were recordedat the location of the GEM planes, where it was found that on average 65 photons passedthrough the GEMs per shower. Given the low efficiency (less than 1%) for photon detection,the probability of detecting a photon from the shower in both GEM planes is near zero. Theaverage multiplicity of electrons (positrons) in the shower was 5 (3.3). The position of the hitwas determined from the average of the hit positions for all charged particles at the GEM.The track of the initial electron, projected to the first GEM assuming no interactions withmaterial, is taken as the predicted position. In Fig. 7.1, the difference between the measuredand predicted position of the hit at GEM1 is plotted. A fit with a Gaussian (the red line inFig. 7.1) showed the position resolution of the CD to be 1.8 mm. The horizontal coordinateresolution of the calorimeter was shown to be 6 mm in Sec. 5.1.

42

-1 -0.5 0 0.5 1Difference between measured and predicted position [cm]

0

50

100

150

200

250

Cou

nts

DifferenceGaussian Fit

Figure 7.1: The difference between the measured and predicted position of the hit (units ofcm) at GEM1 is plotted. The red line is a fit with a Gaussian.

7.2 Determining the track search region in SBS

The resolution in the CD vertical coordinate defines a search region in the first chamber ofthe SBS. A simple Monte Carlo that randomly selected the electron angle in the center ofmass and the interaction point along the 40 cm target, calculated the electron and protonkinematics for an elastic reaction. If the electron passed the center of the CD within±0.54 cm(±3σ of the CD resolution) in vertically and within ±3.0 cm (±5σ of BigCal coordinateresolution) horizontally, the proton was tracked through the SBS magnetic field to the firstchamber using a first-order matrix. The upper left plot in Fig. 7.2 shows the horizontal andvertical distributions of those protons at the first GEM in SBS. It should be noted that thehorizontal band is determined by the range of vertex positions in the target and is not limitedby the BigCal position resolution. The band is tilted due to the range of proton momentaaccepted. In the upper right plot the band has been rotated such that the correlation isminimized. The lower left plot shows a projection of the rotated vertical position with awidth of 4 mm, while the lower right plot in Fig. 7.2 a projection of the rotated horizontalposition with a width of 200 mm.

The x and y distributions correspond to a range of vertical and horizontal angles. In theupper left of Fig. 7.3 the vertical angle is plotted versus the horizontal angle after applyingthe cuts described above, showing a clear correlation between the SBS first track angle andthe position. In the upper right plot this band has been rotated such that the correlation isminimized. The lower left plot shows a projection of the rotated vertical angle with a widthof 1 mrad, the lower right plot a projection of the rotated horizontal angle with a width of

43

34 mrad.These simulations have determined that the area for the initial track search (as described

in Sec. 9.5) could be limited to 200×5 mm2 (horizontal×vertical) with an angular range of30×2 mrad2.

44

Horizontal position [mm] Rotated Horizontal position [mm]

Rotated Horizontal position [mm] Rotated Vertical position [mm]

Rot

ated

Ver

tical

pos

ition

[mm

]

Verti

cal p

ositi

on [m

m]

Coun

tsCoun

ts

Figure 7.2: The upper left plot is the vertical versus horizontal position in mm at the firstGEM in SBS for elastic events. The upper right plot is a rotation of the upper left plot. Thelower left plot is the rotated vertical position and has a width of 4 mm. The lower right plotis the rotated horizontal position and has a width of 200 mm.

45

Rot

ated

Ver

tical

ang

le [m

rad]

Ver

tical

ang

le [m

rad]

Rotated Horizontal angle [mrad]

Rotated Horizontal angle [mrad]

Horizontal angle [mrad]

Rotated Vertical angle [mrad]

Coun

ts

Coun

ts

Figure 7.3: The upper left plot is the vertical versus horizontal angle in mrad at the firstGEM in SBS for elastic events. The upper right plot is a rotation of the upper left plot. Thelower left plot is the rotated vertical angle and has a width of 1 mrad. The lower right plotis the rotated horizontal angle and has a width of 34 mrad.

46

Chapter 8

MC generation of the GEM data

Recommendation #5.5b: Simulate the full chain of track reconstruction in thefirst SBS tracker, starting with the signal readout, taking into account the real-istic pulse lengths and cluster widths.

Response

8.1 Front Tracker simulation framework

The flow diagram of the Front Tracker simulation is shown in Fig. 8.1 and the data structuresare presented in Sec. 8.4:

• GEANT4 is used to generate “real hits” (both for signal and background); each realhit is identified by the parameters listed in Tab. 8.1.

• Real hits are grouped in events (implemented in a TTree structure) which can bechained from different files.

• Real hits can be grouped into tracks by the “track enumerator” module.

• The digitization module (described below) scans the hits and provides “virtual strips”(one or more for each physics hit) (Tab. 8.3); virtual hits (implemented in TTree

structure) can be chained.

• A mixer code combines (sums) signal and background virtual strips charge to generatethe electronics pulses of the real strips (Tab. 8.4).

• The real strips are used to reconstruct the clusters (reconstructed hits) and then thereconstructed tracks by the reconstruction chain described in Sec. 9.4.

47

Figure 8.1: The flow diagram of the Front Tracker simulation framework, with the digitiza-tion module exploded.

48

8.2 Monte Carlo simulation

The Monte Carlo simulation, using the GEANT4 framework, has been designed taking intoaccount modularity and simplicity (for the end user). It includes quite detailed modelsof the target, the dipole magnet with its field clamps, the GEM trackers1 and a smallsilicon detector. Models are individual objects that inherit from a single class; they can beconfigured at run-time through the GEANT4 macro mechanism. Configuration parameterscan be stored in a dedicated flexible database, together with the Monte Carlo results, in aROOT-Tree output file. Figure 8.2 shows part of a possible SBS setup.

The physics processes can currently be selected from the long list of GEANT4 predefinedmodels. For the background estimate the GEANT4 QGSP BERT physics list has beenchosen (it is generally suggested for most GEANT4 applications) with an equivalent energycut of 1 keV (configurable). Figures 8.3 and 8.4 present examples of generated events forthe background calculation project with the magnetic field off and on, respectively.

The Monte Carlo main output is information about hits, including the energy loss in thesensitive materials (e.g. GEM gap) and the position, momentum and time of crossing (orabsorption). Each sensitive detector has a pseudo-unique identifier which tags each hit.

As mentioned, all this information is stored in ROOT-Tree files, together with the config-uration parameters, which are then processed by the Digitizer, which is discussed in detail inthe following section. No digitization is performed in the Monte Carlo itself. In this way, theextra code is configured for higher flexibility; the same Monte Carlo data can be used severaltimes to test different configurations of the GEM readout as well as different backgroundconditions.

8.2.1 Digitization

The digitization module implements the basic physical processes that generate the electronicsignals in the readout strips and the resulting pulses out of the readout electronics. Thestarting information is the energy deposited by the primary particle in the GEM chambersof the front tracker, simulated by the GEANT4 code (shown schematically in Fig. 8.5).

Ionization

Ion pairs are generated randomly (with a uniform distribution) along the primary track atpoints (xi, yi, zi). The pairs generated in the drift gap (between the drift plane and the firstGEM) are multiplied sufficiently to provide an observable signal in the readout plane2.

The average number of ion pairs nion generated by the primary particle is given by:

nion = ∆E/Wi

1The GEM chamber model includes all layers of a triple GEM as well as the electronics and the mechanicalsupports.

2Pairs generated in the other gaps have a lower chance of being detected due the smaller multiplication.

49

Figure 8.2: Part of a SBS configuration: from right to left: the scattering chamber (with thetarget), a small silicon strip detector, the dipole magnet with its field clamps and the GEMtracker with 6 chambers.

Figure 8.3: Events from the GEANT4 background simulation, without magnetic field.

where ∆E is the energy loss by the primary particle, and Wi is the effective average energyneeded to produce one ion pair in the gas (for Ar WAr

i = 26 eV)3. The number of ion pairs

3Replacing ∆E with dE/dx, the specific energy loss, one gets the number of ion pairs per unit path.

50

Figure 8.4: Events from the GEANT4 background simulation, with a 1.7 T magnetic field,which cleans up a lot of low-energy charged tracks.

Geometric Projection

Charge distribution

Drift

GEM Gap

GEM Gap

GEM Gap

Readout

DiffusingDrifting(x,y,z)

(xr,yr,zr)

Figure 8.5: Schematic view of the ionization, diffusion, drift and multiplication processes inthe GEM chamber.

originating the hit nion follows a Poisson distribution with a mean parameter nion.

51

Diffusion and Drift

Drift and diffusion times (velocities) are different for positive ions and electrons. The drifttowards the GEM holes and the readout plane depends on the electrostatic field. A typicalvalue of the average drift velocity is vd = 5 − 6 cm/µs, and the time to reach the readoutplane is given by tro ∼ L/vd, where L is the distance from the center of the drift region to thereadout plane (which is the average distance traveled by the secondary electrons generatedin the drift region)4.

The secondaries will diffuse perpendicular to the electrostatic field and drift along theelectrostatic field direction. The diffusion is basically described by the diffusion coefficientD (depending on the gas), which relates the standard deviation σs of the spatial chargedistribution at time t originated from a point at time 0:

σs(t) =√

2Dt

The diffusion coefficient in argon for electrons is 200 − 300 cm2/s. We can assume thatthe spatial distribution of the charge in the readout plane follows a Gaussian shape with astandard deviation of:

σs = σs(tro = L/vd) =√

2DL/vd (8.1)

Thus, the original ionization point (xi, yi, zi) due to the drifting of the electrons ends upin the projection (xhi , y

hi ) on the readout plane.

8.2.2 GEM multiplication

Each GEM multiplies the secondary electrons (and ions) by an average factor of g (relatedto the first Townsend coefficient α by g = n/n0e

αx, where x is the path where the inelasticprocesses responsible for the multiplication occur). Assuming that the electron attachmentand molecular dissociation are negligible, in a uniform electric field the secondaries followthe Polya distribution [12] quite well:

fPolya(n) =b

n

1

(b− 1)!

(b(n− 1)

n

)b−1exp

(−b(n− 1)

n

)where n is the mean avalanche size (which is g), and b = (1+θ). θ depends on the electrostaticfield and the gas properties by θ = kWiα/E, where k is a constant and E the electrostatic(constant) field. Basically, θ accounts for the variation of α with n: α(n) = α(1 + θ/n). ThePolya distribution becomes a Poisson distribution for b→∞, while it is a Furry distributionfor θ → 0 (or b→ 1):

fFurry(n) =1

nexp

(−nn

).

The latter limit is probably more applicable in a GEM (due to the large electrostatic fieldE in the GEM holes).

Let us consider both cases:4More realistically L is the distance between the ionization point along the track in the drift region and

the corresponding projection in the readout plane and therefore depends on the generated initial pair.

52

Poisson amplification

Assuming that the multiplication in each GEM foil follows a Poisson process, the totalaverage gain G will be the product of the gain in each single foil gi:

G =

nGEM∏1

gi

The distribution of the gain G is not a Poisson distribution (the fluctuation of the mul-

tiplication σG is much larger than√G, but is of the order of the statistical propagation of

the fluctuation); however, it is similar to a Gaussian distribution (central limit theorem).In fact,

σG ∼ G/√g0, (8.2)

where g0 is the mean gain of the first GEM foil (see Fig. 8.6). In this case, we can approx-imate the gain Gi (number of collected electrons from the single ion pair i) by a Gaussiandistribution with mean G and sigma σG:

Gi(xG) = exp{−(xG −G

)2/(2σ2

G)}

(8.3)

Furry Amplification

In this case we assume that the distribution of the electron avalanche after a single GEMfoil follows the Furry distribution (see above). The RMS, after several GEM foils, is similarto the mean value, with the gain parameter replaced by the actual gain (Fig. 8.7), that isσG ∼ G.

In this case the gain Gi is distributed almost like a Furry distribution:

Gi(xG) ∼ exp(−xGG

) (8.4)

8.2.3 Charge Collection

In a first approach, the spatial distribution of the hit charge collected was assumed to bethe sum of Gaussian distributions centered at each projection (xri , y

ri , z

ri )

5 of the original pairproduction point (xi, yi, zi) in the drift gap:

Ghit(x, y) =

nion∑i=1

Gi exp{−((x− xri )

2 + (y − yri )2)} /(2σ2

s(i)) (8.5)

where Gi is given by Eq. 8.3 or 8.4 and σs by Eq. 8.1.

5zri is constant and equal to the position of the readout plane

53

secondaries [x 20.0]0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6

0

5000

10000

15000

20000

25000

30000

35000

40000

45000

secondaries [x400.0]0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8

0

2000

4000

6000

8000

10000

12000

14000

16000

18000

secondaries [x8000.0]0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

0

2000

4000

6000

8000

10000

12000

14000

16000

18000

20000

210

310

10

210

310

RMS vs Mean

Figure 8.6: Distribution of the number of electrons after 1 (top-left), 2 (top-right) and 3(bottom-left) GEM foils, assuming a single secondary pair generated in the drift, a constantgain of 20 per GEM foil and a Poisson distribution; in red a Gaussian fit is shown. In thebottom-right figure, the RMS of the distribution is plotted versus the mean value. Note thatthe predicted RMS value of a pure Poisson distribution would be RMSPoisson =

√8000 = 89.4

whereas the RMS predicted by the simulation is RMSstat ∼ 8000/√

20 = 1788.

54

secondaries [x 20.0]0 0.5 1 1.5 2 2.5 3 3.5 4

310

410

secondaries [x400.0]0 1 2 3 4 5

10

210

310

410

secondaries [x8000.0]0 1 2 3 4 5 6

1

10

210

310

410

210 310

210

310

RMS vs Mean

Figure 8.7: Distribution of the number of electrons after 1 (top-left), 2 (top-right) and 3(bottom-left) GEM foils, assuming a single secondary pair generated in the drift, a constantgain of 20 per GEM foil and a Furry distribution; in red, an exponential fit (y3×GEM =(8.8361± 0.0019) · exp(−(1.0163 ± 0.0013)x/g3)) is shown. In the bottom-right figure theRMS of the distribution is plotted versus the mean value. Note that the predicted RMS of apure Poisson distribution would be RMSPoisson =

√8000 = 89.4, whereas the RMS predicted

by the simulation is RMSstat ∼ 7700, very similar to the mean gain (8000).

55

This assumption did not reproduce a proper charge sharing between the two layers of theCOMPASS-like 2-dimensional readout plane. We therefore replaced the Gaussian distribu-tion by a constant distribution, which reproduced quite well the measured charge distributionin the literature [13, 14].

Therefore, Eq. 8.5 has been replaced by6:

Ghit(x, y) =

nion∑i=1

Gi ·H((f · σ2

s(i))− ((x− xri )2 + (y − yri )

2)), (8.6)

where H() is the properly normalized Heaviside step function (zero for negative values, 1 forpositive) and f = 3.

Implementation details of the response of the strips of the COMPASS-like 2-dimensionalreadout plane are presented in Sec. 8.3.

8.2.4 Pulse Formation and Timing

The shape of the analog pulse from the electronics coupled to a silicon detector is approxi-mated by [15]:

vout = At

T 2P

e−t/Tp

where Tp is the shaping time (∼50 ns, which provides a total width of the signal of about250 ns Fig. 8.8). More realistically, the shape from the electronics coupled to a GEM isrepresented by [16]:

vout = Aτ0 + τ1τ 21

(1.− e−(t−t0)/τ0

)e−(t−t0)/τ1

where t0 is the time shift and τ0 and τ1 the time constants that contribute to the width ofthe pulse. Figure 8.9 compares these two expressions, which appear to differ mainly in thelength of the tail. The normalization factor A is the total charge on the strip (the Ii or Ij ofEqs. 8.7 and 8.8, respectively)7.

In our GEM implementation, the time-dependent pulses are sampled by the APV25 every25 ns. nsample = 3 adjacent samples are transferred from the APV25 to the DAQ systemand are available as raw data from each single strip. If a particle is detected in coincidencewith the trigger, the three samples are synchronized and therefore always represent the samepart of the function (with a fluctuation related to the jitter of the trigger and the function).For background-generated hits, the three adjacent samples are randomly distributed and thesampling is random relative to the starting time of the signal. The simulation takes intoaccount an effective time window ∆t of about −tsignal to nsample · tsample, where tsignal is thewidth of the signal8, while tsample is the sampling time. The background sums to the chargegenerated by the signal.

6The most realistic function is something closer to the uniform distribution with smooth borders.7The normalization factor includes the APV25 and the ADC gains.8The signal width should be estimated as the over-threshold time of the signal, where the threshold is the

sensitivity of the electronics; in first approximation one can assume the Full Width at One-Tenth Maximum.

56

Figure 8.8: APV25 sampler output values with two different fit functions applied (from [15]).

t [ns]0 50 100 150 200 250 300 350 400 450 5000

50

100

150

200Realistic GEM-APV Simon Thesis (2001)

Delta-APV Simon Thesis (2001)

SiD-APV NIMA572(2007)385

APV25 shaped output

Figure 8.9: Different functions (not normalized) modeling the pulse shape coming out of theAPV25; from [15] and [16] (see also Fig. 8.8).

57

8.2.5 Digitization Algorithm

The digitization algorithm implements the above modeled processes. It uses the outputof the GEANT4 MonteCarlo (Tab. 8.1) as well as the geometry parameters and physicsparameters of the GEM tracker (Tab. 8.2) and produces the TTree output (Tab. 8.3).

Here are summarized the main steps of the algorithm, whose flow chart is similar to thatpresented in Fig. 8.10.

1. Project the track segment in the drift gap onto the readout plane.

2. Assume nion (xri , yri ) points are Poisson distributed with a mean nion in the projected

segment.

3. These points are assumed to be uniformly spatially distributed (along the segment)(implementation: extract nion values of length from 0 to Ls. Ls is the length of theprojected segment).

4. Each point is the center of a 2-dimensional uniform distribution extending for 3 · σs(Eq. 8.1).

5. The integral (total charge for each secondary) of the Gaussian is given by G distributedaccording to a Gaussian function with a mean G and σG given by Eq. 8.2 or a Furryfunction (see Eq. 8.4).

6. Sum all uniform spots (Eq. 8.6) centered in the projection of the primary ionizationpoints.

7. Choose a rectangular window around the projected segment as suggested in Sec. 8.2.3.

8. Integrate the window pixels to get the charge of each strip in x and y (note thatintegration along one axis is continuous, while along the other axis it is in steps, seeFig. 8.11 and discussion above).

9. For each strip in the window take three consecutive 25 ns samples of the signal accordingto Eq. 8.2.4 starting at a fixed time (or point in the signal, adding a random Gaussianjitter).

10. Repeat all the above steps for the background hits, except step 9, which will be replacedby:

(a) extract the number of background hits nbck according to a Poisson distributionwith a mean rbck · ∆t where rbck is the background rate; (in principle, spacedependent!) and ∆t the effective time window (see Sec. 8.2.4).

(b) uniformly extract nbck times tribck between −tsignal and nsample · tsample;(c) for each strip, take three consecutive samples starting at the random points ex-

tracted in the previous item, relative to the “0” of the GEM signal.

58

Figure 8.10: The generic digitization flow chart diagram.

59

����������������

����������������

����������������

����������������

����������������

����������������

������������

������������

������������

������������

������������

������������

������������

������������

������������

������������

���������

���������

���������

���������

������������

������������

������������

������������

������������

������������

���������

���������

���������

���������

������������

������������

������������

������������

������������

������������

���������

���������

���������

���������

����������������������������������

��������������������������������������������������������������������

����������������������������������

��������������������������������������������������������������������

Pixel Center (k, m)

0.4 mm

0.4 mmTOP strip (j)

BOTTOM strip (i)

Rectangular Pixel 0.4x0.2 mm2

Project Drift Track

x

y

A

B

yL

yU

xL xUa

b

Figure 8.11: Readout 2D strip plane modeled by a 2D regular array of equidistant rectangularbins (assuming a perfect balance of the signal between x and y); the bin size is the strip pitchalong one axis (a) and the semi-pitch along the other (b). The segment A − B representsthe projection of the drift track onto the readout plane. The (xL, yL) and (xU , yU) are thelower and upper limits of the area involved in the charge collection (a definition is given inthe text).

60

8.3 Charge Collection Implementation

As represented in Fig. 8.11, the x-y strips of the COMPASS like 2-dimensional readout canbe modeled by a 2-dimensional regular grid of size Lx × Ly, each “pixel” being addressedby the indices k,m running in the ranges k = [0, Lx/a − 1] and m = [0, Ly/b − 1]). In thisrepresentation, the horizontal strips (top layer, the y axis) are continuous while the verticalstrips (bottom layer, the x axis) are interleaved with the horizontal strips.

Each strip can be represented by an index i (and j for the other axis). In this scheme,the j strip is formed by the pixels ∀k,m = 2j, while the i strip is formed by the pixelsi = k,∀(2m+ 1). Each pixel has its center at (xk, ym) = (x0 + a · k, y0 + b ·m), a and b arethe pitch along the two directions and x0, and y0 are the offsets that define the origin ofthe coordinates (for example, x0 = a/2 and y0 = b/2). The Gaussian sum of Eq. 8.5 can bedefined in a rectangular window (Fig. 8.11) that includes the segment projection containingthe (xri , y

ri ) points. The coordinates of the lower-left (xL, yL) and upper-right (xU , yU) of this

rectangle are:xL = xrA − fσmaxs , yL = yrA − fσmaxs

andxR = xrB + fσmaxs , yU = yrB + fσmaxs

where σmaxs =√

2DLmax/vd (Lmax is the distance between the drift foil and the readoutplane) and xA,B, yA,B are the minimum and maximum values of the xri and yri coordinates,while f is the coverage factor (f = 2 should be reasonable).

For implementation purposes, instead of (xL, yL) and (xU , yU), one considers the clos-est (safer) pixel centers defined by (kL,mL) and (kU ,mU) as: kL = floor ((xL − x0)/a),mL = floor ((yL − y0)/b) and kU = ceil ((xU − x0)/a), mU = ceil ((yU − y0)/b) (with theconstraints on k and m range of validity).

The above indices are used to define a 2D histogram H2(nk, xl, xr;nm, yb, yu) where nk =kU−kL+1 and nm = mU−mL+1 and borders: xl = x0 +a(kL−1/2), xr = x0 +a(kU +1/2)and yb = y0 + b(mL−1/2), yu = y0 + b(mU +1/2) and the bin content is given by the charge:

C(k,m) =

∫ xk+a/2

xk−a/2dx

∫ ym+b/2

ym−b/2dy ·Ghit(x, y)

and therefore the charge collected by the i-strip is given by:

Ii =∑∀m

C(k = i, 2m+ 1) (8.7)

and for the j-strip it is given by:

Ij =∑∀k

C(k,m = 2j) (8.8)

61

8.4 Data structure implementation

The different data structures refer to each other as shown in Fig. 8.1 by the “ref to” dashedlines, in order to permit a comparison of the reconstructed tracks to the physics tracks; theimplementation is done using the Entry of the TTree.

Table 8.1: GEANT4 Monte Carlo physics hit parameters definition; these data are usedduring digitization; when applicable, the parameters are evaluated in the local chamber.NOTE: The names of the variables will be adapted to the ROOT code convention.

Physics HithcID chamber indexhtID particle ID entering the drift gap (htID = 1 is the primary particle)hPar particle code(hx, hy, hz) entrance point at the drift gap(hxe, hye, hze) exit point at the drift gaphedep energy deposited in the drift gap (∆E)(hmx, hmy, hmz) entrance momentum at the drift gap(hmxe, hmye, hmze) exit momentum at the drift gap(hxro, hyro, hzro) entrance point in the readout foil

Table 8.2: Geometry and physics parameters of the GEM chamber used in the digitization.

Lmax drift to readout planes distanceWi effective average ionization energy of the gas mixtureD diffusion coefficient of the gas mixturevdrift mean drift speed of the secondary electronsgi gain of each GEM foil (due to gas mixture and field in the GEM holes)

62

Table 8.3: Parameter definitions of the digitized virtual strip

Virtual Stripchamber chamber indexplane axis indexns number of strips influenced by hit(s)strip[ns] strip index arraycharge[ns] total charge in stripadc[ns][3] ADC sampled chargesType type of strip (signal=0x1, background=0x2 ...)pHit reference to the physics hit entry (or index)pTrack reference to the physics track entry (to be implemented)

Table 8.4: Parameter definitions of the mixed real strip (sums of virtual strips). Onlya one-dimensional array is used, due to some limitation/or not understood behavior inROOT/TTree

Real Stripdigi.gem.nch (nch for short) number of strips with signaldigi.gem.chamber[nch] chamber indexdigi.gem.plane[nch] plane index of the stripdigi.gem.strip[nch] strip address in a given axisdigi.gem.adc1[nch] ADC first sampled valuedigi.gem.adc2[nch] ADC first sampled valuedigi.gem.adc3[nch] ADC first sampled valuedigi.gem.type[nch] type of strip, registerdigi.gem.charge[nch] total charge collected in stripdigi.gem.time1[nch] time of first sample

63

Chapter 9

Tracking Efficiency

Recommendation #5.5c: Demonstrate the expected tracking efficiency and thelevel of contamination by false tracks using an algorithm for hit recognition andprojection matching.

Response

9.1 The GEM tracker concept

The first and currently most advanced GEM-based tracker was constructed by the COMPASScollaboration [17, 18]. The key features of the COMPASS GEM tracker will be reproducedin the SBS tracker. Therefore, we start with a brief description of the COMPASS tracker.

9.1.1 The COMPASS GEM tracker

The COMPASS GEM detectors consist of three GEM amplification stages, stacked on topof each other, and separated by 2 mm tall thin spacer grids. Such a design, developed forCOMPASS, guarantees safe and stable operation without electrical discharges in a high-intensity particle beam, and has been adopted by various other experiments. The detectorsare operated in an Ar/CO2 (70/30) gas mixture, chosen for its large drift velocity, lowdiffusion, non-flammability, and non-polymerising properties. The electron cloud emergingfrom the last GEM induces a fast signal on the readout anode, which is segmented into stripswith a pitch of 400 mm each. For each particle trajectory, one detector consequently recordstwo projections of the track with highly correlated amplitudes, a feature which significantlyreduces ambiguities in multi-hit events. The active area of each GEM detector is 31×31cm2. The central region, where the beam crosses the detector, is deactivated. The materialbudget for one detector, corresponding to the measurement of two projections of a particletrajectory, amounts to 0.4% of a radiation length in the center, and to 0.7% in the periphery.

Two GEM detectors are mounted back-to-back, forming one GEM station. One detectoris rotated by 45◦ with respect to the other, resulting in the measurement of a charged

64

particle trajectory in four projections (labeled XY and UV). The detection efficiency for aparticle trajectory in at least one of the two projections, averaged over all GEM detectorsin COMPASS, was determined to be 97%. An offline clustering algorithm combines hitsfrom adjacent strips to yield an improved value for the position of a particle trajectory.The average number of strips per cluster is 3.1 for the top layer of strips and 3.6 for thebottom layer, consistent with the lateral diffusion of the charge cloud in the GEM stack. Theresolutions for all GEM planes in the spectrometer are found to be distributed around anaverage value of 70 µm, which includes a contribution of overlapping clusters due to pile-upin high intensity conditions.

In total, 11 GEM detector stations, i.e. 22 detectors, are installed in COMPASS. De-pending on the position in the spectrometer, particle rates as high as 25 kHz/mm2 areobserved close to the central inactive area, equivalent to a total collected charge of morethan 2 mC/mm2.

9.1.2 The SBS GEM tracker

The SBS spectrometer will be used with several configurations of the detectors. The GEp(5)configuration, which will operate at the highest luminosity, is discussed below. There arefour trackers in GEp(5): the high-resolution front tracker located behind the 48D48 magnet,two low-resolution trackers in the polarimeter, and a low-resolution one-direction tracker infront of the electromagnetic calorimeter.

The front tracker will be composed of six planes of GEM chambers, separated by 15cm. Each chamber will consist of three modules of 40 × 50 cm2, forming a total active areaof 40 × 150 cm2. The readout is organized via X and Y strips with a 0.4 mm pitch andAPV25 based Front-End-Electronics. Custom VME electronics will be developed to processdata and filter the out-of-time hits. More details on the physical design and the readoutof the GEM module is given in Sec. 6. The projected rate of background hits in the fronttracker is close to that in the COMPASS experiment, in which according to Ref. [18] thestrip occupancy reached 22%. The material budget for the SBS GEM chamber structurehas been reduced compared to the COMPASS one by 10% (in terms of radiation length).However, for the MC study presented here we have used a GEM structure almost identicalto the COMPASS design.

9.2 Intensity of the background particles

The data from the GEM tracker were simulated by using a GEANT4 model of the tracker(Sec. 8) and the spectra of the background particles, which were generated in advance.This section describes a GEANT3 model for the equipment setup which was used for thecalculation of the background and a number of its important features, allowing a reductionof the background yield.

65

9.2.1 The experimental layout

The flux of the particles incident on the GEM detector was calculated for the configurationpresented in the Review Meeting on 1/22/2010 with the exception of the SBS spectrometerangle. The spectrometer central angle was changed to 16.9◦ from the previously used valueof 12◦, following the PAC35 approval of the measurement of the proton electric form factorto a maximum momentum transfer of 12 GeV2. The front tracker (FT) with dimensions 40× 150 × 75 cm3 in Fig. 9.1 has been placed at a distance of 344 cm from the target.

40−cm long LH2 target

Vacuum snout

Beam

48D48 Magnet

Field clamp

Lead

Lead

FT

Lead

Coil

Helium bag

Figure 9.1: The layout of GEp(5) for the MC simulation of the background of in the planeof the Front Tracker (FT).

A plan view of the experimental layout is shown in Fig. 9.1. The beam of 11 GeVelectrons with a current of 75 µA impinges on the 40-cm long liquid hydrogen target. Thetarget is a 2.5-cm diameter Al cylinder with 0.8 mm side walls and 0.1 mm beam windowsfilled with liquid hydrogen at a density of 0.072 g/cm3. The scattering chamber is extendedup to the field clamp by means of a wide vacuum snout. This configuration avoids a directview of the beam line components from the detector. Specifically, in this layout there isno beam line pipe from the scattering chamber until the beam enters the massive materialof the field clamp. The main idea of the beam line concept is to make the first element ofthe beam line downstream of target sufficiently thick to absorb the electromagnetic shower.For example, a lead insert was placed between the field clamp and the yoke of the 48D48magnet to contain the shower originated in the material of the field clamp. With such adesign the shower created by a high-energy electron scattered from the target to the wall

66

of the beam line will be absorbed and background particles will not reach the detector. Athin-walled helium-filled container is installed between the vacuum window of the snout andthe detector. This improves the angular resolution of the SBS spectrometer. Behind the48D48 magnet, the beam line is covered by an additional lead pipe to contain the showerinduced by stray electrons which strike the beam line vacuum tube inside the 48D48 magnetyoke. Because of the strong magnetic field in the dipole (1.4 T), the low-energy chargedbackground does not reach the detector.

9.2.2 Results of Monte Carlo simulation

The Monte Carlo code for simulation of the background production was written by V. Nelyu-bin in the framework of GEANT3. This code is based on a model presented in the previoussection. The intensity of the photons and other particles was calculated at the Front TrackerFT (Fig. 9.1).

Background flux at the magnet exit aperture

It is interesting to first make a background estimate using the simplified approach that thebackground is produced only in the hydrogen of the target, and that all charged particleswith a momentum below 1 GeV/c are swept away by the 48D48 magnet. We made such anestimate using the DINREG MC calculation performed by P. Degtearenko. The pion andthe photon spectra are shown in the left and the right panels, respectively, of Fig. 9.2.

From these plots we find that, at the detector location, the photon intensity (for energyabove 10 keV) is 150 MHz/cm2 and the pion intensity (for energies above 1 GeV, which canreach at least a part of the detector) is 8 kHz/cm2. Such an estimate shows that in spiteof a low detection probability, the photons induce a much higher GEM detector rate thanthe pions. This estimate provides a lower limit for the photon flux in the real experimentbecause the experimental setup includes the target cell, beam line and many additionalelements which lead to an increase in the rates. The total rates of particles at the locationof the FT are shown in Table. 9.1. The rate of hadrons is not included in the table because,according to our previous estimate, it is relatively low.

Particle Intensity, MHz/cm2

photon 255.0electron 0.16

Table 9.1: The particle rates at the detector location.

The photon energy spectrum is shown in Fig. 9.3 for the energy interval from 0 to 1 MeVand also for the interval from 0 to 100 MeV (with 1 MeV per bin). About 30% of the totalphoton flux corresponds to energies below 100 keV. In spite of this low-energy dominancephotons with energies above 1 MeV are also important because the probability of photon

67

e + H + X at Ee = 11 GeV (2840.0 mg/cm2 target)

T (MeV)

Points and solid line: GDINR M.C. calculation

Ne -1

dN

/dT/

d# (

elec

tron-1

MeV

-1 s

r-1)

0.0o 10.0o(1) 10.0o 45.0o(2) 45.0o 75.0o(3) 75.0o 105.0o(4)105.0o 135.0o(5)135.0o 180.0o(6)

10-9

10-8

10-7

10-6

10-5

10-4

10-3

10-2

10-1

1

10

10-1

1 10 102

103

104

e + H + X at Ee = 11 GeV (2840.0 mg/cm2 target)

(degrees)

Right scale: Detector Load (events/sec)Assuming beam current 10 ∀Aand detector solid angle 0.1 sr

Ne -1

dN

/d#

(el

ectro

n-1 s

r-1)

T 0.10 MeV(1) T 0.32 MeV(2) T 1.00 MeV(3) T 3.16 MeV(4) T 10.00 MeV(5) T 31.62 MeV(6) T 100.00 MeV(7) T 316.23 MeV(8) T 1000.00 MeV(9) T 3162.28 MeV(10)

10-7

10-6

10-5

10-4

10-3

10-2

10-1

1

10

10 2

0 20 40 60 80 100 120 140 160 180

10 6

10 7

10 8

10 9

10 10

10 11

10 12

10 13

10 14

10 15

γ

γ

−>

−>

θ

< Θ <

< Θ << Θ <

< Θ << Θ << Θ <

<<<<<<<<

<<

e + H + X at Ee = 11 GeV (2840.0 mg/cm2 target)

T (MeV)

Points and solid line: GDINR M.C. calculation

Ne -1

dN

/dT/

d# (

elec

tron-1

MeV

-1 s

r-1)

0.0o 10.0o(1) 10.0o 45.0o(2) 45.0o 75.0o(3) 75.0o 105.0o(4)105.0o 135.0o(5)135.0o 180.0o(6)

10-9

10-8

10-7

10-6

10-5

10-4

10-3

10-2

10-1

1

10

10-3

10-2

10-1

1 10 102

103

104

105

e + H + X at Ee = 11 GeV (2840.0 mg/cm2 target)

∃ (degrees)

Right scale: Detector Load (events/sec)Assuming beam current 10 ∀Aand detector solid angle 0.1 sr

Ne -1

dN

/d#

(el

ectro

n-1 s

r-1)

T 0.01 MeV(1) T 0.10 MeV(2) T 0.32 MeV(3) T 1.00 MeV(4) T 3.16 MeV(5) T 10.00 MeV(6) T 31.62 MeV(7) T 100.00 MeV(8) T 316.23 MeV(9) T 1000.00 MeV(10) T 3162.28 MeV(11)

10-7

10-6

10-5

10-4

10-3

10-2

10-1

1

10

10 2

0 20 40 60 80 100 120 140 160 180

10 6

10 7

10 8

10 9

10 10

10 11

10 12

10 13

10 14

10 15

<

π

π

−>

−>

< Θ << Θ << Θ << Θ << Θ << Θ <

<<<<<<<<<<

Figure 9.2: The photon and the pion yields from a 40-cm long target with an 11-GeV electronbeam.

induced ionization in the GEM sensitive volume increases by a factor of 5 to 10 with thephoton energies (Fig. 9.4).

Spectra and origin of the background

We would like to trace the locations where the background is produced. The obvious place isthe 40-cm long liquid hydrogen target, LH2. The target has a direct view of the detector (FTin Fig. 9.1). Additional background was produced when an electron scattered from the LH2target at a significant angle interacts with the material of the beam line or the spectrometercomponents. The intensity of the background has been carefully calculated. As expected,photons represent the dominant type of particles at the FT location. With our design ofthe beam line, the main part of the photons which reach the detector are emitted from thetarget area as illustrated in Fig. 9.5.

68

Eγ , MeV

γ-ra

te,

MeV

-1 c

m-2

ele

ctro

n-1

11 GeV, SBS at angle of 16.9o

10-7

10-6

0 0.2 0.4 0.6 0.8 1

11 GeV, SBS at angle of 16.9o

Eγ , MeVγ-

rate

, M

eV-1

cm

-2 e

lect

ron-1

10-11

10-10

10-9

10-8

10-7

0 20 40 60 80 100

Figure 9.3: Photon energy spectrum at the magnet exit aperture.

Photon energy, MeV

Prob

abili

ty, %

Figure 9.4: Probability of photon-induced ionization in a GEM.

It is interesting to note that low energy photons, which caused most of the GEM detectorrate in the present setup (at 16.9◦), are emitted by low-energy electrons and a significantpart of those photons are emitted in the target walls, see the 2-dimensional map and theprojection (the right plot) in Fig. 9.5.

9.3 Event generation and data digitization

An event in the SBS tracker, which has a track of a few GeV proton in the presence ofbackground, is generated using a custom code in the framework of GEANT4 written byE. Cisbani. Only one third of the front tracker, 40 × 50 cm2, has been implemented, butwith all six planes (Sec. 8). The event information contains the proton track coordinates as

69

to SBS

Z, cm

Rate

radius , cm

Eve

nts

0

5000

10000

15000

0 0.5 1 1.5 2- 14 14 0

0

1.2

-1.2

Figure 9.5: Photon vertex coordinate distributions. The left plot shows a distribution alongthe beam line. The central plot shows a 2-dimensional map. The right plot shows the radialvertex distribution.

well as the data from all strips which have a non-zero amplitude in at least one of the threeamplitude samples.

The full machinery of event generation and digitization is presented in Sec. 8. The dataare generated with a uniform time distribution of the background in a wide time intervalcorresponding to a 250 ns integration time of the APV25 electronics (photons were generatedfrom -400 ns to +100 ns, relative to the time of the proton hit). The proton track is generatedwith time variations in accordance with the GEM chamber characteristics. The intensitiesand energy spectra of the background are obtained from the MC presented above in Sec. 9.2.Three amplitude samples with 25 ns intervals are used for data analysis. However, all 10samples within a 500 ns time window are used to study the data.

9.4 Track Reconstruction

9.4.1 Overview

A complete track-reconstruction chain was developed for the analysis of the Monte Carlodata allowing a detailed study of the tracking performance. The algorithm is based onthe existing track-reconstruction software for the Hall A BigBite spectrometer [19]. In thisalgorithm first 2D track projections are found for each tracker coordinate, and then trackprojections are matched and combined to 3D tracks. The 2D pattern recognition relies onthe TreeSearch algorithm [20], a very fast, recursive template matching method. TreeSearchrequires straight tracks and at least three tracker planes per projection, conditions which aresatisfied in the SBS.

The BigBite tracking software was significantly modified for the SBS analysis. Theanalysis of drift times measured by BigBite’s horizontal drift chambers was replaced by

70

algorithms for finding and fitting clusters of active strips in the GEMs; the special handling ofthe left-right ambiguities of drift chambers was removed; and a new algorithm was developedto filter out the out-of-time hits. Additionally, a new algorithm was developed to exploitamplitude and time correlations of GEM hits in shared readout planes. Figure 9.6 shows theflow diagram of the track reconstruction framework. Each component is briefly described inthe following.

Patterns

Hits Bins

Hits in 3D

3D Track

DecoderFinding Building

2D

FittingFinal tracks

3D

Cluster HitpatternTreeSearch

De−Cloning

2D Track

Fitting

Projection

Matching

3D Track

De−CloningCandidates

fileROOT ADC Strips

Roads

Projections

Track

Data

Figure 9.6: The track-reconstruction chain used in the analysis of simulated GEM trackerdata. Blue text indicates the types of data objects passed between the various components.See text for a detailed description of each analysis step.

9.4.2 Simulation Decoder

The simulation output is stored in TTree objects in ROOT files, using the data structuresdescribed in Sec. 8.4. The simulation decoder reads such ROOT files event by event andconverts the data to a crate/slot/channel/hit format that is identical to the way the dataare presented by the DAQ hardware. In this way, exactly the same reconstruction chain canbe used for real and simulated data. Only the type of data decoder needs to be changed toswitch between input data formats.

The simulation decoder also extracts the Monte Carlo truth data (generated track pa-rameters, hit types, etc.) and makes them available for a direct comparison to reconstructedevent data within the Hall A analyzer framework.

9.4.3 GEM Cluster Finding

Tracks passing through the GEMs generate a charge cloud, which is collected by usuallymore than one strip in the readout plane. To reconstruct the physical hits, the softwaremust find the corresponding clusters of strips. The integral of the signal amplitudes of all

71

the strips within each properly detected cluster is then approximately proportional to thetotal charge deposited by the hit.

Before finding the clusters proper, the software first pre-processes the amplitude dataof each individual strip separately. Using the three amplitude samples measured for eachstrip along the time axis, one can efficiently reject noise hits and signals corrupted by pileup.In addition, the signal amplitude proper can be deconvoluted from electronics effects byanalyzing the amplitude ratios of the three samples. Details of this analysis step are presentedin section 9.5.3. Only strips with “clean” signals are used for further analysis. Both theirdeconvoluted signal amplitude and the signal time with respect to the trigger are availableat this point.

As the next step, clusters are found in each GEM readout plane along the coordinateaxis. A cluster is defined as a group of adjacent strips whose signal amplitudes first rise,then peak and drop again as a function of strip position. Once a strip with no signal (i.e.a gap) or a second rise of amplitude with position is detected, or if the pre-set maximumnumber of strips per cluster has been reached, the cluster is complete.

For each cluster, the sum of the strip amplitudes as well as an amplitude-weighted hitposition are calculated. Weighting the strip positions with their respective signal amplitudesyields a very precise estimate of the actual hit position. We achieve a position resolution ofbetter than 50 µm with Monte Carlo data (cf. Fig. 9.10). The position, amplitude and timedetermined for each cluster are stored in Hit objects for further analysis.

9.4.4 Hit pattern Construction

The hit positions found are translated into active bins in a set of binary “hit patterns” foreach coordinate. These hit patterns form the input of the subsequent TreeSearch algorithm.As described in [20], TreeSearch performs pattern matching on hit patterns whose number ofbins doubles with each iteration (search depth). The maximum search depth is best chosenso that the bin width is slightly larger than the expected hit position resolution. For themaximum tracker dimension of 50 cm and an estimated position resolution of ∼50 µm searchdepths of 11 and 12 (corresponding to bins of 244 or 122 µm width, respectively) were used.

9.4.5 TreeSearch

In this analysis, TreeSearch is implemented as a Visitor pattern [21]. The hit pattern “visits”all the elements of the template database. At each search depth, starting with the lowest,the hit pattern of the event is compared with pre-computed templates that correspond tostraight lines. If a match is found, the search proceeds recursively to higher depths (morefinely grained bit patterns). Patterns that match at the deepest level (maximum resolution)are saved. They do not yet represent real track candidates, but rather “regions of interest”,where hits in all the planes are consistent with a straight line within the (relatively coarse)hit pattern resolution.

72

9.4.6 2D De-Cloning & Road Construction

For each real track the output of TreeSearch is almost always a set of similar (usually neigh-boring) patterns which all describe the same track. Such “clone patterns” occur primarilydue to (1) the finite resolution of the position measurement, which may require two adjacentbins to be set if a measured position is near a bin boundary; and (2) the need to allow formissing hits in one or more planes to account for inefficiencies. A single track may activatebetween 0 and 2 bins in each readout plane, resulting in typically 5–10 patterns per track.

To minimize the generation of “clone tracks” (duplicates with very similar parameters),groups (clusters) of patterns that describe the same track must be detected. This is the classicclustering problem of computer science. There are no exact but only statistical solutions,especially since the number of clusters to be found is a priori unknown.

In this analysis, a heuristic clustering algorithm was used in which clusters of patternsare formed from patterns with common hits. Each resulting group of patterns is saved asa so-called Road for further processing. Like Patterns, Roads represent regions of interestwhere a track might exist.

9.4.7 2D Projection Track Fitting

A straight 2D line is fitted to the hits within each Road. If several hits are present in oneor more planes, all combinations are fit, and the best fit is kept. The assumption here isthat the additional hits are noise or, less likely, belong to a different real track. If the χ2

of even the best fit is too large, the entire Road is discarded, providing a certain degree ofnoise filtering. In the end, each “good” Road is associated with a single 2D track projectioncandidate.

9.4.8 Projection Track Matching

To construct a full track, matching combinations of 2D track projections from differentcoordinates must be found. If three or more coordinates are measured by the trackers, forexample with strips in u, v and x directions, this task is straightforward. The measurementin x can be used to correlate the linearly independent u and v measurements. In the case ofGEMs, however, two coordinate measurements, say x and y, are already sufficient. Insteadof using a third coordinate measurement, one can use amplitude and time information ofeach x and y hit in a shared readout plane to correlate the track coordinates. Even if such acorrelation is relatively weak for a single chamber, this becomes quite effective with a largenumber of like chambers.

For instance, with a 50% probability of getting a random amplitude correlation in asingle shared readout plane, the probability that a given pair of x and y tracks with hits in6 planes is randomly correlated is 0.56 ≈ 1.6%. At the same time, requiring hits to have a3.3σ amplitude correlation in all 6 planes results in a 1−0.9996 ≈ 0.6% tracking inefficiency.Any randomly correlated track projections that survive at this stage will be rejected veryeffectively in the subsequent two analysis steps.

73

These numbers do not take into account the effect of missing hits, which may occurdue to detector inefficiencies or clusters corrupted by pileup. The tracking performance wasstudied for the case of allowing an amplitude mismatch in 2, 3, or 4 planes, correspondingto increasing contamination by accidental correlations but decreasing tracking inefficiency.

9.4.9 3D Track Fitting

Each combination of track projections found in the previous step is fit in 3D space, usingthe linear minimization algorithm (Cholesky decomposition) similar to the one implementedin the TLinearFitter class of ROOT [22]. The resulting 3D tracks are sorted by χ2.

A small number of clone tracks may still be present in the output of the 3D fitting stagesince the 2D clustering described earlier is not always fully efficient due to ambiguities. Theseremaining clones are rejected in a final step by disallowing sharing of any hits between finaltracks. If any tracks share one or more hits, only the track with the best χ2 survives.

9.5 Data analysis

The basic logic of the data analysis presented below is based on the time/amplitude andcoordinate information from the tracker. Specifically, it does a search for a straight trackconstrained by additional information about the search area at the moment coinciding withthe trigger. Specific features of the GEM detector used in the analysis are:

• segmented readout planes (0.4 mm strip pitch);

• APV25 clock time (25 ns);

• integration time of the front-end electronics (129 ns);

• correlation between the amplitudes of the x- and y-strips for the same hit;

• amplitude information from each strip recorded in three consecutive moments of timespaced by 25 ns and synchronized with the trigger with a 25-ns jitter.

9.5.1 Input track distributions

For the MC simulation presented below 500 typical events have been randomly generated ina geometric window of 6× 1 mm2 and an angular range of 0.2× 0.3 mrad2 with the protontracks perpendicular to the plane of the tracker. In the GEp(5) experiment the deflection ofthe proton track by the magnet will be just 5◦. For this simulation the detector plane hasbeen rotated to be normal to the average proton trajectory.

The simulation took into account all background interactions which occurred less than400 ns before or less than 100 ns after the trigger. The simulation used a zero value of theamplitude threshold (the DAQ threshold).

74

X [mm]

0

2

4

6

8

10

12

14

MC track XMC track X

Y [mm]

0

10

20

30

40

50

60

MC track YMC track Y

[mrad]0

10

20

30

40

50

60

MC track MC track

[mrad]-20

10

20

30

40

50

60MC track MC track

-5 52.5-2.5 0 -5 52.5-2.5 0

21.50 10.5-0.5-1-1.521.50 10.5-0.5-1-1.5-2

Figure 9.7: The distribution of the simulated tracks.

9.5.2 Signal pulse shape

Figure 9.8 shows the amplitudes in one strip for six possible events.The shape of the pulse characterized by a short rise of 50 ns and a long almost exponential

tail with an effective 129 ns decay time. The examples in Fig 9.8 correspond to:

• Top-left: Off-time hit tail;

• Top-right: On-time hit, likely from a proton track;

• Middle-left: Off-time hit 80 ns ahead of the trigger;

• Middle -right: On-time hit with the pileup on the tail of preceding off-time hit;

• Bottom-left: Off-time hit 25 ns after the trigger;

• Bottom-right: On-time hit and later an off-time at 68 ns after the trigger.

The analysis of the pulse shape using a so-called “deconvolution” is presented in thenext section. Such an analysis allows the extraction of the on-time amplitude (OTA) in

75

Time [ns]-100 -80 -60 -40 -20 0 20 40 60 80 100

Sig

nal

sam

ple

[AD

C c

han]

0

100

200

300

400

500

600

700

800

ev 55 hit=284 chamber=0 Plane=Y strip=526 type= 4

Time [ns]-100 -80 -60 -40 -20 0 20 40 60 80 100

Sig

nal

sam

ple

[AD

C c

han]

0

50

100

150

200

250

300

350

400

ev 10 hit=10 chamber=2 Plane=Y strip=624 type=1

Time [ns]-100 -80 -60 -40 -20 0 20 40 60 80 100

Sig

nal

sam

ple

[AD

C c

han]

0

100

200

300

400

500

600

700

ev 55 hit=279 chamber=0 Plane=Y strip=439 type= 4

Time [ns]-100 -80 -60 -40 -20 0 20 40 60 80 100

Sig

nal

sam

ple

[AD

C c

han]

0

20

40

60

80

100

ev 77 hit=442 chamber=0 Plane=X strip=505 type=5

Time [ns]-100 -80 -60 -40 -20 0 20 40 60 80 100

Sig

nal

sam

ple

[AD

C c

han]

0

100

200

300

400

500

600

700

800

900

ev 55 hit=283 chamber=0 Plane=Y strip=466 type= 4

Time [ns]-100 -80 -60 -40 -20 0 20 40 60 80 100

Sig

nal

sam

ple

[AD

C c

han]

0

20

40

60

80

100

120

140

160

180

200

220

ev 2 hit=3494 chamber=1 Plane=Y strip=623 type= 5

Figure 9.8: Pulse shapes in one strip for different hit times relative to the trigger time (atzero time). Left: background particles; right: protons (upper panel) and protons togetherwith background particles (lower two panels).

the presence of significant pileup. The APV25 chip has a deconvolution hardware functionwhich would allow to determine the OTA on-line and to sparsify the data with a dramaticreduction of the readout volume to just one sample per strip. However, at this stage ofdevelopment we have not decided to use this deconvolution method on-line.

9.5.3 Background filtering

As shown in Sec. 9.2, during the GEp(5) experiment the rate of ionization signals will be2 to 3 MHz per strip. The amplitude of the signal stays about 10% of its initial valueduring 200 ns (Fig. 8.9). Therefore, the raw data have a large occupancy of up to 70%. Atsuch a high occupancy a track search is not possible. The analysis presented below usestwo methods of data filtering that when combined provide much cleaner data for the tracksearch.

Filtering is performed by applying cuts on two functions of the ADC values of the threesignal samples (S−1, S0, S+1), where S0 is an amplitude of a signal (sample) nearest to the

76

trigger time, S−1 and S+1 are the preceding and successive samples, corresponding to times25-ns before and 25-ns after S0, respectively.

It is assumed that the trigger timing is set in such a way that a time between thetrigger and the S0 sample (“clock jitter time”) is uniformly distributed in a range Tcj =−12.5 . . .+ 12.5 ns.

On-time signal amplitude

If the signal start coincides with the trigger and its time dependence (shape) is close to anideal shape of a CR − RC filter (see Sec. 8.2.4) the amplitude could be obtained using thefollowing formula [23]:

D0 = w1 · S+1 + w2 · S0 + w3 · S−1where w1 = ex−1/x, w2 = −2 ∗ e−1/x, w3 = e−1−x/x with x = TS/τ , where TS is samplingperiod (25 nsec in our case) and τ is shaping parameter (50 nsec in our case). It workssufficiently well in spite of the non-ideal pulse shape from the GEM chamber.

Additionally, a linear correction on the dependence of D0 on the clock jitter time wasapplied:

D = D0 · (1 + 1.75 · |Tcj/TS|)

The pulse time derivative

The time derivative provides the second most important rejection parameter:

∆ =

{S+1 − S0 if Tcj ≥ 0 ,

S0 − S−1 if Tcj < 0 .

It is measure of the slope of the pulse shape at the moment of trigger.

For a complete rejection set (∆ + D) the cuts were:

∆ > 15, D > 35

9.5.4 Tracking efficiency

In the future tracking analysis of the GEp(5) experiment the location of the electron trigger,determined by the electron calorimeter and the preceding coordinate detector, will allow,using the elastic kinematics of GEp(5), to identify a track search window of 200 × 5 mm2

with an angular range of 30× 2 mrad2 (see Sec. 7.1), taking into account the resolutions ofthe electron calorimeter and the coordinate detector (at ±3σ) and the 40 cm length of thetarget. Next, a track search is performed as outlined in detail in Sec. 9.5.3 and following. Theaccepted tracks are then backtraced to the target area. Then, from the vertex position foreach identified proton track-candidate the electron track will be traced back to the elelctron

77

hSumPEntries 2958Mean 476.6RMS 425

Amplitude of deconvoluted pulse [ADC chan]0 200 400 600 800 1000 1200 1400 1600 1800 2000

0

50

100

150

200

250

300

350

400

450

500

hSumPEntries 2958Mean 476.6RMS 425

hSumBEntries 2734Mean 93.49RMS 224.5strip with background only

strip with proton and background

hSumPEntries 2958Mean 535.6RMS 427.2

Amplitude of deconvoluted pulse [ADC chan]0 200 400 600 800 1000 1200 1400 1600 1800 2000

0

10

20

30

40

50

60

70

80

hSumPEntries 2958Mean 535.6RMS 427.2

hSumBEntries 2734Mean 295.2RMS 325.6

proton

background

Figure 9.9: The event distribution of the on-time amplitude (the D rejection parameter).The results shown are for the full expected background. Histograms from six strips (one pereach Y-plane) were added to form each of two histograms. One histogram shows the stripsalong the proton track, the other the strips far away from the proton track. The left plotshows the distribution for all events, the right plot the distribution for events after the cuton the pulse time derivative (∆) has been applied.

calorimeter, again using the elastic kinematics, and the predicted position checked with thatof the electron trigger hit. Finally, the knowledge of the vertex position will reduce theeffective track search window to 30× 2 mm2 and 7× 2 mrad2 for the final search of a protontrack. The parameters of the search area were obtained from the MC data discussed inSec. 7.1.

We used this search area information for the evaluation of the reconstruction accuracyand efficiency in the SBS front tracker, using the MC data of the SBS tracker. Each strip ofthe GEM tracker with a signal was checked by the filter algorithm presented in Sec. 9.5.3.The remaining hits were combined in clusters and submitted to the TreeSearch reconstructionchain, described in Sec. 9.4. After the track was reconstructed, it is defined by its coordinatesin the plane of the first chamber (x, y) and its direction (φ, θ). The differences between thereconstructed track and the MC-track were used to calculate the residuals which characterizethe spatial and angular resolutions of 30 and 47 µm for x and y, respectively, and 0.18 and0.19 mrad for θ and φ, respectively, as shown in Fig. 9.10. Because the resulting resolutionsare much better than the projected ones for the SBS chamber (70µm) we assumed forkinematic analysis the 70 µm resolution (also used in Sec. 4.1 for the SBS optics calculation).

9.6 Conclusion

Table 9.2 and Figs. 9.10 and 9.11 summarize the results of the tracking analysis presented inthis chapter as a function of the background level, where 100% corresponds to the experimen-tal conditions expected during the GEp(5) experiment. The filtering procedures presentedin Sec. 9.5.3 are clearly extremely effective. The strip occupancy that is 70% without any

78

x [mm]-0.4 -0.2 0 0.2 0.40

5

10

15

20

25100 % background

MC-XtrackX

y [mm]-0.4 -0.2 0 0.2 0.4

MC-YtrackY

[mrad] 0

5

10

15

20

25

30

35MC-track

[mrad] 0

5

10

15

20

25

30

35

40

MC-

track

210-1-2 210-1-2

Figure 9.10: The track reconstruction quality. The rms for the x-coordinate is 30 µm, forthe y-coordinate 47 µm. The rms for the θ-coordinate is 0.18 mrad, for the φ-coordinate0.19 mrad.

filtering at the highest rate drops to 18% with the signal-amplitude D filter and then to lessthan 11% by additionally applying the pulse-time-derivative ∆ filter. The fraction of eventsthat are within 3σ of all four coordinates (locations in x and y and directions in φ and θ) tothe MC-track, is the track-reconstruction efficiency. After both filters have been applied atracking efficiency of just over 90% is obtained at the highest rate.

Part of the reconstructed events have an additional track whose parameters are withinthe search window described above. At the highest background intensity less than 5%of events have a second track within the kinematic range allowed for elastic scattering.Whereas the occupancy and the tracking efficiency show a more or less linear dependenceon the background rate, the multitrack probability appears to increase dramatically the ratedoubles. The false tracks can be filtered out in the off-line analysis by applying for example

79

Background contribution [%]0 20 40 60 80 100

Str

ips

occ

up

ancy

[%

]

0

10

20

30

40

50

60

70

80Raw

D

+ D∆

Occupancy for various filter options

Figure 9.11: The mean strip occupancy as a function of the background level. Raw:before filtering (all signals above threshold are read out); D: after the cut on the decon-voluted pulse amplitude only; ∆+D: after the cuts on both background rejection parameters.

Table 9.2: The strip occupancy, tracking efficiency and false tracks, as a function of thebackground level.

background level [%] 0 10 30 50 100Raw occupancy [%] 0.18 14.0 36.2 48.0 69.7after cut on D [%] 0.17 2.4 6.8 10.2 17.5after cut on (∆ + D) [%] 0.16 1.6 4.0 6.3 10.6Tracking efficiency [%] 98.6 98.2 97.2 93.9 90.5Multitrack probability [%] 0 0 0 0.2 4.7

a missing-mass cut or, almost equivalently, a χ2 cut on the kinematical parameters (the 3σkinematical window discussed above). Because the false tracks are distributed uniformlyover the ±3σ range in each parameter they have on average a χ2 of 12, where the real trackshave on average a χ2 of 4. The application of a cut at χ2 = 8 in the off-line analysis willselect the real track in 90% of double-track events and a false track in 5% of these events.Thus, the resulting efficiency will be appr. 90% for real tracks, with a contamination byfalse tracks of appr. 0.3%.

In summary, the studies presented in this chapter indicate that a tracking efficiency ofclose to 90% and a high resolution of the proton reconstruction in the GEp(5) experimentcan be achieved with the GEM coordinate detector of the proposed configuration and thedeveloped algorithm of the track reconstruction.

80

Background contribution [%]0 20 40 60 80 100

Trac

kin

g e

ffic

ien

cy [

%]

80

85

90

95

100

105

D

+ D∆

Efficiency for various filter options

Figure 9.12: The track reconstruction efficiency as a function of the background level. D:cut on deconvoluted pulse amplitude only; ∆ + D: with the additional cut on the pulse timederivative.

81

Chapter 10

Cost Estimate

The budget

Recommendation #6: The SBS Collaboration should develop a bottoms-up costestimate for the project. Include all required components and activities. Makesure that realistic yields for delicate production items are taken into account.

Response

Early in 2011 the SuperBigbite project was submitted to DOE as a Major Item ofEquipment (MIE) proposal. In preparation for that the budget was reevaluated following abottoms-up procedure. It includes modifications to the magnet and its support structure,development of the GEMs and all other aspects of the project. The budget, including year-to-year expenditures on specific items, is available in a spreadsheet format which can be usedfor inquiries about specific budget questions. That spreadsheet with the budget and somesample queries to the budget database can be seen at:http://www.jlab.org/~lerose/sbs/MIE-Feb-17-2011.xlsx

The budget includes expenditures, worker salaries, all relevant overheads, contingencies, andescalation at a rate of 3.5% compounded annually. Contingencies run between 10% and 30%depending on the estimated risk. The JLab overhead is 41% unless 41% is greater than 20k$; then it is limited to $20 k$.

Figure 10.1 shows the total cost year by year for the project. In the table shown inFig. 10.2 those costs are broken down by Institution. Note that the Hadron calorimeter,which will also be used in non-SBS experiments, has not been included in the MIE proposal,but will be funded directly by JLab.

82

Figure 10.1: Timeline for SBS expenditures.

Figure 10.2: MIE year-to-year costs broken down by Institute.

83

Chapter 11

Experimental Tests

11.1 Experimental tests of GEM detector

Recommendation #7: Remaining uncertainties in background rates and elec-tronics performance can be reduced by performing experimental tests undersimilar conditions. This should be done as soon as possible since the resultscould lead to modifications of the segmentation scheme and the readout rateswhich need to be known before the start of mass production.

Response

11.1.1 GEM tests during the PREX experiment

A set of five prototype 10 × 10 cm2 GEM chambers with x − y planes and a readout strippitch of 0.4 mm have been constructed by the UVa group. These chambers were testedat very high rate conditions during the PREX experiment in Hall A (March - June, 2010)when they were located behind the VDCs of the HRS spectrometer. During these teststhe GASSIPLEX frontend cards were used for the readout electronics. A good correlationbetween the tracks projected from the VDCs and the GEMs was obtained during thesetests resulting in a preliminary resolution (from residuals) of ∼ 70 µm; this is similar to theposition resolution expected in the front GEM trackers for the GEp(5) setup.

11.1.2 GEM tests in Hall A during Spring 2011

A stack of three UVa GEM detectors was placed at the back of the E−plane trigger scintil-lators of Bigbite during the spring of 2011 run. BigBite was located at ∼ 95◦. The expectedrate on the GEM trackers (primarily due to soft photons) was calculated using the simula-tion code of P. Degtyarenko and convoluting the number of photons with an energy above100 keV with the photon detection efficiency of the GEM chambers. Assuming a photondetection efficiency of 0.33% for the first GEM chamber the average rate expected in this

84

test configuration was ∼35 kHz/cm2 at an incident beam current of 40 µA on a 20 cm LH2

target. The tests were still done with the GASSIPLEX frontend cards with a variety oftargets and beam currents. The data are currently being analyzed.

11.1.3 Future plans for GEM tests

An order has already been placed to acquire several of the frontend cards and the custombuilt (by the INFN-Rome group) VME64X/ADC controllers for the APV25-S1 chips thatwill be used for GEp(5) running. These frontend electronics are expected to be delivered inSummer 2011. This will give a reasonable chance to test the GEM chambers with frontendelectronics to be used in the experiment during the last 6 months of 6-GeV running. We arediscussing what will be the best place to test the GEM chambers.

85

Bibliography

[1] M. Capeans et al., Construction of GEM detectors for the COMPASS experiment -Production Guide, CERN-EP/TA1 - Technical Note TA1/00-03, February 17, 2001.

[2] T. Hilden, TOTEM GEM assembly and quality control procedures, presentation at the“GEM & Micromegas detector design & assembly training” within the RD51 collabo-ration, CERN, February 17-20, 2009.

[3] C.W. de Jager et al., Electromagnetic Form Factor of the Nucleon using the 12-GeVCEBAF, funding proposal submitted to the US Department of Energy, November, 2009.(http://www.jlab.org/~bogdanw/EMFF.pdf)

[4] M. C. Altunbas et al., Aging measurements with the Gas Electron Multiplier (GEM),Nucl. Instrum. & Methods A 515, 249 (2003).

[5] F. Bagliesi et al., Nucl. Instrum. & Methods A 617, 134 (2010).

[6] B. Surrow, Nucl. Instrum. & Methods A 617, 196 (2010).

[7] http://leszek.home.cern.ch/leszek/GEM_at_CERN_seminar.pdf.

[8] M.J. French et al., Nucl. Instrum. & Methods A 466, 359 (2001).

[9] P.W. Cattaneo, Solid-State Electronics, 54, 252 (2010).

[10] The Super-Bigbite Spectrometer for Jefferson Lab Hall A (available at:http://hallaweb.jlab.org/12GeV/SuperBigBite/SBS-CDR/).

[11] J.C. Santiard et al., Gasplex, a low-noise analog signal processor for readout of gaseousdetectors, CERN-ECP/94-17.

[12] A.H. Cookson and T.J. Lewis, Variations in the Townsend first ionization coefficient forgases, Brit. J. Appl. Phys. 17, 1473 (1966); http://iopscience.iop.org/0508-3443/17/11/312.

[13] A. Sharma, 3D simulation of charge transfer in a Gas Electron Multiplier (GEM) andcomparison to experiment, Nucl. Instrum. & Methods A 454, 267 (2000) .

86

[14] C. Altunbas et al., Construction, Test and Commissioning of the Triple-GEM TrackingDetector for COMPASS, CERN-EP/2002-008.

[15] M. Friedl et al., Nucl. Instrum. & Methods A 572, 385 (2007).

[16] F. Simon, Commissioning of the GEM Detectors in the COMPASS Experiment, PhDThesis, Physik-Department E18, Technische Universitat Munchen (2001).

[17] B. Ketzer et al., Nucl. Instrum. & Methods A 535, 314 (2004).

[18] P. Abbon et al., Nucl. Instrum. & Methods A 577, 455 (2007).

[19] J.-O. Hansen in Hall A Status Report 2008, p. 23, available at http://hallaweb.jlab.org/publications/AnnualReports/AnnualReport2008.pdf

[20] M. Dell’Orso and L. Ristori, Nucl. Instrum. & Methods A 287, 436 (1990).

[21] E. Gamma, R. Helm, R. Johnson, J. Vlissides, Design Patterns: Elements of ReusableObject-Oriented Software, Addison-Wesley Professional Computing Series, 1994.

[22] R. Brun and F. Rademakers, ROOT - An Object Oriented Data Analysis Framework,Proceedings AIHENP’96 Workshop, Lausanne, Sep. 1996, Nucl. Instrum. & MethodsA 389, 81 (1997). See also http://root.cern.ch/.

[23] S. Gadomski et al., Nucl. Instrum. & Methods A 320, 217 (1992).

87

Additional references:

[24] The KLOE-2 Collaboration, Technical Design Report of the Inner Tracker for theKLOE-2 experiment, LNF-10/3(P) February 10, 2010, http://arxiv.org/abs/1002.2572; A. Balla et al., Status of the Cylindrical-GEM project for the KLOE-2 InnerTracker, arXiv:1003.2770v1 March 19, 2010.

[25] G. Cates, K. de Jager, B. Wojtsekhowski et al., Electromagnetic Form Factors of theNucleon using the 12-GeV CEBAF, http://www.jlab.org/~bogdanw/EMFF.pdf.

[26] E. Brash, M. Jones, E. Cisbani, M. Khandaker, L. Pentchev, Ch. Perdrisat, V. Punjabi,B. Wojtsekhowski et al., Jefferson Lab experiment E12-07-109,http://www.jlab.org/exp_prog/proposals/07/PR12-07-109.pdf.

[27] R. Gilman, B. Quinn and B. Wojtsekhowski et al., Jefferson Lab experiment E12-09-019,http://www.jlab.org/exp_prog/proposals/09/PR-09-019.pdf.

[28] G. Cates, S. Riordan, and B. Wojtsekhowski et al., Jefferson Lab experiment E12-09-016, http://www.jlab.org/exp_prog/proposals/09/PR-09-016.pdf.

[29] G. Cates, N. Liyanage, Z.E. Meziani, G. Rosner, B. Wojtsekhowski, X. Zheng et al.,Jefferson Lab experiment E12-06-122, http://www.jlab.org/exp_prog/proposals/

06/PR12-06-122.pdf.

[30] G. Cates, E. Cisbani, G. Franklin, and B. Wojtsekhowski et al., Jefferson Lab experi-ment E12-09-018, http://www.jlab.org/exp_prog/proposals/09/PR-09-018.pdf.

[31] Qweak experimental layout, http://www.jlab.org/qweak/images/qweak_lg.jpg.

[32] http://rd51-public.web.cern.ch/RD51-Public/

[33] M. Alfonsi et al., Aging measurements on triple-GEM detectors operated with CF4-basedgas mixtures, Nucl. Phys. B (Proc. Suppl.) 150 (2006) 159-163.

[34] B. Wojtsekhowski, Experimental program of the Super Bigbite, report to the Technicalreview, Nov. 17, 2008, http://www.jlab.org/~bogdanw/SBS-CDR/Exp-program.pdf.

88