Calibration of the Surface Array of the Pierre Auger Observatory
X. Bertou, P.S. Allison, C. Bonifazi, P. Bauleo, C.M. Grunfeld, M. Aglietta, F. Arneodo, D. Barnhill, J.J. Beatty, N.G. Busca, A. Creusot, D. Dornic, A. Etchegoyen. A. Filevitch, P.L. Ghia, I. Lhenry-Yvon, M.C. Medina, E. Moreno, D. Nitz, T. Ohnuki, S. Ranchon, H. Salazar, T. Suomijärvi, D. Supanitsky, A. Tripathi, M. Urban, L. Villasenor
CCalibration of the surface array of the Pierre Auger Observatory
X. Bertou, a P. S. Allison, b ∗ C. Bonifazi, c P. Bauleo, d C.M. Grunfeld, e M. Aglietta, f F. Arneodo, g D. Barnhill, h J.J. Beatty, b N.G. Busca, i,j,k
A. Creusot, l D. Dornic, l A. Etchegoyen, m A. Filevitch, m P.L. Ghia, f,g
I. Lhenry-Yvon, l M.C. Medina, m E. Moreno, n D. Nitz, o T. Ohnuki, h S. Ranchon, p H. Salazar, n T. Suomijärvi, l D. Supanitsky, m A. Tripathi, h M. Urban, p and L. Villasenor, q for the Pierre Auger Collaboration † a Centro Atómico Bariloche (CNEA), S.C. de Bariloche, Argentina b Department of Physics, Ohio State University, 191 W. Woodruff Ave., Columbus, OH 43201, USA c CBPF/IN2P3-CNRS, Rua Xavier Sigaud, 150 Rio de Janeiro, Brazil d Department of Physics, Colorado State University, Fort Collins, CO 80523, USA e Universidad Nacional de la Plata, Facultad de Ciencias Exactas, Departamento de Física and IFLP/CONICET, C.C. 67, (1900) La Plata, Argentina f Istituto di Fisica dello Spazio Interplanetario, INAF, and INFN, Torino, Italy g INFN, Laboratori Nazionali del Gran Sasso, Assergi, Italy h University of California, Los Angeles (UCLA), USA i Department of Astronomy & Astrophysics, The University of Chicago, Chicago, IL 60637-1433, USA j Kavli Institute of Cosmological Physics, The University of Chicago, Chicago, IL 60637-1433, USA k Fermi National Accelerator Laboratory, Particle Astrophysics Center, Batavia, IL 60510-0500, USA l Institut de Physique Nucléaire d’Orsay, Université Paris-Sud et IN2P3-CNRS, 91406 Orsay Cedex, France m Laboratorio Tandar, Comisión Nacional de Energía Atómica and CONICET, Av. Gral. Paz 1499 (1650) San Martín, Buenos Aires, Argentina n Benemérita Universidad Autónoma de Puebla (BUAP), Ap. Postal J-48, 72500 Puebla, Puebla, Mexico o Physics Department, Michigan Technological University, Houghton, MI 49931, USA p Laboratoire de l’Accélerateur Linéaire, IN2P3-CNRS et Universite Paris-Sud, Centre Scientifique d’Orsay, Bat 200, B.P. 34, 91898 Orsay Cedex, France q University of Michoacan, Morelia, Michoacan, Mexico
Abstract
The Pierre Auger Observatory is designed to study cosmic rays of the highest energies ( > eV). The ground array ofthe Observatory will consist of 1600 water-Cherenkov detectors deployed over 3000 km . The remoteness and large number ofdetectors require a robust, automatic self-calibration procedure. It relies on the measurement of the average charge collected by aphotomultiplier tube from the Cherenkov light produced by a vertical and central through-going muon, determined to 5 −
10% atthe detector via a novel rate-based technique and to 3% precision through analysis of histograms of the charge distribution. Theparameters needed for the calibration are measured every minute, allowing for an accurate determination of the signals recordedfrom extensive air showers produced by primary cosmic rays. The method also enables stable and uniform triggering conditions tobe achieved.
PACS : 96.50.S − ; 96.50.sd; 29.40.Ka Keywords : Ultra high-energy cosmic rays; Water-Cherenkov detectors; Calibration; Atmospheric muons
Published in Nucl. Instrum. Meth. A as DOI:10.1016/j.nima.2006.07.066
At energies above 10 eV, the cosmic ray flux is very low( ∼ − sr − yr − ) requiring a very large area, sparse, sim-ple, and reliable design, and the ability to identify rare showercandidates from a large background. The Pierre Auger Obser-vatory is a hybrid device optimized for energies above 10 eVconsisting of an air fluorescence detector as well as a surface de-tector (SD) using water-Cherenkov tanks. The water-Cherenkovdetector method has been used prominently in previous cosmicray air shower experiments (e.g. Ref. [1]). The SD obtains ameasurement of the Cherenkov light produced by shower par-ticles passing through the detector at ground, and reconstructsthe air shower by fitting the observed signal as a function oflateral distance from the shower core. The Cherenkov light is * Corresponding author, tel. + [email protected] † Pierre Auger Observatory, Av. San Martín Norte 304, (5613) Malargüe,Argentina. measured in units of the signal produced by a vertical and cen-tral through-going (VCT) muon, termed a vertical-equivalentmuon (VEM).The SD consists of 1600 water tanks, with 10 m watersurface area and 1.2 m water height and three 9 in Photo-nis XP1805PA/1 photomultiplier tubes (PMTs) looking into aTyvek ® reflective liner through optical coupling material [2].The signal from the 3 PMTs is digitized by local electronics,and the data are sent to a central data acquisition system (CDAS)when requested.The total bandwidth available from each SD to the CDAS isapproximately 1200 bits per second [3] which implies that thecalibration must be done by the local electronics. The localprocessor is an 80 MHz PowerPC 403GCX lacking floating-point hardware, which forces the calibration to be as simpleas possible. Finally, the remoteness of the detectors impliesthat the calibration procedure must be robust, accepting thepossibility of failures of individual PMTs, to allow for recovery1 a r X i v : . [ a s t r o - ph . H E ] F e b f these stations in data analysis.The SD electronics uses six 40 MHz AD9203 10-bit flashanalog-to-digital converters (FADCs) to digitize the signalsfrom the 3 PMTs. Two signals are taken from each PMT –one directly from the anode, and the other from the last dynode,amplified and inverted by the base electronics to a total of nom-inally 32 times the anode. The two signals are used to provideenough dynamic range to cover with good precision both theparticle flux near the shower core ( ∼ µ s) andfar from the shower core ( ∼ µ s). The two signalsare called simply the anode and dynode, respectively. The sig-nal recorded by the FADC is referred to in units of channels(ch), with a range of 0 − − The primary signal calibration information required from theSD is the average charge measured for a VCT muon, namedthe vertical-equivalent muon (VEM, or 𝑄 VEM when needed forclarity). During shower reconstruction, the signal recorded bythe tanks is converted into units of VEM, and the shower char-acteristics, i.e. total energy and arrival direction, are fit using alateral distribution function and energy conversion based uponhybrid analysis using the florescence detector. The conversionto units of VEM is done either to provide a common referencelevel between tanks or to calibrate against the detector simula-tions for other Monte Carlo-based studies. Therefore, the goalof the calibration is to obtain the value of 1 VEM in electronicsunits (i.e. integrated channels).In addition, to maintain a uniform trigger condition for thearray, the station must also be able to set a common triggerthreshold in detector-independent units. This will allow fora tank-independent analysis of the acceptance of the array bymodeling the trigger [4].There are several quantities which are strongly related toa VEM, but are determined with different methods. Thesequantities are listed in Table 1 for easy reference.
Table 1Reference for calibration termsSymbol Definition SectionVEM or 𝑄 VEM
Charge deposited in PMT by light 2from VCT muon 𝑄 peakVEM Peak in a charge histogram 2.1 𝑄 estVEM On-line estimated value of 𝑄 peakVEM 𝐼 VEM
Pulse height of light from VCT muon 2.2 𝐼 peakVEM Peak in a pulse height histogram 2.2 𝐼 estVEM On-line estimated value of 𝐼 peakVEM Atmospheric muons passing through the detector at a rate ofapproximately 2500 Hz give an excellent method for measuring1 VEM precisely, but the SD in its normal configuration hasno way to select only VCT muons. However, the distributionof the light from atmospheric muons also produces a peak in acharge distribution [5]. This peak ( 𝑄 peakVEM ) is at approximately1.09 VEM for the sum of the 3 PMTs and ( . ± . ) VEM foreach PMT, both measured with a muon telescope providing the trigger in a reference tank [6]. The difference between thesetwo cases is due to the fact that the sum of the PMTs measuresthe total signal, whereas the individual PMTs primarily measurethe portion of the signal deposited closest to them. Examplecharge and pulse height histograms produced by a SD are shownin Fig. 1. The peak produced by the response of atmosphericmuons in the tank is clearly visible. The relation of 𝑄 peakVEM to 𝑄 VEM can be understood using a simple geometrical model [7].The shift observed is caused by the convolution of photoelectronstatistics on an asymmetric peak in the track length distributionand local light collection effects.
The SD uses two triggers to identify shower candidates [8]. Thefirst is a simple threshold trigger, which is satisfied when thesignal from all 3 PMTs exceeds a set threshold, and is designedfor signals close to the shower core. The second is a time-over-threshold trigger, which requires that the signal from 2of the 3 PMTs exceed a much lower threshold than the firsttrigger for a number of time bins within a given time window,and is designed for signals far from the shower core. Thesetriggers are set in electronics units (channels) – a measure ofthe current from the PMT – so the station must have a referenceunit for current as well. Atmospheric muons again providethis reference, as the same mechanism (see Section 2.1) thatproduces a peak in the charge histogram also produces a peakin a pulse height histogram ( 𝐼 peakVEM ), which is then used as thecommon reference unit for threshold levels. This peak, like thecharge histogram peak, is related to the peak current producedby a vertical through-going muon ( 𝐼 VEM ).The target trigger threshold is 3 . 𝐼 peakVEM for the simple thresh-old trigger, and 0 . 𝐼 peakVEM for the time-over-threshold trigger.The conversion from electronics units to 𝐼 peakVEM must be con-tinually updated in order to maintain the proper trigger level.The accuracy of this determination does not have to be high– the trigger units are quantized in channels – and the targettrigger level of 0 . 𝐼 peakVEM ( (cid:39)
10 ch – see Section 3) for the lowerof the two triggers implies that the precision of the on-line cal-ibration does not need to be much better than 10% (1 part in 10channels) before the quantization of the trigger dominates.In addition, the initial end-to-end gains of the 3 PMTs – thatis, 𝐼 peakVEM – must be roughly equivalent. This ensures that thesignals recorded from the PMTs are similar in amplitude, andsets the proper dynamic range and signal size for the electronics. There are three main steps to the calibration to VEM units.(1) Set up the end-to-end gains of each of the 3 PMTs to have 𝐼 peakVEM at 50 ch.(2) Continually perform a local calibration to determine 𝐼 peakVEM in channels to adjust the electronics-level trigger. Thiscompensates for drifts which occur after step 𝑄 peakVEM to high accuracy using chargehistograms, and use the known conversion from 𝑄 peakVEM to Based on more recent data from Ref. [6] and a similar setup with a differenttank.
The end-to-end gains (i.e. 𝐼 peakVEM in electronics units) of eachof the 3 PMTs are set up by matching a point in the spectrumto a measured rate from a reference tank (see Ref. [6]). Thereference tank is calibrated by obtaining a charge histogram andadjusting the PMT high voltage until the peak ( 𝑄 peakVEM ) of thethree histograms agree. The singles rate spectrum of each of thePMT (i.e. no coincidence between the PMTs required) is thenobtained as a reference. A point on the spectrum convenient asa trigger threshold was chosen as a target calibration point for alltanks – that is the singles rate of a PMT at 150 ch above baselinewas required to be 100 Hz, which corresponds to a trigger pointof roughly 3 𝐼 peakVEM . This choice sets up each of the PMTs to haveapproximately 50 ch/ 𝐼 peakVEM .When the local station electronics is first turned on, each ofthe 3 PMTs is forced to satisfy this condition by adjusting thehigh voltage until the rate is 100 Hz at a point 150 ch abovebaseline. This balances the PMTs to approximately 5%. ThePMTs have a range of temperature coefficients so that subse-quent drifts from the initial settings are inevitable. Thus, highprecision is not required. For a sample of 661 tanks, the meanRMS spread in 𝐼 peakVEM between the 3 PMTs was 4.4%.The end-to-end gain measurement implies that the PMTs inthe SDs will not have equivalent gains – indeed, even PMTswithin the same tank may not have equivalent gains. If a watertank produces more photons per vertical muon than an averagetank, then the PMTs in the tank will be at a lower gain than anaverage tank to compensate. Likewise, if a PMT has a worseoptical coupling than the others in the same tank, resulting infewer photons seen per vertical muon, the PMT will be runat a higher gain. Thus, we would expect to see an inverserelationship between the gain of each PMT and the number ofphotoelectrons ( 𝑛 pe ) per VEM for all the PMTs in the SD. This isshown in Fig. 2. The inverse relationship is quite clear, showingthat the initial end-to-end gain setup is operating correctly. Thechoice of 50 ch/ 𝐼 peakVEM results in a mean gain of approximately3 . × for a mean 𝑛 pe / VEM ∼
94 pe.
Once the gains of the 3 PMTs are set up, the drifts of the valueof 𝐼 peakVEM in electronics units for each detector must be compen-sated to ensure that the surface array triggers uniformly. Thiscompensation is done via adjusting the trigger levels based ona continual on-line calibration. The PMT high voltage is notchanged during normal operation, which implies that the dy-namic range of the SD will be slightly non-uniform. For normaloperation, this non-uniformity is minimal ( ∼ >
20 ch) from the nominal 𝐼 peakVEM of 50 ch are re-initialized following the procedure in Section 3.1.The average value of 𝐼 peakVEM for the PMTs of the SD is currently(46 ±
4) ch.The value of 𝐼 peakVEM as defined in Section 2.2 is not obtainedon-line since this would increase the dead time of the detectorto unacceptable levels. Instead, the trigger levels are set withrespect to an estimate of 𝐼 peakVEM . This estimate ( 𝐼 estVEM ) is definedimplicitly for a given PMT by requiring that the rate of eventssatisfying a “calibration trigger” be 70 Hz. An event satisfies thecalibration trigger if the signal is above 2 . 𝐼 estVEM for the givenPMT and above 1 . 𝐼 estVEM for all three PMTs. The value of therate (70 Hz) was obtained from the reference tank.To obtain the value of 𝐼 estVEM , a 𝜎 - 𝛿 convergence algorithmis used, where a test value ( 𝐼 estVEM ) is altered by an adjustment 𝛿 if a measured value (the rate) is outside of a bound 𝜎 . Thisalgorithm is implemented as follows:(1) Start with a value of 𝐼 estVEM =
50 ch.(2) Measure, for each PMT, the rate of events satisfying thecalibration trigger by counting these events for a time 𝑡 cal ,initially set to 5 s.(3) If, for a given PMT, the rate is above 70 + 𝜎 Hz, increase 𝐼 estVEM by 𝛿 . Likewise, if the rate is below 70 − 𝜎 Hz,decrease 𝐼 estVEM by 𝛿 , with 𝜎 = 𝛿 = 𝜎 away from70 Hz, adjust 𝐼 estVEM by 5 ch in the appropriate direction, set 𝑡 cal to 10 s, 𝛿 = 𝑡 cal <
60 s, increase 𝑡 cal by 5 s. If 𝛿 > . 𝛿 by 0.1 ch, and repeat from step (2).As the calibration trigger is a single PMT trigger within a3-fold coincidence, a small drift in the calibration trigger rate3igure 2: Number of photoelectrons ( 𝑛 pe ) at the first dynode versus PMT gain. As the end-to-end gain setup is designed toproduce equivalent FADC channels for a charge deposition of 1 VEM, an inverse relationship is expected between gain and 𝑛 pe . 𝑛 pe at the first dynode is calculated as described in Section 3.2. The mean gain for the PMTs in the SD is 3 . × , and the mean 𝑛 VEM /VEM is 94 pe.of one PMT should not affect the rate of another. Step (4)allows the algorithm to switch back to a coarser tracking modeto minimize the effect that one PMT can have on the other two.In practice, changes of less than 10% in a period of 𝑡 cal do notaffect the other PMTs significantly.A minimum value of 𝛿 = . 𝐼 estVEM tocompensate for drifts in the baseline of each channel as small as0.1 ch without significantly affecting the error in the estimate.The value of 𝜎 = ∼ × the Poisson fluctuationover that time period, and is reasonable given the requirementof <
10% accuracy.A simple test of the success of the convergence algorithmis to look at the trigger rates for the simple threshold trigger,which is just a 3-fold coincidence trigger at 3 . 𝐼 peakVEM . On areference tank, with 3 PMTs tuned to equal 𝐼 peakVEM values, thisgives a rate of ∼
20 Hz. The 3-fold coincidence rates for 21tanks before and after the convergence algorithm is applied isshown in Fig. 3. The rapid convergence to ∼
20 Hz shows thatthe method described enables the uniform trigger levels to beestablished rapidly.A comparison of the converged 𝐼 estVEM value with values ob-tained from a pulse height histogram gives 𝐼 estVEM = ( . ± . ) 𝐼 peakVEM . The systematic offset is due to a slight inaccuracyin the required calibration trigger rate (i.e. 2 . 𝐼 peakVEM results in70 Hz) and is unimportant.The on-line calibration also estimates 𝑄 peakVEM as well ( 𝑄 estVEM )by computing the charge of pulses with a peak of exactly 𝐼 estVEM ,and using a 𝜎 - 𝛿 convergence algorithm on 𝑄 estVEM , determinedinitially from the charge of the first pulse. A comparison ofthe converged 𝑄 estVEM and 𝑄 peakVEM determined from a peak fitto the charge histograms yields 𝑄 estVEM = ( . ± . ) 𝑄 peakVEM .During operation, 𝑄 estVEM is used to monitor the status of the de-tector continuously and to provide a cross-check on the 𝑄 VEM histogram measurement. Since 𝑄 VEM is just the number ofphotoelectrons per muon ( 𝑛 pe ) times the PMT gain, the dyn-ode/anode ratio, and the electronic gain, 𝑄 estVEM can be used tocalculate 𝑛 pe for all detectors as well. The history over the last 7 𝑡 cal (60 s) periods of the adjustmentsto 𝐼 estVEM is included with each event, along with 𝐼 estVEM , 𝑄 estVEM ,and the last 70 Hz rates for each of the 3 PMTs. The ratio of 𝐼 peakVEM to 𝐼 estVEM was found to have a very slightpressure dependence, which is expected since the online cal-ibration uses the rate of atmospheric muons at 1 . 𝐼 peakVEM todetermine 𝐼 peakVEM . The dependence is clearer for technical rea-sons for 𝑄 estVEM / 𝑄 peakVEM , and is shown in Fig. 4. The correlationis 0.1% per g/cm . The typical pressure change of the SD overone year is about 30 g/cm , implying a maximum 3% yearlyvariation in the trigger level. 𝑄 peakVEM determination from charge his-tograms The SD electronics has a separate trigger designed specificallyfor collecting high-rate data with fewer bins (20 bins instead of768 for the event data) at low threshold (0 . 𝐼 estVEM ). Once thecalibration procedure has stabilized 𝐼 estVEM , this trigger is enabledand sets of histograms of various quantities are collected over60 s intervals – approximately 150 000 entries per histogram.These histograms are sent to the CDAS along with any eventsthat are requested – therefore, each event has a high-statistics setof charge and pulse height histograms from the previous minuteaccompanying the data.The histograms created every minute are: • Charge histograms for each individual PMT. • Charge histogram for the sum of all 3 PMTs. • Pulse height histograms for each individual PMT. • Histograms of the baseline of each FADC channel.The average of all pulse shapes with an integrated charge of(1 . ± . ) 𝑄 estVEM is also sent. An example of the histograms andpulse shape average sent with each event is shown in Fig. 5.4igure 3: Convergence of the 3-fold coincidence trigger at 3 . 𝐼 estVEM to ∼
20 Hz after the convergence algorithm based on the2 . 𝐼 estVEM singles rate for 21 SD stations (station ID is on the right). The convergence algorithm was turned on at 𝑡 ≈
20 min. Thedrop to 0 Hz was caused by the re-boot of the SD stations to enable the convergence algorithm.Figure 4: Correlation of the ratio 𝑄 estVEM / 𝑄 peakVEM to atmosphericpressure as measured by a weather station located at the LosLeones fluorescence site. Note 1 hPa = .
020 g/cm in atmo-spheric depth. The effect ( ∼ .
1% per g/cm ) on the trigger levelover the course of a year is approximately 3%.During data analysis, the second peak of the individual chargehistograms (Fig. 5c) is fit by a quadratic function to obtainthe value of 𝑄 VEM used to convert the integrated signal intounits of VEM. The agreement of 𝑄 estVEM and 𝑄 peakVEM is a goodindication that this peak is resolvable for all the SD stations.The 𝑄 VEM obtained from charge histograms can also be cross-checked using measurements of the charge deposited in the PMTby the Cherenkov light from a decay electron from a stoppedmuon [9]. These two measurements were found to agree to0.5%, well within the 4% uncertainty of the 𝑄 VEM measurementfrom the decay electron.
In addition to the primary conversion from integrated channelsto VEM units, the calibration must also be able to convert theraw FADC traces into integrated channels.There are two primary parameters needed for this. 1. The baselines of all six FADC inputs.2. The gain ratio between the dynode and the anode (calledthe “dynode/anode ratio”).The baseline is computed from each of the 100 Hz calibrationtriggers obtained over a 60 s interval ( ∼ 𝐷 / 𝐴 ) The dynode/anode ratio ( 𝐷 / 𝐴 ) is slightly more complicatedto measure. The only pulses available to measure 𝐷 / 𝐴 aremuon-like pulse shapes – either from atmospheric muons orfrom an onboard LED flasher (used for linearity measurements).A muon-like signal can be described essentially as a fallingexponential after the peak, with a typical decay constant of ∼
60 ns (see Fig. 5d). The signal/noise of the sum thereforedecreases as bins farther from the peak are included in thesummation.The nominal gain between the dynode and the anode is 32 –that is, 5 bits of overlap out of a 10 bit FADC, giving 15 totalbits of dynamic range. For a signal which is nearly saturatedon the dynode ( ∼
950 ch with a 50 ch baseline), the anode signalwill be merely 30 ch above baseline. The RMS noise of theanode and dynode channels is ∼ . − . ∼
60 ns will only be above thenoise level on the anode for four bins (200 ns), as compared to17 bins for the dynode.Ideally, the best measurement of 𝐷 / 𝐴 would occur simplyby taking the peak of the dynode divided by the peak of theanode, and averaging over many samples. Unfortunately, thedynode signal is not simply the anode signal multiplied by 𝐷 / 𝐴 – the dynode is amplified by two Analog Devices AD8012 am-plifier stages, each of which has a phase delay of approximately2 − − ( . ± . ) 𝑄 estVEM . The RMS spread in the 𝐼 estVEM values of the 3 PMTs was 3.3% (the mean for a randomsample of 661 tanks was 4.4%). The baselines are not subtracted in the charge or pulse height histograms.be summing the signal and dividing the sum of the anode by thesum of the dynode. This, however, is also not possible, as theerror associated with the RMS noise of the dynode and anodebecomes quite significant. Structured noise (below the RMSnoise level) due to channel-to-channel crosstalk or other tem-porally correlated noise sources prevents obtaining an accurate( < 𝐷 / 𝐴 measurement even with large statistics. 𝐷 / 𝐴 is therefore measured by modelling the anode signalshape ( 𝐴 ) from the dynode ( 𝐷 ) as 𝐴 ( 𝑡 ) = 𝑅 (( − 𝜀 ) 𝐷 ( 𝑡 ) + 𝜀𝐷 ( 𝑡 + )) (1)where 𝑡 is the time bin, 𝑅 is 𝐷 / 𝐴 , and 𝜀 is the fractional bin offsetof the dynode. 𝑅 and 𝜀 are determined using 𝜒 minimization. 𝐷 and 𝐴 are determined with about 100 pulses taken within3 min. Here, 𝜀 is known to be positive, but is allowed to varyfor the fit. This procedure also has the advantage of measuringthe phase delay and any time dependence it may have.An example of the fit (and the fit region) is shown in Fig. 6,for a version of the front end electronics with higher noise characteristics than the production version. The two pulseswere generated by a resistively-divided pulse generator, whichhad no phase delay between the dynode and the anode. Forthese pulses, the “dynode/anode” ratio was 33.9 by design, andmeasured independently with an oscilloscope. This proceduregave a dynode/anode ratio of 33.3, accurate to within 2%. Thefitting procedure correctly gave a very small value for 𝜀 forthis fit ( < − ) . For actual stations, however, 𝜀 is measuredto be on average 0.23, with an individual precision of 0.04.This corresponds to a delay of 5.8 ns, in agreement with ourexpectations. 𝐷 / 𝐴 determined with this method are also in goodagreement with direct measurements of the 𝐷 / 𝐴 on the PMTbase. However, estimating the accuracy of this comparison isquite difficult as the dependence of 𝐷 / 𝐴 on the high voltage ofthe PMT is poorly measured.6igure 6: Example of the 𝐷 / 𝐴 fit to a resistively-divided pulse with the preproduction front end electronics. The pulse generatorhad an intrinsic 𝐷 / 𝐴 of 33.9, and the 𝐷 / 𝐴 fit method gave a 𝐷 / 𝐴 of 33.3, within 2%. The excess in the anode seen near 300time bins is ∼ 𝐷 / 𝐴 even inthe presence of < The main calibration goal for the surface detector is to convertthe integrated flash ADC signal into vertical equivalent muon(VEM) units, and to provide a stable and uniform trigger forthe detector. The conversion to VEM units is done by deter-mining 𝑄 VEM through their relation to a peak in charge ( 𝑄 peakVEM )histograms, which is determined through an independent ex-periment. 𝑄 peakVEM is measured with a high-statistics (150 000entries) charge histogram every minute, and agrees with an in-dependent local software estimate to 3%.Conversion of the anode signal requires the determinationof the dynode/anode ratio ( 𝐷 / 𝐴 ), which is done by averaginglarge pulses and performing a linear time-shifted fit to determineboth the 𝐷 / 𝐴 and the phase delay between the two signals.This method, when performed on two resistively divided signalsdetermined the 𝐷 / 𝐴 to 2%.Uniform trigger levels are provided by estimating 𝐼 peakVEM – thepeak in a pulse height histogram – via a 𝜎 - 𝛿 convergence algo-rithm on a 70 Hz singles rate inside a 100 Hz 3-fold coincidence.This measurement is precise to 6%. The use of a rate to deter-mine 𝐼 peakVEM introduces a small systematic pressure dependencein the trigger level of approximately 0.1% per g/cm leading toless than a 3% effect over the course of a year.The calibration parameters mentioned here are determinedevery 60 s and returned to the central data acquisition system(CDAS) with each event and stored along with the event data.Each event therefore contains a large amount of informationabout the state of each surface detector in the minute precedingthe trigger, allowing for an accurate calibration of the data. References [1] M. Lawrence, R.J.O. Reid, and A.A. Watson, J. Phys. G 17 (1991) 733.[2] J. Abraham et al., Nucl. Instr. Meth. A 523 (2004) 50.[3] P.D.J. Clark and D. Nitz, Proceedings of 27th ICRC, Hamburg, vol. 2, 2001,p. 765.[4] D. Allard et al., Proceedings of 29th ICRC, Pune, vol. 7, 2005, p. 71.[5] P. Bauleo et al., Nucl. Instr. Meth. A 406 (1998) 69. [6] M. Aglietta et al., Proceedings of 29th ICRC, Pune, vol. 7, 2005, p. 83.[7] A. Etchegoyen et al., Nucl. Instr. Meth. A 545 (2005) 602.[8] D. Nitz for the P. Auger Collaboration, IEEE Trans. Nucl. Sci. NS-51 (2004)413.[9] P. Allison et al., Proceedings of 29th ICRC, Pune, vol. 8, 2005, p. 299.[1] M. Lawrence, R.J.O. Reid, and A.A. Watson, J. Phys. G 17 (1991) 733.[2] J. Abraham et al., Nucl. Instr. Meth. A 523 (2004) 50.[3] P.D.J. Clark and D. Nitz, Proceedings of 27th ICRC, Hamburg, vol. 2, 2001,p. 765.[4] D. Allard et al., Proceedings of 29th ICRC, Pune, vol. 7, 2005, p. 71.[5] P. Bauleo et al., Nucl. Instr. Meth. A 406 (1998) 69. [6] M. Aglietta et al., Proceedings of 29th ICRC, Pune, vol. 7, 2005, p. 83.[7] A. Etchegoyen et al., Nucl. Instr. Meth. A 545 (2005) 602.[8] D. Nitz for the P. Auger Collaboration, IEEE Trans. Nucl. Sci. NS-51 (2004)413.[9] P. Allison et al., Proceedings of 29th ICRC, Pune, vol. 8, 2005, p. 299.