CATIROC: an integrated chip for neutrino experiments using photomultiplier tubes
Selma Conforti, Mariangela Settimo, Cayetano Santos, Clément Bordereau, Anatael Cabrera, Stéphane Callier, Cédric Cerna, Christophe De La Taille, Frédéric Druillole, Frédéric Dulucq, Victor Lebrin, Frédéric Lefèvre, Gisèle Martin-Chassard, Frédéric Perrot, Abdel Rebii, Louis-Marie Rigalleau, Nathalie Seguin-Moreau
PPrepared for submission to JINST
CATIROC: an integrated chip for neutrino experimentsusing photomultiplier tubes
Selma Conforti 𝑎, ∗ Mariangela Settimo 𝑏, ∗ Cayetano Santos 𝑐 Clément Bordereau 𝑑 AnataelCabrera 𝑒 Stéphane Callier 𝑎 Cédric Cerna 𝑐 Christophe De La Taille 𝑎 Frédéric Druillole 𝑐 Frédéric Dulucq 𝑎 Victor Lebrin 𝑏 Frédéric Lefèvre 𝑏 Gisèle Martin-Chassard 𝑎 Frédéric Perrot 𝑐 Abdel Rebii 𝑐 Louis-Marie Rigalleau 𝑏 Nathalie Seguin-Moreau 𝑎 𝑎 OMEGA, Ecole Polytechnique-CNRS/IN2P3, Paris, France 𝑏 SUBATECH, IMT Atlantique, Université de Nantes, CNRS-IN2P3, Nantes, France 𝑐 Astro-Particle Physics Laboratory, CNRS/CEA/Paris7/Observatoire de Paris, Paris, France 𝑑 Univ. Bordeaux, CNRS, CENBG, UMR 5797, F-33170 Gradignan, France 𝑒 IJCLab, Université Paris-Saclay, CNRS/IN2P3, 91405 Orsay, France
E-mail: [email protected] , [email protected] Abstract: An ASIC (Application Specific Integrated Chip) named CATIROC (Charge AndTime Integrated Read Out Chip) has been developed for the next-generation neutrino experimentsusing a large number of photomultiplier tubes (PMTs). Each CATIROC provides the time and thecharge measurements for 16 configurable input channels operating in auto-trigger mode. Originallydesigned for the light emission in water Cherenkov detectors, we show in this paper that its usecan be extended to liquid-scintillator based experiments. The ∼ a r X i v : . [ phy s i c s . i n s - d e t ] F e b ontents – 1 – Introduction
Even though neutrinos were discovered over half a century ago, neutrino physics is still one of themost active fields in physics. A precise measurement of the neutrino oscillations and of the neutrinomass ordering is requiring a huge exposure and challenging performances for the new experiments.Multi-tons experiments using thousands of photomultiplier tubes (PMTs) are currently under designand construction (e.g., JUNO [1, 2], HyperK [3], KM3NeT-ORCA [4],...). For such experiments,flexible integrated systems with fast response, high performance in charge and time determinationand compact data stream become more and more crucial. The typical requirements are listed inTable 1.The Jiangmen Underground Neutrino Observatory (JUNO), currently under construction in thesouth of China, aims to determine the neutrino mass ordering, namely the sign of the atmosphericmass splitting and to perform precision measurement of the oscillation parameters. With its20 kilotons of liquid scintillator (LS), it will be the biggest detector of such a type ever built, andit will be competitive for astrophysical neutrinos, as Supernovae, geo-netrinos and for rare-eventsphysics. The detector consists of a spherical array of 18000 20-inch PMTs and about 26000 3-inchPMTs (also named “small PMT” or “SPMT”). The core of the SPMTs readout is the CATIROCchip, a 16-channel front-end ASIC (Application Specific Integrated Chip) designed by the Omegalaboratory in AustriaMicroSystems (AMS) SiGe 0.35 𝜇 m technology to readout PMTs in large-scaleapplications. CATIROC is an upgraded version of PARISROC2 [6] chip conceived in 2010 in thecontext of the R&D PMm (square meter PhotoMultiplier) project [7].CATIROC, was initially developed for large water Cherenkov detectors. Thus, its applicationin liquid-scintillator based experiments requires some additional considerations. In fact, differentlyfrom the Cherenkov light emission which arrives to the PMTs in a few nanoseconds, the processesof excitations and re-emission of photons in liquid scintillator determine a signal spread on longertime windows (up to hundreds of nanoseconds depending on the LS specific recipe). The timeseparation between consecutive hits in the same PMT is thus significantly large and the matchingbetween the dead time and the signal time distribution becomes critical.In this paper we point out some new features of CATIROC that were not examined in previouscharacterization of the chip, as in [5] or in the datasheet. We perform dedicated measurements toensure that the detection of multiple hits, which arrive with a time separation typical of the LS,(from tens of ns up to hundreds of ns), is not biased. A dead time on the time scale of tens ofns is emerged from these measurements. Moreover, we extensively study the charge pile-up (orcharge acceptance) in the case of two hits arriving shortly in time, especially close to this triggerdead time window. We prove that some signal loss or underestimation will occur if the CATIROCconfiguration is not properly adapted to the physics of the detector. Several CATIROC configurableparameters have been tested to find the optimal ones which mitigate the charge biases and triggerdead time effects. Finally, we show that the application of CATIROC to a liquid scintillator isactually possible only when the rate of multiples hits on the same PMT is sub-dominant. This isthe case of PMTs operating in “photon-counting" mode as, for example, for the primary physicsgoal of the JUNO SPMT sub-system [8, 9]. Because of the small photocathode surface, each 3-inchPMT will rarely detect more than one photo-electron (PE) from neutrino interaction events and willthus suit the CATIROC features. For 1 MeV positron uniformly distributed in the detector, about– 2 –% of the SPMTs have more than one photon hitting the PMT photocathode (4% at 10 MeV). Forthem, the time separation between hits in the same PMT is within 250 ns in 99% of the cases. Forcompleteness, we also provide a description of the major features of the CATIROC, its charge andtime response, its trigger efficiency and the performance attainable even in the cases of PMTs withmultiple-hits (i.e., atmospheric muons crossing the detector).Requirement value sectionNumber of channels O(10 ) 1Trigger Threshold < 𝜇 s)) 5Dynamic range application specific (1 to ∼
100 PE) 6Charge resolution < <
1% 6Cross talk O(10 − ) 6Charge integration window application specific ( ∼
20 ns) 6.2Time resolution O(100 ps) 7
Table 1 . Typical requirements for a PMT-based experiment. Some of the requirements are related tothe physics goals (e.g., the expected signal intensity and variation for the charge dynamic range, theexpected trigger rate for the dead time, the time profile for the time accuracy), the type of experiment(Cherenkov or liquid scintillator which affects the time integration and dead time) or the detectorperformance (e.g., detector size and material, PMT signal width and transit time spread, trigger condi-tions). For these cases we indicate between parentheses the expected values for the JUNO-SPMT application.
The manuscript is organized as follows. The CATIROC design is described in Section 2. The maintests to characterize the ASIC performance are presented in Section 3 which includes the triggerefficiency, the dead time contributions, the charge linearity and resolution and the time resolution.For each specific performance test, the experimental requirements are discussed in the relatedsub-section. In Section 8 we performed tests of CATIROC with 3-inch PMTs. The suitability ofCATIROC to fulfill the requirements of liquid scintillator experiments is finally summarized inSection 9.
CATIROC is conceived as an autonomous and flexible system made of a chip operating in auto-trigger mode and providing a measurement of the arrival time and the integrated-charge for 16input signals. An adjustable gain is featured for each input channel to compensate for the gainvariation between the PMTs. A shift register is used to send the configuration parameters (hereafternamed slow control parameters) inside the chip. There are 328 slow control parameters which areloaded serially to control the chip. The architecture of the ASIC is shown in Fig. 1, distinguishingin violet, orange and green the circuits responsible for the trigger operation, and time and chargemeasurements, respectively. In the following we describe these circuital parts with a level of detailssufficient for a comprehensive view of the chip functionalities and performance. Other technical– 3 –nformations are given in the CATIROC data sheet which can be provided on request.Each of the 16 input channels has two low-noise preamplifiers (PA), for the high (HG) and the lowgain (LG) paths with a fixed ratio LG/HG of 10. The two preamplifiers have an input capacitance(C in ) of 5 pF and 0.5 pF, respectively. A feedback variable capacitor ( 𝐶 𝑓 ) tunable with a 8 bits slowcontrol parameter, is used to further vary the gain of each channel independently (see Figure 1).The gain is obtained from the ratio C in /C 𝑓 and can theoretically vary in a 8 bits range with thesmallest step of 0.008 pF and within the limit of the signal saturation (see Section 6). The use ofthe HG and LG paths allows to achieve a dynamic range from 160 fC up to 70 pC (from 0.3 PEto145 PEs at a PMT gain of 3 × and with a preamplifier gain of 20). After the preamplificationphase, the signal feeds a slow and a fast channels used respectively for a charge measurement andfor timing and trigger. Fast channel
It is responsible for the trigger formation and the time measurement (violet andorange circuital parts in Fig. 1). It comprises a fast shaper (FSH), followed by a low-offsetdiscriminator which allows to auto-trigger on signals above a set threshold.The fast shaper is a band-pass (CR/RC) circuit with a time constant of 5 ns. It filters possiblenoise contributions that would also alter the minimum signal threshold. It is coupled to the HG
TopManager FIFO+SCA STATE MACHINE 10-bitsADCCOUNTER 26-bitTIMESTAMPCOUNTER READ OUT 2 DATAOUTPUT i npu t s P A L G P A HG SS H L G SS H HG F S H MUX SCALGSCAHG D I S CR I CHARGEHG orLGMUX ADCCHARGEADCTIMEMUX D I S CR I ChargeThreshold* 16 OUTPUTSSSHTO EXT ADCChargeHG orLGRamp charge T r i gg er T h re s h o l d * MaskTrigger
MUX TRIGGERSYSTEM TACRAMP1TACRAMP2 MUX
STARTrampSTOPramp
Out Ramp TACfromdigitalblock40MHz160MHzRamp timeOut RampTAC
DELAYCELL
MUX Ext HoldTriggerdelayed
AnalogchargeLGAnalogcharge HG
16 OUTPUTSTRIGGERSNOR 1610-bit
DACTRIGGER
DACTRIGGER
External Trigger*ChargeThreshold
RAMPCHARGEADC
RAMPTIMEADCBANDGAPCfCf* Common for the 16 chsCinCin50 W externaleachchannel Figure 1 . CATIROC simplified block schematic. In violet, the part responsible for the trigger formation;in orange and green the part dedicated to time and integrated-charge measurements and in red the twopreamplifiers; in blue and black the digital parts of the circuit and the block commons to all channels. – 4 – elayed triggertriggerfromfast channel delayed triggerto SCAdelay5ns to ~ 280 nsSSH amplitude stored in SCA C1 (ping)C2 (pong)Switched Capacitor Array (SCA)
Figure 2 . The Switched Capacitor Array (SCA) schematic. The trigger signal output is delayed by a user-configured time to match the maximum SSH voltage value. This value is stored in the first capacitor of theSCA (C1 or ping) and then digitized. The next signal follows the same path but will be stored in the secondcapacitor (C2 or pong). preamplifier to produce a trigger on input charges as low as 50 fC. For each channel, a logicalgate system is used to select between the trigger formed by the input signal and an external trigger(injected through an external system as for example an FPGA). In addition, an internal structureallows to mask the internal trigger in each channel.The discriminator output is followed by a system responsible for the time measurement and forstoring the analog signal charge of the “Slow channel” discuss below. The time is composed of twoterms: 𝑇 tot [ ns ] = CoarseTime × [ ns ] − FineTime [ ns ] . (2.1)where the “coarse time” results from a 26-bit Gray Counter with a resolution of 25 ns (40 MHzclock) and the “fine time” uses a Time to Analog Converter (TAC) and a 10-bit ADC to provide atime measurement inside the 25 ns coarse time clock with a resolution (in RMS) of less than 200 ps. Slow channel
This part of the circuit (green in the figure) is responsible for the charge measure-ment and consists of a slow shaper (SSH) and a Switched Capacitor Array (SCA) system. The SCAis made of two capacitors working in the so-called “ping-pong mode": the analog signal is heldalternatively in two capacitors before the digitization phase, effectively reducing the dead time ofthe circuit. Small differences between the two capacitors will reflect on the measured charge value,but they are constant and can be fully characterized as discussed later on in the paper. The chargemeasurement process is sketched in Fig. 2 to help the reader understanding some crucial aspects forthe application of CATIROC to liquid scintillator experiments (Section 6.2). The circuit is repeatedtwice, for the HG and the LG channels, but only one of these two analog signals is selected anddigitized by an internal 10-bit Wilkinson ADC. The choice of the HG or LG charge is done by adiscriminator that compares the HG SSH signal with a threshold (named “charge threshold”) set byan internal 10-bit DAC and common to the 16 inputs.All the channels are handled independently by the digital part, and only those channels that fireda trigger are digitized, transferred to the internal memory and then sent out in a data-driven way.– 5 –hen the charge and the time are converted and stored in the registers, the digital part stops theconversion and starts the readout (RO) of the data. The coded data are sent in parallel in two 8channels serial links to be readout. The 𝑇 DeadTime thus is given by the convertion and readout timesand is estimated up to 9.3 𝜇𝑠 if 8 channels per link are hit and around 6.8 𝜇𝑠 if only one channel ishit.A specific input line (named “incalib”) allows the injection of the input charge in more than onechannel in parallel. This is possible thanks to an array of 16 switches managed by slow control.Finally, in addition to these main features, other outputs are accessible, as the 16 triggers, theanalog signals (slow and fast shapers, preamplifiers), the digital ones. These outputs are meant fordebugging but a possible application of the trigger analog signal to reduce the effective dead timewill be discussed in the next section. Figure 3 . Top: Evaluation board of CATIROC. It makes use of a general purpose communication interface(USB), for data transferring to a computer, coupled to a programmable logic array (FPGA) Altera CycloneIII for the implementation of custom functionalities. Bottom: the data output format used in this paper. Inaddition to the standard CATIROC output (3 bits for channel number, 26 bits for coarse time, 10 bits forthe fine time, 10 bits for the analog charge and 1 bit for the gain), two bits are used to define the event type(standard CATIROC output, a trigger rate monitor and a debugging output), 11 bits for an event counter andan additional bit for the channel number for future uses on multi-chip boards. – 6 – ime [s]10 - - - · A m p li t ude [ V ] - - - - Width: 10 nsEdges: 5 nsAmplitude: 850mVFrequency: 10 kHz
AmplitudeWidth
Figure 4 . Example of the typical pulse injected with the signal generator, having a width of 10 ns and theedge times (calculated between 10% and 90% of the signal) of 5 ns. The amplitude, the edge times, frequencyand signal width have been varied depending on the performed tests.
The evaluation board used to characterize the CATIROC ASIC is shown in Fig. 3 (top). It givesaccess to all the 16 input channels, to the “incalib” line and all CATIROC probes to monitor theanalog processing of the input signal and to inspect the behavior of the digital ASIC section. Aprogrammable logic array (FPGA) handles the communication with the control software, sets upof the ASIC parameters and provides the two 40 and 160 MHz clocks used by CATIROC ensuringlocal synchronisation in multiple-chip read-out boards. Moreover the FPGA reads out the ASICoutput providing a package data customized as shown in Fig. 3 (bottom).The tests shown in the next sub-sections are performed using a signal generator which providespulse waveforms with edge times ≤ × ) with the current configuration. For the studiesin this paper a few other parameters in the slow control (as the discriminator threshold) havebeen varied depending on the performed test and will be explicitly indicated in the text when needed.In the following, we summarize the tests performed with CATIROC with a focus on the onesinteresting for the application to scintillation light emission. More specifically, these tests include:– 7 – trigger threshold [DACu]600 650 700 750 800 850 900 950 e ff i c i en cy c ha r ge [ p C ] = 99% S ˛ = 90% S ˛ = 50% S ˛ Figure 5 . Left: Trigger efficiency as a function of the threshold for various injected charges up to 0.48 pC anda preamplifier gain of 20. The pedestal is on the right side and the larger inputs go to the left side of the figurebecause the fast shaper provides negative signals with a pedestal at around 1.8V [5]. The correspondingsignal value in PE, is also indicated, considering a gain of 3 × . Right: typical charges that can be detectedwith a trigger efficiency of 99% (dotted), 90% (long dashed) and 50% (solid line) for a given discriminatorthreshold. The case of 0.3 PE (i.e., 0.16 pC at a gain of 3 × ) is shown as short dashed line for illustrativepurposes. the trigger efficiency versus the discriminator threshold to verify the capability of observing afraction of PE; the charge resolution and linearity over a wide signal range; the characterization ofthe charge biases induced by the circuit for signals arriving very close in time; the time resolutionand possible dead times. A typical requirement of PMT-based experiments is the capability of measuring signals whoseamplitude can also correspond to a fraction of a photo-electron with a dead time small enough notto affect our capabilities of reconstructing the event of interest for our physics goal.The trigger efficiency is defined as the ratio between the number of trigger counts in the output ofCATIROC (“trigger output”) and the number of injected pulses for a given threshold. The efficiencyversus the discriminator threshold (also called S-curve) is shown in Fig. 5 (left) for different inputsignals (0.16 pC, 0.32 pC and 0.48 pC). Each thin line is one of the 16 input channels. The black linesrefer to the case in which no signal is injected and the trigger is fired on the pedestal. The value ofthe discriminator threshold is configured by a 10-bit Digital-to-Analog Converter (DAC), coveringa range between 1 and 1.9 V, with a slope of 0.9 mV/DACu (Threshold [ mV ] = + . × DACu).The right plot shows a projection of the S-curves and gives the minimum charge corresponding tothree fixed reference values of signal efficiency (99%, 90% and 50%) as a function of the triggerthreshold. Any signal above these three lines will have a trigger efficiency larger than the referencevalues. A threshold of 0.3 PE (equivalent to 0.16 pC for the case considered in the figure) istypically considered in experiment using PMTs and is thus attainable with CATIROC. However, thediscriminator threshold is common to all channels. This implies that an optimal value needs to be– 8 – signal generator rate [kHz]0 50 100 150 200 250 a c qu i r ed da t a r a t e [ k H z ] Figure 6 . Trigger rate as measured with CATIROC versus the signal generator frequency, with red squaresfor the case of one channel hit and blue dots for the case of all channels hit at the same time tuned for each CATIROC during the readout boards and PMTs configuration.
Two dead time contributions have been identified which may have an impact for CATIROC appli-cation to liquid scintillator-based experiments.A first term, on a few microseconds scale, is the 𝑇 DeadTime already described in Section 2 which is dueto the digitization of the charge and of the fine time. This dead time 𝑇 DeadTime has been calculatedup to 6.8 𝜇 s and 9.3 𝜇 s depending on the number of data streams to read out (see Section 2) and itis mitigated further by the use of a SCA with two layers of charge holding (ping and pong). Thisimplies that up to two signals can be read out without losses if they arrive within 𝑇 DeadTime . Athird signal in the same window would be lost in the data stream. A check of this dead time hasbeen performed by measuring the rate of events in the CATIROC data output for increasing inputsignal frequencies. Fig. 6 shows the rate of events observed with CATIROC as a function of thegenerator frequency for the case of signals injected in one or all channels. A saturation is observedat generator frequencies of ∼
180 kHz and 120 kHz for the two cases respectively, in agreementwith expected values.The second dead time contributor, hereafter called 𝑇 TrigDeadTime , is due to the handling of the triggerwhich is delayed to store the analog charge in the analog memory and is used to create the TACramp. To measure this dead time a burst of two consecutive pulses is injected with a period 𝑃 = Δ 𝑇 ) varying between 15 ns and hundreds of ns. Theobtained efficiency curve is shown in Fig. 7. A sigmoid function 1 − ( + exp ( 𝑥 − 𝑝 )/ 𝑝 ) − ) is fit to data. The best-fit parameter p0 indicates the time Δ 𝑡 , which corresponds to 50% triggerefficiency. The p1 parameters is the inverse of the slope of the curve : the value 𝑝 ± 𝑝 [ns] D time from previous hit 0 20 40 60 80 100 120 140 160 180 t r i gge r e ff i c i en cy / ndf c – – c – – c – – c – – Figure 7 . Trigger efficiency as a function of the time separation between two consecutive hits : an hitarriving within about 60 ns from the previous one, will not produce any independent trigger signal and theassociated charge may sum up to the previous hit. the interval with trigger probability between 23% and 73%. This curve has the same behavior inthe three circuits tested for this study. Two pulses arriving with Δ 𝑇 larger than 90 ns will always bedetected as separate signals and the charge will be measured in the digital part of CATIROC. Thisis not always the case below 90 ns. For a Δ 𝑇 ≤
60 ns, the second pulse is not handled by the digitalpart of the trigger and in practice the board does not produce an independent charge measurementin the CATIROC output. As we will discuss in Section 6.2, in this case, the charge of the secondpulse sums up to the first hit, depending on their relative time difference, so that the informationis not completely lost (charge acceptance). Moreover, as mentioned in the previous section, thedirect signal from the discriminator, which is not affected by the digital treatment of the trigger, canbe directly read by the FPGA or with an oscilloscope. An example of the discriminator output isshown in Fig. 8 (left) for two input pulses arriving at 20 ns and 30 ns apart, proving the capability ofseparating signals below the 𝑇 TrigDeadTime . The only limitation comes from the discriminator signalwidth (8, right) which can increase with the input charge from 12 ns up to 24 ns. By identifying thetwo edges of the signal in the FPGA, we can count the actual number of input signals and retrievethe signal charge (up to 1 pC, i.e. 2 PE for a gain of 3 × ) based on the discriminator signal width.It is thus a good complement in the trigger dead time region and provides independent check forthe charge acceptance in Section 6.2. A major difference in the use of CATIROC for liquid scintillator based detectors is that the expectedsignals are spread in time (up to hundreds ns) compared to the Cherenkov ones. In the previoussection we have pointed out a trigger dead time, the 𝑇 TrigDeadTime , affecting consecutive hits arrivingwithin less than ∼
90 ns. Moreover, we mentioned that the charge is measured from the digitized– 10 – s] m time [ A m p li t ude [ V ] - - - - T = 20 ns D s] m time [ A m p li t ude [ V ] - - - - T = 30 ns D input charge [pC] 0.2 0.4 0.6 0.8 1 1.2 T r i gge r w i d t h [ n s ] Figure 8 . Left: discriminator analog signal as observed with an oscilloscope for two pulses injected with atime difference of 20 ns (top) and 30 ns (bottom), smaller than the 𝑇 TrigDeadTime . Right: the width of eachtrigger signal increases with the input charge up to 1 pC (about 2 PE for a gain of 3 × ). value of the slow shaper (SSH) amplitude at the peaking time. The slow shaper is an RC filter ofsecond order with a bi-polar signal which may introduce distortions in the charge measurementsfor pulses arriving close in time. A dedicated validation is thus needed to ensure the capabilityof CATIROC to detect all hits and to retrieve the full charge without any significant bias. In thissection, we thus initially prove the charge linearity following the classical approach with inputpulses of increasing amplitude. We then inject a train of two pulses performing a scan of their timeseparation from hundreds of ns to 15 ns. This test mimics the case of two hits arriving close in timeand allows to investigate all possible charge distortion introduced by the combination of the SSHshape and the trigger dead time, for different RC values of the SSH. For some of these effects wepropose a correction method or we comment on the impact on physics measurements. The linearity of the charge measurement has been tested for all channels for signals ranging between0.1 and 70 pC. Fig. 9 shows one example channel, for the HG (red) and the LG line (blue) and forthe two SCA capacitors, ping (filled markers) and pong (empty markers). Measurements repeatedon a second test board and using an independent setup, provide compatible results in terms ofresolution and LSB (Least Significant Bit, given in fC/ADCu). As reference, the input charge scaleis also given in PE units, assuming a PMT gain of 3 × in the charge-to-PE conversion. Theresults of the linear fit for HG and LG are reported in Table 2, together with a list of the usedCATIROC configuration parameters. A measurement of the pedestal is independently performedby measuring the charge distribution with an external trigger (generated by the FPGA): the meancharge is consistent with the intercept (p0) in Table 2 and the RMS is about 1.5 ADCu, whichprovides a signal over noise ratio (SNR) around 40 for a PMT gain of 3 × and the calibrationparameters in the table. The RMS of the charge distribution has been measured equal to 2 ADCuwhen operating in HG mode and 1 ADCu in the LG case, which corresponds to 0.015 pC and 0.7pC, respectively. These charge resolutions correspond to 3% for a gain of 3 × (i.e. 0.48 pC) and– 11 –re within typical PMT-based experiments requirements for which single photo-electron resolutionsare expected around 30% [9].The calibration curve shown in Fig. 9 is obtained in the case of RC=50 and a preamplifier gainof 20. The typical signals of a 3-inch PMT have a width of 10 to 20 ns. The impact of the RCconstant on the integration window can be visualized in Fig. 10, where we also tested the chargelinearity against the signal width for different RC. For a fixed amplitude of the injected signal, weincrease the width of the pulse and we measure a deviation from linearity starting at about 20, 25and 55 ns, respectively, for the three considered RC constants. For these configurations we checkedthat no cross-talk signals where observed in other channels than the injected one, over a million oftriggered pulses and up to ∼
60 pC input charges. input signal [pC]0 10 20 30 40 50 60 m ea s u r ed c ha r ge [ A DC u ] · input signal [PE (at G = 3 0 20 40 60 80 100 120 channel 0 HG - ping LG - pingHG - pong LG - pong
Figure 9 . Linearity of one of the channels for the high-gainin red, and the low-gain in blue. Empty and filled markersdenote the two capacitors (ping/pong) in the analog memory.
ConfigurationsSSH RC 50PA gain 20Discr. thr 900 DACuHG/LG thr 720 DACuSSH Peak time 26 ns 𝜎 charge , HG 𝜎 charge , LG . ± . . ± . . ± . . ± . < < Table 2 . Summary of the CATIROC configura-tions and of the results of the charge linear fit.
For what concerns the preamplifier, its gain directly affects the measured charge as an effectivemultiplication factor as shown in Fig. 11. In the figure the ratio between the measured charge atgain G and the charge measured at the reference case G=10 is plotted against the nominal gain. Evenif the gain can extend more, only the interval up to 60 looks exploitable with small saturation effects.This is sufficient for typical applications where the PMTs can be grouped to have similar gains at thenominal high voltage value. For completeness, we also indicate the maximum variations observedamong three tested CATIROCs (gray band) and within the 16 channels of a same CATIROC (red).In practice, the effect of this variation is negligible as the preamplifier can be fully characterizedfor each ASIC separately during the production test phase of the chips or of the readout boards.– 12 – signal width [ns]0 20 40 60 80 100 r e sc a l ed c ha r ge [ a . u ] Figure 10 . Charge linearity as a function of the signal width for different RC constants of the slow shaper.
Nominal PA gain G 0 20 40 60 80 100 c ha r ge m u l t i p li c a t i on f a c t o r w r t t o G = variation within a CATIROCvariation among CATIROCs Figure 11 . Charge multiplication factor (effective gain) vs preamplifier (PA) nominal gain 𝐺 . The averagegain linearity is shown as dark solid line with a gray band indicating the maximum variation between differentCATIROCs and a red shaded band indicating the variation within channels of the same CATIROC. As previously mentioned, the charge is measured from the digitized value of slow shaper (SSH)amplitude at the peaking time (see Fig. 12, top). In this section we study the possible effectsintroduced by the slow shaper for three possible configurations of RC (25, 50, 100). One ofthese effects is due to the undershoot visible in the figure: a dashed line is drawn to indicate theexpected baseline. We show in Fig. 12 (bottom) the example case of two identical pulses arrivingat Δ 𝑇 =
180 ns: the two SSH overlaps and the amplitude of the second shaper is biased by the– 13 –
Time (ns)0 50 100 150 200 250 SS H a m p li t ude ( D A C u ) RC=25RC=50RC=100 s) m Time (0 0.5 1 1.5 2 s i gna l ( V ) Real data (oscilloscope waveform) = 180 ns pulses T D RC = 50,
Figure 12 . Top: Scan of the SSH vs time (measured from the digitized charge value) for three different RCvalues zoomed around the peaking time. The undershoot tail starts (ends) around 50 (250 ns), 100 (400 ns)and 250 (1000 ns) for RC=25, 50 and 100, respectively. Bottom: case of two signals arriving close in timewith the two shapers overlapping. The waveforms are measured with an oscilloscope from one of the probeoutputs available for debugging on the evaluation board. The amount of overlap and the possible effects onthe measured charge (pile-up, signal loss or signal underestimation) depends on the RC shape and on the Δ 𝑇 between the two pulses. undershoot tail of the first one, producing an underestimation of the charge. A second possible bias,which may potentially engender a signal loss, occurs when two signals arrive with a time separationof the order of the 𝑇 TrigDeadTime or smaller.For the studies presented in this section, we inject a burst of two identical pulses with a period– 14 – time separation [ns]10 20 30 40 50 60 70 80 90 100 a c ha r ge a cc ep t an c e SS H pea k t i m e ( RC ) SS H pea k t i m e ( RC ) RC = 50, W = 4ns RC = 100, W = 4nsRC = 50, W = 10ns RC = 100, W = 10ns
Figure 13 . Charge acceptance for RC=50 (circle) and RC=100 (square) as a function of the separation timebetween pulses. Empty (filled) markers refers to points taken with a signal width of 10 (4) ns.The dashedlines indicates the SSH peak time for both RC values.
P = 1 ms and a time separation between pulses Δ 𝑇 varied from 15 to 1000 ns . To simplify thedescription, we indicate with 𝑞 ∗ and 𝑞 ∗ the charge measured by CATIROC and 𝑞 = 𝑞 the “true”corresponding values. We distinguish two cases : • Δ 𝑇 ≤ 𝑇 TrigDeadTime : only the first hit of each burst is triggered. Its measured charge 𝑞 ∗ mayinclude a fraction of the second charge of the second pulse : 𝑞 ∗ = 𝑞 + 𝛼 ( Δ 𝑇 , RC ) · q (6.1) 𝑞 ∗ = ≤ 𝛼 ( Δ 𝑇 , RC ) ≤ 𝛼 willdepend on the time overlap between the two pulses and on the RC shape. It is shown in Fig. 13for RC=50 and RC=100 . The empty markers refers to injected signals with the same width(10 ns) used so far. However, to explore the time separation <
20 ns, we additionally testedthe case of signal width of 4 ns. For signal separations smaller than 15 ns we extrapolateto 𝑓 = Δ 𝑇 = 𝑇 SSHpeak < Δ 𝑇 < 𝑇
TrigDeadTime will not be detected and at the same time their charge does notmodify the peak amplitude of the SSH for the first hit (i.e. 𝛼 = • Δ 𝑇 > 𝑇
TrigDeadTime ( undershoot bias ) : The two hits are triggered independently but, asillustrated in Fig. 12 (bottom), the undershoot acts as an offset for the second SSH. This The signal generator does not allow for shortest signal thus limiting the pulse separation to twice the signal width The case RC=25 is not considered here since the signal generator would not allow to investigate pulses separationsmuch smaller than the 𝑇 SSHpeak ∼
18 ns. – 15 – charge [ADCu]80 100 120 140 160 E n t r i e s Figure 14 . Charge distribution for first hit (solid line) and second hit (thin dotted), for Δ 𝑇 =
100 ns. The reddashed line shows the charge of the second hit after applying the correction for the bias introduced by theSSH undershoot (see text). induce an underestimation of the measured charge ( 𝑞 ∗ < 𝑞 ) : 𝑞 ∗ = 𝑞 𝑞 ∗ = 𝑞 − 𝜅 ( Δ 𝑇 , RC ) ∗ q (6.2)where 𝜅 ( Δ 𝑇 , RC ) < 𝜅 . This correction isworking perfectly as illustrated on Fig. 14. As described in Section 2, the time measurement is the combination of a 26-bit coarse time anda fine time measured with a TAC and digitized by a 10-bit Wilkinson ADC. A scan of the TACramp is done spanning the clock cycle (25 ns) in steps of 0.1 ns as shown in Fig. 15 (top). Anexample of the TDC measured values (thin gray line) and the linear fit (thick red line) is shownin the top panel. The parameters of the linear fit to the average over hundred independent runs,gives 973 . ± . − . ± .
01 TDCu/ns for the intercept and the slope, respectively.For each measurement, the residuals from the fit function distribute with an RMS (over the fulltime window) of about 0.15 ns. Moreover, these residuals (thin colored lines in Fig. 15, bottom)exhibit a modulation due to the clock coupling in the substrate. The average curve (solid thick line)ranges between + − TDCu, which corresponds to − . + . ns, with variations of ± ∼
54 ps)between different runs. As emphasized by the figure, the observed modulation is well reproduciblebetween different runs and may be corrected offline, which is an interesting possibility for specific– 16 – F i ne t i m e [ T DC u ] single run (x100)linear fit time [TDCu]0 200 400 600 800 1000 r e s i dua l s ( f i ne t i m e - li nea r f i t ) [ T DC u ] - - - - - single run (x100)average over 100 runs r e s i dua l s ( f i ne t i m e - li nea r f i t ) [ p s ] - - - Figure 15 . Top: TAC response (after digitisation) obtained from a scan in time with step size of 100 ps.The measurement is repeated for 100 runs. Bottom: Residuals with respect to the fit of each scan curve (thincolored lines) and for the average value (black solid). applications aiming at a time resolution below 100 ps. Such a study is out of the scope of this paper,because they are much smaller than the typical PMT time transit spread ( 𝜎 ) which is typically of1.5-2 ns [9, 11]. In the validation board used for this paper, the voltage levels of the TAC rampare not well adapted to the ADC input ones and the digital output spans a range smaller than theavailable bits (970 ADCu instead of 1024 ADCu). This reduces the LSB from the nominal valueof 24 ps/TDCu to 27 ps/TDCu. Though it has negligible effects for the results shown here and forsystems dominated by PMT resolutions, this non-adaptation could be solved by adding an externalresistor on the board.A check of the time resolution ( 𝜎 T ) of the system is also obtained by measuring the time difference, 𝑑𝑇 , between two consecutive signals injected in the same 𝑖 -th channel and the time difference– 17 – [ns] i,0 dT0.5 - E v en t s [ no r m a li z .] - - - ch2 - ch0ch4 - ch0ch7 - ch0 ch14 - ch0ch0 : consec. evts = 67 ps s Figure 16 . Example distributions of 𝑑𝑇 𝑖, 𝑗 between the 𝑖 -th channel and channel 𝑗 = 𝑑𝑇 (subtracted of the signal generator period) obtained from thetime difference between consecutive signals injected in the same channel ( 𝑖 = 𝑗 = 𝑑𝑇 𝑖 𝑗 measured for the same signal injected in the channels 𝑖 and 𝑗 using a passive signal splitter(FANOUT). In the first approach, the 𝑑𝑇 distribution is expected to be centered at the signalgenerator period, with a spread that is related to the time resolution convoluted with the jitter of thesignal generator. An example is shown in Fig. 16 where the thick blue distribution represent thecase of channel 0. The signal generator period is subtracted for plotting reasons. The distribution isfitted with a gaussian function giving a 𝜎 of 67 ps. This results is obtained similarly for all channelsand is mostly affected by the modulation of the TAC value mentioned above. The second approachis introduced to get rid of the systematics due to the signal generator, even if it will be then sensitiveto possible systematics between different channels. The distributions of 𝑑𝑇 𝑖 𝑗 in Fig. 16 are thenexpected to be centered at zero with a measured spread ( 𝜎 𝑖 𝑗 ) given by the time resolutions 𝜎 𝑖 and 𝜎 𝑗 of the two channels by: 𝜎 𝑖 𝑗 = √︃ 𝜎 𝑖 + 𝜎 𝑗 ≈ √ 𝜎 T under the assumption that 𝜎 𝑖 ≈ 𝜎 𝑗 = 𝜎 T . We conservatively take the spread 𝜎 𝑖 𝑗 as time resolutionper channel. The distributions 𝑑𝑇 𝑖, ( 𝑖 ≠
0) are not perfectly centered at 0 which can be explained asan effect of the FANOUT (tested by mixing the channels in inputs of the FANOUT) and a dispersionof the TAC ramps of different channels. With both approaches, the value of 𝜎 𝑖 𝑗 is within 150 ps,compatible with the results from the TAC ramp residuals.The accuracy of the time measurement is limited by the “time walk” of the fast shaper, which isbasically the time at which the fast shaper signal exceeds the trigger threshold. The time walkdepends mostly on the signal amplitude. In this study the time walk has been measured injectinga charge with a pulse generator and comparing the trigger time in CATIROC and the signal time.Fig. 17 shows the difference between these two times as a function of the injected charge for atrigger threshold of 0.3 PE. In this configuration, the time walk of the CATIROC circuit has been– 18 – Input charge [pC]0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 D e l t a T i m e [ n s ] Time walk = 4.6 ns
Figure 17 . Difference between the trigger time and the pulse generator signal as a function of the inputcharge. A time walk of 4.6 ns is measured, for a trigger threshold of 160 fC (about 0.3 PE at 3 × andpreamplifier gain of 20). measured to be around 5 ns. It should be noted that the time walk depends on both the PA gainvalue and the DAC threshold value and that a correction of the time walk can be applied based onthe input rise time and the measured charge. In this section we demonstrate the CATIROC performance when employed to read the signals froma 3-inch PMT. The aim of this section is not to provide a complete study of the PMT responsebut rather to validate the charge spectrum measured by CATIROC and point-out possible effectsintroduced by the circuit. To validate the ASIC response we firstly measure the single photo-electron(SPE) spectrum with an oscilloscope and then we compared the obtained PE position and resolutionwith the one measured by CATIROC.For this purpose, we set up a test-bench consisting of a light-tight box (50 × ×
25 cm ) containingup to two 3-inch PMTs (model HZC XP72B22 [11]) powered with a negative high voltage (HV).A simple approach to measure the SPE spectrum is to look at the dark noise, which is mostlydue to thermo-ionic emission from the photocathode, leakage current between the cathode and thedynodes and ionization from residual gases. To ensure that a steady dark current is reached, dataare always acquired at least 3 hours after any manipulation requiring the PMT exposition to light.The PMT spectrum measured with an oscilloscope is shown in Fig. 18 (top), for the case of a PMTwith a high voltage of -950 V. The first peak (black markers) is obtained integrating the trace ina time window of 50 ns before the trigger and thus provides a measurement of the pedestal. Thesecond peak (red) is the PMT signal integrated in a time window of 50 ns after the trigger. Thetotal PMT spectral response, can be described by a simplified function [10]:– 19 – charge [pC]0.4 - - E n t r i e s / ndf c ped N 9.9 – ped m – - ped s – spe N 10.0 – spe m – spe s – - signal regionpedestal regionsum signal + pedestal charge [ADCu]60 80 100 120 140 160 180 E v en t s - - - - pingpong HV = 950 V – [ADCu] = 62.042 ped m – [ADCu] = 101.15 m – [ADCu] = 1.868 ped s – [ADCu] = 12.9 s Figure 18 . Top: Example of the PMT spectrum measured with an oscilloscope: in black and red the chargeintegrated in the pre-trigger (baseline) and in the signal regions (PE), respectively; in light blue the sum ofthe two distribution. The fit with the function (8.1) is shown as dashed red line. Bottom: example of thespectrum of a 3-inch PMT measured with CATIROC. Data acquired with the ping and pong capacitors canbe reconstructed separately guaranteeing that the resolutions are not degradated. f ( 𝑖 ) = 𝑁 Ped √ 𝜋𝜎 Ped exp (cid:32) − ( 𝑆 𝑖 − 𝜇 Ped ) 𝜎 (cid:33) + ∑︁ 𝑗> 𝑁 𝑗 √ 𝜋𝜎 𝑗 exp (cid:32) − ( 𝑆 𝑖 − 𝜇 𝑗 − 𝜇 Ped ) 𝜎 𝑗 (cid:33) (8.1)The first term is the gaussian fit of the pedestal, while the second term describes the PE spectrumin its general form, summing over 𝑗 > 𝑗 =
1– 20 – charge [ADCu]0 100 200 300 400 500 600 700 800 900 1000 E v en t s - - - - - - HV = 850 V 900 V 950 V1000 V1100 V1200 Vpingpong
PMT high voltage [V]850 900 950 1000 1050 1100 1150 1200 ga i n PMT 1, Ch 0PMT 1, Ch 3PMT 2, Ch 0 pingpong
Figure 19 . Top: charge spectrum measured for different HV. Bottom: Gain measured with CATIROC fortwo PMTs. The conversion from ADC to charge is done using the linear fit in Fig. 9. For a given PMT, themeasured gain is independent of the selected channel (red vs black) and of the use of ping or pong capacitors(empty and filled marker). For comparison, a second PMT (in blue) is tested using the same channels. (SPE). In the equation, 𝑆 𝑖 is the value of the 𝑖 -th ADC bin, 𝑁 Ped , 𝜇 Ped , 𝜎 Ped are the parameters ofthe gaussian fit to the pedestal distribution; 𝑁 𝑗 , 𝜇 𝑗 and 𝜎 𝑗 gives the amplitude, position and spreadof the 𝑗 -th photo-electrons. Hereafter, we denote the SPE mean position as 𝜇 SPE = 𝜇 − 𝜇 Ped . Theresults of the fit of equation 8.1 to the data are reported in figure, with an SPE position of 0.30 pC(which corresponds to a gain of 1 . × ) and a sigma of 31%.The charge spectrum measurement is repeated using the same PMT and the CATIROC evaluationboard . The slow control parameters are configured as for Fig. 9, with the exception of the trigger threshold that is here setto 950 DACu to observe also the pedestal peak. – 21 –he results are shown in Fig. 18 (bottom) with the fit function (eq. 8.1, with 𝑗 < =
1) drawn asa red dashed line. The SPE position, calculated after pedestal subtraction, is 39 ADCu, whichcorresponds to 0.3 pC when calibrated with the LSB in Table 2. The relative SPE resolution is 33%,well compatible with the result from the oscilloscope. As clearly visible in the figure, the spectraldistributions are slightly different for ping and pong data. This difference, mostly due to the offsetin the pedestal, is stable against the HV and is corrected once the proper calibration for ping andpong is applied (see Fig. 19).The SPE spectrum measured for different HV values is shown in Fig. 19 (top) for one PMT. TheSPE mean position ( 𝜇 SPE ) and spread ( 𝜎 SPE ) rise with HV as a consequence of the increasing PMTgain. At large HV, the SPE spectrum is better fitted including an exponential tail but we leave thisstudy to a dedicated paper on the PMT performance.Fig. 19 (bottom) shows the gain of two PMTs as a function of the applied HV. The measured gainfor ping and pong perfectly overlaps after calibration. Moreover, we confirm that the measuredgain is the same in different channels using the respective calibrations (red vs black). Moreover, wechecked that the pedestal is suppressed for increasing discriminator threshold without affecting theSPE distribution and we find out that with a threshold of 850 DACu we can cut out all signals witha charge smaller than 70 ADCu.Finally, we want to discuss briefly the effect of a clock-coupling which is visible as “wiggles” in themeasured charge spectrum (Fig. 18, bottom). This effect is well reproducible and is emphasized inFig. 20 (top) when comparing the bin content in each bin and the fit function in equation 8.1 for thecase 𝑗 =
1. Its trend can be modeled with a function 𝑁 𝑖 𝑁 fit , i = 𝑘 × sin ( 𝛼𝑥 + 𝜙 ) (8.2)where 𝑁 is the measured value and 𝑁 fit the one expected from the fit, and 𝑘 , 𝛼 and 𝜙 are theamplitude, frequency and phase of the modulation term. To evaluate the impact of the wiggles inthe determination of the PE position, a toy Monte Carlo test was performed (see Fig. 20, bottom).A Gaussian distribution with mean ( 𝜇 true ) and standard deviation ( 𝜎 true ) is simulated (blue shadedarea). The modulation effect is then applied, using eq. 8.2. The distorted distribution (solid black)is then fit with a Gaussian function and the derived 𝜇 PE and 𝜎 PE compared to the fit parametersfrom the gaussian fit of the true distribution. A discrepancy smaller than 0.01 ADCu and 0.3% for 𝜇 and 𝜎 , respectively is obtained and is mostly independent of the 𝜇 true and 𝜎 true , in the range ofthe typical values for PMTs. The CATIROC ASIC, developed for PMT-based applications, has been tested for its future appli-cation in Cherenkov and liquid scintillator experiments. The main features of CATIROC − theauto-trigger mode, the compressed digital output and the flexibility offered by the slow-controlparameters − are of high interests for next-generation experiments which are all facing systems witha large number of channels. In such cases, a full readout and storage of the signal waveform (e.g.,Flash ADC trace) is impracticable. The small amount of data to be transferred from the CATIROCand the number of input channels in a single chip allows to develop a simplified electronics. An– 22 – ntries 1000k 0.186 – a – f – -
250 260 270 280 290 300 charge [ADCu]0.8 - - - - - ] f i t r e s i dua l s [ y / y Entries 1000k 0.186 – a – f – - E n t r i e s ModifiedGaussSimpleGauss
Figure 20 . Top: Plot emphasizing the wiggles observed in the SPE spectrum and the fit function (reddashed line) with a sinusoid function. Bottom: a toy Monte-Carlo test applying the wiggles modulations ona gaussian SPE distribution, assuming pedestal subtracted. In shaded blue the original distribution and inblack after the modulation. The impact on the reconstructed mean SPE position is negligible (compare thetwo fits as red and blue dashed lines). – 23 –xample application of CATIROC is the readout of the 3-inch PMTs by the JUNO liquid-scintillatorneutrino experiment, under construction in China [1].The present paper gives a complete description of the CATIROC features and provides also afirst proof of its relevant use in a liquid scintillator experiment. Among the major features ofCATIROC described here, particular attention have been be payed to the dead times induced bythe handling of the triggers ( 𝑇 TrigDeadTime in sections 5 and 6.2) and by the digitization of thecharge ( 𝑇 DeadTime ∼ 𝜇 s with the mitigation offered by the SCA system). For those input pulsesnot producing an independent charge measurement because of 𝑇 TrigDeadTime , we proved that theircharge can be recovered as it summed up to the previous hit according to the charge acceptance inSection 6. Moreover we mentioned the possibility of using the analog signal from the discriminatorto count independent triggers, even when arriving within the 𝑇 TrigDeadTime . The dead times and theeffects of the slow-shaper are the most critical points for the application of CATIROC to liquidscintillator detectors. This is doable for PMTs operation in photon-counting mode for most of thephysics goals. For normal triggered pulses, we demonstrate the charge linearity for each of the inputchannels for both HG and LG circuits, which ensures a wide operation range, up to 70 pC, withoutsignal saturation. The achieved charge resolutions are about 15 fC in HG and 73 fC in LG mode.The trigger threshold can be set sufficiently low to measure charges down to 0.16 pC ( ∼ × ) which is the typical choice for PMT-based experiments. An optimizedvalue can be selected during the readout board validation and later on during the commissioningphase. The time resolution has been measured with independent approaches, achieving resultsbetter than 150 ps, with the possibility to improve it further correcting for non-linearities in theTDC ramp. Whereas this correction is not needed for experiments using PMTs (which have a timeresolution of few ns by their own), it offers an opportunity for the use of CATIROC beyond PMTdetectors (e.g. SiPM) and which require a resolution better than 100 ps. CATIROC offers someadditional flexibility which may be of convenient to adapt to the specific requirements of differentexperiments. The configurable preamplifiers for each of the 16 input channels allow to compensatefor differences in the gain of the PMT. This is particularly important when dealing with a very largenumber of PMTs, when a strict PMT selection is difficult to be handled.Finally, we presented extensive tests with some 3-inch PMTs in order to compare the CATIROCresponse to the one obtained with an independent readout system. Further measurements, donewith a large sample of JUNO PMTs and front-end boards confirmed these results but will be partof a future paper. Acknowledgments
The authors thank all the JUNO-SPMT collaborators, Jacques Wurtz and Andrea Triossi for supportand many fruitful discussions.
References [1] JUNO collaboration,
Neutrino physics with JUNO , arXiv:1507.05613[2] JUNO collaboration,
JUNO conceptual design report , arXiv:1508.07166.[3] Hyper-Kamiokande Proto-Collaboration,
Hyper-Kamiokande Design Report , arXiv:1805.04163. – 24 –
4] KM3NeT Collaboration,
KM3NeT 2.0 : Letter of Intent for ARCA and ORCA , J. Phys. G: Nucl. Part.Phys. 43 (2016) 084001.[5] S. Blin et. al.
Performance of CATIROC: ASIC for smart readout of large photomultiplier arrays ,JINST 12 C03041.[6] S. Conforti Di Lorenzo et al.,
PARISROC, an autonomous front-end ASIC for trigger-less acquisitionin next generation neutrino experiments , Nucl. Instum. Meth. A 695 (2012) 373.[7] B. Genolini et al.,
PMm2: large photomultipliers and innovative electronics for the next-generationneutrino experiments , Nucl. Instrum. Meth. A 610 (2009) 249, arXiv:0811.2681.[8] Miao He,
Double calorimetry system in JUNO , Radiation Detection Technology and Methods, 2017,1 (2) : 21.[9] Jilei Xu for the JUNO collaboration,
The JUNO double calorimetry system , Proceeding at NuFact2019, 26 - August 31, 2019, Daegu, Korea.[10] E.H. Bellamy, et al. , Absolute calibration and monitoring of a spectrometric channel using aphotomultiplier