没有合适的资源?快使用搜索试试~ 我知道了~
资源推荐
资源详情
资源评论
Eur. Phys. J. C (2018) 78:392
https://doi.org/10.1140/epjc/s10052-018-5852-7
Regular Article - Experimental Physics
Improving sensitivity of a BEGe-based high-purity germanium
spectrometer through pulse shape analysis
M. Misiaszek
1
, K. Panas
1,a
, M. Wojcik
1
, G. Zuzel
1
,M.Hult
2
1
M. Smoluchowski Institute of Physics, Jagiellonian University, Lojasiewicza 11, 30-348 Kraków, Poland
2
European Commission, DG JRC, JRC-Geel, Retieseweg 111, 2440 Geel, Belgium
Received: 23 December 2017 / Accepted: 26 April 2018
© The Author(s) 2018
Abstract We performed Pulse Shape Analysis to sepa-
rate single-scattered gamma energy deposition events from
multiple-scattered photons in a high-sensitivity γ -ray spec-
trometer. The spectrometer is based on a Broad Energy High
Purity Germanium detector and the developed technique uses
multivariate analysis by an application of the Multi-Layer
Perceptron Neural Network. A very good separation of the
single-site- and multi-site events was achieved leading to a
significant reduction of the background level of the investi-
gated spectrometer – the double escape peak, rich in single-
site events, was reduced by 95%, while the full energy peaks
lost at most 25% of their counts. The peak to Compton ratio,
calculated for the 2614.5 keV gamma line from
208
Tl, was
improved by 114.3%.
1 Introduction
For germanium semiconductor detectors a single energy
deposition by γ -ray is often described as a single-site event
(SSE). Multi-site events (MSEs), like e.g. multiple Comp-
ton scatterings, have several interaction sites separated by a
distance of about 1 cm. The differences and discrimination
between SSEs and MSEs is of primary interest for exper-
iments like Gerda [1,2] looking for the neutrinoless dou-
ble beta (0νββ) decay with application of the High Purity
Germanium (HPGe) detectors. Events from the hypothetical
0νββ decay would ionize the detector’s active volume by
means of two electrons (with a range of less than 1 mm in
germanium) and therefore belong to the first category. The
background due to external γ -rays is typically of multi-site
type, because γ -rays with energies in the range of ∼ 1MeV
deposit their energy mainly via multiple Compton scattering
with a mean free path of a few centimeters.
a
e-mail: krzysztof.panas@doctoral.uj.edu.pl
For applications in γ -ray spectrometry the situation is
reversed: SSEs should be rejected and MSEs preserved. The
SSEs are of background type, because they are mostly single
Compton scattered events with the scattered photon escap-
ing the crystal, while the full energy peaks (FEPs) regis-
tered by the detector (and used to evaluate the activities of
radionuclides) contain mainly MSEs.
1
The developed proce-
dure shall therefore make peaks more distinguished by reduc-
ing the flat Compton continuum and make possible to eval-
uate peaks from radioisotopes, which would be otherwise
under spectrometer’s minimal detectable activity (sensitiv-
ity).
Discrimination between SSEs and MSEs for the Broad
Energy Germanium (BEGe) detectors was extensively stud-
iedintheframeofGerda experiment [3–7], also with respect
to applications in ultra-low background γ -ray spectrometry
[8]. The approach that was worked out was to look for a ratio
of the maximal current signal amplitude (A) to the corre-
sponding event energy (E) – the so-called A/E cut. This is
a one-parameter method, where the cut value is determined
according to the calibration data, obtained usually by irradi-
ating the spectrometer with a
228
Th source (due to the emitted
high energy γ -ray from the decay of
208
Tl – a more detailed
explanation can be found in Sect. 3.2). Measurements car-
ried out for various BEGe detectors demonstrated very good
performance of the A/E method, however, its applications to
the coaxial detectors did not provide satisfactory results.
The main goal of our work was to find an alternative, effi-
cient and stable Pulse Shape Analysis (PSA) procedure to
distinguish between MSEs and SSEs in BEGe type detectors.
As it will be shown, our method does not need any corrections
like e.g. the ones related to the dependence of the A/E classi-
1
For typical detectors used nowadays in γ -ray spectroscopy (crystal
volume of several hundreds of cm
3
), and for energies of some hundred
keV and higher, the full energy peaks contain mostly events, which were
caused by a few Compton scatterings followed by the photoelectric
effect.
123
392 Page 2 of 11 Eur. Phys. J. C (2018) 78:392
Fig. 1 Schematic drawing of the 50% relative efficiency BEGe detec-
tor applied in the Ge-5 spectrometer and used in the presented study
fier on energy [5]. To achieve this we applied methods imple-
mented in the Toolkit for Multivariate Analysis (Tmva)[9].
Tmva provides a Root-integrated machine-learning envi-
ronment for the processing and parallel evaluation of multi-
variate classification and regression techniques [10]. Tmva
includes classifiers like Projective Likelihood (PL), Multi-
Layer Perceptron Neural Network (MLP, see Sect. 4.1 for
more details), Boosted Decision Trees (BDT) and Support
Vector Machine (SVM). Each of the implemented methods
provides training, testing, performance evaluation algorithms
and visualization scripts. The training and testing is per-
formed with the use of user-supplied data sets in the form
of Root trees or text files.
2 Experimental setup
To study the new PSA technique we collected data using
the BEGe-based γ -ray spectrometer (Ge-5) operated by the
JRC-Geel (formerly Institute for Reference Materials and
Measurements) in the HADES underground laboratory. The
rock overburden of 500 m water equivalent provides attenua-
tion of the muon flux by about four orders of magnitude [11].
The spectrometer is based on a 50% relative efficiency p-type
BEGe diode installed in a standard vacuum cryostat produced
by Canberra (model BE5030). A massive low-radioactivity
shield surrounds the detector in order to reduce the environ-
mental background and makes the spectrometer sensitive to
very weak radioactivity. A sketch of the Ge-5 crystal is shown
in Fig. 1.
To train the Tmva-based methods and to evaluate their
performance, apart from the energy information (as obtained
usually from the system based on multi channel analyzers),
the waveforms from the preamplifier have to also be regis-
tered. They were acquired with the Struck SIS3302, a 16-bit
100 MHz Flash Analog to Digital Converter (FADC) – the
total length of each waveform was 30 µs. Noise and events
with dubious quality (trigger position much more before/after
pre-trigger, pile-upped events with multiple slopes) were
excluded form further analysis. For pulses, which survived
Fig. 2 Energy resolution of the peaks in the
228
Th spectrum. The out-
lying peak at 2103 keV is a single escape peak from 2615 keV
208
Tl
line. Its abnormally high FWMH value is due to the Doppler broadening
[12]
the quality cuts, the first step was to perform the energy recon-
struction. The energy value for each pulse was determined
using the trapezoidal filter described in [13], followed by
a calibration using a linear function. The energy resolution
values for the selected peaks from the
228
Th decay chain
spectrum are plotted in Fig. 2.
3 Pulse shape analysis
The data features that are used to train machine learning
models have a huge influence on the final performance. In
our case of the PSA, employing neural networks procedure
to the
228
Th data, the task of feature selection was limited
to a vector of the sampled waveforms of the preamplifier
output voltage. After series of dedicated tests, we found that
considering only samples close to the point of the maximal
current as the input parameters give us substantial discrimi-
nation effect. The shape of the pulse in the range of limited
samples manifests the type of interaction in the detector due
to the differences in speed and possible fluctuations of the
charge collection over time.
In the first step, after initial data pre-processing and energy
reconstruction, we were performing pulse shape normaliza-
tion. The sampled amplitudes were divided by the corre-
sponding reconstructed energy values in order to remove
potential energy dependency of the signal. Next, normalized
input variables were extracted from each rising edge and the
pulses (edges) were digitally differentiated and smoothed in
order to find the moment of the maximal values of the current
signals. 31 single amplitudes in total were chosen for each
pulse for the PSA: 15 before, 1 at and 15 after the moment
123
剩余10页未读,继续阅读
资源评论
weixin_38734361
- 粉丝: 6
- 资源: 904
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
最新资源
资源上传下载、课程学习等过程中有任何疑问或建议,欢迎提出宝贵意见哦~我们会及时处理!
点击此处反馈
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功