Digitized two-parameter spectrometer for
neutron-gamma mixed field
Zdenek Matej, Jiri Cvachovec, Vaclav Prenosil, Frantisek Cvachovec and Sergey Zaritski
Abstract—This paper shows the results of digital processing
of output pulses from combined photon-neutron detector using
a commercially available digitizer ACQUIRIS DP 210. The
advantage of digital processing is reduction of the apparatus
in weight and size, acceleration of measurement, and increased
resistance to pile-up of pulses. The neutron and photon spectrum
of radionuclide source 252
Cf is presented.
I. INTRODUCTION
THE recent development in the area of digital data processing
enables dramatic reduction of the spectrometric
system; this is achieved by converting the time course of
the detected pulse into numeric samples at the output of the
digitizer. Further data transmission and processing can be done
exclusively digitally. The above described procedure has a lot
of advantages:
• Down-sizing of the spectrometric devices
• Increased measurement rate (i.e. a shorter and thus
cheaper measurement is sufficient)
• Possibility of advanced pulse shape analysis – e.g. evaluation
of superimposed impulses caused by fast sequence
of particle detections.
• In certain cases, another characteristic of the output
impulse can be determined – the characteristic duration
(time constant) of the output pulses.
The last mentioned parameter sometimes carries very useful
information about the nature of the detected particle. This
applies e.g. to organic scintillator of type stilbene or NE-213.
The distinguishing of nature of the particles (in our case neutron/photon)
is based on the different relation of the probability
of triplet state excitation in the molecule of the scintillator
(the de-excitation generates the slow part of the fluorescence)
to the linear stopping power of the detected charged particles.
The neutron scintillation (a neutron is detected by means of a
scattered proton) has a longer decay than the photon impulse
(a photon is detected by means of a scattered electron). On the
other hand, for proportional detectors filled with hydrogen the
rise time of a neutron impulse is shorter than that of a photon
Manuscript received on 16/05/2011. The work presented in this paper
has been supported in the frame of specific research of the Department
of Mathematics and Physics, University of Defense Brno in 2010, specific
research of the Faculty of Informatics, Masaryk University in 2010 and
research project SPECTRUM, No. TA01011383 of the Technology Agency
of the Czech Republic.
Z. Matej, Faculty of Informatics, Masaryk University, Botanicka 68a, 602
00 Brno, Czech Republic. E-mail: xmatejz@mail.muni.cz.
J. Cvachovec and V. Prenosil is with Masaryk University, Brno, Czech
Republic.
F. Cvachovec is with University of Defence, Brno, Czech Republic.
S. Zaritski is with Kurchatov Institute, Moscow, Russia.
0 20 40 60 80 100
Time in ns
-3
-2.5
-2
-1.5
-1
-0.5
0
VoltageinV
Fig. 1. Example time courses of neutron (solid) and photon (dashed) pulses.
impulse; this is again caused by different linear stopping
power of the secondary charged particles (proton, electron). By
adjusting the load resistance at the output of the detector, we
can achieve a state where the speed of the increasing part of the
output voltage impulse (rise time) reflects the differences in the
time course of the current impulses. This different information
carried by the time courses can be utilized to distinguish the
neutron impulse from the photon one. The time courses of a
neutron and photon pulse are shown in Fig. 1. In this case,
the scintillation time constant is significantly smaller then the
time constant of the circuit consisting of a working resistance
and parasitic capacitance. We can see that a neutron pulse rises
2-3 ns longer than a photon pulse.
In the past, analog circuits were used to distinguish the
courses. Since 2000, attempts have been made to evaluate
both parameters (energy, type of particle) from the digitized
output impulse. It is usual to arrange the measured data into
a 3-D representation where one axis (e.g. x) represents the
rise time (type of the detected particle), another (y) stands
for amplitude (energy), and the last one (z) records counts
of impulses of the same parameters. This distribution, which
we measured with ACQUIRIS DP 210 digitizer, is shown in
Fig. 2. We can clearly distinguish neutrons and gamma part in
this distribution. Moreover, other types of charged particles,
noise, disturbances can be identified. Further processing leads
to spectral energetic distribution.
II. EXPERIMENTAL DATA PROCESSING
The digitizer ACQUIRIS DP 210 converts the initial part
of each input pulse (i.e. output signal from the detector) into
100 samples with a repetition frequency of 1 or 2 GHz.
20
40
60
Amplitude in a.u.
25 50 75 100
Rise Time in a.u.
0
50
100
Count
0
50
100
Count
Fig. 2. 3-D distribution of output pulses from stilbene detector. The x-axis
represents pulse rise time, y-axis amplitude, and z-axis count of pulses of the
same parameters corresponding to 5 and 95 percent of the amplitude. Gamma
part on the left, neutron part on the right.
These 100 values (obtained as discrete data) represent the
most important part of the pulse containing information on
the amplitude (energy) and the timing of the rising part of the
pulse (nature of the particle). The usually measured neutron
spectra in the interval 0.5 to 10 MeV require the recording
of about one million pulses. Since the neutron pulses are
sampled simultaneously with the photon ones without knowing
the nature of the particle and the ratio of neutrons to photons in
the mixed fields is more than ten, we receive relatively large
amounts of data. The data is transported to memory, where
it is stored for off-line processing because on-line processing
cannot be performed quickly enough.
Further data processing is performed so that for each pulse
(100 numbers) we find the largest number – amplitude – and
then we determine the time interval τ between samples.
According to that τ we should ideally be able to divide
all pulses into two groups – neutrons and photons. However,
reality is more complex and the result is a distribution as
shown in Fig. 2. It is obvious that the n/g pulses can be easily
distinguished for higher amplitudes but human interaction is
still required for lower energies. This distribution is then split
into neutron and photon parts; further, it is adjusted into a form
required by the codes solving the Fredholm integral equation
for the evaluation of the final product, i.e. the neutron and
photon energy spectrum.
The processing of the pulses with the current digitizer is
not ideal because of its low dynamic range of quantization
(8 bits). For neutrons from energy interval 0.5 to 10 MeV,
the ratio between the lowest and the highest pulse amplitudes
is ca. 1:200. We solved this problem by taking the measurements
at three different voltage levels for the photomultiplier;
the resulting amplitude distributions have been merged by a
software routine.
III. THE EVALUATION OF THE ENERGY SPECTRUM
When evaluating experimental data from the detector, we
meet the linear inverse problem that can be formulated in case
of neutron spectrum as follows: Let g(Ep) and A(En, Ep) be
(continuous) functions. A (continuous) function f(En) is to
be found such that
g(Ep) = A(En, Ep)f(En)dEn. (1)
The above equation models the process of measurement using
various devices or e.g. the output of a graphical device. It is
a Fredholm integral equation of the first kind where
• A(En, Ep) is the kernel, a characteristic of the measuring
device, often called the response function. Its discrete
approximation is determined by Monte Carlo modelling
and measurement of monoenergetic neutron sources.
• g(Ep) is the measured proton spectrum, i.e. experimental
data. (Neutrons are detected by means of protons.)
• f(En) is the result to be found, i.e. the neutron spectrum.
In general, equation (1) has no analytical solution and thus
must be solved in a discrete form:
g = Af (2)
where g = (g1, . . . , gm)T
, A is a m × n matrix, and
f = (f1, . . . , fn)T
(T
means transposition). For stilbene
and NE-213 detectors m = n. Our options for stilbene are
m ≈ 60, 120, 240. We solve (2) by the maximum likelihood
method [1]. It is a standard statistical tool for point estimations.
For the maximizing of the likelihood function, we use a
general iterative algorithm called Expectation Maximization,
originally developed for image reconstruction in astronomy,
medicine etc.
IV. MEASUREMENT RESULT AND ENERGY SPECTRUM
EVALUATION FOR NEUTRONS AND PHOTONS FROM THE
252
CF SOURCE
By the above described procedure we measured and evaluated
neutron and photon spectrum of the 252
Cf source.
The choice of this radiation source producing mixed field is
not random. It is an artificial radionuclide and the neutrons
originate from spontaneous fission of the Cf nuclei. Its neutron
spectrum is widely regarded as standard and its shape can be
used for testing correctness of the experimental method and
accuracy of the evaluation process. It is very close to fission
spectrum. Fig. 3 shows the neutron spectrum and Fig. 4 the
photon spectrum of this source (measured with stilbene 45 x
45 mm).
V. DISCUSSION
The first measurement results of energy spectra show
prospects of the experimental methods used and the possibility
of further improvement. The methodology of discrimination of
impulses belonging to different types of radiation (in our case,
neutrons and photons) requires more precise mathematical
methods for finding the characteristics in which the pulses
of the two types of radiation differ. These can be various, e.g.
Fourier analysis, the use of neural networks, machine learning
etc. We have recently used the Support Vector Machines
(SVM) technique. It is a machine learning method which seeks
optimum classification of the training data into two categories
2 4 6 8 10
Energy in MeV
0.001
0.01
0.1
SpectralFluxDensityina.u.
Fig. 3. Neutron spectrum ϕ(En) from spontaneous fission of 252Cf.
0 1 2 3 4 5 6
Energy in MeV
0.01
0.1
1
10
SpectralFluxDensityina.u.
Fig. 4. Photon spectrum ϕ(En) from spontaneous fission of 252Cf.
separated by a hyperplane. Simply said, this hyperplane maximizes
the minimum distance between those two data sets. As
SVM is a binary classifier, it can be used to distinguish neutron
and proton pulses. Preliminary results, shown in Fig. 5, look
very promising. Fig. 5(a) depicts training data representing
well separated neutron and photon pulses corresponding to
higher energy values. Fig. 5(b) shows the result of the SVM
classification / separation. You can compare these results with
Fig. 5(c) that illustrates the classical separation according
to pulse rise time (e.g. between 5 and 95 percent of the
amplitude).
Our device used for sampling the pulses is currently not
high-end. The main problem is its small dynamic range (8-bit
quantization), which forced us to measure the spectra in more
steps and then combine the resulting parts in a complicated
way. We want to achieve substantial improvements in our
results in the above outlined areas.
REFERENCES
[1] J. Cvachovec, F. Cvachovec, Maximum likelihood estimation of a neutron
spectrum and associated uncertainties, Advances in Military Technology,
2008, vol. 1, no. 2, p.67-79.
[2] F. Cvachovec, J. Cvachovec, P. Tajovsky, Neutron response function for
stilbene detector in energy range 0.5 to 20 MeV, International Workshop
on Neutron Field Spectrometry in Science, Technology and Radiation
Protection (NEUSPEC 2000), June 04-08, 2000, Pisa, Italy.
Fig. 5. (a) Training data, (b) Result of SVM separation, (c) Classical approach
based on pulse rise time