Design and Calibration of an Optically Segmented Single Volume Scatter Camera for Neutron Imaging
A. Galindo-Tellez, K. Keefe, E. Adamek, E. Brubaker, B. Crow, R. Dorrill, A. Druetzler, C.J. Felix, N. Kaneshige, J.G. Learned, J.J. Manfredi, K. Nishimura, B. Pinto Souza, D. Schoen, M. Sweany
PPrepared for submission to JINST
Design and Calibration of an Optically Segmented SingleVolume Scatter Camera for Neutron Imaging
A. Galindo-Tellez, 𝑎, K. Keefe, 𝑎 E. Adamek, 𝑎 E. Brubaker, 𝑏 B. Crow, 𝑎 R. Dorrill, 𝑎 A. Druetzler, 𝑎 C.J. Felix, 𝑎 N. Kaneshige, 𝑎 J.G. Learned, 𝑎 J.J. Manfredi, 𝑐 K. Nishimura, 𝑎 B. Pinto Souza, 𝑎 D. Schoen, 𝑎 and M. Sweany 𝑏 𝑎 Department of Physics and Astronomy, University of Hawai‘i at M¯anoa,Honolulu, HI 96822, USA 𝑏 Sandia National Laboratories,Livermore, CA 94550, USA 𝑐 Department of Nuclear Engineering, University of California, Berkeley,Berkeley, CA 94720, USA
E-mail: [email protected]
Abstract: The Optically Segmented Single Volume Scatter Camera (OS-SVSC) aims to imageneutron sources for non-proliferation applications using the kinematic reconstruction of elasticdouble-scatter events. Our prototype system consists of 64 EJ-204 organic plastic scintillatorbars, each measuring 5 mm × ×
200 mm and individually wrapped in Teflon tape. Thescintillator array is optically coupled to two silicon photomultiplier ArrayJ-60035 64P-PCB arrays,each comprised of 64 individual 6 mm × × Na and Sr, respectively, reporting an average resolutionof ( . ± . ) mm for energy depositions between 900 keVee and 1000 keVee. We furtherdemonstrate a position calibration method for the internal bars of the matrix using cosmic-raymuons as an alternative to emission sources that cannot easily access these bars, with an averagemeasured resolution of ( . ± . ) mm for depositions between 900 keVee and 1000 keVee.The coincident time resolution reported between pairs of bars measured up to 400 ps from muonacquisitions. Energy and position calibration values measured with muons are consistent with thoseobtained using particle emission sources.Keywords: Neutron detectors (fast neutrons), Search for radioactive and fissile materials, Detectoralignment and calibration methods (lasers, sources, particle-beams) Corresponding author. a r X i v : . [ phy s i c s . i n s - d e t ] F e b ontents Neutron scatter cameras use the kinematic information from two elastic scatter interactions inhydrogenous material to reconstruct the neutron source direction and energy. This technique hasbeen proposed for non-proliferation applications to image Special Nuclear Materials with neutronenergies in the ∼ MeV range (e.g., [1–3]), as well as in astronomical applications to image neutronsof a few to hundreds of MeV [4–6].While traditional neutron scatter cameras have used a dual plane design that limits acceptance,omnidirectional imaging has been obtained in a portable form factor with a height offset cylindricalconfiguration [7]. Despite such improvements, double scatter imaging generally suffers from poor– 1 –fficiency due to the low probability of multiple neutron scatters occurring in spatially separatedscintillation cells.Multiple concepts have been proposed to enhance the efficiency over previous designs, witha focus on moving toward a single volume scintillation region, rather than on discrete, spatiallydistant scintillators. A compact active volume dramatically improves the likelihood that a neutronwill scatter twice within the active scintillation volume.Truly monolithic designs propose to reconstruct multiple interaction positions within a singlevolume [8, 9]. A second technique proposes closely packed but optically segmented scintilla-tors [10]. In both cases, fine spatial resolution of ∼
10 mm and fine temporal resolution of < The imaging methodology of the OS-SVSC requires neutrons to scatter at least twice in differentscintillator bars to allow an estimate of the incident neutron direction and energy [10]. In the firstinteraction, a neutron scatters off a proton and deposits part of its energy in the bar, which can beobtained by measuring the intensity of the light detected. Our collaboration has performed protonlight yield measurements [11] which are used to convert the measured light into proton recoilenergies. The outgoing energy of the second neutron scatter, 𝐸 𝑛 (cid:48) , is calculated using the distance 𝑑 and the time of flight Δ 𝑡 between the two scatters: 𝐸 𝑛 (cid:48) = 𝑚 𝑛 𝑣 = 𝑚 𝑛 (cid:18) 𝑑 Δ 𝑡 (cid:19) , (1.1)where 𝑚 𝑛 is the mass of a neutron and 𝑣 is the speed of the scattered neutron. This non-relativisticexpression is valid for MeV-scale neutrons. The initial incident energy of the neutron 𝐸 𝑛 can beobtained by adding up the outgoing energy with the energy lost due to proton recoil 𝐸 𝑝 in the firstscatter as: 𝐸 𝑛 = 𝐸 𝑛 (cid:48) + 𝐸 𝑝 . (1.2)The incident neutron direction is confined on a cone centered on the vector connecting the twoscatters with opening angle 𝜃 , as: cos 𝜃 = √︂ 𝐸 𝑛 (cid:48) 𝐸 𝑛 . (1.3)The physical process is illustrated in Fig. 1. An image of the neutron source can be obtainedby overlapping back-projected cones to a plane placed at the estimated neutron source distance, orby implementing more sophisticated iterative image reconstruction methods.– 2 – igure 1 . Neutron double-scatter event illustration in the OS-SVSC. The projected cone of one event isobtained using the time-of-flight of the neutron between two bars and the intensity of the light in the bar ofthe first scatter. This section describes the design of the OS-SVSC, including scintillators and their surface wrapping,photodetectors, electronics and readout, and mechanical design. A primary motivation for many ofthe design choices was to use either existing in-house or readily available commercial parts in orderto accelerate design and testing efforts. While this did lead to the first realization of a prototype,there are some detrimental impacts of these decisions, which are explained in detail in this section.
The active volume of the OS-SVSC is composed of an array of 64 rectangular 5 mm × ×
200 mm EJ-204 scintillator bars from Eljen [12], each individually wrapped with three layersof Teflon tape. The decision to use this scintillator and surface treatment was motivated by ourcollaboration’s previous measurements of interaction position and timing resolution using a varietyof combinations of scintillating materials and surface treatments. A number of scintillators andsurface wrappings were shown to give combined resolutions of better than 10 mm, but the enhancedlight output of EJ-204, combined with the ease of consistently wrapping Teflon relative to ESR, ledto our particular chosen combination [13].From experience with our single bar measurements across two different sites and experimentalsetups, there can be significant variation in the optical properties of Teflon due to variationsin wrapping technique, the number of layers, and the underlying thickness and quality of thecommercial Teflon tape. The standardization of the wrapping procedure, number of layers, andadoption of a minimum specification for procured Teflon (MIL SPEC T-27730A), all led to improvedreproducibility of results and were adopted for the design of this detector. Three layers werenecessary to achieve the desired thickness of approximately 0 . .2 Photodetectors, Optical Coupling, and Signal Conditioning The scintillator bars are optically coupled on each end to pixels of the ArrayJ-60035 64P-PCBarrays (ArrayJs), using individually cut squares of 5 mm × × . × × Design features, such as optical coupling, electrical connections/cabling, enclosure configuration,and internal support details, were designed to match the ArrayJ dimensions and layout. Anillustration of the OS-SVSC detector internals is shown in Fig. 2.The internal component supports and the aluminum enclosure were designed and fabricatedin-house. The design of such components was motivated by the need to maintain alignment ofindividual pixels with the bars and to provide consistent and moderate contact pressure between thebars, the optical interface material, and each pixel’s glass cover. With this, we assure maximumlight transmission while reducing mechanical stress on the ArrayJ, which could lead to cracking ofthe glass. Two lattice supports were 3D printed with polylactic acid filament to align the Teflonwrapped bars with the SiPMs. In addition, two SensL SiPM array supports were 3D printed tomaintain the ArrayJs in placement. The support designs are shown in Fig. 2. All four supportsare kept aligned using four threaded rods, which run the length of the detector, and spacers (tubesaround the threaded rods) between all the supports.– 4 – igure 2 . Detector internals consisting of 64 bars (8 × Figure 3 . Mechanical enclosure design for OS-SVSC prototype: (a) disassembled and (b) assembled.
Aluminum was chosen for the enclosure material for its low cost, and compatibility withcomputer numerical control machining, as well as to reduce the potential for neutron scatter in theenclosure. The top and bottom sections have slots that match the supports to hold them in place.Rabbet joints and blind screw holes are used to minimize external light penetration. To avoidoverheating issues and the need for active cooling, the electronics and power supply are connectedto the ArrayJs with Samtec EQCD cables and reside outside of the enclosure. The assembledmechanical components without scintillator bars and ArrayJs are shown in Fig. 3. Calibrations areperformed with the detector in the horizontal orientation shown.
The OS-SVSC design places constraints on the performance requirements of the electronics and dataacquisition (DAQ) systems. It requires that the 8 × × O (100) ps coincidenttiming typical of the SiPMs [13]. The digitization of each ArrayJ is carried out by independentstandard control readout (SCROD) board stacks. For synchronizing the two SCRODs, a thirdexternal board called CAJIPCI, (Clock and JTAG interface by peripheral component interconnectbus) is utilized to distribute a global clock. The SCRODs and CAJIPCI boards were previouslyutilized in the electronics system of the mini-TimeCube experiment and are described in more detailelsewhere [19]. Figure 4 . Diagram of the front-end electronics and DAQ chain. The SCRODs connect via Samtec connectorto the interposer cards from Ultralytics and are synchronized by the CAJIPCI board, responsible for theglobal trigger and clock. The user configures and reads data from the SCRODs via two parallel ethernetfibers.
The front-end electronics functionality lies in the ice radio sampler (IRS), a family of application-specific integrated circuits (ASICs) developed for projects that require fast sampling and deepbuffering [19]. A diagram of the electronics is shown in Fig. 4. Features added for this systeminclude improved networking interfaces control (containing MicroBlaze based networking), high-speed deserialization on trigger monitoring, adapter cards for carrier boards, Samtec connectors,and interposer cards for recording individual channels from the ArrayJ. named after the Joint Test Action Group which codified it. – 6 – .4.2 Data Acquisition The user control between a PC and the SCRODs utilizes a User Datagram Protocol (UDP) interfaceto an embedded software system. Each SCROD, responsible for digitizing 64 channels on one edgeof the detector, uses three UDP ports. Each port can be configured as input or output on an assignedIP address. The first output port is dedicated to the FPGA configuration, ASIC configuration, anddigital-to-analog converter voltage settings. A second input port is used for digitizing each channelin an event. The third input port reads out trigger information, that indicates which channels wetriggered during the event. Since UDP packet transmission is unreliable, the collected data must beimmediately saved to prevent buffer overflow errors and eliminate packet loss. Each packet is sentwith a unique message identification number in order to ensure each one can be uniquely identifiedand that events are complete with no missing packets.After a completed run, the trigger information and digitized data are stored in independentbinary files which are then combined into a single ROOT file [20] for further analysis. Electronicscalibrations explained in Sec. 4.1 are applied during the conversion from binary to ROOT dataformats. In addition, metadata are stored separately for each board stack’s register configuration.
Before describing the techniques and results of the fully assembled OS-SVSC prototype characteri-zations, we first note some measurements on crosstalk investigated further based on initial anecdotalobservations. A waveform coming from a crosstalk event shows a negative voltage value when therise of a positive pulse in a neighbor channel is present for both the FOUT and the standard output(SOUT). Particularly, a damped sine wave as the voltage returns to the baseline is observed for theFOUT. Examples of electrical crosstalk waveforms from the FOUT and SOUT are shown in Fig. 5aand Fig. 5b, respectively.To understand these effects, we set up a collimated, pulsed laser (ultrashort pulses down to50 ps at 408 nm) attached to a 2D motor stage, incident on the face of one ArrayJ connected to theUltralytics adapter card. For each acquisition set, a DRS4 evaluation board [21] reads out fourpixels at 5 .
12 GSa / s. The laser provides an external trigger and tours each of the 64 pixels. Werecorded 1,000 signals per illuminated pixel. We acquired sixteen acquisition sets of four channelsto measure the 64 pixels from the ArrayJ. The FOUT and the SOUT were recorded in separateacquisition tests, utilizing the commercial adapter board ArrayX-64 A.3 from Ultralytics for readingthe SOUT.The maximum pulse height illuminated by the laser and the negative pulse height seen ineach pixel are measured to quantify the crosstalk, as characteristic negative pulses are present oncrosstalk waveforms for both outputs. The correlation matrices that describe crosstalk in percentageare shown in Figs. 6a, and 6b, for the FOUT and SOUT, respectively. It should be clarified that theartifacts seen in channels from pixel 4 of the FOUT and pixel 12 of the SOUT are due to damagefound in their correspondent Ultralytics cards.The correlation matrices reveal that the crosstalk is localized in physical groupings of channelsin the Samtec QSE connectors used to extract signals from the SensL ArrayJ. Likewise, the matricesshow specific groups into quadrants of the ArrayJ. The crosstalk distribution along the pixels forthe FOUT and the SOUT are consistent with each other. When a pixel is illuminated, the impacted– 7 – a) (b) Figure 5 . (a) Electrical crosstalk when using the FOUT recorded in pixels 54 and 64, with pixel 61 underillumination, and (b) electrical crosstalk when using the SOUT recorded in pixels 40 and 39, with pixel 46under illumination. channels are the ones with pins located in the same line of the connector, according to the layoutschematic of the ArrayJ, shown in Fig. 6c. Furthermore, the highest crosstalk levels measuredare located at the two adjacent pins to the illuminated pixel pin. A still detectable crosstalklevel is measured on pins that share the same bar ground from the same section in the connector.Nevertheless, the crosstalk does not extend to independent bar grounds or different connectors.The maximum measured crosstalk reached up to 16 . . In this section, we describe the offset and timing calibrations of the electronics, and the waveformprocessing techniques applied to the calibrated waveforms.
Fabrication variations from the IRS ASIC individual capacitors and comparators generate sample-to-sample differences creating a unique offset voltage value on each sample. The subtraction of– 8 – a) (b)(c)
Figure 6 . (a) Correlation matrix from the FOUT and (b) correlation matrix from the SOUT. Both correlationmatrices describe the response in all pixels when one of them is illuminated by the laser at a time. The numberof each pixel corresponds to the same defined in the ArrayJ datasheet [22]. The z-scale corresponds to thepercentage of the maximum height of the negative pulse seen in each pixel, with respect to the maximumpulse height of the pixel under illumination. (c) Connector schematic for the ARRAYJ-60035-64P, obtainedfrom [22]. – 9 –hese values allows for acquiring proper waveforms. The pedestal value is determined as the averagevalue of the measurement when random sampling is performed with no external input. An examplepedestal subtracted noise trace and corresponding noise histogram is shown in Fig. 7. Typical noisevalues for pedestal-subtracted traces range from 1 . ∼ (a) (b) Figure 7 . (a) Example pedestal-subtracted waveform with no input signal. (b) ADC count distribution forthe noise waveform.
The ASICs at the core of the readout electronics utilize delay lines to toggle an array of switchedcapacitors. Process variations in the fabrication of individual transistors along the delay line result invarying timing separations from sample to sample, which must be adequately calibrated to achievetiming resolutions on the order of 100 ps or better [23]. A matrix-based time-period calibrationmethod was utilized to determine the sampling delays of the 32,768 storage cells (2 ASICs × ×
128 channels) [24]. Two input pulses were sent through an SMA calibrationinput on each carrier with a known delay time. The pulse rise times were chosen to roughly matchthose seen in a typical SiPM pulse. The precision of the solution was verified by calculating theperiod of the input pulse with the calibrated timing values for each sampling cell. An example ofthe distribution of the pulse period measurement is shown in Fig. 8.An external ROOT file stores the calibration values along with the output data during theacquisition process. Fig. 9 summarizes the resulting timing resolutions from the characterizedchannels. The average timing resolution for each channel is 𝜎 ≈
30 ps to 90 ps. Variations areattributed in part to jitter introduced from the FPGA and PCB in the distribution of timing strobes.In addition, a number of internal timing parameters must be tuned for each ASIC in order to avoidtiming overlaps or gaps at the wraparound locations of the sampling array. Note that the timingresolutions here are all based on pairs of identical pulses measured within a single channel, they donot represent expected coincidence time resolutions between channels measuring a wider varietyof pulse shapes.
The waveforms recorded from the IRS3D have 256 samples for the particle emission sources testsand 384 samples for muon acquisitions. In both cases, the acquisition was performed at 2 .
73 GSa / sand the FOUTs of the ArrayJs were stored. Any residual baseline offset is eliminated by calculating– 10 – igure 8 . Example distribution of measurements of the pulse period of calibration signals on a channel aftertiming calibration is performed. The injected pulse separation is 12 . 𝜎 =
42 ps.(a) (b)
Figure 9 . (a) Timing Calibration values vs. channel number, (b) Timing calibration value distribution. the mean of the signal base using the first 64 samples and subtracting that value from all thewaveform samples. A 5-sample moving average filter was applied to smooth the signal. In somecases, ASIC waveforms exhibited artifacts with two or more spikes, while other waveforms showedsignificant electrical crosstalk, as discussed in Section 3. About 12 . Previous work [13] employed
Cs and Na sources for energy calibrations, and either collimated Sr source or the back-to-back gammas from a Na source to scan each bar along its length andgenerate a calibrated response of amplitude- and time-based functions from SiPM waveforms taken– 11 – a) (b)
Figure 10 . Example of the waveform processing. (a) is a raw waveform coming from the digitizer, and (b)corresponds to the same waveform after the signal processing. at each end of a bar. However, in the OS-SVSC design it is not possible to apply this technique foreach bar, because the 28 outer bars “shield” the remaining 36 bars, preventing particles from thecalibration source to reach the inner bars without scattering in the outer bars.In order to calibrate the position in the inner bars, we have developed a calibration techniquefor the OS-SVSC taking advantage of cosmic ray muons, which are the most abundant naturallyavailable charged particles at sea level [25]. This method utilizes the outer bar position calibrationsto construct a “muon telescope”. We use the calibration from Sr scans of the outer bars toreconstruct the muon trajectory, then project it through the array to calibrate the inner bars.With the condition that all channels reading the bars from the same column must pass athreshold of 30 ADC ( ∼ .
31 mV) to acquire a muon event, we expect to observe only paths ofmuons running through a full column of bars. The geometry of the OS-SVSC, combined withthe trigger condition, allows us to record muons with angles of up to 76 . ° with respect to thezenith. However, the cosmic muon intensity follows a cos 𝜃 distribution, where 𝜃 is the zenithangle: therefore, a small number of muon events will have an angle higher than 60 ° with respect tothe zenith.Muon data can also be used to calibrate energy. Muons have an average energy of approximately4 GeV at level sea and deposit ∼ / cm in the scintillator [25]. The advantage of the triggercondition allows us to assure that the recorded energy comes from the muon passing through thescintillator and not from muon decays in the inner bars, so as to not add spurious energy to themeasurements. In addition, as these muons travel at almost the speed of light, we can utilize timedifferences among pairs of bars to obtain time resolutions.This section describes the techniques for calibrating bars for energy, position, and time usingmuons, along with the traditional energy and position calibrations using particle emission sources.– 12 – igure 11 . Illustration of a muon passing through a column of bars. The trigger is activated when a pulse oneach channel from a vertical column of bars passes a threshold. The triggered bars are marked in red color. The tested column of bars is located at the edge of the detector in order to compare muon calibrationswith particle source calibrations, and includes bars numbered 8 (top), 16, 24, 32, 40, 48, 56, and64 (bottom). Unfortunately, the electronics channel that reads bar 8 was not functional. Hence, thecalibrations are reported with a primary focus on using seven of the eight available bars.
We evaluate and compare two techniques for energy calibration. The first is a technique where thegeometric mean of the pulse heights on each end of the given bar is measured and fit to a predictedspectrum model of a Na gamma-ray source around a Compton edge region. The technique isdescribed in more detail in [13] and is readily available for bars on the edge of the array. Thethreshold for these acquisitions is 20 ADC ( ∼ . ∼ ×
350 mm above the top bar, allows muons with large zenith angles to cross the bar set. The inputwas generated in CORSIKA [27] version 73900 from primary protons with zenith angles from 0 ◦ to 70 ◦ , azimuth angle set at 0 ◦ , a slope of the energy spectrum of −
2, and an observation altitude atsea level.In Table 1, we use the energy conversion obtained from the Na source measurements to– 13 – igure 12 . Example of an energy calibration. The expected model of a Na gamma-ray source in theCompton edge region 340 .
67 keV fits the amplitude spectrum from bar 56. Residuals are included.
Table 1 . Comparison of MPVs for real and simulated muon data, using the energy calibrations derived fromthe Na Compton edge for a column of bars on the edge of the detector.
Barnumber Energy conversionusing Compton edge(keVee/mV) Muon energy deposition MPV (keVee)
Data Simulations16 20.50 968 . ± . . ± .
824 20.51 984 . ± . . ± .
232 24.0 985 . ± . . ± .
640 19.57 986 . ± . . ± .
548 18.04 987 . ± . . ± .
256 18.43 985 . ± . . ± .
764 19.35 929 . ± . . ± . .
83 % and thedifferences with the corresponding simulated MPV ranged from 3 % up to 7 .
97 %. By knowingthese differences, the MPV from the real muon data can be used to determine energy conversion– 14 –alues for the rest of the bars.
The interaction position along a bar can be determined by using either the time or amplitudemeasurements from the bar’s ends, i.e., using the functions 𝑡 − 𝑡 or ln 𝐴 𝐴 , where 𝑡 and 𝑡 are thearrival times of the pulses and 𝐴 and 𝐴 are the pulse heights measured on the two ends of the bar.A detailed description of both approaches can be found in [13]. The experimental setup consists of a lead collimated 𝛽 -emitting Sr source, supported by a bracketmounted to a 2D Velmex XSlide (XN10-0120-M01-71) and driven by two stepper motors (PK245-01AA). The lead collimator limits the range of incident positions of betas within 2 mm on the bars.The source position, controlled via USB, was moved along each target bar length in steps of 5 mm;5,000 events were acquired per source position. (a) (b)
Figure 13 . Calibration example for the amplitude-based position resolution measurement, bar 40. (a) Thelogarithm of the ratio of the amplitudes as a function of the source position and (b) the standard deviationof the logarithm of the ratio of the amplitudes as a function of the source position. Only depositions from900 keVee to 1000 keVee are included.
The distributions of 𝑡 − 𝑡 and ln 𝐴 𝐴 were fit with Gaussian functions for every source position.Plots of the mean and sigma values as a function of the position of the source are used to obtain theposition resolution 𝜎 𝑧 , calculated as 𝜎 avg 𝑝 , where 𝜎 avg is the average of the standard deviation as afunction of 𝑧 and 𝑝 is the slope of the first-order polynomial fit to the mean as a function of 𝑧 . Anexample of amplitude-based position calibrations plots for one of the tested bars is shown in Fig. 13.An error of 1.44 mm contributes to the position resolution measurements due to the collimator width.Table 2 summarizes position resolutions when using amplitude-based measurements, time-basedmeasurements and their combination using the best linear unbiased estimator (BLUE) [28].– 15 – able 2 . Summary of the position resolution results when using a Sr source. Only depositions from900 keVee to 1000 keVee are considered. The errors represent the statistical uncertainty from the fit of theparticular distribution.
Bar ID 𝜎 𝐴𝑧 (mm) 𝜎 𝑡𝑧 (mm) 𝜎 𝑧 (mm)16 15 . ± .
08 37 . ± .
34 14 . ± . . ± .
07 43 . ± .
38 13 . ± . . ± .
13 21 . ± .
25 13 . ± . . ± .
06 41 . ± .
45 11 . ± . . ± .
06 31 . ± .
59 8 . ± . . ± .
07 33 . ± .
56 8 . ± . . ± .
07 27 . ± .
41 8 . ± . We tested the muon calibration technique with the same column of edge bars calibrated with particleemission sources. Only amplitude-based measurements for the ∼ 𝑧 axis create the muon trajectory, as illustrated in Fig. 14a. These two interactionpositions are obtained using the logarithm of the ratio of the amplitudes measured from muon eventsand the 𝑝 slope value of the calibrations using particle sources, as shown in Fig. 14b. The muontrajectory is then projected to the rest of the bars. We directly associated the interaction positionsof the inner bars with the corresponding amplitude measurements. An example of a 2D distributionrepresenting such an association for bar 24 is shown in Fig. 14c. Each 2D distribution is then binnedalong the interaction position axis (with a bin width of 6 mm). Every bin is projected along the logamplitude ratio axis to generate sets of 1D distributions. The mean and standard deviation valuesfrom Gaussian fits of the distributions allowed to determine the position resolution via the sameapproach described in Section 5.2.1. An example for depositions from 900 keVee to 1000 keVee isshown in Fig. 15.The reconstructed interaction positions from the top and bottom bars have associated errors fromthe previous calibrations, propagated to the muon trajectory and, consequently, to the reconstructedpositions of the inner bars. We implemented an algorithm based on the Monte Carlo (MC) methodto approximate the uncertainty due to the error in the estimated muon trajectory. Simulated truemuon trajectories from the top and bottom bars’ positions were randomly generated following theexpected angular cosmic-ray muon distribution. A second track, which represents the measuredmuon path, is generated by smearing the interaction positions from the top and bottom bars, usingthe actual amplitude-based position resolution values reported in Table 2. For each trial, the trueand the estimated muon trajectories were projected to the rest of the bars. Fig. 16 shows an exampleof an MC trial for an individual bar. 𝜎 trajectory represents the uncertainty of the estimated muontrack due to the errors present in the top and bottom bars interaction positions. Values of 𝜎 trajectory are obtained for each bar using 100,000 MC trials fitted with Gaussian functions. The width of the– 16 – a)(b) (c) Figure 14 . (a) Illustration of a muon passing through the studied bar set, (b) relation between the logarithmof the ratio of the amplitudes vs. the reconstructed interaction position for the top bar (16) using the 𝑝 slopevalue, (c) relation between the logarithm of the ratio of the amplitudes vs. the estimated interaction positionfor an inner bar (24). Gaussian fit values, along with the amplitude-based position resolutions using muons, are shownfor each bar in Table 3.
The same muon data used for position calibrations were utilized to obtain the time resolution amongthe bars and to determine time deviations. The measured time difference between the interactionsoccurring in two bars, Δ 𝑇 𝑖, ref , can be defined as: Δ 𝑇 𝑖, ref = ( 𝑡 + 𝑡 ) ref − ( 𝑡 + 𝑡 ) 𝑖 + 𝛿 𝑖, ref (5.1)where ( 𝑡 + 𝑡 ) ref is the average time of the reference bar, ( 𝑡 + 𝑡 ) 𝑖 is the average time of the 𝑖 th bar, and 𝛿 𝑖, ref is the time difference between the 𝑖 th bar and the reference bar due to electronics and physicalfactors, like cable lengths, pixel time responses, and ASIC fabrication effects, among others. The– 17 – a) (b) Figure 15 . (a) The logarithm of the ratio of the amplitudes as a function of the probable muon 𝑧 -positionand (b) the standard deviation of the logarithm of the ratio of the amplitudes as a function of the probablemuon 𝑧 -position for bar 40. Only depositions from 900 keVee to 1000 keVee are included. Figure 16 . Illustration of 𝜎 trajectory . The red star represents the true interaction position. The black starrepresents the estimated interaction position from the muon trajectory generated with smeared top and bottombars interaction positions. The black dot represents a measurement obtained during this event. Table 3 . Summary of the position resolutions when calibrating the inner bars with muon data 𝜎 𝐴 𝜇 𝑧 and theMC errors for inner bars with “true” position resolutions of 15 mm. Only depositions from 900 keVee to1000 keVee are considered in the real data. The errors on the resolutions are statistical. Bar ID 𝜎 𝐴 𝜇 𝑧 (mm) MC 𝜎 𝑡𝑟 𝑎 𝑗𝑒𝑐𝑡𝑜𝑟 𝑦
24 15 . ± .
79 13 . ± . . ± .
70 10 . ± . . ± .
52 8 . ± . . ± .
56 7 . ± . . ± .
96 7 . ± .
02– 18 –xpression is similar to the one reported in [29]. Nevertheless, our measurements only report ontime, no geometric information related to the interaction position along the bars or the distancebetween the bars was considered. (a) (b)(c)
Figure 17 . (a) Distributions of ( 𝑡 + 𝑡 ) ref − ( 𝑡 + 𝑡 ) 𝑖 for the studied bar set. (b) Distributions of Δ 𝑇 𝑖, ref for thestudied bar set. (c) Distribution of Δ 𝑇 𝑖, ref for bar 48. Bar 40 was defined as the reference bar, hence Δ 𝑇 , =
0. The assignment for a referencebar was made with no specific preference on the placement or the performance of such a bar. Thenon-centered positions of the peaks for the ( 𝑡 + 𝑡 ) ref − ( 𝑡 + 𝑡 ) 𝑖 distributions, shown in Fig. 17a, suggestevidence of the delay time 𝛿 𝑖, ref due to physical factors among the pairs of bars. It is possible tointer-calibrate bars for the same column when aligning those peaks to zero, thus obtaining 𝛿 𝑖, ref asa relative time offset. The alignment is shown in Figs. 17a and 17b. The standard deviation of aGaussian function fit from Δ 𝑇 𝑖, ref represents the time resolution, as is shown in Fig 17c for bar 48.The time resolution results, summarized in Table 4, show values of about 400 ps, with variationsamong the channels of 10 % or less. – 19 – able 4 . Summary of the 𝜎 Δ 𝑇 𝑖, ref results when using cosmic-ray muons. The errors are statistical from theGaussian fit. Bar ID 𝜎 Δ 𝑇 𝑖, ref (ns)16 0 . ± . . ± . . ± . . ± .
048 0 . ± . . ± . . ± . A number of limitations of the current OS-SVSC prototype have motivated significant designchanges for our next iteration. The presence of substantial electrical crosstalk described in Section 3has led to work toward custom arrays of individual SiPMs, as these allow more control over layoutand component choices (e.g., connectors), which can help to limit electrical crosstalk effects. Theability to design increased spacing between pixels relative to the densely packed pixels of thecommercial array also allows some opportunity for the mitigation of potential optical crosstalk.Such custom arrays can solve some other design issues as well. For example, by assembling SiPMpixels into individual 2 × × .
62 % when extrapolating the measurements using the BLUE method shown inthis work with a power-law compared to the previous characterization effort, for depositions from300 keVee to 400 keVee [13]. Furthermore, an increase of 54 .
66 % was found when comparing our– 20 – a) (b)
Figure 18 . (a) Position resolution results summarized for bars along one edge of the detector, includingtime-based, amplitude-based analyses, and the BLUE combination for source calibrations, and amplitude-based measurements for the muon calibration. 𝜎 𝑡𝑟 𝑎 𝑗𝑒𝑐𝑡𝑜𝑟 𝑦 is included as a contribution due to the error inthe estimated muon trajectory, and (b) position resolution as a function of the energy of a bar calibrated ina previous effort reported on [13], vs. bar 56 from the OS-SVSC prototype, both using the BLUE method.The threshold value (30 ADC) impacted the position resolution of bar 56 in the energy bin from 300 keVeeto 400 keVee, causing such an error. measurements with extrapolated previous data for depositions from 900 keVee to 1000 keVee, asis shown in Fig 18b. This degradation may be attributed to several factors, including the use ofdifferent readout electronics, effects of electrical and optical crosstalk, and difficulty controlling thequality and uniformity of optical coupling while working with the full array of bars.The timing-based position resolutions turned out to be significantly worse than amplitude-based resolutions. The timing resolutions are significantly worse than those measured with thetiming performance of the electronics, shown in Section 4.1. Since timing calibrations were carriedout on a single SCROD, we attribute much of this difference to jitter on the distribution of timingstrobes from the CAJIPCI board that is used to synchronize both SCRODS. In addition, inconsistentdigitization window-pairs between the SCRODs could also contribute. The time resolutions fromSection 5.3 showed time differences up to 400 ps, a value that will limit the ability to analyze eventswith a very short time of flight between bars, in particular those where the neutron double-scatteroccurs on immediately contiguous bars. Despite these limitations, we expect that we can stillperform meaningful neutron double scatter imaging for events where the two interaction positionsare at least 6 mm apart, with the neutron traveling at the expected velocity of ∼ . 𝑐 . We have described in detail the design of the OS-SVSC prototype. The chosen components (selectedbased on previous studies of scintillator bars, existing off-the-shelf solutions, and home-made pieces)allowed for the relatively fast delivery of a compact detector for testing.– 21 –alibrations to the electronics allowed recording proper waveforms to calibrate energy andposition using particle emission sources and energy, position, and time using a “muon telescope”.The muon energy calibration method was repeatable and reliable despite a systematic decreasewith respect to the particle emission source calibration. The position resolution calibration resultsshow better performance when using amplitude-based than when using time-based measurements.Amplitude-based results are in a reasonable range for performing further work on neutron imag-ing. Studies of muon-based position calibrations indicate a promising path toward full detectorcalibration to be used for interaction reconstructions. Namely, the edge bars can be calibrated foran amplitude-based response using sources. In turn, the calibrated edge bars are used to providetracking information for muon trajectories to central bars to regenerate amplitude-based calibrationcurves. Results indicate position resolutions for such measurements to be less than 17 .
74 mm. Thecomparison with the calibration using emission sources showed consistency. The time calibrationmethod allowed for determining the time deviations between pairs of bars needed for double-neutronscatter detections. The method can be extrapolated further to the full detector by rotating it ∼ ° and setting the calibrated bars as the referenced ones.Our collaboration is continuing to pursue measurements on this prototype OS-SVSC, intendingto demonstrate neutron imaging. A parallel effort is also underway to realize a second prototypewith a modular design that addresses the limitations described in Section 6. Acknowledgments
Sandia National Laboratories is a multimission laboratory managed and operated by NationalTechnology & Engineering Solutions of Sandia, LLC, a wholly owned subsidiary of HoneywellInternational Inc., for the U.S. Department of Energy’s National Nuclear Security Administrationunder contract DE-NA0003525. This paper describes objective technical results and analysis. Anysubjective views or opinions that might be expressed in the paper do not necessarily representthe views of the U.S. Department of Energy or the United States Government. Document releasenumber SAND2021-1287 O.The authors would like to thank the US DOE National Nuclear Security Administration, Officeof Defense Nuclear Nonproliferation Research and Development for funding this work.
References [1] P. Marleau, J. Brennan, K. Krenz, N. Mascarenhas, and S. Mrowka,
Advances in Imaging FissionNeutrons with a Neutron Scatter Camera , (2007) 1082-3654. doi:10.1109/NSSMIC.2007.4436310[2] N. Mascarenhas, J. Brennan, K. Krenz, P. Marleau, and S. Mrowka, Results with the Neutron ScatterCamera , IEEE Transactions on Nuclear Science. , (2009) 1269-1273.doi:10.1109/TNS.2009.2016659[3] A. Poitrasson-Rivière et al. , Dual-particle imaging system based on simultaneous detection of photonand neutron collision events. Nucl. Instrum. Methods Phys. Res. A , (2014) 40-45.doi:10.1016/j.nima.2014.05.056[4] R. T. Grannan et al. , A large area detector for neutrons between 2 and 100 MeV. Nucl. Instrum.Methods Phys. Res. A ,
103 (1) (1972) 99-108. doi:10.1016/0029-554X(72)90465-X – 22 –
5] A. D. Zych et al. , Large area double scattering telescope for balloon-borne studies of neutrons andgamma rays. IEEE Transactions on Nuclear Science , NS-22 (1975) 605-610.doi:10.1109/TNS.1975.4327713[6] C. B. Wiinderer et al. , SONTRAC-A low background, large area solar neutron spectrometer. AIPConf. Proc. ,
450 (1) (1998) 73-76. doi:10.1109/CHERBS.1997.660251[7] J. E. M. Goldsmith, M. D. Gerling, and J. S. Brennan,
A compact neutron scatter camera for fielddeployment , Rev. Sci. Instrum. , (2016) 083307. doi:10.1063/1.4961111[8] J. Braverman et al. , Single-Volume Neutron Scatter Camera for High-Efficiency Neutron Imaging andSpectroscopy , [arXiv:1802.05261][9] G. Jocher, J. Koblanski, V.A. Li, S. Negrashov, R.C. Dorrill, K. Nishimura, M. Sakai, J.G. Learned,and S. Usman, miniTimeCube as a neutron scatter camera , AIP Advances. , (2019) 035301.doi:10.1063/1.5079429[10] K. Weinfurther, J. Mattingly, E. Brubaker, and J. Steele, Model-based design evaluation of a compact,high-efficiency neutron scatter camera. Nucl. Instrum. Methods Phys. Res. A , (2018) 115-135.doi:10.1016/j.nima.2017.11.025[11] T. Laplace et al. , Low energy light yield of fast plastic scintillators. Nucl. Instrum. Methods Phys.Res. A , (2020) 161444. doi:10.1016/j.nima.2018.10.122[12] Eljen Technology, General Purpose Plastic Scintillator EJ-204 (Jan 2016). Retrieved from https://eljentechnology.com/products/plastic-scintillators/ej-200-ej-204-ej-208-ej-212 .[13] M. Sweany et al. , Interaction position, time, and energy resolution in organic scintillator bars withdual-ended readout. Nucl. Instrum. Methods Phys. Res. A , (2019) 451-462.doi:10.1016/j.nima.2019.02.063[14] M. Janecek, Reflectivity Spectra for Commonly Used Reflectors IEEE Transactions on NuclearScience.
59, 3 (2012) 490-497. doi:10.1109/TNS.2012.2183385[15] Eljen Technology, Silicone Rubber Optical Interface EJ-560 (2016) Retrieved from https://eljentechnology.com/products/accessories/ej-560 [16] S. Dolinsky, G. Fu and A. Ivan,
Timing resolution performance comparison for fast and standardoutputs of SensL SiPM. IEEE Nuclear Science Symposium and Medical Imaging Conference(NSS/MIC) (2013) 1 - 6. doi:10.1109/NSSMIC.2013.6829520[17] M. Sweany, P. Marleau, C. Allwork, G. Kallenbach, and S. Hammon,
Characterization of a siliconphoto-multiplier array with summing board as a photo-multiplier tube replacement in organicscintillator assemblies. Nucl. Instrum. Methods Phys. Res. A , (2020) 163118.doi:10.1016/j.nima.2019.163118[18] SensL Technologies Ltd., How to Evaluate and Compare Silicon Photomultiplier Sensors,
Rev 1.1 (2015). Retrieved from https://sensl.com/downloads/ds/WP-Evaluate_and_Compare_SiPM.pdf[19] V. A. Li et al. , Invited Article: miniTimeCube. Rev. Sci. Instrum. , (2016) 021301. doi:10.1063/1.4942243[20] R. Brun and F. Rademakers, ROOT - An Object Oriented Data Analysis Framework.
Nucl. Instrum.Methods. Phys. Res. A, (1997) 81-86. doi:10.1016/S0168-9002(97)00048-X. See also http://root.cern.ch/ .[21] M. Bitossi, R. Paoletti, and D. Tescaro,
Ultra-fast sampling and data acquisition using the DRS4 – 23 – aveform digitizer. IEEE Transactions on Nuclear Science. ,
63, 4 (2016) 1558-1578.doi:10.1109/TNS.2016.2578963[22] SensL Technologies Ltd.,
ArrayJ Series: Silicon Photomultiplier (SiPM) High Fill-Factor Arrays.
Rev 6
Thelarge analog bandwidth recorder and digitizer with ordered readout (LABRADOR) ASIC Nucl.Instrum. Methods Phys. Res. A , (207) 447-460. doi:10.1016/j.nima.2007.09.013[24] B. Cheng, L. Zhao, J. Qin, H. Chen, Y. Guo, S. Liu, Q. An, An overdetermined linear equations-basedtime calibration method for fast sampling ASICs Nucl. Instrum. Methods Phys. Res. A , A, Vol. ,(2019) 71-76. doi:10.1016/j.nima.2018.11.029[25] S. Cecchini, M. Spurio,
Atmospheric muons: experimental aspects.Geosci. Instrum. Method. DataSyst. , (2012) 185–196. doi:10.5194/gi-1-185-2012[26] J. Allison et al. , Recent developments in Geant4. Nucl. Instrum. Methods Phys. Res. A , (2016)186-225. doi:10.1016/j.nima.2016.06.125[27] D. Heck, J. Knapp, J. N. Capdevielle, G. Schatz, and T. Thouw, CORSIKA: A Monte Carlo Code toSimulate Extensive Air Showers. FZKA 6019 (1998). Retrieved fromhttps://web.ikp.kit.edu/corsika/physics_description/corsika_phys.pdf[28] L. Lista,
Combination of measurements and the BLUE method. EPJ Web Conf. , (2017) 11006.doi:10.1051/epjconf/201713711006[29] k. Zhu et al. , Calibration of large neutron detection arrays using cosmic rays Nucl. Instrum. MethodsPhys. Res. A , (2020) 163826. doi:10.1016/j.nima.2020.163826(2020) 163826. doi:10.1016/j.nima.2020.163826