Improved Pulse Shape Discrimination in EJ-301 liquid scintillators
IImproved Pulse Shape Discrimination in EJ-301 Liquid Scintillators
R.F. Lang a , D. Masson a , J. Pienaar a , S. R¨ottger b a Department of Physics and Astronomy, Purdue University, West Lafayette, USA b Physikalisch-Technische Bundesanstalt, Braunschweig, Germany
Abstract
Digital pulse shape discrimination has become readily available to distinguish nuclear recoil and electronic recoil eventsin scintillation detectors. We evaluate digital implementations of pulse shape discrimination algorithms discussed inthe literature, namely the Charge Comparison Method, Pulse-Gradient Analysis, Fourier Series and Standard EventFitting. In addition, we present a novel algorithm based on a Laplace Transform. Instead of comparing the performanceof these algorithms based on a single Figure of Merit, we evaluate them as a function of recoil energy. Specifically, usingcommercial EJ-301 liquid sctintillators, we examined both the resulting acceptance of nuclear recoils at a given rejectionlevel of electronic recoils, as well as the purity of the selected nuclear recoil event samples. We find that both a StandardEvent fit and a Laplace Transform can be used to significantly improve the discrimination capabilities over the wholeconsidered energy range of 0 −
800 keV ee . Furthermore, we show that the Charge Comparison Method performs poorlyin accurately identifying nuclear recoils.
1. Introduction
Liquid scintillators such as EJ-301 (which is similarto NE-213 and BC-501) are very popular for neutron de-tection as they can easily be shaped into the desired sizeand geometry of a given application and offer fast timingperformance. However, since such liquid scintillators arealso sensitive to gamma rays, pulse-shape discrimination(PSD) techniques are essential in order to correctly iden-tify neutron interactions in the detector.The ability to discriminate nuclear recoil (NR) eventsfrom electronic recoil (ER) events originates in the partic-ular production mechanisms of scintillation light in organicliquid scintillators. These liquids are aromatic compoundswhich have planar molecular structures built up from ben-zenoid rings. Such structures allow for extended group-ings of conjugated molecular bonds between unsaturatedcarbon atoms [1]. This results in some of the valence elec-trons of the carbon atoms being delocalized in π -molecularorbitals. It is the excitations of these π -electronic statesthat create the fluorescence observed in organic scintil-lators. During these excitations, π -electrons can be pro-moted from the ground state S to excited singlet states S n or triplet T n states. For low excitation densities, all ex-cited singlet states above the first excited singlet states S decay rapidly and non-radiatively to the lowest excited sin-glet state. This state then decays exponentially producingfluorescence in the process.In contrast, the decay of the triplet state is governed bythe diffusion time-scale of the triplet exciton and results in Email addresses: [email protected] (R.F. Lang), [email protected] (D. Masson), [email protected] (J. Pienaar), [email protected] (S. R¨ottger) delayed fluorescence in which the intensity does not decayexponentially. NRs exhibit greater energy-loss rates andthus have higher densities of triplet states. Pulses fromthe ionization tracks of these particles exhibit higher yieldsof delayed fluorescence, hence decaying more slowly thanthose of ERs. Scintillation light from EJ-301 has threemain decay components: 3 . γ + FWHM n (1)where peak separation refers to the distance between the Preprint submitted to Elsevier May 16, 2017 a r X i v : . [ phy s i c s . i n s - d e t ] M a y igure 1: The irradiation setup of the EJ-301 detector (left) corresponding to the 0 ◦ orientation in Table 1. The shadow cone is visible towardthe right. The neutron beam enters the setup from the right. center of the neutron and gamma distributions in a his-togram of the discrimination parameter, and FWHM i isthe full-width half maximum of the respective distribu-tions. Hence, the FOM does not provide any informa-tion on the energy dependence of the performance of PSDtechniques. This precludes a comparison of the variousalgorithms across different authors that may use differentenergy thresholds in the calculation of their FOM, and ad-ditionally, may mask performance issues of the algorithmsin particular at low recoil energy. Therefore, we exam-ined the energy-dependent ability of PSD techniques todiscriminate between ER and NR events. Furthermore,we determined the efficiency of EJ-301 for detection ofneutrons as a function of energy.
2. Setup
The fast neutron detector used in this work is a 3” cellof EJ-301 liquid organic scintillator optically coupled toa fast photomultiplier tube (PMT), type 9821KB man-ufactured by ET Enterprises. The detector response toneutrons was characterized at the Physikalisch-TechnischeBundesanstalt (PTB) in Braunschweig, Germany, using adeuterium ion beam hitting a Ti( H) target. The deu-terium ion beam energy (3.356 MeV) was chosen to pro-duce (2 . ± . H(d,n) He nuclear reaction, in thedirection of the ion beam. The detector was placed 3 mfrom the target. The output of the PMT was connectedto a CAEN DT5751 digitizer, which samples at 1 GHzwith a resolution of 10 bits. This digitizer has a 1 V dy-namic range. A 1 MeV ee pulse from an ER event in thePMT produces a 550 mV signal.Data were collected at three different nominal beamcurrent settings to study the effect of neutron flux on theperformance of the detector. The detector was placed suchthat the neutron beam was parallel to the normal of thefront face, defined as an angle of 0 ◦ . The distance betweenthe front face of the detector and the active layer of thetarget was (3000 ±
2) mm (k=2 [12]) for all measurements.Additional data were taken at each setting with a shadowcone, made of iron and polyethylene, placed between the target and the detector to measure the in-scatter of neu-trons, as illustrated in Figure 1. At the highest nominalbeam current, data was also collected with an angle of 90 ◦ between the direction of the ion beam and the front face ofthe detector. In the 90 ◦ orientation the detector is rotatedsuch that the neutron flux is incident on the side of thedetector, rather than the front face.These datasets are listed in Table 1 with their knownfluxes as measured using calibrated detectors at PTB.Dataset 4 has a greater flux than dataset 1, despite thebeam conditions being the same, due to the greater cross-sectional area the detector presents to the neutron beamin this orientation. The known flux in Dataset 4 is slightlyhigher than can be attributed to geometric factors alone,as the nominal beam charge for Dataset 4 is 5.6% greaterthan in Dataset 1. Table 1: Data for the irradiation of the detector in the neutron fieldwith a mean energy of 2 . Data Set Current Orientation Nominal Charge[ µ C] Flux [s − ]1 1.5 µA ◦ ± nA ◦ ± nA ◦ ±
154 1.5 µA ◦ ± The response of EJ-301 to ERs is known to be linear.Data presented in this work is therefore given in termsof the electron recoil equivalent energy keV ee . This en-ergy scale is set using the Compton backscatter edge ofgammas from Co,
Cs and Mn, measured from datacollected with the detector in the experimental hall. Thebackground rate of ER events in the experimental hall wasmeasured during an overnight measurement.A total of 80 million waveforms (amounting to 85 GB)were collected from the neutron source, background, andcalibration gamma-sources, and stored for offline process-ing.
3. Discrimination Algorithms
As EJ-301 features different decay constants for NRand ER signals, a variety of methods can be used to dis-2riminate the corresponding waveforms. Five PSD al-gorithms were implemented in a C++ program to per-form offline analysis of the data and compute discrimi-nation parameters for each waveform. These algorithms,described in detail below, are the Charge ComparisonMethod (CCM), Pulse Gradient Analysis (PGA), FourierSeries Expansion (FSE), Laplace Transform (LAP), and afit to standard events (SEF). Typical scintillation pulseslast for 0 . A m p li t u d e [ m V ] Integration threshold σ noise levelBaselineregion Trigger thresholdIntegralstartFastintegralwindowSlowintegralwindow PGA sample Figure 2: A typical gamma event of energy 100 keV. Shown are thetrigger and integration thresholds, the ends of the Fast and Slow in-tegral windows, the location of the sample used in the PGA method,the region used for baseline calculations, and the noise level.
The Charge Comparison Method (CCM) [6–8, 13–17]predates modern digital computing and was first imple-mented via passive electronics [4]. In this method, thebaseline-subtracted waveform is integrated over two timewindows of different lengths, called slow and fast or long and short , respectively. The start of these integral win-dows is the onset of the pulse, which is defined here as thepoint at which the waveform exceeds 3 σ of the baselineRMS (as shown in Figure 2). The lengths of the two win-dows are generally set to match the decay modes of thedetector. As a NR pulse will decay more slowly than anER pulse, the slow integral value I slow will be larger forNR waveforms than for ER, while the fast integral values I fast are typically comparable for both ER and NR wave-forms. We have optimized these times according to thetraditional Figure of Merit [18] and found that values of50 ns for the fast window and 310 ns for the slow win-dow result in optimal discrimination. The discrimination parameter is the ratio of the two integral values,PSD CCM = I slow I fast . The Pulse Gradient Analysis method (PGA) [9] com-pares the relative height of the peak H peak to that of asample H sample a set time after the peak, here 50 ns. Thissecond sample is averaged with the neighboring 10 sam-ples to reduce noise. As ER pulses decay more quicklythan NR pulses, the gradient between the peak and thissecond sample should be larger for the ER than for the NR.We define the discrimination parameter for this method asthe ratio between the two amplitudes,PSD PGA = H sample H peak . The third is a method based on a Fourier series ex-pansion of the waveform ( f ( t ) = (cid:80) n A n exp( iω n t ); A n = (cid:82) T d t f ( t ) exp( iω n t )). Using an approach similar to [19],the difference between an even order coefficient A n (oddorder A n +1 ) and the zeroth coefficient A (first coefficient A ) is normalized to the zeroth (first) coefficient and thensummed: F even = (cid:88) n A n − A A and F odd = (cid:88) n A n +1 − A A . The expansion is computed to the 30 th order. The dis-crimination parameter is then defined as the ratio betweenthese two parameters:PSD FSE = F odd F even . We also evaluate a discrimination algorithm based onthe Laplace transform L{ f } = (cid:82) ∞ d t f ( t ) exp( − st ). A 9-point moving average is calculated over the trailing edgeof the waveform, and the smoothed pulse is then trans-formed. The transformed waveform is integrated over twofrequency ranges, 0 .
01 GHz ≤ s < . . ≤ s < L low and L high . The frequency ranges are chosen such thatthe contributions from the 32 ns and 270 ns decay modesare maximized in each respective range. The discrimina-tion parameter for this method is defined as L high − L low .However, since this parameter varies significantly with en-ergy, we choose to re-scale it according toPSD LAP = I − A log( B + L high − L low )in order to aide visualization, with the integral of the pulsegiven as I . Parameters A and B are constants chosen as1/30 and 2, respectively, simply to linearise the plot.3
50 100 150 200Time [ns]050100150200250300350400 A m p li t u d e [ m V ] τ Λ GammaNeutron
Figure 3: Standard nuclear recoil (red dotted line) and electronicrecoil (blue solid line) events after baseline subtraction. The free fitparameters are a horizontal shift τ and a vertical scale factor Λ Prior to analysis, 10 events identified as NR or ER byCCM are selected in a narrow energy region around theCompton edge of the 662 keV gamma from Cs. Av-eraging these waveforms together forms the standard NRor ER event. Both these standard events are then fittedto a given waveform [20, 21]. While the baseline is fixedto the average of the first 40 ns of the waveform, the freeparameters of the fit are a horizontal shift τ N,E and a scal-ing factor Λ
N,E , see Figure 3. The ER/NR discriminationparameter is then defined as the difference between the chi-squared value χ N,E of each fit, normalized to the verticalscaling fit parameter Λ:PSD
SEF = χ N Λ N − χ E Λ E .
4. Analysis
The collected data were each processed using the 5 PSDmethods discussed in Section 3. Rejection cuts for ERevents are defined using the datasets taken with variousgamma-sources using histograms of the discrimination pa-rameter versus energy. In each energy bin, we considerthe one-sided 95%, 96%, 97%, 98% and 99% ER rejectionquantiles. Application of the resulting rejection cuts to theovernight background data confirm their performance onneutron datasets. These rejection cuts are robust againstchanges in detector orientation and data acquisition rate.The event distribution from neutron data are shownin Figure 4 for all PSD methods together with the one-sided 99% ER rejection cut. For ease of comparison, allPSD parameters were defined such that the NR band isthe upper event population. CC M F S E Energy [keV ee ] L A PP G A Energy [keV ee ] S E F Counts/bin
Figure 4: Plots of various discrimination parameters versus energyfor the CCM (a), FSE (b), LAP (c), PGA (d), and SEF (e) algo-rithms. The 99% rejection cut of electronic recoils is also shown ineach case. The upper populations are the respective nuclear recoilbands. .2. Neutron Flux Using the rejection levels defined from the ER band,events are tagged as being either an electronic or nuclearrecoil.Figure 5 shows the live time-corrected NR energy spec-trum, measured from dataset 2, after 99% rejection of ERsusing the CCM algorithm as an example. The expecteddouble-peaked structure due to neutron double scatterevents within the detector volume is evident. The recoilspectrum extends to just below 1 MeV ee , as expected from2 . ee which will be discussed in section 5.
200 400 600 800 1000
Energy [keV ee ] -2 -1 R a t e [ s − k e V ee − ] Total FluxBackscattered FluxRecoil Spectrum
Figure 5: Nuclear recoil spectra as measured in the EJ-301 detectorusing the CCM algorithm. The measured backscatter flux (dotted,red) is subtracted from the total measured flux (dashed, blue) in or-der to obtain the recoil spectrum of 2 . Data collected under the same conditions as dataset 2,but with the shadow cone between the detector and thetarget is also shown. The observed spectrum is consistentwith a range of energies from neutrons scattered off of theair in the experimental hall. The absence of a similar NRspectrum in background data (with no ions on the target),also shown, confirms that these events are related to thebeam.Since we are interested in the efficiency of the EJ-301detector with regards to detecting 2 .
5. Results
Rather than reducing the performance of different PSDalgorithms to a single Figure of Merit value (Eq. 1), we in-vestigate their energy-dependent behaviour. Specifically,for a given ER discrimination cut, we compare the effi-ciency of various algorithms using the number of acceptedneutrons as a function of energy. In order to quantify the acceptance of neutrons, a pureband of NRs directly from the beam is required. How-ever, given the reduced performance of all algorithms atlow energies, the subsequent overlapping of the NR andER band, as well as the background from scattered neu-trons, no such pure sample is available. We thus invokea simple statistical algorithm as follows. We use the datataken with the neutron beam incident on the detector toproduce the event distributions in the space of PSD pa-rameter
P SD i versus energy E , as shown in Figure 4.These distributions are corrected for the livetime of eachdataset and the integrated beam current, to obtain thetime-normalized event density (cid:37) sig+bck ( P SD i , E ). Simi-larly, datasets in which the shadow cone was present re-sult in a time-normalized background density of both ERand NR events (cid:37) bck ( P SD i , E ). Subtracting these twoevent densities results in an event density (cid:37) sig ( P SD i , E ) = (cid:37) sig+bck ( P SD i , E ) − (cid:37) bck ( P SD i , E ) that is assumed to berepresentative of the pure band of NRs directly from thebeam. This 2D-histogram of event density is then pro-jected onto the energy axis to obtain a ”pure” spectrumof all 2 . ee at 95% rejection of ERs. The acceptances of the LAPand FSE algorithms match each other at this rejectionlevel. As the rejection level of ERs is increased to 99%,the acceptance of the LAP algorithm at higher energies ismarginally but consistently higher than that of the SEFalgorithm. Of note here is that the performance of the tra-ditional CCM algorithm decreases as the rejection criteriafor ER events becomes more stringent. Specifically, at 95%rejection of ERs, its acceptance is consistent with that ofthe SEF, FSE and LAP algorithms, and better than thePGA algorithm. However, at 99% rejection of ERs, theCCM algorithm has the lowest acceptance below 80 keV ee . At low energy ( <
100 keV ee ), there is considerable over-lap between the NR and ER bands. Therefore, we not onlywant to consider the ability of an algorithm to accept neu-trons, but also the purity of the resulting NR spectrum.To this end, we construct a reference sample for each algo-rithm i that contains events that are NRs with high con-fidence, using those events that are identified by all other5 .00.20.40.60.81.0 A cc e p t a n c e CCMPGAFSESEFLAP
20 40 60 80 100 120 140
Energy [keV ee ] A cc e p t a n c e CCMPGAFSESEFLAP
Figure 6: Energy dependence of the fraction of nuclear recoil eventsthat are accepted by each algorithm after 99% (top) and 95% (bot-tom) rejection of electronic recoils. algorithms j (cid:54) = i as a NR. As a function of energy, we thencalculate how many of the events in the reference samplewere also tagged as a NR by the algorithm under investi-gation. The fraction of events in the reference sample thatan algorithm has tagged as a NR is shown in Figure 7.
20 40 60 80 100 120 140
Energy [keV ee ] N e u t r o n I d e n t i f i c a t i o n CCMPGASEFFSELAP
Figure 7: The fraction of NRs identified by a given algorithm from areference sample. The reference sample consists of events identifiedas NRs by all algorithms except the algorithm under consideration.The shaded region indicates the systematic uncertainty.
The inability of the traditional CCM algorithm to iden-tify NR events in a reference sample below 100 keV ee isstriking. In contrast, the SEF algorithm performs bestat low energies and correctly identifies more than 95% ofNR events down to energies below 50 keV ee . For the SEFalgorithm, this fraction is flat down to energies well be-low the point at which the acceptance of neutrons for allalgorithms has fallen below 50% (compare Figure 6). In- triguingly, while the LAP algorithm does provide lowerefficiency at low energies, it provides a slightly higherlevel of confidence that an event is truly a neutron above ∼
75 keV ee . In contrast to [9], we find that the perfor-mance of the PGA algorithm is inferior to those of allother algorithms above 80 keV ee . This observation can beattributed to the fact that the PGA algorithm is basedon only two samples of the recorded waveform, thus beinghighly susceptible to electronic noise.The performance of algorithms which perform poorlyat low energies, such as the CCM algorithm, can bias theselection of the reference neutron sample. To estimate thesystematic uncertainty in the fraction of identified NRs,we neglect one algorithm, and repeat the procedure above,i.e. we construct reference samples from three of the fourremaining samples, and examine which fraction of NRsare identified by the fourth method. In this way we obtainfour neutron identification curves for each method, fromwhich we determine the standard deviation. This is takenas a measure of the systematic uncertainty of each methodand is shown as the shaded regions around each curve inFigure 7. Table 1 lists the total expected neutron flux of 2 . ◦ at selected en-ergy thresholds. We find that the efficiency measured inthe detector is inconsistent with both the overall efficiencydescribed by the NEFF7 code, as well as the functional de-pendence of the efficiency on the threshold. Independentof the PSD algorithm, the measured efficiency in the EJ-301 detector is lower and less threshold-dependent thanthat described by the NEFF7 code. We thus concludethat the NEFF7 code can currently not be used to obtainaccurate detection efficiency information for EJ-301 liquidscintillator cells.The PGA algorithm has the lowest efficiency above30 keV ee , consistent with it having a poorer ability to ac-cept neutrons at a given rejection level of ERs. The tradi-tional CCM algorithm has poor efficiency below 60 keV ee when compared to the three newer algorithms FSE, SEFand LAP. The LAP algorithm has the highest efficiency atall energy thresholds. This observation can be ascribed toits higher acceptance of NRs at higher energies. The LAPalgorithm is better at discriminating between the NR andER band at higher energies, as the ER band of the SEFalgorithm spreads as the energy increases (compare Fig-ure 4), while the ER band of the LAP algorithm has amore consistent width.6
00 200 300 400 5000.10.20.30.40.5 E ff i c i e n c y LAPNEFF7
20 40 60 80 100
Energy [keV ee ] E ff i c i e n c y CCMPGAFSESEFLAP
Figure 8: (Top) Comparison between the threshold-dependent ef-ficiency of the SEF algorithm and the prediction from the NEFF7code in EJ-301. The detector is orientated at 90 ◦ . (Bottom) The ef-ficiency of all algorithm as a function of threshold. Note the differentenergy ranges. The processing speed of each algorithm was mea-sured on a dataset of approximately 500 000 events (about550 MB), both with and without the C++ compiler opti-mizations available in the GNU compiler collection version4.8.3, using a standard desktop computer. The results areshow in Table 2. From this we see that the increased low-energy performance of the SEF algorithm comes at a veryhigh computational cost.
Table 2: Processing rates of the different algorithms, given in unitsof waveforms/sec, using a standard desktop computer.
Algorithm Rate withoutoptimizations Rate withoptimizationsPGA 1 . × . × CCM 168 000 1 . × FSE 4 300 84 000LAP 1 100 16 000SEF 110 230
6. Conclusions
We have compared the performance of five differentpulse shape discrimination algorithms using a commercialliquid scintillator cell. The studied algorithms include theavant-garde algorithms Standard Event Fit SEF, Fourier-Series Expansion FSE, and Laplace Transform LAP in ad-dition to the traditional Charge Comparison Method CCM and Pulse-Gradient Analysis PGA. The energy-dependentbehaviour of all five algorithms was discussed as a bettermeans of describing PSD algorithms than the Figure ofMerit previously used in the literature.Specifically, we considered the ability of each algorithmto accept NRs as function of recoil energy, as well as thepurity of the resulting NR sample. We find that at 99%rejection of ER background events and above 80 keV ee , thePGA algorithm accepts the fewest number of NRs and isthe least likely to identify a NR event from a reference sam-ple. The CCM algorithm performs better than the PGAmethod, but a marked deterioration in the performance ofthe CCM algorithm is observed as the rejection level ofER events becomes more stringent.Both the SEF algorithm and the LAP algorithm dis-play improved performance compared to the traditionalmethods considered. The SEF algorithm is more likely toaccept NRs from a reference sample at low energies, and itprovides a higher acceptance of NR events below 80 keV ee .The LAP method however provides a slightly higher effi-ciency overall for the detection of NRs in EJ-301 due to abetter-resolved ER band at higher energies.
7. Acknowledgements
This work was supported by grant
8. ReferencesReferences [1] F.D. Brooks. Development of organic scintillators.
Nucl. Instr.Meth. , 162(1-3):477–505, 1979.[2] F. T. Kuchnir and F. J. Lynch. Time dependence of scintilla-tions and the effect on pulse-shape discrimination.
IEEE Trans-actions on Nuclear Science , 15(3):107–113, 1968.[3] F.D. Brooks. A scintillation counter with neutron and gamma-ray discriminators.
Nucl. Instr. Meth. , 4(3):151–163, 1959.[4] T.K. Alexander and F.S. Goulding. An amplitude-insensitivesystem that distinguishes pulses of different shapes.
Nucl. Instr.Meth. , 13:244–246, 1961.[5] J.M. Adams and G. White. A versatile pulse shape discrim-inator for charged particle separation and its application tofast neutron time-of-flight spectroscopy.
Nucl. Instr. Meth. ,156(3):459–476, 1978.[6] Y. Kaschuck and B. Esposito. Neutron/ γ -ray digital pulse shapediscrimination with organic scintillators. Nucl. Instr. Meth.Phys. Res. A , 551(2-3):420–428, 2005.[7] D. Cester et al. Pulse shape discrimination with fast digitizers.
Nucl. Instr. Meth. Phys. Res. A , 748:33–38, 2014.[8] C. Liao and H. Yang. N/ γ Pulse shape discrimination compar-ison of EJ301 and EJ339A liquid scintillation detectors.
Ann.Nucl. Energy , 69:57–61, 2014.[9] B. D’Mellow, M. D. Aspinall, R. O. Mackin, M. J. Joyce, anda. J. Peyton. Digital discrimination of neutrons and γ -rays inliquid scintillators using pulse gradient analysis. Nucl. Instr.Meth. Phys. Res. A , 578(1):191–197, 2007.
10] K. A. A. Gamage, M. J. Joyce, and N. P. Hawkes. A compari-son of four different digital algorithms for pulse-shape discrim-ination in fast scintillators.
Nucl. Instr. Meth. Phys. Res. A ,642(1):78–83, 2011.[11] S. Yousefi, L. Lucchese, and M. D. Aspinall. Digital discrim-ination of neutrons and gamma-rays in liquid scintillators us-ing wavelets.
Nucl. Instr. Meth. Phys. Res. A , 598(2):551–555,2009.[12] International Organisation for Standardization. Evaluation ofmeasurement data - guide to the expression of uncertainty inmeasurement, 1995.[13] D Wolski, M Moszy´nski, T Ludziejewski, A Johnson,W Klamra, and ¨O Skeppstedt. Comparison of n- γ discrimi-nation by zero-crossing and digital charge comparison methods. Nucl. Instr. Meth. Phys. Res. A , 360(3):584 – 592, 1995.[14] H Klein and S Neumann. Neutron and photon spectrometrywith liquid scintillation detectors in mixed fields.
Nucl. Instr.Meth. Phys. Res. A , 476(12):132 – 142, 2002. Int. Workshop onNeutron Field Spectrometry in Science, Technology and Radi-ation Protection.[15] G Ranucci, A Goretti, and P Lombardi. Pulse-shape discrimi-nation of liquid scintillators.
Nucl. Instr. Meth. Phys. Res. A ,412(23):374 – 386, 1998.[16] P.A. S¨oderstr¨om, J. Nyberg, and R. Wolters. Digital pulse-shape discrimination of fast neutrons and γ -rays. Nucl. Instr.Meth. Phys. Res. A , 594(1):79 – 89, 2008.[17] G. Rannucci An analytical approach to the evaluation of thepulse shape discrimination properties of scintillators.
Nucl. In-str. Meth. Phys. Res. A , 354:389 – 399, 1995.[18] J.R.M. Annand. A fast module for pulse shape analysis.
Nucl.Instr. Meth. Phys. Res. A , 262(2):371 – 377, 1987.[19] G. Liu, M. J. Joyce, X. Ma, and M. D. Aspinall. A digitalmethod for the discrimination of neutrons and γ rays with or-ganic scintillation detectors using frequency gradient analysis. IEEE Transactions on Nuclear Science , 57(3):1682–1691, 2010.[20] C. Guerrero, D. Cano-Ott, M. Fern´andez-Ord´o˜nez, E. Gonz´alez-Romero, T. Mart´ınez, and D. Villamar´ın. Analysis of theBC501A neutron detector signals using the true pulse shape.
Nucl. Instr. Meth. Phys. Res. A , 597(23):212 – 218, 2008.[21] S. D. Ambers, M. Flaska, and S. A. Pozzi. A hybrid pulseshape discrimination technique with enhanced performance atneutron energies below 500 kev.
Nucl. Instr. Meth. Phys. Res.A , 638(1):116 – 121, 2011.[22] V.V.l Verbinski et al. Calibration of an organic scintillator forneutron spectrometry.
Nucl. Instr. Meth. , 65(1):8–25, 1968.[23] A. Aksoy et al. Response-function measurement of an NE213scintillator using the 2H(d, n)3He reaction.
Nucl. Instr. Meth.Phys. Res. A , 337:486–491, 1994.[24] A A Naqvi, F Z Khiari, A Aksoy, A Coban, and M A Al-Jalal.Light yield measurements of a NE213 detector for 2.8-14.8 MeVneutrons.
Nucl. Instr. Meth. Phys. Res. A , 325(3):574–577,1993.[25] G. Dietze and H. Klein. NRESP4 and NEFF4 - Monte Carlocodes for the calculation of neutron response functions and de-tection efficiencies for NE 213 scintillation detectors, 1982., 325(3):574–577,1993.[25] G. Dietze and H. Klein. NRESP4 and NEFF4 - Monte Carlocodes for the calculation of neutron response functions and de-tection efficiencies for NE 213 scintillation detectors, 1982.