Waveform discrimination analysis applied to PoGOLino data

(1)

Waveform discrimination analysis applied to PoGOLino data

Victor Sedin

sedom@kth.se

SA104X Degree Project in Engineering Physics, First Level Supervisor: Elena Moretti

Department of Physics School of Engineering Sciences KTH, Royal Institute of Technology

Stockholm, Sweden

May 21, 2013

(2)

Abstract

PoGOLite (Polarised Gamma-Ray Observer Lite) is a balloon borne experiment meant to measure the polarisation of the electromagnetic radiation from specic astrophysical objects. The main target of interest for PoGOLite is the Crab pulsar. A considerable particle background was recorded during the short ight of PoGOLite in July 2011. In order to further study the particle background before the relaunch of PoGOLite, a smaller experiment named PoGOLino was initiated. For both experiments, methods to increase the signal to noise ratio in the data collected are important. The aim of this project has been to propose new methods for increasing the signal to noise ratio in collected data, and quantitatively compare them with the previous methods. On a data sample where the original ratio of signal to noise was 0.1, the prior methods were able to increase the ratio to 1.13. A signal to noise ratio of 1.98 was achieved on the same data sample, when executing two algorithms developed during the course of this project in sequence.

(3)

Chapter 1 Introduction

The study of astrophysical objects through measuring electromagnetic radiation has been conducted for a long time, yet the polarisation has not been widely studied. However many interesting astrophysical objects are expected to emit radiation with a high degree of polarisation, as well as time and energy variations in the polarisation. Measurements of the polarisation can be of value to evaluate the current models of these objects. PoGO- Lite (Polarised Gamma-Ray Observer Lite)¹ is a balloon borne instrument for measuring polarisation[1]. The main target to be observed during ight is the Crab pulsar, a well known astrophysical object. At present there are three main models proposed for the mechanism of electromagnetic emission from the Crab[2]. The expected intensity of their emission is very similar. However, the expected polarisation diers between them. Mea- surements of the polarisation from PoGOLite will hopefully lead to a better understanding of pulsars of this kind. PoGOLite has been developed and built by a collaboration headed by a team at KTH in Stockholm, Sweden[3].

A common problem for measurements in space is the large disturbance of the particle background, which consists of neutrons, charged particles and gamma-rays. During the short ight of PoGOLite in 2011, measurements conrmed neutrons to be the dominant source of background[4]. They also conrmed that previous measurements regarding neutrons in the atmosphere at dierent latitudes are not applicable to the conditions at 40km altitude, 67^◦ latitude which is the case when launching from Esrange (European Space and Sounding Rocket Range) in Kiruna².

In order to study these special conditions a small experiment called PoGOLino (Italian for smaller PoGO) was initiated. The objective of the PoGOLino project is to measure the amount of particles as a function of the altitude, in preparation for the ight of PoGOLite.

On board PoGOLino there are detectors for neutrons as well as other particles, in a particular conguration that mimics that of PoGOLite. PoGOLino and PoGOLite both use scintillators to detect incoming particles. In order to distinguish between signal and background, dierent kinds of scintillating materials are used, each with a characteristic decay time. The analysis of the characteristic scintillator output signals called waveforms, makes active rejection of background possible.

PoGOLino was successfully launched in march 2013 and brought back unharmed after data was recorded[5]. it is important that eective tools are available for the current data

1Lite because it is a lighter version of a previously planned instrument. The Gamma-Ray part is because of historical reasons even though the instrument is meant to detect X-rays.

2Near the Swedish northern border.

(5)

analysis, hence the scope of this project has been to create and evaluate methods that can be useful for sorting waveforms depending on their characteristic times.

Focus has been directed towards developing a waveform discrimination tool for data originating from one of detectors on board PoGOLino. Similar detectors are used in PoGOLite, which makes results from the project applicable to the background rejection analysis of PoGOLite data.

The goals and data collection principles of PoGOLino are covered primarily in chapter 2, followed by an explanation of the particle environment at high altitude. Special attention is given to explain the atmospheric conditions in the polar region due to their signicance for the project. In chapter 3 the relevant equipment and the choice of the detector positions are described, as well as how the on board software works. The specic problem of distinguishing between dierent waveforms is described in detail in chapter 4, the chapter will also cover how this problem was dealt with prior to my work. Chapter 5 contains the project results, and the methods introduced in chapter 4 are used to evaluate the eectiveness of the analysis tools developed in the project. The conclusions of the project are presented in chapter 6.

(6)

Chapter 2 Particle environment at high altitudes

The scientic goals of the PoGOLino project are explicated in the rst section of this chapter, including a short exposé on their importance for future balloon borne experiments at high latitudes. In the subsequent section, is a condensed account of the cosmic ray interactions in the atmosphere. Special attention is given to eects stemming from the geomagnetic eld of the earth. The last section gives an account of the neutron environment in the atmosphere.

2.1 PoGOLino goals

Various studies have been conducted measuring the neutron ux at dierent altitudes[6].

However, none of them are applicable to 40km altitude, 67^◦ degree latitude. Hence, the main goal of PoGOLino is to bring back data, to study the amount of neutrons and other particles as a function of the altitude.

Extensive simulations of the neutron background are run in preparation for the launch of PoGOLite, and measurements that can verify the validity of these simulations on the amount of neutrons are very helpful. To limit the incoming neutron ux, PoGOLite has a passive shield of polyethylene¹. Simulations of neutrons traversing materials are known to be dicult, so verication of the GENT 4 simulations[7] concerning the functionality of the polyethylene shield are important. Lastly the ight data PoGOLino collects is valuable to test the data analysis tools on.

2.2 Cosmic ray interactions in the atmosphere

Conditions for high altitude particle detectors are very dierent from conditions for earth bound particle detectors. The main dierence being that high altitude detectors are subject to constant bombardment of cosmic rays. It is commonplace to divide cosmic rays into the two subcategories of primary and secondary cosmic rays. Primary cosmic rays are ones that come directly from the astrophysical source they originate from. The primary cosmic rays consist of about 86% protons, 11% α -particles and about 2% electrons[8]

when they reach the atmosphere. When the primary cosmic rays reach the atmosphere they interact with it through a number of dierent interactions, known as cosmic showers.

An example of a particle shower interaction in the atmosphere can be seen in gure 2.1.

1More about the polyethylene shield in chapter 3.

(7)

Figure 2.1: Incoming primary cosmic rays interact with the atmosphere in a process called a cosmic shower. The particles resulting from the interactions are shown, this is one of many possible variations of the cosmic showers. Reprinted from [9].

Secondary cosmic rays are the result of these cosmic showers. The amount of detected particles increase as the detector approaches the altitude where cosmic showers occur, and decreases once the detector has passed that altitude, as demonstrated in gure 2.2.

Figure 2.2: The count of particles in the atmosphere as a function of pressure (solid curve) and of altitude (dashed curve). The maximum at around 20km is known as the Pfotzer peak. After the Pfotzer peak the amount of particles detected falls o quickly and stabilizes at around half of the peak value.

Reprinted from [10].

(8)

This can be compared to gure 2.3 depicting data collected during the short maiden voyage of PoGOLite in 2011, which was aborted due to a helium leak. The Pfotzer peak can be seen in these data as well, but appears at a dierent altitude.

Figure 2.3: The gure shows the particle count as a function of altitude in the data recorded by PoGOLite during the short ight in 2011. The Pfotzer peak can be seen at about 30 kilometres altitude.

Reprinted from[4].

As mentioned the primary rays consist mostly of charged particles, and as such they are eected by the geomagnetic eld, as illustrated in gure 2.4.

If a charged particle comes in towards earth near the equator where the eld lines of the geomagnetic eld are directed tangentially to the earth surface, they will need to be highly energetic not to be deviated by the magnetic eld. Near the poles where the magnetic eld is directed towards the earth's surface much less energetic primary cosmic rays will enter the atmosphere. The minimum energies required at dierent locations for a particle to be able to pass the geomagnetic eld barrier can be seen in gure 2.5. The geomagnetic eect explains why data collected at other latitudes will not provide useful information about the particle background environment when launching from Esrange, Kiruna.

2.3 Neutron environment at high altitude

PoGOLino will ascend to about 30km above sea level, a large part of the neutrons at this altitude are created by cosmic showers. Although the resulting particles from the particle showers have the highest probability to be emitted forward, they don't necessarily continue on the same direction as the original particle. Neutrons created in these showers can be backscattered upwards, and thus particles originating from a lower altitude will also hit the PoGOLino detectors from below.

The number of scattered neutrons is proportional to the amount of particles partaking in interactions in the atmosphere. Therefore more neutrons will be found at high latitudes such as Kiruna, which has a low geomagnetic cut-o energy, see gure 2.5.

(9)

Figure 2.4: A schematic illustration of the earth's magnetic eld. As can be seen it is a dipole eld and the geomagnetic poles are on an angle relative to the earth's rotational axis. However, the eld lines near the poles are directed into the earth, which leads to more particles being able to enter there.

Reprinted from [11].

Figure 2.5: The lines on the world map show the cut o energies which are due to the geomagnetic eld stopping low energy charged particles. The energies are specied by the numbers next to their respective lines, given in GeV. Observe that cut o energies are lower near the magnetic poles. Reprinted from [12].

(10)

Chapter 3 PoGOLino

3.1 PoGOLino set up

PoGOLino consists of an aluminium vessel closed by two end plates, inside of which electronics and detectors are mounted on four long metal rods. A large polyethylene shield is screwed into place to one of the end plates, see gure 3.1 for an overview of PoGOLino with its case removed.

Figure 3.1: In the picture on the left PoGOLino can be seen with its protective case removed, on the right is a schematic of the equipment. The schematic is reprinted from [13]

As showed in gure 3.1 there are two detectors placed inside the polyethylene shield to register particles that pass through the shield. At the other side of the vessel there is a neutron detector as can be seen in gure 3.1. This detector is not behind a polyethylene shield. By analysing data from detectors with and without polyethylene around them, the eects of the polyethylene shield can be isolated.

(11)

3.2 Detector description

PoGOLino contains three PDCs (Phoswitch¹ Detector Cells), each PDC consists of three scintillators layered in a sandwich conguration. On one side the PDCs taper into a cylinder which allows them to be tightly tted against a PMT (Photomultiplier Tube)². Inside PoGOLino are two dierent types of PDCs. The rst kind is designed for neutron detection. The neutron detector is layered with a 5mm piece of LiCaAlF6 (Lithium Calcium Aluminium Fluoride) between two pieces of BGO (Bismuth Germanium Oxide, Bi₄Ge₃O₁₂). The Lithium consists of 50% 6Li and 50% 7Li. The 6Li has a high neutron capture cross section of 940barn, and is highly sensitive to thermal neutrons[14].

The other type of PDC is sensitive to dierent kinds of particles, including X-rays and neutrons. However, contrary to the rst type, the latter class of PDC does not have high eciency for neutrons. This detector is layered with a 5mm thick EJ-204

plastic[20] between two pieces of BGO. In the context of PoGOLino and PoGOLite the EJ-204 plastic scintillator is called the fast scintillator because of its short decay time of 0.7ns.

As explained in chapter 3.1, one of the neutron detectors is covered by a polyethylene shield. The purpose of this shield is to shift the energy spectrum of incoming neutron towards lower energies. This is necessitated by the fact that high energy neutrons cannot be distinguished from X-rays in the fast scintillator, unlike charged particles that deposit much more energy in the scintillators and and are consequently recognised. Results from simulations of the eects of a 10cm polyethylene shield on an incoming neutron energy spectrum are shown in gure 3.2. The gure illustrates how the shield has made interactions with the neutron scintillator more probable by shifting the energy spectrum.

Figure 3.2: Simulation results of the eects of the polyethylene shield on the energy spectrum of incoming neutrons. The black line represents all the events that are registered. The red line represents the neutrons initial energy, and the blue line represent their energy after traversing the polyethylene.

Reprinted from [4], data taken from from[15].

1Short for Phosphor sandwich.

2Model: Hamamatsu R7899EGKNP.

(12)

3.3 Read out electronics and logic

When an interaction excites the scintillator material in a PDC, light from the ensuing deexcitations are registered by the PMT. The PMTs are connected to a FADC (Fast Analog to Digital Converter) board that handles the trigger logic, which is the decision of which data to save. This logic is coded in VHDL³ (Virtual Hardware Description Language) code. If between two clock cycles, the dierence in ADC (Analog-to-Digital Converter) channels is greater than a set value, a trigger is issued. When the trigger has been issued 10 clock cycles before the trigger and 40 clock cycles after are recorded, hence 50 clock cycles of data are collected which constitutes a waveform. Events with energy exceeding a set threshold are vetoed. This process is executed during data collection, and is therefore called the online analysis as opposed to the consecutive oine addressed in this project. The online analysis is necessitated by an otherwise unsustainable demand on data storage.

3Developed at Hiroshima by Hiromitsu Takahashi with assistance of Takafumi Kawano, Takzumi Hirano and Merlin Kole.

(13)

Chapter 4 Methods of waveform discrimination

This chapter will begin by introducing the terminology thus far undened followed by an account of the importance of dependable methods for waveform discrimination. Section 4.1 introduces the hitherto utilised methods for waveform discrimination. The two suc- ceeding sections introduce tools for the evaluation of waveform discrimination methods.

From now on the EJ-204 plastic scintillator will be referred to as the fast scintillator, and the waveforms resulting from interactions in this material will be referred to as fast waveforms. The waveforms resulting from interactions in the BGO will be referred to as slow waveforms. Algorithms created to remove noise from the data will be referred to as cuts.

Since the online analysis which was introduced in section 3.2 is applied simultaneously as data is being collected it must be fast and therefore cannot be too sophisticated. For this reason data that has passed the rst sorting the online analysis will contain undesired waveforms. Further renement is accomplished by the oine analysis, which works with data already processed by the online analysis. The purpose of the oine analysis is to increase the signal-to-noise ratio of the data, without the time constrains of the online analysis.

Dierences in the scintillator materials lead to dierently shaped waveforms. The fast waveform has a rise time of ∼ 0.1µs, while the rise time of the slow waveform is ∼ 0.3µs, which is illustrated in gure 4.1.

The dierence in rise times is utilised by the oine analysis, which distinguish between signals on the basis of the shape of their waveforms. Figure 4.2 depicts a fast waveform, the kind of waveform that constitutes the signal and which the oine analysis will try to distinguish. If the oine analysis is eective, there will be a higher prevalence of these waveforms as opposed to background than in the original data. For comparison a slow waveform can be seen in gure 4.3.

Another kind of waveforms of relevance because of its high prevalence in PoGOLino data is the superimposed waveform, which occurs when both the BGO and the fast scintillator are excited within one CPU (Central Processing Unit) clock cycle. Equation (4.1) shows that light travels the 5mm between the layers in the PDCs in much less time than one clock cycle of the CPU onboard PoGOLino¹.

T = 1

f = {f = 37.5M Hz} = 1

37.5 ∗ 10⁶ ≈ 27ns (4.1)

1Light travels 5mm through a material with reective index n in ≈ _2,997∗10^n∗5∗10⁻³8 m^m ≈n*0.017ns, which

(14)

Figure 4.1: The fast plastic scintillator has a shorter rise time than the BGO. In this plot the polarity is negative[16]. Reprinted from [17].

Where T is the time of one clock cycle, and f is the frequency of the CPU (37.5Mhz). So if a photon interacts with the BGO and then travels to the fast scintillator and interacts with it, then a fast waveform will be superimposed on a slow one. An example of a superimposed waveform can be seen in gure 4.4.

Figure 4.2: A fast waveform, these are the waveforms the oine analysis will try to single out. This example is fairly clean, as can be seen by the rather smooth shape of the curve.

(15)

Figure 4.3: The gure shows a slow waveform, resulting from an interaction in the BGO. This example is also fairly clean, as can be seen by the rather smooth shape of the curve.

Figure 4.4: A superimposed waveform, here both types of material in a PDC have been excited within the duration of a single clock cycle.

(16)

4.1 Previous work on waveform discrimination

Previous to this project a continuous eort has been made to improve the oine analysis.

The most used method is based on two values, called the fast output and slow output.

These values are calculated as in equations (4.2) and (4.3) [17]:

Fast output = max

1≤i≤46 {v[i + 4] − v[i]} (4.2)

and

Slow output = max

1≤i≤35 {v[i + 15] − v[i]} (4.3)

Where v[i] is the amplitude of the waveform in the i:th sample point, 1 ≤ i ≤ 50. In this section it is important to know that in PoGOLite more data points are saved before the trigger. The principles are the same, but the sample point numbering is altered for PoGOLino.

Once these fast outputs and slow outputs are calculated the events can be plotted in a histogram, see gure 4.5.

Waveforms that look alike will cluster together in the histogram branches. Events can be selected depending on whether or not they satisfy equation (4.4).

k_l1∗ Fast output + m_l1 < Slow output < k_l2∗ Fast + m_l2 (4.4) Where l ∈ N and kl1, k_l1, m_l1, m_l2 are constants dierently set for the upper and lower boundaries of the l:th section of the histogram. The dotted blue and red lines in gure 4.5 show two areas that satisfy equation (4.4) with the constants set the following way [17]: k11 = 1.0, m₁₁ = −40, k₁₂ = 1.0, m₁₂ = 140, k₂₁ = 2.1, m₂₁ = −300, k₂₂ = 2.8, m21 = 100.Notice that when using equations (4.2) and (4.3) both the slow and fast output will equal similar values in the fast waveforms ². Which is why the fast branch can be seen to have a 1-to-1 inclination in gure 4.5.

The method has been rened after the realisation that better separation is achieved when the slow output is commenced from the same data point as the fast output. Meaning that the fast output is rst calculated with equation (4.2) after which the slow output is calculated with equation (4.5)

Slow output = v[j + 15] − v[j] (4.5)

where : j = i if and only if v[j + 4] − v[j] = max

1≤i≤46 {v[i + 4] − v[i]}

where i and j are data points, and v[k] is the amplitude in the k:th data point.

Increased separation of the histogram branches could be seen when the cut was applied on data after these modications. In spite of the work done so far on the waveform analysis, the members of PoGOLite collaboration are of the opinion that it is desirable to further improve the waveform analysis.

2Because both the slow output and the fast output will use the data point corresponding to the highest eligible value of the waveform as their second value, and the baseline as their rst

(17)

Figure 4.5: Events are plotted in a histogram depending on their fast output and slow output. The red dotted lines denes what events survive the cut (These data are from PoGOLite and slow scintillator refers to a type of plastic scintillator not used in PoGOLino). Reprinted from [17].

(18)

4.2 Analysis methods

In order to nd the optimal cut, it is necessary to nd a systematic way of evaluating the eectiveness of cuts. In this section the tools used and the principles followed when evaluating new cuts will be covered.

A useful tool is data from irradiation of PDCs by a known radioactive source in a laboratory environment. The reason being that the signal portion of this data can be known beforehand. So since the purpose of a cut is to remove background, it can be known be known beforehand what would be sorted as signal in a awless cut. The eectiveness of a cut can be evaluated by its approval rate of data and how well this performance approximates the calculated values of radioactive signal and background, respectively.

The way to control this is to plot a spectrum in the form of a histogram, where the number of waveforms with specic peak amplitude is plotted against the ADC-value of the waveform peak.

Numerous such data les were recorded, measuring dierent radioactive isotopes to make sure that the specic energies are not important for the eectiveness of the algorithms.The used radioactive sources and their characteristic emission peaks are given in table 4.1, and how the plastic scintillator interacts with the photons of dierent energies is shown in gure 4.6.

Element Isotope γ emission E (keV)

Cesium Cs-137 ∼660

Cobalt Co-60 ∼ 1173, 1333

Sodium Na-22 ∼ 511, 1022, 1274

Table 4.1: Table of the names and emission energies for the radioactive sources that were used to make data samples for the analysis.

With the information from 4.1, and the material properties of the EJ-204 plastic scintillator, which are shown in gure 4.6 it can be predicted what forms of interactions will occur in the scintillator when it is irradiated by a specic radioactive source.

Histograms check the separation between branches of data after each cut is applied.

A good cut will have a large separation between the branches populated by fast or slow waveforms and preferably have the signal events placed closely together. Furthermore each cut was tested on two test samples, one of signal only and one of background only, which will be introduced in the next section.

4.3 Selection of test samples

Themain quantitative evaluation of the eectiveness of cuts was to apply them to two test samples, consisting of manually sorted waveforms. In order to create these test samples several thousands of waveforms were checked one by one. The clean fast events were saved to one data le, and the background waveforms were saved into another data le.

To reduce the human factor the data was re-sorted repeatedly, and an increasing rate of background was discarded each time. To get waveforms from a wide energy range, and in order to remove systematic errors this process was employed on data samples from dierent radioactive isotopes. The data les were then put together to what was the

(19)

Figure 4.6: Mass attenuation coecient for photon interactions in the EJ-204 plastic scintillator. Base

gure reprinted from [17], with data from [18].

nal, clean and background test samples. The nal sample with signal contained 1000 waveforms, and the nal background sample contained 10,000 waveforms. If an idea for a cut did not perform well on the test samples it was disqualied.

(20)

Chapter 5 Analysis description and application on data

In this chapter, the algorithms developed during the course of this project, and results from when the algorithms are applied to data are presented. In section 5.1 the algorithms are dened and outlined in detail. In section 5.2 the algorithms are applied to dierent sets laboratory data. Lastly, in section 5.3 the algorithms are evaluated by the methods introduced in sections 4.2 and 4.3 to produce the nal results of this project.

5.1 Cut optimisation for fast waveform selection

5.1.1 Polished fast vs slow cut

This project has used equations (4.2) and (4.5) to determine the fast output and slow output. An extra restriction was added, the fast output must correspond to i = 10 ± 2 in equation (4.2). If an event satises this condition, then the event continues to the next step of the analysis, otherwise it is discarded. This restriction proved useful and is therefore implemented in all cuts. Any results achieved in this project will be evaluated in comparison to the results achieved by this version of the fast vs slow cut.

5.1.2 Peak vs area cut

This algorithm works by rst calculating the peak ADC, as dened in equation (5.2), Peak := j where j satisfies v[j] − v[j − 4] = max

1≤i≤46 {v[i + 4] − v[i]} (5.1) Peak ADC := v[j] with j as in equation (5.1) (5.2) Where i, j ∈ N , i, j ≤ 50 and v[k] is the amplitude of the waveform at the k:th clock cycle. Thereafter it calculates the integral from equation (5.3) on page 18 which goes from the peak (as dened in equation (5.1)) until the end of the waveform.

Area = Z t2

t1

v(t)dt (5.3)

(21)

Where v(t) is measured in ADC channels and t is time. The set value t1 is the time of the peak and t2 the time at the end of the waveform. Waveforms are then classied as signal or background depending on how the peak ADC and the magnitude of the area are related. The integral in equation (5.3) is numerically calculated with Root[19], a data analysis framework for C++, developed at CERN (Organisation Européenne pour la Recherche Nucléaire, previously Conseil Européen pour la Recherche Nucléaire).

The concept of the peak vs area cut can be understood by examining gure 5.1.

Where the purple dot indicates the peak, and the red area indicates the area as dened by the area in equation (5.3). For fast waveforms the number of ADC-channels at the peak is large, and the area is small, for slow waveforms it is the opposite.

Figure 5.1: In this gure the peak is marked by a purple circle, and the integral calculated from the peak to the end of the waveform is marked by the red area. The dierence between the waveform amplitude at the peak, as well as the dierence in size of the area between the two waveforms can easily be seen.

Observe that even though the peak ADC value corresponds to the highest amplitude of a fast waveform, it is not necessarily the case for other waveforms. For BGO for example, the value is taken when the waveform has reached about half of its maximum value.

5.1.3 Negative slopes cut

This algorithm utilises the fact that a fast waveform will have a certain number of negative amplitude changes between two consecutive clock cycles. If there are more or less, it is either a dierent kind of waveform or a sign that something out of the ordinary has occurred. More precisely, the algorithm calculates the sum in equation (5.4).

Neg =

49

X

i=0

q_i where (q_i = 1 if (v[i + 1] − v[i]) < a

q_i = 0 if (v[i + 1] − v[i]) ≥ a (5.4) Where a is a chosen number of ADC channels, and v[i] is the amplitude at the i:th data point. After evaluation of dierent cut limits, it was concluded that demanding neg>33 for fast waveforms was optimal. When that is set as the cuto limit, this cut is called the neg>33 cut.

(22)

5.2 Application of the selection algorithms

In this section histograms with events plotted versus the cut variables for the algorithms introduced in the preceding section are shown. The data consists of several laboratory samples of the radioactive isotopes displayed in table 4.1 on page 16 merged together.

By inspecting the histograms one can see that the cuts are not aected by the energies of the photons that induced the waveforms. To see the histograms for each isotope individually, see the appendix. Because waveforms with a similar appearance will have a similar relationship between their cut variables they will cluster together in distinct populations referred to as branches when plotted in the aforementioned histograms.

A histogram with the fast output on the horizontal axis and the slow value on the vertical axis is displayed in gure 5.2. The fast branch containing the waveforms originating from the plastic scintillator have gathered in the population which has been isolated in gure 5.3, by using equation (4.4). There are two other distinct branches. The slow branch with a slow output about two times as larger its fast output containing slow waveforms originating from the BGO. And then there is the branch between the slow and the fast branch, which contains the superimposed waveforms.

Figure 5.2: Data from the dierent radioactive isotopes merged together. The events are plotted depending on their fast output and slow output. This illustrates how the peak vs area cut distinguishes between waveforms. As can be seen the peak vs area cut is not aected by the fact that the data comes from waveforms from a wide energy range.

The histogram for the peak vs area cut, with the peak ADC value on the horizontal axis and the area on the vertical axis is shown in gure 5.4. The branches of data are the same as for the fast vs slow cut, and as with the fast vs slow cut there is no problem with the wide energy range of the data set. But the signal contain fewer events, as can be seen by the number of entries in the data displayed in gure 5.5. In the next section results that show that the reduced number of signal events according to the peak vs are cut compared with the fast vs slow cut is due to increased background rejection and not because of reduced signal recognition.

Lastly, the same three populations can be seen in gure 5.6 on page 23, with the slow waveforms having the fewest number of negative slopes (count peaking at 13 negative

(23)

Figure 5.3: These are the events that are considered to constitute the signal, according to the fast vs slow cut.

slopes), the superimposed waveforms having more (count peaking at 29) and the fast waveforms having the highest number of negative slopes (count peaking at 34). Here the red line marks the cut o, before 34 negative slopes. The waveforms corresponding to the last two bins are considered signal, the rest of the events are considered background.

(24)

Figure 5.4: Data from the dierent radioactive isotopes merged together. The events are plotted depending on their peak ADC value and area, illustrating how the peak vs area cut distinguishes between waveforms. As can be seen the peak vs area cut is not aected by the fact that the data comes from waveforms from a wide energy range.

Figure 5.5: These are the events that are considered to constitute the signal, according to the peak vs area cut, these events are somewhat fewer than the events considered to constitute the signal according to the fast vs slow cut.

(25)

Figure 5.6: This gure illustrates how the neg>33 cut distinguishes between waveforms. The events that are considered to constitute the signal according to this cut are the events plotted to the right of the red bar.

(26)

5.3 Results

5.3.1 Evaluation of spectra

This section compares histograms with event count plotted versus peak ADC values. In the cases when no cut (except for the demand that the trigger is located in clock cycle i = 10 ± 2) has been applied to the data, the histograms can look suspicious with large peaks at low ADC values. However the denition of the peak ADC (see equation (5.2)) correspond to the highest ADC-value for fast waveforms, but not necessarily for other waveforms. Therefore the spectra can not be interpreted before the cuts for background reduction are applied to the dataset.

Sodium

From 22N a the emission peaks are 511 keV, 1022 keV and 1274 keV. The amount of photons emitted corresponding to the rst peak is signicantly more than the other two peaks. As shown in gure 4.6 on page 17, the plastic scintillator will interact with photons with either of these energies through Compton scattering[21]. Because there is a signicantly larger amount of BGO than plastic scintillator, and because BGO has a larger cross section for interaction with photons of these energies[18] a large portion of the data will originate from the BGO and consist of background. Therefore an eective cut will remove most of the background events in the data, and the histogram should show a smooth curve peaking at an ADC-value corresponding to 511 keV. The smooth slope is to be expected because of the angular dependence of the kinetic energy for the Compton scattered electrons.

It is quite easy to qualitatively say from inspecting gure 5.7 that all the cuts increase the signal to noise ratio, however it appears that the neg>33 cut does the best job, while the fast vs slow cut and the peak vs area cut are more comparable, even if the peak vs Area cut seems to be somewhat more eective.

Cesium

Since the emission peak for137Csis at 662 keV the photons will interact with the plastic scintillator through Compton scattering. Therefore the histograms for 137Cs, if a cut is eective, is quite similar to the 22Na, except for where the slope should peak, and since there is no other emission peak with a higher energy there should be no signicant amount of events at higher ADC values than where the peak is located. These expectations can be controlled by inspection of gure 5.8.

Since the two peaks at lower ADC-channels are removed by all the cuts, they all seem to be eective at removing background. But as was seen in the sodium histograms, the neg>33 cut seems to give a histogram with the most even slope, and with the most events removed. Once again the shapes achieved after applying the fast vs slow and peak vs area cuts are very similar.

(27)

Figure 5.7: The four histograms show data after four dierent cuts have been applied. Sodium emits mostly 511 keV photons, but also 10022 keV and 1274 keV photons which are outside of the energy range of the equipment. A perfect cut would show a large, smooth slope with its peak at around 900ADC(cross checked with Cesium data), and then a smaller amount of events up to the ADC limit. The neg>33 cut gives a result which looks much like that, followed by the peak vs area cut and then the fast vs slow cut.

(28)

Figure 5.8: The four histograms represent the data from laboratory irradiation with Cesium, after four dierent cuts have been applied. Because of the single gamma emission energy of 662 keV of the Cesium it easy to predict how signal should appear in the histograms. A smooths slope from Compton scattered events, peaking at around 1100ADC-channels (cross checked with Sodium) should appear. The bump in the slope seen in these histograms, after the cuts have been applied seem to indicate that there is room for some optimisation regarding how the branch distinctions are drawn. However the result indicates the same order of eectiveness among the cuts. With the neg>33 cut providing the best result, followed by the peak vs area cut and then the fast vs slow cut.

(29)

5.3.2 Application on test samples

The selection algorithms were applied one by one to the test samples introduced in section 4.3, and the results can be seen in table 5.1.

Cut Nothing Fast vs Slow Peak vs Area neg>33

Signal 1000 974 932 909

Background 10,000 860 784 532

Ratio 0.1 1.13 1.19 1.71

Table 5.1: The table shows the ratio between signal and background when no cuts are applied and when the studied cuts are applied. In each column the number of signals that passed the cut (true positive) and the number of background waveforms that passed cut (false positive) are listed.

The algorithms were applied then in combination to the test samples, and the results in table 5.2 were achieved.

Cut Nothing Fast vs Slow and neg>33 Peak vs Area and neg>33

Signal 1000 893 849

Background 10,000 457 428

Ratio 0.1 1.95 1.98

Table 5.2: The table shows the results obtained when the cuts were applied in combination to the tests samples. As can be seen, the resulting signal to noise ratio was higher when the cuts were run in combination, particularly the neg>33 combined with the Peak vs Area cut achieved the highest signal to noise ratio.

It is clear from the results that a better signal to noise ratio is achieved when the algorithms are run in combination, and also that the algorithms developed during the course of this project give a higher signal to noise ratio, compared with the algorithm used before. These results are in line with the results shown in section 5.3.1.

(30)

Chapter 6 Conclusions

The PoGOLite experiment will give a deeper understanding of astrophysical objects, primarily the Crab nebula and Crab pulsar. The theoretical models of how the Crab pulsar emits electromagnetic radiation have distinctly dierent predictions for the polarisation.

The imminent PoGOLite polarimetry may come to conclude which theoretical model is correct.

The predicted severe particle background at ight altitude of PoGOLite was conrmed during the short ight of PoGOLite in 2011, which prompted the PoGOLino project whose purpose is to retrieve experimental data to further study the environmental particle background. PoGOLino was successfully launched in March 2013 and the data retrieved at its return is currently being analysed.

Much like PoGOLite, PoGOLino uses scintillator detectors with dierent decay times to actively reject background. If a particle interacts with the detector dedicated to signal detection a fast waveform is produced. On the contrary if a particle interacts with the scintillator dedicated to background rejection a slow rising waveform is produced.

The ability to eectively dierentiate between waveforms is a crucial part of the data analysis of both PoGOLino and PoGOLite, and the scope of this project has been to improve that process. During the course of this project a number of new selection methods have been developed to distinguish between signal and background.

It could be concluded that all of the selection methods increased the signal-to-noise ratio when applied to laboratory irradiation data from radioactive samples. However the new methods proposed seemed to improve the signal to noise ratio more than the previously adopted methods.

Application of all the selection methods to a pure signal sample and a pure background sample, conrmed an improvement respect to the method previously adopted by the PoGOLite collaboration.

The new proposed methods, the peak vs area cut and the neg>33 cut raised the signal-to-noise ratio of the test samples from 0.1 to 1.19 and 1.71 respectively. When run in combination they achieved a signal-to-noise ratio of 1.98, these results were an improvement over the signal-to-noise ratio of 1.13 achieved when applying the fast vs slow cut which is the currently used method.

Excitingly PoGOLino was launched during the fabrication of this thesis, and on that account the proposed continuation is to test the selection methods of this project on the data collected during the ight of PoGOLino. After that an interesting extension would be to test the methods on data from PoGOLite.

(31)

Chapter 7 Appendix

7.1 Histograms of individually plotted radioactive iso- topes

Figure 7.1: Data from laboratory irradiation with Cesium plotted in a fast output vs slow output histogram.

(32)

Figure 7.2: Data from laboratory irradiation with Sodium plotted in a fast output vs slow output histogram.

Figure 7.3: Data from laboratory irradiation with Cobalt plotted in a fast output vs slow output histogram.

(33)

Figure 7.4: Data from laboratory irradiation with Cesium plotted Peak ADC-channels versus area histogram.

Figure 7.5: Data from laboratory irradiation with Sodium plotted in a Peak ADC-channels versus area histogram.

(34)

Figure 7.6: Data from laboratory irradiation with Cobalt plotted in a Peak ADC-channels versus area histogram.

(35)

List of Figures

2.1 One of many kinds of particle shower interactions. . . 5

2.2 Data which shows a clear Pfotzer peak. . . 5

2.3 PoGOLino particle count rate data from the 2011 ight. . . 6

2.4 Schematic illustration of the geomagnetic eld. . . 7

2.5 Geomagnetic cut o energies drawn on map of the earth. . . 7

3.1 Photograph of PoGOLino placed next to a schematic of PoGOLino. . . . 8

3.2 Simulated energy spectrum of neutrons, before and after polyethylene. . . 9

4.1 Rise times of BGO and the fast scintillator in one common plot. . . 12

4.2 A fast waveform. . . 12

4.3 A BGO waveform. . . 13

4.4 A superimposed waveform. . . 13

4.5 Events plotted in a histogram based on their fast output and slow output. 15 4.6 Events plotted in a histogram based on their fast output and slow output. 17 5.1 Illustration that explains the peak vs are cut. . . 19

5.2 Events plotted in a histogram based on their fast output and slow output. 20 5.3 Signal events according to the fast vs slow cut, from the three radioactive isotopes together. . . 21

5.4 Events plotted in a histogram based on their fast output and slow output. 22 5.5 Signal events according to the peak vs area cut, from the three radioactive isotopes together. . . 22

5.6 A histogram for the neg>33 cut, with events from the data le with the three radioactive samples placed together. . . 23

5.7 Four histograms of counts per bin versus peak ADC, displaying data from laboratory irradiation with Sodium. . . 25

5.8 Four histograms of counts per bin versus peak ADC, displaying data from laboratory irradiation with Cesium. . . 26

7.1 Histogram plotted with Cesium data, axis: Fast and Slow. . . 29

7.2 Histogram plotted with Sodium data, axis: Fast and Slow. . . 30

7.3 Histogram plotted with Cobolt data, axis: Fast and Slow. . . 30 7.4 Histogram plotted with Cesium data, axis: Peak ADC-channels and Area. 31 7.5 Histogram plotted with Sodium data, axis: Peak ADC-channels and Area. 31 7.6 Histogram plotted with Cobolt data, axis: Peak ADC-channels and Area. 32

(36)

List of Tables

4.1 Information about sources used for laboratory tests. . . 16 5.1 Results from when cuts were applied one by one to the test samples. . . . 27 5.2 Results from when cuts were applied in combination to the test samples. 27

(37)

Bibliography

[1] M. Pearce, M. Kiss, PoGOLite: Opening a new window on the Universe with polarised gamma-rays, Proceedings of Imaging 2006, Stockholm, June 2006.

[2] J. Dyks, et al, Relativistic Eects and Polarization in Three High-Energy Pulsar Models, Astrophysical Journal 606:1125-1142, May 2004.

[3] The PoGOLite collaboration website, http://www.particle.kth.se/

pogolite/, Accessed, 31 March 2013.

[4] M. Kole, PoGOLite: 2011 ight results and 2012 pre-ight predictions, Licentiate Thesis 2012.

[5] Oscar Clein Center's website, http://okc.albanova.se/blog/

pogolino-successfully-launched/, Accessed 31 march 2013.

[6] Altitude and Latitude Variations in Avionics SEU and Atmospheric Neutron Flux, IEEE TRANSACTIONS ON NUCLEAR SCIENCE VOL. 40, NO. 6 DECEMBER 1993.

[7] The Geant 4 toolkit website, http://geant4.cern.ch/, Accessed 18 May 2013.

[8] D.H Perkins, Particle Astrophysics, 2nd edition, section 6.1, page 148, Particle and Astrophysics department, Oxford University, ISBN: 0198509529.

[9] Sciencenews homepage, special credit to E. Feliciano, http://www.sciencenews.

org/view/access/id/341881/description/PARTICLE_SHOWER/, with original image taken from http://imagine.gsfc.nasa.gov/docs/science/know_l1/

cosmic_rays.html, Accessed, 31 march 2013.

[10] G. Pfotzer, Z. Physik 102, 23, 1936.

[11] T. Hurst, Magnetic eld - The magnetic eld and its direction, Te Ara - the Encyclopedia of New Zealand, updated 14-Nov-12,http://www.TeAra.govt.nz/

en/diagram/9213/earths-magnetic-field Accessed 14 april 2013, base image taken from http://visibleearth.nasa.gov/.

[12] D.F. Smart, M.A. Shea, Fifty years of progress in geomagnetic cuto rigid-ity determinations, D.H Perkins, Advances in Space Research, Volume 44, Issue 10, November 16 2009.

[13] M.Kole, PoGOLino, PoGOLite collaboration internal document, available upon

(38)

[14] H. Takahashi et al, A Thermal-Neutron Detector with a Phoswich System of LiCaAlF6 and BGO Crystal Scintillators on board PoGOLite, As presented at the IEEE conference 2011.

[15] T.W Armstrong et al, Calculations of Neutron Flux Spectra Induced in the Earths Atmosphere by Galactic Cosmic Rays", Journal of Geophysical Research, 1973.

[16] T. Tanaka et al, Data Acquisition System for the PoGOLite Astronomical Hard X-ray Polarimeter, Nuclear Science Symposium Conference Record, (2007) 445.

[17] M. Kiss, Pre-ight development of the PoGOLite Pathnder, PhD thesis, April 2011.

[18] M.J. Berger, et al, National Institute of Standards and Technology XCOM: Pho- ton Cross Sections Database http://www.nist.gov/pml/data/xcom/index.cfm, Accessed May 20 2013.

[19] Root homepage, http://root.cern.ch/drupal/, Accessed 23 April 2013.

[20] EJEN Technology product data sheet, http://www.eljentechnology.com/

images/stories/Data_Sheets/Plastic_Scintillators/EJ204\%20data\

%20sheet.pdf, Accessed 21 April 2013.

[21] Compton. Arthur H, A Quantum Theory of the Scattering of X-Rays by Light Elements, Physical Review 21 May 1923.

Waveform discrimination analysis applied to PoGOLino data