summer 2013 reu research summary: microboonenevis labs, columbia university august 2013 summer 2013...

Nevis Labs, Columbia University • August 2013

Summer 2013 REU Research Summary:MicroBooNE

M. Phipps

Abstract

This paper is a final write-up of my work at Columbia University/Nevis Labs as part of the NSF-sponseredREU program. The paper begins with a discussion of the physics and scientific goals of MicroBooNE. Itthen moves on to a summary of my contributions to the project. These include commissioning of the TPCreadout equipment, calibration of the PMT readout system, and construction of an interactive monitoringGUI for both TPC and PMT readout. Notable results include the successful resolution of the following:DMA timeout errors, the PMT multiple PE discrepancy and the ringing signal induced by the splitter.There is also an exhaustive analysis and documentation from the TPC crate tests and the Bo PMT readoutstudies, and both of these are available in extended form on MicroBooNE’s DocDB.

MicroBooNE will be the firstappearance-based, liquid argon TPCexperiment of its kind, and a com-

bination of high particle resolution and largestatistics will allow us to address many of theopen oscillation questions in the 100-2000 MeVregime. The objectives for this experimentinclude both scientific and long term techno-logical goals. These include the following:

I. MicroBooNE Objectives

I.1 LArTPC development

From a research and development standpoint,MicroBooNE plays a crucial role for futurelarge scale, next generation LArTPC experi-ments, in particular LBNE. LArTPC technologyoffers heightened resolution and track identifi-cation ability, especially in discriminating be-tween electron and photon events and discern-ing potential proton decay, but its feasibilityand implementation at this scale must still betested.

Prior to MicroBooNE, ArgoNeuT was thelargest LArTPC in the United States, and ithad only 480 readout channels; MicroBooNEwill have 8,256 channels, and this type of leapraises a number of challenges. These include:

• Achieving and sustaining liquid argonpurity within a medium-sized, non-

evacuable cryostat. If trace amounts ofoxygen or water molecules are presentin the liquid argon, the drifting electronswill recombine and never reach the read-out plane, decreasing the resolution ofthe detector. [2]

• Successfully storing long readout wiresat cryogenic temperatures. The readoutwires stored in the cryostat have low vol-ume to surface area and will cool to thecryogenic temperatures almost immedi-ately. Since the frame holding the wireswill cool more slowly, this introducesheavy tension on the wires and a riskof the wires stretching or breaking. [1]

• Accurately predicting the costs of con-struction, installation and maintanenanceof such a project. In March of 2012, theDepartment of Energy halted funding ofLBNE due to cost concerns. Fundingwas reissued later that year, but only af-ter project leaders were forced to makepainful cuts. If MicroBooNE is able tominimize its costs, it may increase the po-litical feasiblity of next generation, longbaseline experiments.

• Reducing the oxygen deficiency hazardin an underground storage facility [2]

1


• Developing a LArTPC software frame-work capable of handling large scaledata analysis. Foremost among these arerather complicated tracking algorithmsthat accurately interpret particular events.[5]

I.2 Resolving the MiniBooNE lowenergy excess

The choice of LArTPC in MicroBooNE wasstrongly motivated by its ability to differenti-ate between photon and electron events and, inthe process, resolve the low energy excess seenin MiniBooNE. The excess appeared in the 200-475 energy range with a 3σ significance andcan be traced to either single electron or singlephoton production through neutrino scatteringon carbon (CH4).

MiniBooNE, which was originally taskedwith investigating the 3 σ ν̄e excess found byLSND, took data from 2002-2012, searching forCherenkov and scintillation light with 1,520photomultiplier tubes housed in 800 tons ofmineral oil. It did observe an excess consistentwith LSND’s findings in the ν̄µ → ν̄e channel,but in the process, it also observed 3 σ low en-ergy excess in the ν̄µ → ν̄e channel. This excessis shown in Figure 1, and its resolution is oneof the primary scientific goals of MicroBooNE.

Figure 1: MiniBooNE low energy excess in ν mode[2]

The shape of this excess is consistent with asource of νe and ν̄e charged-current scatteringor νmu neutral-current scattering of a final statephoton. If the excess is due to electrons, Mi-croBooNE expects to see a spectrum like that

shown in Figure 2, while a photon related ex-cess, would produce a spectrum like that seenin Figure 3.

Figure 2: If the low energy excess is attributableto electrons, this is the expected shape of a Micro-BooNE low energy excess [2]

Figure 3: If the low energy excess is attributable tophotons, this is the expected shape of a MicroBooNElow energy excess [2]

With LArTPC, electrons can be distin-guished from photons by pinpointing the ion-ization process at the beginning of each track.Photons would deliver twice as much chargeas electrons, since their conversions result inpair production (See Figure 4). Dependingupon which explanation MicroBooNE sup-ports, there are numerous theoretical possibil-ities. These range from the mundane (back-ground noise) to the spectacular (beyond theStandard Model explanations, like sterile neu-trinos). For a thorough discussion, see [3]or [4].

2


Figure 4: Electron events are distinguished fromphoton events by the ionization rate at the begin-ning of the track. Photon like events result in pairproduction that produce twice as many events butoccur later in the track. If MicroBooNE discoversan excess, it is expected to be 5 σ for an electronexcess or 4 σ for a photon excess [2]

I.3 Measurement of neutrino crosssections

Measurements of final state neutrino scatteringin liquid argon will become exceedingly impor-tant as next generation LArTPC experimentsattempt to increase the precision and accuracyof various oscillation parameters, and Micro-BooNE will make the first such cross-sectionmeasurements.

Figure 5 shows the cross sections expectedover the 2-3 year Phase I run time. It includesa rich variety of scattering measurements, fromthe primary CCQE mode down to rare NC andCC kaon channels. The high statistics CCQEmeasurements (in the BNB energy range of 100MeV to 3 GeV) will be vital to helping suppresssystematics in future experiments. The highefficiency of the detector at low energies willallow nuclear effects to be constrained and re-duce the CCQE cross sectional uncertainty [5].Measurements will also be made of coherentpion production in charged current and neutralcurrent interactions, and kaon production byneutrinos, providing additional backgroundsfor pion decay experiments and neutrino me-

son production measurements. Furthermore,MicroBooNE will be able to measure the elasticscattering cross section ratio, which will helplead to a determination of ∆ s, the fraction ofnuclear spin carried by strange quarks [2].

Figure 5: Expected cross sections to be found inMicroBooNE [2]

I.4 Cosmic physics: supernovae ob-servation

MicroBooNE is principally designed to studybeam interactions, and as such, it is small andsurface-based and somewhat limited in its abil-ity to investigate cosmic rays. One exceptionlies in supernovae detection. If a galactic su-pernova occurs, MicroBooNE expects to detectmore events than observed by Super K fromsupernova 1987a. And until SNO+ and NOνAare turned on, MicroBooNE will be the onlyneutrino detector in North America capableof observing a significant number of events.Such events are expected to lie around 10 MeV,within the upper range of MicroBooNE’s read-out ability. [2]

II. Commissioning uBooNEreadout equipment

For a full, detailed write-up of the commission-ing of the readout system, see MicroBooNE-

3


doc-2734-v2 on the MicroBooNE DocumentDatabase.

II.1 Background on the readout sys-tem

In order to achieve the physics goals describedabove, MicroBooNE will employ the largestLArTPC readout system built to date. It willbe housed in a 38,000 gallon cryostate, locatedslightly upstream from MiniBooNE’s locationalong the BNB beamline at FNAL. The cryostatwill be filled with pure liquid argon and storedat a 87 K, with less than a 0.1 K temperaturegradient. An illustration of the TPC frame isshown in Figure 6.

Figure 6: TPC frame located inside the cryostat

In this illustration, the cathode plane is lo-cated along the right side of the TPC frame andset at -128 kV. The electric field lines run acrossthe width of the frame to the anode on the farleft side of the frame. When a minimum ion-izing particle (MIP) interacts and ionizes theLAr, free electrons drift toward the anode at avelocity of 1.6 mm

µ s.Before they reach the anode, these electrons

are detected by three consecutive wire readoutplanes. The two inner planes have 2400 wireseach with a 0.3 mm pitch and are oriented at60 ◦ from vertical. These planes are tasked withreading out an induction signal. The outermostplane has 3456 wires and is oriented vertically.

Its signal derives from the collection of thedrifting electrons. The wires themselves have adiameter of 150 microns and an image of thewire planes is shown in Figure 7. Their orienta-tion was chosen in order to allow accurate onedimensional tracking.

Figure 7: TPC wire readout planes already installedat MicroBooNE

The triggers for the TPC system come froma series of 8", bi-alkali PMTs that operate at64 MHz with a 60 ns rise time. Calculationsshow that 40 PMTs at regularly spaced inter-vals should receive about 2 photoelectrons per1 MeV energy loss from an MIP, and since theirsignal will derive primarily from scintillationlight, it will arrive before the drifting electronsin the TPC. This makes them well suited fortiming and trigger purposes.

The PMTs will be arranged just outside thereadout planes, and Figure 8 shows a schematicof this placement.

4


Figure 8: PMTs along the TPC readout frame

In order to read out the 8,256 wires fromthe TPC, a total 130 FEM/ADC modules areneeded. Another two FEM/ADC modules areneeded to read out the 40 PMTs, with the sig-nal from each PMT split between a high andlow gain channel. This is done to ensure thePMT readout system can resolve energy rangestypical for interactions from both the beamlineand supernovas.

The readout processes for the two types ofdetectors are shown in Figures 9 and 10.

Figure 9: An illustration of the TPC readout system

Figure 10: An illustration of the PMT readout sys-tem

The 130 TPC FEM/ADC modules arestored in 9 different crates. The boards ineach crate are connected through a commonbackplane board that is kept at high voltage.The readout from each TPC board is controlledand transmitted through a controller and XMITboard flanking the two ends of each crate. Like-wise, the two PMT FEM/ADC modules arestored in a single crate, along with shaper, con-troller and XMIT boards.

This readout equipment was tested andcommissioned at Nevis, and the results arethe subject of the rest of this section.

II.2 Backplane Tests

Backplane tests were conducted to ensureproper connections between the board andbackplane in each slot of the crate. This alsoserved to test the upper limit of how manyboards the voltage settings and token passingscheme could service.

Before each backplane test, voltage pointswere tested in an empty crate to make surethey matched expectations. This was done toprevent damage to boards from any impropervoltage setting. Voltage and current readingswere made at multiple points during this pro-cess and those readings are available in theextended version of this write-up.

The crate was then fully populated with18 ADC+FEM boards, with a crate controllerin the second slot from the left and the XMITin the farthest slot to the right. Fake data wasloaded onto the FEMs and at least 50,000 eventswere read out and checked against the initialword sequence. Data during this initial testwas read out through the XMIT neutrino path.

All boards booted properly and passed thistest, however a few boards had persistent DMAtimeout errors. These results are summarizedin Table 1 (and DMA timeout errors are dis-cussed below).

5


Table 1: Particular crates and slots that causedDMA timeout errors. Note that since this problemwas slot dependent we cannot be sure that thereare not other boards that may show errors whenloaded into a critical slot. However, final crate con-figurations (as discussed in the following section)were tested and returned no errors. We were alsoable to isolate the cause of DMA timeout errors andprevent them from happening by attaching a surgeprotector to the power source of the clock fanout.This is discussed extensively in section II.4.

The next test used the same board configu-ration, but this time readout was through thesupernova path with Huffman compression.The randomly generated (fake) event data waskept fixed from event to event. As such, onlyfour boards were read out at a time to allow thereadout electronics and DAQ server to keep upwith the data rates. The configurations for thefirst two subruns are shown in Figures 11 and12.

Figure 11: Crate setup for the supernova backplanetest, first subrun of four boards

Figure 12: Crate setup for the supernova backplanetest, second subrun of four boards

All boards passed this test, except FEM33.FEM33 had several data mismatches that gotprogressively worse with time. This was likelyan isolated, temperature dependent issue, sinceFEM33 later passed this test with no furtherproblems. However, it may be worth notingthat FEM33 is used in the final configuration ofcrate 8, slot 15 (L→R). This board performedwell in the final assembly tests discussed below.

II.3 Final Assembly Tests

The final assembly tests involved populatingthe crates in the final configurations to beused in MicroBooNE and performing variouschecks before shipment from Nevis to Fermi-lab. To simulate the conditions of the actualexperiment, a trigger board was temporarily in-stalled to act as an external trigger to a pulser(Systron-Conner 100A) and function genera-tor (Tektronix AFG 3022B), which in turn in-jected charge onto the ASICs. This signalwas then digitized and read out through theADC+FEMs.

In the first set of tests, the crates were testedfor slow readout through the controller; in thesecond set, readout was performed throughthe XMIT. All crate configurations passed thesetests. Boards booted properly, data integritywas maintained, pulses and proper baselineswere observed in all channels and input voltageand output ADC values showed the expectedlinearity prior to saturation.

The results from these tests are shown anddiscussed in the following figures/captions.

6


Figure 13: Representative plot of a time dependentreadout from a single channel. At 0 mV, a pulse isnot visible, as expected, and the channel is main-tained at an appropriate baseline ADC. This resultwas seen across all bipolar channels in all detectorsfor this input voltage.

Figure 14: Representative plot of a time dependentreadout from a single channel at an input voltageof 1200 mV. A pulse is clearly registered for a fi-nite number of samples with ADC max amplitudeproportional to Vin, as expected, and the channel ismaintained at an appropriate baseline ADC acrossthe rest of the samples. This result was seen acrossall channels in all detectors for this input voltage.

Figure 15: Baseline RMS per FEM per channel inunits of ADC, with warm ASICs connected andoutliers labeled. Results are shown before repairswere made to outlier boards and tests were run atroom temperature in a non-shielded environment.After repairs, the faulty channels were fixed.

Figure 16: Representative linearity plot for an Op-tion 1 board. These boards feature 32 inductionplane channels and 32 collection plane channels.Readout for the two planes is set at different base-lines, with the collection plane at an ADC countat 512 and the induction plane at 2048, in orderto take full advantage of the unipolar differentialsignals from the collection plane. This particularplot shows a linearity fit from x = 0 to x = 600mV, with all 64 channels overlayed. As expected, alinear relationship is observed between input volt-age and output ADC, until saturation is reached.All channels across all Option 1 boards showed thisrelationship. Note, ADC max = 4096 but satura-tion in this plot occurs at a lower value because theASICs saturate before the ADC.

7


Figure 17: Representative linearity plot for Option2 boards. In these boards, all 64 channels are re-served for readout from the induction plane, and assuch, the baseline for each channel is set at 2048.This particular plot shows a linearity fit from x =0 to x = 600 mV with all 64 channels overlayed.As expected, a linear relationship is again observedbetween input voltage and output ADC, until satu-ration is reached at ADCmax = 4096. All channelsacross all Option 2 boards showed this relationship.

Figure 18: Distribution of slope (output ADC/inputvoltage) per FEM averaged over all channels, withoutliers labeled. Results are shown before repairswere made to outlier boards. After repairs, the faultychannels were fixed.

Figure 19: Average slope for Option 1 boards as afunction of channel. All 64 channels are expectedto have approximately the same slope. The averageslopes across all channels were observed within 0.2units of each other. However, there is a clear (butslight) deviation between the upper and lower 32channels. There is also a noticeable dropoff in chan-nel numbers divisible by 16. These correspond tochannels receiving data from a single ASIC chip.

Figure 20: Average slope for Option 2 boards as afunction of channel. All 64 channels are expectedto have approximately the same slope. The averageslopes across all channels were observed within 0.2units of each other. However, there is a clear (butslight) deviation between the upper and lower 32channels. There is also a noticeable dropoff in chan-nel numbers divisible by 16. These correspond tochannels receiving data from a single ASIC chip.

II.4 DMA Timeout Errors

DMA (Direct Memory Access) timeout errorsrefer to an exception thrown during the read-out process of a particular event. Under thetoken passing scheme, data is passed in smallbatches to the XMIT from one particular boardat a time. It is then written directly to theDAQ machine without tying up computerCPU. When a DMA timeout error is observedthis process hits a roadblock, possibly due toa batch of data stuck in the XMIT or FEMnever being written to DAQ. This halts the to-ken passing process and the testing algorithmthrows an exception.

These errors initially appeared during thetesting of individual boards. A number of so-lutions were proposed and implemented withvarying degrees of success. The most critical fixinvolved reinstalling firmware on the boards.However, the problem reappeared during thebackplane tests with fully loaded crates.

Eventually the problem was isolated to par-ticular boards in particular slots across mul-tiple crates. If either the slot or board waschanged, the problem could be suppressed.

8


Upon returning to the problematic configura-tions, DMA timeout errors were always ob-served if enough events were processed; itusually took between 5,000-20,000 events toobserve a DMA timeout.

Eventually, A DMA timeout was observedat the same moment that a fan in the samerack as the clock fanout was shut off (see Fig-ure 21 for a diagram of the two racks). Thisgave us the idea of trying to induce a DMAtimeout by turning on and off the fan. Almostevery time the switch was flipped a timeoutwas immediately induced. Upon attaching ascope to the clock we were able to witness avoltage spike at the moment the fan’s powerswitch was flipped. Furthermore, by connect-ing a surge protector to the common powersource we were able to eliminate the problem.Long runs of 200,000 events were taken withno errors and our method of inducing errorsno longer worked. No DMA timeouts havebeen ovserved since this fix.

Further diagnostic tests are being per-formed on the final crate (the only crate tohave problematic boards) in order to fully un-derstand this problem. It will be important tounderstand why these errors only afflict partic-ular boards.

Figure 21: Sketch of the locations of the tested crateand the clock fanout crate that is causing the DMAtimeouts

II.5 Crate Status (August 7, 2013)

As of August 7, 2013, the status of each of thenine crates can be seen in Table 2. In Figure 22,the four crates just shipped are shown sealed

and wrapped for delivery, and in Figure 23they are shown as later installed at DAB.

Table 2: Status of each individual TPC crate as ofAugust 7, 2013.

Figure 22: First four crates sealed before beingshipped

Figure 23: Successful installation of the first TPCcrates to arrive at DAB, Fermilab

9


III. PMT Readout Calibration

Studies

For a full, detailed write-up from these readoutstudies, see MicroBooNE-doc-2954-v1 on theMicroBooNE Document Database.

In order to calibrate the PMT readout sys-tem before MicroBooNE comes online, a 220liter, liquid argon cryostat located at PAB wasused as a prototype for two sets of readoutstudies: one performed in April of 2013 andthe other in August.

The goals of these studies included:

• Verifying the ADC/PE values for thePMT readout system using external trig-gers

• Determining PMT gain for nominalADC/Npe

• Confirming the linearity of pulse ampli-tude and area measurements across thefull ADC scale

• Determining noise levels for the Bo PMTreadout system

These issues were originally discussed inthe May 14 readout analysis by Georgia Kara-giorgi and Leslie Camilleri (MicroBooNE-doc-2543-v2). Since work this summer picked upwhere that work left off, a summary of the pre-vious results, including unresolved issues, isprovided in Table 3.

Table 3: Summary of results of the original PMTreadout study in LAr with an LED light source andEXT=LED trigger (MicroBooNE-doc-2543-v2)

After this initial paper was written, a num-ber of issues were confronted in both hardwareand analysis and a second round of data wastaken in August to explore these issues further.The most important results from these studiesinclude:

• Correction of the ringing artifacts seen inthe April data.

• Resolution of the multiple PE discrep-ancy between scope-based analog dataand the discrete ADC-based data.

• Confirmation of pulse amplitude andarea across full ADC scale in both roundsof data

III.1 Results

III.1.1 April 2013

The statistics from five LED runs are shown inTable 4. The number of events refers to the to-tal number of triggers received, while the HGand LG events are the number of PMT pulses

10


read out within 20 samples of the end of thetrigger readout. The majority of the bad eventswere improperly formatted by the FPGA andmissing pulse header information.

Table 4: Run Statistics

III.1.2 August 2013

Table 5 shows the statistics from each of the 10LED data files from this run. Data was takenover a longer period of time than it was inApril, and this was reflected in the increasednumber of events per file. There was also agreater number of total pulses read out ("To-tal Discriminator Fires"), stemming from thedecrease in the discrimator settings (from 10ADC to 4 ADC in both channels). This ledto an increase in noise that needed to be sup-pressed before analysis. Most of the noise waseliminated by enacting a simple timing cut onthe HG and LG channels, requiring the pulseto fall within 20 readout samples of the trigger.

The problems causing the bad events inApril appear at least partially resolved. Therate of bad events in April was about 1:7; now,that has been reduced to about 1:41. The re-maining problems are split between pulses thatare simply missing the closing words and thosethat are missing channel headers. Detailed ex-amples of these bad events can be found inthe full write-up at MicroBooNE-doc-2954-v1.These bugs in the FPGA coding need to beidentified and corrected for future tests.

Table 5: Statistics from the August LED runs

III.1.3 Splitter Problem

Figures 24 and 25 highlight the ringing prob-lem that plagued the April data. The pulsesshowed oscillatory behavior above and belowbaseline across the readout window follow-ing any relatively large signal from the PMT.Since the solution to this problem was not im-plemented in Bo until August, all April dataanalysis was performed within the bounds ofsystematic error that this ringing caused.

Figure 24: LED1600: Single Event from the Aprilreadout test

11


Figure 25: LED 1600: 100 pulses from the LED1600 run are overlayed in this plot. Pulses of thesame event show up in the same color across chan-nels.

This effect was not anticipated, and deter-mining its source involved simulating complexshaper and splitter circuits. The cause and solu-tion were both discovered by Bill Sippach whowas able to show that the ringing was causedby secondary signals reaching the shaper af-ter the original pulse. This was happeningbecause part of the main pulse was reflectedbackward when it reached the splitter due tothe impedance differential. These secondarypulses would travel back toward the PMT andreturn to the splitter with a slight time offset,causing a wave like pattern in the readout. Inorder to avoid this ringing pattern, either thecapacitance of the splitter had to be increasedor the length of the cable between the PMT andsplitter decreased. Or more succintly, to avoidringing the following condition had to be met:

τcapacitor = RC >> τcable =L

vpulse(1)

In this case, vpulse refers to the velocity ofthe pulse in the cable and is approximately0.66 ft/sec. An increased capacitance can bethought of as decreasing the magnitude of thereflection, while a shortened cable is akin to asmaller temporal gap between reflections. Soneither solution would completely eliminatethe problem, but the ringing approaches zeroas the two parameters are properly adjusted.

This problem was further investigated byDavid Caratelli, Victor Genty and GeorgiaKaragiorgi at Nevis by building a simple RCcircuit and recreating the conditions presentin Bo (MicroBooNE-doc-2621-v1). They foundSippach’s scenario was plausible and beganproposing new capacitor designs for Bo andMicroBooNE (see MicroBooNE-doc-2937-v2 orMicroBooNE-doc-2804-v1). Since there was afast approaching lower boundary to the lengthof the cable (a somewhat fixed distance be-tween the PMT and splitter), the prefered ap-proach was increasing the capacitance of the

splitter. For the August data shown in this re-port, an 18.5 m cable was used with a 30 nFsplitter – a factor of 3 greater capacitance thanthe splitter used in April. As Figure 26 shows,this greatly minimized the ringing problem inthe August data. The new splitter is shown inFigure 29.

Figure 26: LED4100: Single event from the Augustreadout test

Figure 27: New splitter: 0.18 gain reduction; 30 nF

III.1.4 Single PE and Noise Measurements

Single PE measurements were made to confirma PMT HV value of 1080 V (in the April read-out studies) resulted in the nominal ADC/PEof 20 in the HG channel and 2 in the LG chan-nel. This time, all noise and bad events were

12


excluded and more good events were includedthan in the previous analysis. Baseline valuesare also calculated and now set on an event-by-event basis. These improvements shouldimprove both the accuracy and precision ofour results.

In the previous analysis (MicroBooNE-doc-2543-v2) we found an HG ADC/NPE of about18, which fell just short of expectation. Thistime around we found a value of 25.3, whichexceeded expectation. The current results aremore accurate, and the reason for the linger-ing discrepancy is likely the proximity of thesingle PE peak to the discriminator threshold(Figure 28). The discrimator threshold of 10ADC lies along the lower left tail of the singlePE peak and effectively biases our mean ADCamplitude upward. The new analysis with amean slightly exceeding expectation reflectsthis nicely.

Figure 28: Single PE max amplitude from a run atLED 1450, with EXT triggering on visible LED.Discriminator settings for this run are 10 ADC or(0.5 PE), which is conservatively high and effec-tively reduces the lower tail of our peak, biasing themean of the distribution upward.

Noise measurements for the 10 ADC dis-criminator setting using the April data and thenew readout procedures are shown in Figure29. The data for this plot was taken using LEDtriggering but with the actual LED inputs tothe cryostat disconnected, preventing the PMTsfrom detecting LED light. Noise levels for theLG signal were below the readout threshold–and that plot was excluded – but noise from theHG signal was almost perfectly consistent with1 PE, showing an ADC/PE of 20.12. This re-sult is encouraging and acts as a counter checkto the single PE ADC measurement shown inFigure 28.

Figure 29: Noise measurement made with HG ADCdata: 20.12 ADC/PE ... Excellent agreement withnominal value.

III.1.5 Multiple PE Measurements

In the previous readout analysis (MicroBooNE-doc-2543-v2), there were two concerns with theconsistency of the multiple PE measurementsfrom the April data.

• There appeared to be a discrepancy be-tween the PE measurements made in theHG and those made in the LG channelsfor a given LED voltage. Under nominalinput voltage, we expect the HG channelto show 20 ADC/PE and the LG channelto show 2 ADC/PE. However, the chan-nels derive from the same signal, andas such, the number of calculated photo-electrons should be identical for the twochannels.

• There also appeared to be a discrepancybetween the number of photoelectronscalculated from ADC data (Figures 33and 34) and the number calculated fromthe oscilloscope (Figures 30 and 31).

These issues were investigated further byrewriting the readout script to exclude badevents and noise from the analysis, to set base-line values on an event-by-event basis, to recre-ate all plots, and to recalculate all measure-ments for the analog and discrete data. Theresulting analog scope plots are shown in Fig-ures 30 and 31. The fits and NPE calculationsfrom these plots changed only slightly, and theraw PMT NPE measurements again showedexcellent agreement with the shaper derivedNPE. This relationship is plotted in Figure 32.

13


Figure 30: Scope Data: Raw PMT derived NPE

Figure 31: Scope Data: Shaper derived NPE

Figure 32: Scope Linearity Test: This figure showsa linear, almost 1-to-1 relationship between NPEcalculated from the scope: raw PMT and shaperhigh gain data. This implies that the two can beused interchangably for calibrating ADC processeddata.

The un-saturated, discrete ADC plots thatresulted from our new analysis are shown in

Figures 33 and 34. As the LED voltage in-creased and the NPE rose, the distributionshould resemble a Gaussian, however the tailsin the two channels were often repressed due tounder- and over-saturation, and the HG chan-nel showed greater noise due to lower statisticsat an increased distribution width. The linear-ity of the mean of the max ADC distributionand the NPE at each LED voltage is shown inFigure 35.

Figure 33: LED 1550: LG 1.28 ADC/PE; HG 12.3ADC/PE

Figure 34: LED 1600: LG 1.30 ADC/PE; HG 12.8ADC/PE

14


Figure 35: NPE Linearity Test: High gain scopeand ADC data are plotted in green and low gainscope and ADC data in orange. The dots correspondto scope data while the stars correspond to ADCdata.

The linear fit performed in Figure 35 is al-most perfect for the scope data, but the ADCdata still showed deviation. Table 6 presentsa comparison of the linear fits for the data inFigure 35.

Table 6: Comparison of linear fits for scope andADC-derived NPE for old and new data.

We expect the high gain data (from thescope and the ADC) to fit a straight line with aslope of 20 ADC/PE, while the low gain datashould produce a slope of 2 ADC/PE. Thiscomes from our nominal setting of 20 ADC/PEand the 10:1 ratio expected for the high andlow gain signals.

The new analysis showed a drop in slopefor the high gain shaper fit, leaving us a lit-tle less in line with expectation. On the otherhand, the slope for every other fit got closer toexpectation or remained the same upon reanal-ysis of data.

The R2 value for these linear fits showedmixed results in the ADC data, while R2 forthe scope data remained around 1.0. In general,

the new analysis suggests greater amounts offluctuation in the ADC readout than the scopemeasurement, and as such, we should expectless precision in the ADC multiple PE measure-ments.

We also expect an overlap of the scope andADC fits in each channel. This deviation wasmentioned above as a motivating factor for thisreanalysis, and though we were not able to getrid of the deviation, we did minimize it acrossboth channels. Table 7 shows the differencein slopes between new and old data for eachchannel.

Table 7: Deviation of the slope in ADC/PE betweenold and new data.

While the new analysis still suggests dis-crepancies, notice that in the two ADC LEDfiles shown, neither the low gain nor the highgain ADC signal was under or over saturated.In those two files (LED1550 and LED1600), thenumber of low gain and high gain PE werealmost identical – as they should be – and theanalog scope PE calculations for the two wereexactly 63% lower. For the next file, LED1650(not shown), the high gain signal was saturated,but for the low gain signal, the scope data Npewas 70% lower than the ADC calculation – verysimilar to the discrepancy in the previous twofiles. On the other hand, the ADC/PE for thoseeach channel of those data sets showed almostidentical deviation from their nominal values(HG: 20 ADC/PE and LG: 2 ADC/PE). Thissuggests a systematic explanation for the dis-crete and analog discrepancy. It is not clearwhat that explanation could be, but it appearsto bias the ADC PE calculations up by a factorof 1.57. These results are summarized in Table8.

15


Table 8: Deviation of the slope in ADC/PE betweenold and new data.

One potential explanation is due to the con-servatively high discriminator settings for theseruns. At a threshold of 10 ADC, the low gain 1PE events are repressed completely (since weexpect 2 ADC/PE), while the lower tail of thehigh gain events are repressed. This wouldbias our HG ADC/NPE up from its expectedvalue of 20 to somewhere in the mid 20s – likewe found in this new analysis.

The scope was not used for PE calculationsduring the August run and the discriminatorthresholds were lowered, so this conclusioncannot be tested with new data. However, aswill be mentioned below, the multiple PE ADCmeasurements in the August data are consis-tent across HG and LG channels and suggestthat this discrepancy may have already beenfixed or, at worst, be innocuous.

The following plots show the ADC derivedNPE calculations for the August data.

Figure 36: LED 3700: LG: 1.97 ADC/NPE; HG:17.9 ADC/NPE

Figure 37: LED 3800: LG: 2.01 ADC/NPE; HG:18.6 ADC/NPE

For the plots that suffered no form of satura-tion, the ADC/PE were in relatively consistentagreement with nominal expectations. FromLED values of 3700- 4100, the low gain mea-surements were very close to 2.0 ADC/PE andthe HG measurements just short of 20 ADC/PE.This suggests that HV input voltage of 1120V used for the August studies leaves us veryclose to nominal operation. It also suggeststhat, whatever the problem with the multiplePE measurements from April, it is now fixed.

A plot of the linearity of the max ADC toNPE values is shown in Figure 38. As men-tioned above, we expect the slope shown inthis plot to increase by 20 ADC/PE for thehigh gain channel and 2 ADC/PE for the lowgain.

Figure 38: NPE Linearity Test: Linearity was wellmaintained until the last three LED input voltageswhen the LED saturated.

These results again show greater consis-tency with expectations than we saw in April.

16


The low gain slope is now very close to 2ADC/PE, while the high gain slope is morein line with 20 ADC/PE than it was in April.A comparison between April and August mul-tiple NPE measurements is shown in Table 7.

It should be noted that the discriminatorsettings were lowered from 10 ADC for bothchannels in April to 4 ADC for both channels inAugust. This change surely biased our singlePE measurements from April upward, sincethe lower tail was missing due to the conserva-tively high discriminator threshold.

Table 9: Linearity Comparison: April Data andAugust Data

In general, the August ADC data showedbetter linearity than the April data, as evidentin the R2 values of the respective linear fits.And even though we do not have scope datato compare against, the goodness of fit andthe proximity to expectation of the new datashould resolve most questions about the consis-tency of the multiple PE ADC measurements.

III.1.6 Linearity of Pulse Amplitude and In-tegral at Varying NPE

In making our NPE calculations, we assumedthat the distributions of pulse height and pulsearea were equally valid methods of coming upwith mean and RMS measurements. In orderto test this, we plotted these values for eachpulse at different LED voltages versus one an-other. This plot using April data is shown inFigure 39.

As expected, the relationship is linear untilsaturation is reached at about 2048 ADC abovebaseline. This follows from the fact that LEDgenerated PMT pulses are typically narrowerthan the signal shaper time.

Figure 39: April data: Linearity test between ADCmax and the pulse integral (defined as primary pulseabove baseline) for each event in each channel of alldata files are shown in this figure.

The linearity of pulse area and height wasagain tested. Like we found in April, the twoquantities were indeed linear across each chan-nel at each LED voltage.

Figure 40: August data: Linearity test betweenADC max and pulse integral. These results mirrorthose seen in April.

IV. GUI uBooNE readout status

When MicroBooNE comes online in Spring2014, it will be important, at times, to querythe status of individual TPC and PMT readoutboards. This could be necessary when trou-bleshooting particular readout problems or aspart of a rigorous online monitoring system.In order to do this, a readout routine mustbe implemented that interacts with individualboards through the PCIe cards in the DAQmachine. The signal received then needs tobe displayed in a manner accessible to bothshifters and experts.

17


The solution developed this summer was aROOT-based GUI that initially shows the gen-eral status of the boards in each crate througha simple red-green color coded system (Fig-ure 42). If more details, are desired, icons canbe clicked on this initial window to displaya second, more comprehensive window withthe status for all boards within that particularcrate (Figure 43). This system has the benefitof being simple and intuitive for shifters, butdetailed enough to satisfy experts.

The final code for this GUI is object-oriented and scalable for more general LArSoftimplementation but is likely one step short ofits final version. In order to fit seamlessly withthe MicroBooNE monitoring tools, the ROOTdependancy may need to be removed.

Figure 41: Initial welcome screen that appears whenthe application is loaded. The MicroBooNE logoshown on this GUI was created by David Kaleko

Figure 42: The same welcome screen after the "Re-fresh" icon is clicked. In this case, each crate con-tained at least one board with errors.

Figure 43: After clicking the "Show Boards" iconfor a particular crate in the previous window, thiswindow appears that shows in-depth status infor-mation for all the boards in the crate.

V. Acknowledgments

I would like to thank everyone at Nevis forthe help and support this summer. Grace wasinstrumental in so many ways, and Amy didan amazing job arranging housing, funding,transportation and much more. With all thecrate tests this summer, Nancy went out of herway to change the HV whenever a new cratewas installed, and our success with these testswould not have been possible without her. Billwas always available and willing to help withany ROOT, C++, compiler, or network issues– something for which I and all the other REUstudents were grateful.

I also owe an immense debt to everyone onthe MicroBooNE team. David was the first per-son who I worked with extensively this sum-mer, and he did an amazing job of getting meup to speed on the readout electronics and ofdoing so much preliminary work on the cratetests. Leslie was there all summer and had adirect role in all my work; without his help,the results written about in this paper wouldnot have been possible. Kazu is one of the bestprogrammers I have ever had the chance towork with, and it would take another 20 pagepaper to list everything I learned from him.Georgia was my mentor this summer, and mydebt to her is immense. She believed in me,was always there when I needed help and the

18


results discussed in this paper would not havebeen possible without her. Finally, I want tothank Mike for giving me this opportunity. Heoversaw all the results and his input was vital,but without his initial support I would havenever been at Nevis in the first place. So I owethese results to him and the rest of the team,and, moving forward, hopefully they’re of helpto the MicroBooNE Collaboration.

Last but not least, I want to thank the NSFfor funding and everyone who oversaw thisprogram and made it possible. This includesMike and John Parsons. The way Dr. Parsonsand Mike (and everyone else involved) orga-nized this program was flawless. They gave usan opportunity and then put us in a positionto succeed, and that is all you can ever ask.

References

[1] LBNE Collaboration. Lbne reconfiguration: Steering committee report. 2012.

[2] MicroBooNE Collaboration. The microboone technical design report. 2012.

[3] J.M. Conrad. Sterile neutrino fits to short baseline neutrino oscillation measurements.arXiv:1207.4765v1, 2012.

[4] Lewis Conrad and Shaevitz. The lsnd and miniboone oscillation searches at high δ m.arXiv:1207.4765v1, 2012.

[5] Georgia Karagiorgi. Microboone: Searching for new physics in the neutrino sector with a100-ton-scale liquid argon tpc. 2013.

19

summer 2013 reu research summary: microboonenevis labs, columbia university august 2013 summer 2013...

Documents