Analysis of the first Data from the SiPM-Camera of the Air Cherenkov Telescope IceAct at the South Pole by Lasse Halve Bachelor Thesis in Physics presented to the Faculty of Mathematics, Computer Science and Natural Sciences of the RWTH Aachen University created at the III. Physikalisches Institut B Prof. Dr. Christopher Wiebusch in April 2016
2
Advisor: Dr. Jan Auffenberg III. Physikalisches Institut B RWTH Aachen First Examiner: Univ.-Prof. Dr. Christopher Wiebusch III. Physikalisches Institut B RWTH Aachen 3
4
Contents 1 Introduction 7 1.1 The IceCube Neutrino Observatory...................... 8 1.2 IceAct at the IceCube detector........................ 11 2 The IceAct prototype at the South Pole 13 2.1 Structure of the IceAct system........................ 13 2.1.1 SiPM-Camera.............................. 14 2.1.2 Custom Power Supply Unit...................... 15 2.1.3 Data Acquisition............................ 15 2.2 Commissioning of the prototype........................ 16 3 Introduction to the data 17 3.1 First analysis and comparison of data inside and on top of the ICL.... 18 3.1.1 Analysis of the baselines in the waveforms.............. 18 3.1.2 Fourier-analysis of the baselines.................... 20 3.2 Features in waveforms recorded on the ICL................. 21 4 Analysis of SiPM-pulses 27 4.1 Search for signals................................ 27 4.2 Extraction of SiPM-signals.......................... 28 5 Examination of analysis results for data recorded on the ICL 31 5.1 Trigger rates.................................. 31 5.2 Baseline RMS distribution........................... 33 5.3 Position of SiPM-signals............................ 35 5.4 Distribution of integrated signals....................... 37 5.5 Integral and min_value distributions at one channel............ 38 6 Conclusion and outlook 41 7 Bibliography 43 8 Appendix 45 8.1 Python scripts for the analysis of SiPM-pulses................ 45 8.2 Additional plots................................. 54 9 Acknowledgments 57 5
6
1 Introduction The detection of cosmic rays has been an important field in astronomy in the last decades and still is today. Cosmic rays are highly energetic particles hitting the Earth and are most probably produced in astrophysical events such as supernova remnants and black holes [1]. However the origin of highest-energy cosmic rays remains unknown. The reconstruction of source locations has two main preconditions. On the one hand detectors need to be able to detect cosmic ray particles directly or indirectly. On the other hand the direction of these particles arriving at Earth needs to be correlated to their sources position. The detection of charged particles in cosmic rays can not provide precise information of the location of the source, because charged particles are deflected by electromagnetic fields on their way to Earth. This effect is illustrated in figure 1. Photons from cosmic ray sources can interact with matter on their path to the Earth and complicate the reconstruction of their source s position. The ideal particle for positional analyses of cosmic ray sources are neutrinos. They are not electrically charged and are therefore not deflected. Additionally neutrinos only interact with matter via Weak Interaction and have very small interaction rates. Thus neutrinos are not attenuated on their way to Earth. Figure 1: Sketch of different comic ray particles arriving at Earth [2]. Heavy nuclei and protons are deflected by electromagnetic fields. The interaction of photons with matter is not illustrated in this figure. Neutrinos propagate without deflection or attenuation. Since neutrinos have low interaction rates, very big volumes are necessary to detect them. This requirement is fulfilled by the IceCube Neutrino Observatory, which is introduced in the next chapter. 7
1.1 The IceCube Neutrino Observatory The IceCube Neutrino Observatory [IceCube] is a high-energy neutrino telescope located at the geographic South Pole. It uses the Antarctic ice as an active volume for detection of Cherenkov light produced by charged particles. The design of IceCube is illustrated in figure 2. Its main goal is the detection of astrophysical neutrinos and potentially the location and nature of their sources [3]. Figure 2: Design of the IceCube detector [4]. Strings are deployed in a depth of 1450m to 2450m. An array of cosmic ray detectors[icetop] is located on the surface. Both detectors are connected to the IceCube Laboratory on the surface. The detector consists of 86 strings lowered into holes, which are melted into the ice. 60 Digital Optical Modules[DOMs], each housing a Photo-Multiplier-Tube, are attached to each string. The DOMs are able to detect Cherenkov light produced by charged particles passing the active volume. This light is used for analyses of the particles passing through the detector. 8
Cherenkov radiation is produced when charged particles travel faster than the speed of light in the medium they are moving in. Note that this does not violate special relativity, claiming particles cannot travel faster than the speed of light in vacuum. The speed of light in ice is 75% of the speed of light in vacuum. The effect is illustrated in figure 3. The particle polarizes the surrounding atoms and molecules, which then emit dipole radiation. This radiation results in a cone shaped front of electromagnetic waves [5]. Figure 3: Two dimensional sketch of the Cherenkov radiation effect [5]. The pink line symbolizes the path of the charged particle, the circles illustrate wavefronts emitted by polarized surrounding atoms and molecules. The cone shaped wavefronts are displayed as black lines. The orange arrows illustrate the propagation of the Cherenkov radiation. The IceCube detector is optimized to detect up-going particles, that passed through Earth before entering the detector. Interactions of neutrinos with atoms of the ice produce charged particles. Because of the high density of the Antarctic ice, mostly muons reach the detector. These muons emit Cherenkov radiation, which is detected by the DOMs. The direction of the muons passing through the detector can be reconstructed and enables analyses of the location of the neutrinos source. The main background for neutrinos searches are atmospheric muons produced in airshowers in the southern hemisphere. These showers are produced by particles e.g. protons entering and interacting with the atmosphere. Some secondary particles, especially muons, are able to reach the detector as illustrated in figure 4. Muons produced in air showers at the northern sky do not reach the detector, since it is shielded by Earth. One way to suppress the background caused by atmospheric air showers is to deploy a veto system on the surface of the South Pole [7][8]. This system needs to be able to detect atmospheric air showers in coincidence with events detected in IceCube. One possible veto system is an array of telescopes detecting Cherenkov light produced by particles of airshowers. The effect is the same utilized in IceCube, but with the atmosphere as active medium instead of Antarctic ice. This veto system would enable IceCube to detect astrophysical much more sensitively. 9
Figure 4: Sketch of possible paths of cosmic ray particles detected in IceCube [6]. Astrophysical Neutrinos can penetrate Earth and then be detected. Most particles produced in atmospheric air showers in the northern hemisphere cannot pass through Earth. Muons produced in air showers at the southern sky are detected in IceCube and represent the main background for neutrino searches. A surface veto is not implemented yet, but will be a part of an extension of IceCube. A first prototype of an Air Cherenkov Telescope[IceAct] was designed and built at the RWTH Aachen University[RWTH] and deployed at the South Pole. The topic of this thesis is analysis of first data recorded by this prototype. 10
1.2 IceAct at the IceCube detector IceAct is a proposed extension of the IceCube experiment for the second generation of the detector, IceCube-Gen2. The concept of IceAct is illustrated in figure 5. IceAct is an array of Air-Cherenkov-Telescopes based on SiPM-cameras designed to withstand the harsh weather conditions at the South Pole [9]. The telescopes are suitable for detection of Cherenkov light produced by airshowers by using the atmosphere as active medium. In coincidence with IceCube the telescope array will be an effective veto system for atmospheric muons [10]. Figure 5: Concept of a surface veto [11]. The orange strings under the surface represent the current detector. The black strings illustrate a possible expansion of Ice- Cube. The yellow arrows symbolize particles passing the detector. The black dots in the surface represent possible locations for IceAct telescopes. In December 2015 a first prototype telescope, designed at the RWTH, was deployed at the South Pole. The assembly of the prototype is described in detail in the next chapter. In this thesis first dark noise data recorded by the prototype is analyzed. The operability of the telescope is discussed, the collected data is characterized and first solutions to revealed problems are proposed. 11
12
2 The IceAct prototype at the South Pole This chapter gives an introduction to the prototype telescope IceAct deployed at the South Pole. Goal of the commissioning of the IceAct prototype is detecting Cherenkov light caused by atmospheric air showers at the geographic South Pole in coincidence with IceTop and IceCube. 2.1 Structure of the IceAct system The assembly of the telescope is illustrated in figure 6. The design is based on the FAMOUS telescope developed at the RWTH [12], but most of the components were altered to fulfill the requirements of detecting Cherenkov light under the harsh weather conditions of the South Pole [9]. Figure 6: Assembly of the IceAct system as deployed at the South Pole [13]. The carbon fibre lens tube houses a Fresnel lens, which is protected by glass. The telescope is mounted on a wooden rack, which houses the electronics. The telescope is based on a Fresnel lens held in place by a carbon-fiber lens tube and protected by a glass plate. The camera consists of seven SiPM-pixels prefaced by Winston- Cones and is located at the back of the tube. The camera is connected to an electronics box including two drs4 evaluation boards [14] and a mini PC recording and storing the data. Everything is constructed to withstand the harsh weather conditions. 13
2.1.1 SiPM-Camera Central system of the telescope is the Silicon Photo-Multiplier[SiPM] based camera shown in figure 7. The camera houses seven SiPM pixels prefaced by Winston Cones focusing the incoming light onto the Silicon PMs. Figure 7: The camera of the IceAct prototype [15]. The SiPMs are located on the board on the right, the detached Winston Cones prefacing the SiPMs are shown on the left. The SiPMs connect to electronics, that are shown in the background. The SiPMs used in the camera are SensL C-Series MicroFC-300xx-SMTs [16]. These SiPMs consist of several diodes run in Geiger mode. Single diodes fully discharge when hit by a photon resulting in a characteristic signal at the output of the channel. Multiple photons can trigger discharge of an integer number of diodes, resulting in an increased signal. Additionally secondary particles exiting a discharging diode can trigger another diode, this effect is called crosstalk. 14
2.1.2 Custom Power Supply Unit As the gain of SiPMs is temperature dependent on a scale of 10 3 V K with applied voltages of 20V-50V [16], their power supply has to be very precise to compensate for changes in temperature. The power supply unit[psu] used in the assembly of the telescope was developed at the RWTH Aachen University[RWTH] for a wide range of applications of SiPM based systems. The PSU supports up to 64 individual channels, with temperature sensors at each pixel and a precision of 10 5 in the applied voltage [17]. The PSU guarantees a constant gain on each channel and allows readouts of the currents, voltages and temperatures at each channel. Since the properties of SiPMs vary individually, the PSU has to be calibrated by setting parameters of the temperature dependence individually for each pixel. The calibration of the camera was conducted in Aachen at T 20. 2.1.3 Data Acquisition The main piece of the Data Acquisition[DAQ] hardware are two drs4-evaluation boards designed by the Paul-Scherrer-Institut [18][14]. These boards store waveforms of the voltages with a length of t waveform 1µs and a time resolution of t res = 1ns. Each board connects to four channels of the camera, the middle pixel is connected to both boards to be able to detect malfunctioning boards. The boards are able to trigger on signals at single channels as well as on a trigger logic at more channels. When triggered, both boards record the following voltages for a time of t delay 400ns and store the whole waveforms on a hard drive. Additionally a timestamp indicating the beginning of the event and the trigger settings are stored. The DAQ does not store participation of individual channels to the trigger. 15
2.2 Commissioning of the prototype The IceAct telescope was deployed at the South Pole in December 2015. The commissioning was conducted by Leif Rädel of the IceCube group of the RWTH (second to the right in figure 8). Figure 8: The IceAct telescope on top of the roof of the IceCube Laboratory. The deployment team, John Kelley, John Felde, Leif Rädel (RWTH), Aongus O Murchadha (from left to right). [19] The aperture of the telescope is covered by a lightproof foil to be able to perform dark noise runs. First runs were conducted inside of the IceCube Laboratory[ICL] before mounting the telescope on the roof of the ICL. Runs on the ICL are performed regularly, the telescope will be opened in late April 2016. 16
3 Introduction to the data This chapter introduces the used data sets, especially the waveforms. Additionally first analyses of the waveforms are conducted and the impact of the position of the IceActtelescope is discussed. A raw waveform is shown in figure 9, where the voltage of one channel is shown in dependence of time. This waveform was recorded inside of the IceCube Laboratory[ICL]. Figure 9: Example waveform recorded inside of the ICL. Time is displayed on the x-scale in ns, voltage is shown in mv on the y-scale. The first 500ns of the waveform is background, SiPM signals lie at 500ns to 700ns. The exact position of signals is discussed later. The last 300ns show the relaxation of the voltage. There are three sections of the waveforms as indicated in figure 9. The section of the first 500ns is background, scattering around a value of U base 0mV with a RMS of σ base 1mV. The signal always lies in a window of t signal = 500 700ns and consists of an abrupt drop in voltage. The position of the signals is discussed in more detail in chapter 5.3. The last part of the waveforms is the relaxation after the signal at t recovery = 700 1000ns. 17
3.1 First analysis and comparison of data inside and on top of the ICL This chapter describes the results of first simple analyses of the waveforms and compares datasets, that were recorded inside of the ICL and on the roof of the ICL. The datasets used were recorded with the same telescope settings, in particular with the same trigger settings. 3.1.1 Analysis of the baselines in the waveforms The median and the RMS of the voltages in the background of 8000 waveforms are calculated. The number of waveforms is chosen out of convenience and does not have a large impact on the information the plots provide. The median and RMS are calculated for each channel separately, figures 10 and 11 show the histograms for median and RMS on the ICL and inside the ICL of one channel. 2000 histogram of medians on board 0 on channel 3 in the ICL on the ICL 1500 # 1000 500 0 14 12 10 8 6 4 2 0 2 4 background median[mv] Figure 10: Histogram of background medians with 8000 entries. The green plot refers to data recorded on top of the ICL, it shows a double-peak structure. The blue plot refers to data gathered inside of the ICL, it shows a single peak. The histogram of data recorded on the ICL shows a clear shift towards smaller background levels. The first apparent feature in the histograms of the medians is a shift towards lower medians in data recorded on the roof of the ICL. This probably is an effect of lower 18
1600 1400 histogram of RMS on board 0 on channel 3 in the ICL on the ICL 1200 1000 # 800 600 400 200 0 0 1 2 3 4 5 background RMS[mV] Figure 11: Histograms of background RMS with 8000 entries. The green plot shows data recorded on the ICL, the blue plot refers to data recorded inside of the ICL. The histogram of data recorded in the ICL shows a shift towards bigger background RMS. The histograms show the same behavior at large background RMS. temperatures on the roof, which influences the gain of the SiPMs by altering the breakdown voltage (see chapter 2.1.1). This is not fully compensated by the custom PSU described in chapter 2.1.2. The shift of the baselines results in a lower effective trigger level, which allows smaller pulses to trigger readout of the waveforms. This leads to lower trigger rates in the data recorded inside of the ICL. The trigger rate inside the ICL is f inicl 22Hz, the rate on the roof of the ICL is f onicl 57Hz. Examining the histogram of data recorded on the ICL a double peak structure is apparent. This structure is found at every single channel of the camera and is not a binning effect. The histograms for the other channels are to be found in the appendix in figures 29 and 30. This feature is caused by SiPM-pulses in the background. The waveforms associated with the peak at lower medians show several SiPM pulses in the baseline, while the waveforms associated with the other peak at larger medians show no such signals. The median histogram for the waveforms recorded in the ICL shows a single peak structure. To understand why there is no visible double peak structure here, we have to take a look at the distributions of the background RMS in figure 11. 19
The histograms for the background RMS both show a single-peak structure and have a steep edge towards lower RMS. This limitation is caused by the random noise of the drs4-chips [18] and noise caused by the drs4-evaluation board [14]. The tail towards higher RMS is caused by drifts, see figure 12, and SiPM-pulses in the baseline. The most apparent difference of the baseline RMS histograms in figure 11 is that the distribution of data on the ICL s smallest values are at RMS start,on 0, 6mV, whilst the histogram of the RMS inside the ICL starts at RMS start,in 0, 87mV. This is caused by an oscillation of the voltages in the data recorded in the ICL, which is discussed in the next chapter. These oscillations cause the RMS to be larger in cases without small SiPM-pulses in the baseline, which explains the shift of the left edge of the background RMS histogram in figure 11. Since the RMS of the voltages in the tail of the histogram towards bigger values is dominated by SiPM-pulses, the oscillations do not have a big effect and the tails of both histograms have the same shape and rates. Now the single peak structure in the median histogram of data gathered in the ICL, see figure 10, can be explained. The increased noise and oscillations of the background in data recorded inside the ICL smears out small SiPM pulses. This reduces the effect the SiPM pulses have on the median calculated in the baseline. Therefore a double peak structure probably is still existent, but much harder to detect. Both histograms of the baseline medians show a broad distribution. This is caused by instabilities and the discussed SiPM pulses in the baselines. Baselines tend to drift as shown in figure 12, which is an effect observed on all channels on both boards. 3.1.2 Fourier-analysis of the baselines The second part of the analysis is done by constructing Fourier transformations of the background in the waveforms. This is done to find possible sources of oscillating noise. A Fast-Fourier-Transformation[FFT] algorithm is applied to voltages in the bins 0-400 of the waveforms representing the background. The average absolute value of the Fourier coefficients is calculated for 1000 waveforms per channel on each data set. Figure 13 shows the averaged FFTs for one channel inside of the ICL and on the roof of the ICL. The averaged FFTs of the datasets inside the ICL show peaks at 32MHz and 64MHz on all channels. One additional small peak is located at 48MHz These peaks in frequency do not appear in the FFTs of the dataset on the roof of the ICL. The origin of these peaks probably is noise of the servers and other electronic devices inside the ICL and therefore are not found in data recorded on the roof. These frequencies correspond to a period of T 32 n ns with n = 1, 2. It can be smoothed out as explained in chapter 4.2. 20
2 Raw Waveform on board 0 at channel 2 of event 7 0 2 voltage/mv 4 6 8 10 12 0 200 400 600 800 1000 time/ns Figure 12: Example of a drifting baseline with time in ns on the x-scale and voltage in mv on the y-scale. This waveform was recorded, because another channel triggered readout. A clear drift towards bigger voltages is visible. There is an effect of increased noise at the start of the waveform. The signature of a SiPM pulse can be seen at the end of the waveform. The first analyses of the waveform are very promising regarding the data quality of samples recorded on the roof of the ICL. Oscillations appearing in data recorded inside of the ICL do not appear in waveforms recorded outside of the ICL and the histograms characterizing the baseline of the signals are well understood. 3.2 Features in waveforms recorded on the ICL There are several features in the waveforms recorded on the roof of the ICL occurring regularly. This chapter describes the most common features of these waveforms. One very common feature of the waveforms are spikes in voltages. These are shown in figure 14 and are an inherent problem of the used drs4 chips. Since they have a very distinct signature of two bins of voltages clearly deviating from the surrounding voltages, they can be cleaned easily, see chapter 8.1 for the code, and therefore are no problem for analyses. Another common feature of the waveforms is an increase in noise on all channels at the 21
Mean 2.0Fourier-Spectrum for 1000 events on board 1 at channel 3 Mean 2.0Fourier-Spectrum for 1000 events on board 1 at channel 3 sum of abs values [au] 1.5 1.0 0.5 sum of abs values [au] 1.5 1.0 0.5 0.0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 MHz (a) Averaged FFT for data set inside the ICL 0.0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 MHz (b) Averaged FFT for data set on top of the ICL Figure 13: Averaged Fourier transformations in the background (bins 0-400) inside and on top of the ICL. The frequency is displayed on the x-scale in MHz, the averaged absolute values of the Fourier coefficients is displayed in arbitrary units on the y-scale. The averaged FFT inside the ICL shows three peaks, the averaged FFT on the ICL does not show peaks. same time a large pulse is detected on one channel. One example of this effect is shown in figure 15. As soon as the signal reaches its extremum all other channels experience a burst in RMS. The voltages of this burst do not exceed the trigger level, which is indicated by the horizontal green line in figure 15 and does not show the signature of a SiPM-pulse. Both these criteria prevent this effect to cause accidental analysis of the bumps as described in chapter 4. Also the increase in noise decays on a timescale of t decay 300ns, which lies within the waveform and therefore has no effect on following signal-analyses. The noise bump and the following oscillation are similar to another feature, which appears in datasets recorded with faulty trigger settings, specifically if the trigger level is chosen to lie within the baseline of the waveforms. This can happen easily, because the baselines reach values up to median base,min 10mV as seen in figure 10. At some time in these waveforms there is a burst in noise followed by an oscillation with a frequency of f 2, 5MHz illustrated in figure 16. This effect is seen in 40% of the waveforms in these runs simultaneously on all channels. The origin of these features is not fully comprehended, but since this is an evitable problem created by a faulty trigger setting, these effects can be avoided by using a trigger level with a bigger absolute value. The datasets used in the following analysis all were recorded with trigger settings avoiding this effect. Some signals show instable baselines prior to the signal as shown in figure 17. The baseline starts to drift towards lower voltages at t start 450ns until the signal arrives. This instable baseline would be a problem for the analysis of the signal, since we cannot 22
Figure 14: Example waveform with spikes. Time is displayed on the x-scale in ns. The voltage at one channel is shown on the y-scale in mv. There are two sets of highlighted spikes with a distinct signature of two bins of deviating voltages. predict what the baseline would look like without the signal. Therefore the RMS of the baseline immediately prior to the signal is computed in the analysis described in chapter 4.2 as an indication of quality. Few signals show two pulses overlapping as illustrated in figure 18. The overlap in these pulses results in a bigger effective rise time of the signal, which gives a large RMS of the baseline pre-pulse. The influence of these signals on the analysis is discussed in chapter 5.2. 23
5 Raw Waveform on board 1 at channel 1 of event 4 5 Raw Waveform on board 1 at channel 3 of event 4 0 5 0 voltage/mv 10 15 20 25 voltage/mv 5 10 30 35 15 40 0 200 400 600 800 1000 time/ns 20 0 200 400 600 800 1000 time/ns (a) Large signal at one channel (b) Correlated noise on another channel Figure 15: Example of an increase in noise correlated with a large signal on another channel. Time is displayed on both x-scales in ns, voltage in mv is displayed on the y-scales. The y-scale is not shared and deviates! The plot on the left shows a large SiPM pulse, which triggered the readout of the waveforms. A increase in noise right before the signal can be seen. The plot on the right shows an increase in noise in roughly the same bin as the signal at another channel. 5 Raw Waveform on board 0 at channel 2 of event 2 0 voltage/mv 5 10 15 20 0 200 400 600 800 1000 time/ns Figure 16: Example of a waveform with too low trigger level. Time is displayed on the x-scale in ns, voltage in the y-scale in mv. The trigger level is displayed as a green line. An increase in noise is located at t start 150ns, after that an oscillation with f 2.5MHz starts. 24
0 Raw Waveform on board 0 at channel 0 of event 1 5 10 voltage/mv 15 20 25 30 35 0 200 400 600 800 1000 time/ns Figure 17: Example of an instable baseline before a signal. Time is displayed on the x-scale in ns, voltage is displayed on the y-scale in mv. The baseline before the SiPM-pulse starts to drift towards smaller voltages t drift = 200ns prior to to signal. 0 Raw Waveform on board 1 at channel 2 of event 4171 5 10 voltage/mv 15 20 25 30 35 0 200 400 600 800 1000 time/ns Figure 18: Example of two overlapping pulses. Time is displayed on the x-scale in ns, voltage is displayed on the y-scale in mv. There are three distinct SiPM-pulses in the signal. 25
26
4 Analysis of SiPM-pulses This part of the thesis describes the used methods for the analysis of SiPM-pulses. The goal of this analysis is to set a foundation for a calibration of the individual channels, which requires high quality data. Firstly the algorithm for searching for SiPM-signals is described and secondly the evaluation of the signals is illustrated. 4.1 Search for signals Since the DAQ of the telescope does not store which channels of a event participated in triggering, a software trigger has to be implemented to find these channels. This software trigger uses the trigger level, that was used by the DRS-board hardware and was stored in the DAQ of the telescope. The functionality of this trigger is illustrated in figure 19. 5 0 5 voltage/mv 10 15 20 25 30 0 200 400 600 800 1000 time/ns Figure 19: Example for the software trigger. Time in ns is displayed on the x-scale, voltage in mv is displayed on the y-scale. The trigger level is represented by the horizontal green line, the software trigger recognizes the signal and stores a trigger_index represented by the vertical black line. Before the software-trigger iterates, the waveform is cleaned of spikes, which have been described in chapter 3.2. 27
The trigger operates in a window, that is by default set to the bins 580 to 650, so that the signal is captured and SiPM pulses in the background do not interfere with the search. The software-trigger iterates the voltages in this window until the waveform crosses the trigger level, which is illustrated as a green horizontal line in figure 19. The trigger condition includes that the voltages in a window of 5 bins before the trigger fall below the trigger level. It also includes that all voltages in a window of 5 bins after the trigger exceed the trigger level. If these requirements are fulfilled, the trigger bin is stored as the trigger_index (black line in figure 19) and a flag indicating the current channel participated in the software trigger is set. The comparison of voltages before and after the trigger_index is implemented to guarantee a stable software trigger, which only triggers once per signal. 4.2 Extraction of SiPM-signals The extraction of the signal is only executed if the channel participated in the softwaretrigger of the respective event. For the analysis of pulses a local copy of the raw waveform is made and the waveform is cleaned as following: 1. Possible spikes are declared as invalid entries. 2. The median of the voltages of the baseline in bins 0-200 is calculated and the median is subtracted from the whole waveform. 3. A sliding average with a standard width of 32 bins is applied to the waveform. 4. The edges of the waveform are declared as invalid entries with a width of 32 ns to either edge. It is important to remove the spikes before the waveform is smoothed out, since they would result in a block of voltages shifted from their natural position after smoothing. The correction of the baseline assures that waveforms are comparable. The smoothing of the waveform is advantageous for extracting parameters for the analysis of the SiPMpulses, since oscillations with frequencies of f 30MHz which occur in raw waveforms, see chapter 3.1, are averaged out. Now a smooth waveform is available for evaluation as displayed in figure 20. Analysis is started by searching for the extremum of the signal in the smoothed waveform to locate the pulse (see figure 20b). This search is constrained to a window that starts at the trigger_index (dashed black line in figure 20b) and has a standard width to the right of 32 bins. This window needs to be at least as wide as the smoothing window 28
5 5 0 0 5 5 voltage/mv 10 15 20 voltage/mv 10 15 20 25 25 30 500 550 600 650 700 750 800 850 900 time/ns (a) Raw waveform used in the analysis 30 500 550 600 650 700 750 800 850 900 time/ns (b) Smooth waveform used in the analysis Figure 20: Example of an analysis of a SiPM-pulse. Time in ns is displayed on the x-scales, voltages in mv are displayed on the y-scales. The raw waveform is plotted in the left figure, the smooth waveform is plotted on the right. The blue line in the raw waveform represents the start_index, the black line illustrates the background window, the green line illustrates the integration window. The dashed black line in the right plot represents the trigger_index, the dashed green line represents the min_index, the blue line represents the start_index. to assure the smoothed extremum is captured. The index of the extremum is stored as min_index (dashed green line). Subsequently another index, which marks the start of the signal, is set and stored in front of the min_index. This start_index is computed as start_index = min_index (risetime+smoothing_window). Here the rise time of the SiPM-pulses has an assumed value of 5 bins, which is typical for the SiPM-pulses in the data. It is important to note that the start_index also marks the start of the signal in the raw waveform, as the smoothing projects the extremum of the raw waveform to a later bin with an offset of the width of the smoothing_window. The computation of start_index compensates for this by subtracting the smoothing_window. An example for the analysis of a SiPM-pulse is shown in figure 20, where the blue lines represent the start_index in both waveforms. After extracting the start_index from the smoothed waveform the analysis deals with the raw waveform. In the raw waveform the baseline median is computed in a window left of the start_index with a standard width of 50 bins (see black line in figure 20a). Additionally a RMS is calculated in the same window, both the median and RMS are stored. Subsequently the median is subtracted off the whole raw waveform, setting the baseline in front of the pulse to 0mV. This ensures that the following analysis does not 29
depend on the position of the baseline before the signal, since it is not causally linked to the signal. The RMS of the baseline before the signal is a quality parameter for the signal. The last step in the analysis of the waveform is computing an integral of the signal in the raw waveform. For this an integration window is used, that starts at start_index and has a standard width of 64 bins to the right. This ensures most of the pulse lies in the integration window. The right edge of the integration window is highlighted as a green line. The integral and the width of the integration window both are stored. The correction of the baseline level before the signal is also done in the smooth waveform. Here the baseline median of the smooth waveforms is subtracted. Additional to the integral in the raw waveform the voltage at min_index is acquired and stored as min_value. This is another way to obtain information of the signal, since the smoothed waveform itself represents an integration. 30
5 Examination of analysis results for data recorded on the ICL This chapter discusses the results of the data selection illustrated before. Firstly we will discuss the trigger rates of each channel and try to understand the distribution of the trigger rates, secondly we will examine the distribution of the RMS of the baselines. Thirdly the min_indices, that mark the position of the signals, are examined. Finally we discuss the distribution of the integrated pulses and compare it to the distribution of min_value, which represents another way of pulse-integration. 5.1 Trigger rates To fully understand what causes the trigger rates of the individual channels one needs to understand the trigger settings in the hardware of the telescope as well as the software trigger as discussed in chapter 4.1. One issue of this analysis is that the DAQ at the South-Pole does not store contribution to the hardware trigger of individual channels. Another problem of the DAQ is that the settings of the hardware trigger are not properly stored, which complicates the recovery of the trigger settings. 10 6 Histogram of coincident signals on ICL 10 5 10 4 10 3 # 10 2 10 1 10 0 10-1 0 1 2 3 4 Number of coincident signals Figure 21: Histogram of coincident signals in recorded events. The number of triggering signals in an event is displayed on the x-scale. The y-scale shows counts on a logarithmic scale. The histogram is dominated by single coincident signals, there are some double-coincident signals. There are some events not showing a signal satisfying the software trigger. 31
Figure 21 shows the histogram of coincident signals as constituted by the software trigger. The histogram is dominated by single signals and shows a maximum of three coincident signals in 33 events. This corresponds to a setting triggering on single channels. The bins of two or more coincident signals are dominated by the shared channel on both boards. The waveforms in the bin of zero coincident signals, representing 3.5% of the total events, show real SiPM-pulses not captured by the software trigger. The starting edges of these waveforms have longer rise times and are not monotone in voltages. As a results the software trigger does not recognize a stable trigger and in rejects the waveform. The relative trigger rates shown in figure 22 are computed by dividing the number of events, where the software trigger saw a signal, by the total number of events for each channel, which is n events 622.000 for this run. Additionally the temperatures in figure 22a and the voltages at the channels in figure 22b are displayed. (a) Relative trigger rates with temperature (b) Relative trigger rates with voltage Figure 22: Relative trigger rates of the channels. The figure on the left shows the relative rates and the mean temperatures at each channel. The figure on the right shows the relative rates and the mean voltage applied to each channel. The trigger rates vary with a maximum factor of nearly 40. The trigger rates of the channels vary, with a maximum factor of about 40. The rates neither seem to correlate with the geometric positioning of the channels nor with the temperatures at each channel. They do however seem to be linked to the voltages applied to the SiPMs. Since a higher voltage should result in an increase in gain for each SiPM, this just shows that the SiPMs at each channel need to be recalibrated for temperatures at the South Pole. The calibration for the pixels was performed at T 20 C. It does not guarantee the same linear dependence of the breakdown voltage of the SiPMs at temperatures at the South Pole. Therefore the parameters of the linear dependency used in the power supply [17] are probably not correct. 32
5.2 Baseline RMS distribution Figure 23 shows the distribution of the RMS of the pre-pulse baselines for each channel. The great differences in trigger rates can also been seen in this histogram. The dark green and purple lines show the histograms for the channels connected to the middle pixel. As expected they have the exact same behavior since they both are connected to the same pixel of the camera. # 10 5 10 4 10 3 10 2 Histograms of baseline RMS 0_0 0_1 0_2 0_3 1_0 1_1 1_2 1_3 10 1 10 0 10-1 0 1 2 3 4 5 6 7 8 baseline RMS before peak[mv] Figure 23: Histogram of baseline RMS for all channels. The RMS of the baseline right before the signal are displayed on the x-scale in mv. The counts are displayed on a logarithmic y-scale. All histograms show a peak at low RMS 1mV and an exponential decline towards bigger RMS. They all have a small peak at RMS 3mV. The shared channels show the exact same behavior, they are displayed as dark green and purple line. The vertical red line represents the cut applied at RMS cut = 1.5mV The histograms of the channels look very similar. They all show a peak at low RMS 1mV and an exponential decline towards higher RMS. The most apparent feature of figure 23 is a peak in all channels at RMS 2.5 3mV. This corresponds to larger rise times caused by crosstalk and double pulses in the signal. Figure 24 shows a waveform of low baseline RMS 1.5mV as well as a waveform with high baseline RMS 4mV. The double pulses caused by crosstalk as seen in figure 24b have a distinct signature with a sharp rise time. Additionally the delay between both pulses does not vary a lot. 33
0 Raw Waveform on board 1 at channel 2 of event 773 0 Raw Waveform on board 1 at channel 2 of event 7682 5 5 10 10 voltage/mv 15 20 25 voltage/mv 15 20 30 25 35 30 40 0 200 400 600 800 1000 time/ns 35 0 200 400 600 800 1000 time/ns (a) Waveform with low baseline RMS 1.5mV (b) Waveform with high baseline RMS 4mV Figure 24: Examples of waveforms with low and high baseline RMS. Time is displayed on the x-scale in ns, voltage in mv is displayed on the y-scales, which do not show the same interval. The waveform on the left has a single SiPM-pulse and a low baseline RMS. The plot on the right has two SiPM pulses in the signal and has a larger baseline RMS. This leads to a fairly stable signature resulting in a distinct feature at RMS 3mV. However this feature only includes a small fraction of the waveforms. To guarantee high quality of the data a cut is applied rejecting all signals with a baseline RMS 1.5mV (red line in figure 23). This excludes waveforms with double SiPM-pulses in the signals as well as signals with instable baselines. The fraction of events that are discarded by this cut is 27%. 34
5.3 Position of SiPM-signals This chapter discusses the position of the signals in the waveforms. The position of the signal is determined by min_index as discussed in chapter 4.2. The histograms of min_index are displayed in figure 25 for two channels on each of the two drs4-boards. The histograms of channel 0,1 and 2 on board 0 are very similar to the one shown in figure 25a. They all have a single-peak structure with a tail towards small min_indices and a sharper edge towards larger min_indices. Channel 3 on board 0 however shows a double-peak structure with a shift of t shift 16ns. This structure is also found on every single channel on board 1 including channel 3, which is connected to the same pixel as channel 3 on board 0. The tail towards earlier min_indices is expected, since the software trigger iterates the waveform from small indices towards larger indices. The evaluation of the signals only uses the first trigger found in the window set by the user, which favors early indices. This results in the tail towards earlier indices and a sharper edge towards later indices. As discussed in chapter 5.1 the hardware triggered on single signals for this run. After applying cuts in the histogram of min_index on channel 3 on board 0 selecting left and right peak individually, there are very few events left in any channel but channel 3 on board 1. This is expected since these channels are connected to the same pixel. The histograms of channel 3 on board 1 show a double-peak structure for each cut. The most probable explanation of the double peak structure is an effect caused by the clock on the drs4 evaluation boards. It samples the trigger with f clock = 60MHz corresponding to a period of T 16.7ns. This matches the shift of the peaks in figure 25. If the of the arriving signal lies close enough to the clock tick, clock shifts and jitter can cause the chip to trigger in the current or following clock cycle. This explains the shift of t shift 16ns in the double peak structures. Since clock drifts and jitter vary from channel to channel, the double peak structure does not need to appear on all channels. This is observed on three channels on board 0. The difference in peak widths can also be explained by varying jitter, since bigger jitters would result in larger widths. The results of applying cuts to the min_index histogram on channel 3 on board 0 reinforce this theory. The structure of the histogram on channel 3 on board 1 does not change, when cuts on either of the peaks are applied. This shows that the position of the signals on one board is independent from the position on the other board. To be able to implement a more complex trigger logic the clock inputs and outputs on the evaluation boards were hijacked. The clock connections were used to synchronize the clocks on both boards. Since the board s clocks are not perfectly synchronized and the jitters on both channels are independent the behavior of independent signal positions is to be expected. 35
# 4500Histogram for min_index on board 0 and channel 1 4000 3500 3000 2500 2000 1500 1000 500 0 600 610 620 630 640 650 660 670 bin (a) Min_index histogram at channel 1 on board 0 # 1200Histogram for min_index on board 0 and channel 3 1000 800 600 400 200 0 600 610 620 630 640 650 660 670 bin (b) Min_index histogram at channel 3 on board 0 6000Histogram for min_index on board 1 and channel 1 5000 1000Histogram for min_index on board 1 and channel 3 800 # 4000 3000 2000 1000 # 600 400 200 0 600 610 620 630 640 650 660 670 bin (c) Min_index histogram at channel 1 on board 1 0 600 610 620 630 640 650 660 670 bin (d) Min_index histogram at channel 3 on board 1 Figure 25: Histograms of min_index for two boards and two channels. The bin of min_index is displayed on the x-scale. Channel 3 on both boards is connected to the same center pixel of the SiPM camera, both histograms are displayed on the right. The histograms on the left show channel 1 on both boards, they do not share a pixel. The histogram of channel 1 on board 0 shows a single peak. The histogram of channel 1 on board 1 shows a double-peak structure relating to a timeshift of t shift 16ns. The channels connected to the middle pixel both show a double-peak structure with the same shift. 36
5.4 Distribution of integrated signals This chapter discusses the distribution of signal-integrals of signals with low baseline RMS. The integral histograms present means to validate the calibration of the SiPMs. Ideally they show a so called finger-structure caused by the binary behavior of the SiPMs in the camera. Figure 26 displays the histograms of integrals for all channels. # 10 5 10 4 10 3 10 2 10 1 integral histogram after RMS Cut board 0 channel 0 board 0 channel 1 board 0 channel 2 board 0 channel 3 board 1 channel 0 board 1 channel 1 board 1 channel 2 board 1 channel 3 10 0 10-1 0 150 300 450 600 750 900 1050 1200 1350 1500 1650 1800 1950 2100 2250 2400 integral[duc] Figure 26: Histogram of integrated signals for all channels. The integral of the signals is displayed in digital units of charge[duc] on the x-scale. The counts are displayed on the logarithmic y-scale. Each channel shows a single peak. The channels connected to the middle pixel are displayed as dark green and purple lines. They show the exact same behavior. The heterogeneous distribution of trigger rates is indicated by the varying peak heights. Also the position of the peaks varies. Again the great difference in event rates can be observed in this histogram. The histogram for channel 0 on board 0 contains a lot more events than the histogram of channel 2 on board 0. The histograms for the channels connected to the middle pixel, represented by the green and purple lines, have the exact same shape and position. Every channel shows the same behavior of a single peak, while the positions of these peaks vary. E.g. the integrated pulses on channel 2 on board 0 tend to be bigger than the integrated signals at the channels connected to the middle pixel. The position of the peaks correspond to the gain of the SiPMs at each channel. Figure 26 reveals that the SiPMs need to be recalibrated, because the positions of the peaks vary. This is not surprising, since the SiPMs were calibrated at temperatures very different from those at the South Pole. 37
5.5 Integral and min_value distributions at one channel In this chapter we will discuss a single histogram of integrated pulses and evaluate whether there is a significant difference in the signatures of the distributions of integrated signals and min_value. The extraction of min_value is described in chapter 4.2, it represents a second way of integrating signals in the waveforms. Figure 27 displays the integral distribution, figure 28 shows the distribution of min_value for the same channel. 10 4 histogram for integrals on board 1 and channel 2 10 3 10 2 # 10 1 10 0 10-1 0 150 300 450 600 750 900 1050 1200 1350 1500 1650 1800 1950 2100 2250 2400 integral[duc] Figure 27: Histogram of integrated SiPM pulses for channel 2 on board 1. The integrated signals are displayed in digital units of charge[duc] on the x-scale. The counts are displayed on a logarithmic y-scale. The red bars in the plot represent the statistical errors. The histogram shows a large peak at 1200duC and a small peak on the flank at 650duC. Ideally one would expect a so called finger-spectrum with characteristic equidistant peaks in either histogram. Each peak in a finger-spectrum corresponds to an integer number of avalanches in the SiPM triggered by photoelectrons or crosstalk, see chapter 2.1.1. Neither the integral nor min_value distribution show the signature of a finger-spectrum, both histograms show a single peak with fairly smooth flanks. The integral histogram has a small feature at 650duC, which could be an indication of a finger peak, the histogram of min_value only shows some steep steps. The hardware trigger level used in this run is trigger_level = 20mV. The smallest SiPM found in the background show a rising edge of 1.5 2mV height. Therefore the signals evaluated in the analysis relate to many discharged diodes in the SiPM. The uncertainty 38
10 4 histo for min_values on board 1 and channel 2 10 3 # 10 2 10 1 10 0 0 4 8 12 16 20 24 28 32 36 40 44 48 52 56 peak height[mv] Figure 28: Histogram of and min_value on channel 2 on board 1. Min_value is displayed as peak height in mv on the x-scale. The counts are displayed on a logarithmic y-scale, the red bars represent the statistical errors. There is one large peak at 23mV, but no distinct other peaks can be found. of the pulse-height associated with each discharge of a diode sums up to larger errors when dealing with many discharged diodes. This results in smeared out peaks in the histograms of integrated pulses and min_index, which makes them hard to detect. In order to record a finger spectrum runs with lower trigger level will be conducted as discussed in the next chapter. In conclusion neither of these histograms shows the signature of a finger-spectrum and the features that could be interpreted as signs of a peak mostly lie within their respective errors. The method of integrated signals is the more promising one, since its distribution shows features similar to a finger-spectrum. The distribution of min_value does not show signs of a finger-spectrum. 39