Comparison of Different Trigger and Readout Approaches for Cameras in the Cherenkov Telescope Array Project
aa r X i v : . [ a s t r o - ph . I M ] J u l ND I NTERNATIONAL C OSMIC R AY C ONFERENCE , R
IO DE J ANEIRO T HE A STROPARTICLE P HYSICS C ONFERENCE
Comparison of Different Trigger and Readout Approaches for Cameras in theCherenkov Telescope Array Project
M. S
HAYDUK , S. V OROBIOV , U. S CHWANKE , R. W ISCHNEWSKI FOR THE
CTA C
ONSORTIUM . DESY Zeuthen Humboldt University, Berlin [email protected]
Abstract:
The Cherenkov Telescope Array (CTA) is a next-generation ground-based observatory for g -rayswith energies between some ten GeV and a few hundred TeV. CTA is currently in the advanced design phase andwill consist of arrays with different size of prime-focus Cherenkov telescopes, to ensure a proper energy coveragefrom the threshold up to the highest energies. The extension of the CTA array with double-mirror Schwarzschild-Couder telescopes is planned to improve the array angular resolution over wider field of view. We present an end-to-end Monte-Carlo comparison of trigger concepts for the different imaging cameras that will be used on theCherenkov telescopes. The comparison comprises three alternative trigger schemes (analog, majority, flexiblepattern analysis) for each camera design. The study also addresses the influence of the properties of the readoutsystem (analog bandwidth of the electronics, length of the readout window in time) and uses an offline showerreconstruction to investigate the impact on key performances such as energy threshold and flux sensitivity. Keywords: gamma-rays, Cherenkov telescopes, trigger, readout.
The emerging success of the very high energy ground-based g -ray astronomy was prominently ensured by theImaging Cherenkov telescopes technique [1],[2]. The cur-rently operating Cherenkov telescope experiments likeH.E.S.S. [3], MAGIC [4] and VERITAS [5] have al-ready proved to be very capable instruments to study thevery high energy astrophysical phenomena both in ourgalaxy and in extragalactic sources. The next generationground-based g -ray experiment - Cherenkov Telescope Ar-ray (CTA) Observatory [6] is currently in the preparatoryphase. It will provide an order of magnitude higher sen-sitivity and extend the observable g -ray energy range upto hundreds of TeV. The CTA will comprise about 60Cherenkov telescopes with different sizes of the reflectorand will be extended with an array of double-mirror tele-scopes based on Schwarzschild-Coude optical design.The Monte Carlo studies for CTA done with the firstmassive production of simulations [7] were following aconventional majority/next-neighbor logic for the singletelescope trigger, requiring for any pixel that some numberof its direct neighbors must have signal of a certain ampli-tude within a given coincidence time. Moreover, the read-out system concept and signal extraction methods wererather following the present generation H.E.S.S array.The current hardware developments in CTA are ongo-ing under considerations of reliability and cost, while keep-ing the requested performance. These developments areaiming to optimize many of telescope components, that re-sulted in several design options for telescope triggers andreadout systems . The proper implementation of all thesenovel designs in the Monte Carlo simulations provides im-portant information to decide what will be finally built. The detailed Monte-Carlo simulations of the camera trig-ger and data acquisition system were performed with trigsim package. The software package details and resultsof first studies are presented in [8]. There are several trig-ger designs currently considered in CTA, as described in[7]:
Majority Trigger:
Each of the analogue pulses comingfrom photomultipliers in a predefined overlapping cameraregion is fed into the comparator, which produces the dig-ital signal if the initial pulse exceeds the adjustable refer-ence amplitude. Then the sum of these digital signals againcan be passed to the comparator with a certain threshold tocount the number of pixels and issue the final camera trig-ger
Analogue Sum Trigger:
The analogue pulses from allpixels in a predefined overlapping camera region (see ex-amples in Fig.1 ) are added, regardless to their ampli-tude and then compared to the reference threshold. In or-der to avoid that photomultiplier after-pulses dominate thesummed signal, the amplitude of analogue pulses is lim-ited to a certain value before summation (clipping thresh-old).
Binary Trigger:
The analogue signal from the photosen-sors are passed through comparators at regular time inter-vals, transforming the camera image to a binary pattern.This pattern can be processed with flexible trigger FPGA-based classification algorithms, considering the space andtime properties of the data. With additional thresholds, thecamera image can be converted to the several-bit patternand the trigger approach can emulate on-line image pro-cessing, similar to image cleaning procedures used in theoff-line data analysis.
Digital Trigger:
This concept is the interplay betweenthe binary trigger algorithm and the fully analogue ap-proach, described above. The signal from the photosensoris digitized by a Flash-ADC and the digital signal is passedto the flexible FPGA-based trigger logic. In this FPGA
Comparison of Different Trigger and Readout Approaches for Cameras in the CTA Project”33 ND I NTERNATIONAL C OSMIC R AY C ONFERENCE , R
IO DE J ANEIRO
Figure 1 : Example of the Middle Size Telescope (MST)camera geometrical structure. Each hexagon represents a 7-pixel-cluster. The dashed red upper circles indicate 3 neigh-bored supercluster (each made of 7 next neighbor clus-ters, corresponding to 49 pixels, which is the region pro-cessed by the trigger cluster-FPGA.). The baseline triggerpatches geometries: Singlets, Doublets and Triplets (con-tained within grey contours) are shown and also given inthe inset.module, the large variety of trigger logic schemes couldbe implemented, including a digital sum trigger or a dig-ital majority trigger. This trigger approach has an elegantfeature that the trigger Flash-ADC data is used as well asthe event data, reduces the amount of front-end electron-ics components (discriminators and comparators), but cur-rently lacks cost-effective solutions for designs faster than250MHz.All of these concepts were implemented in trigsim andstudied in detail by comparing telescopes collection areas.The influence of the camera electronics analog bandwidthon the g -ray collection areas and the optimal analog sig-nal pulse width values were discussed in [8]. The lowestenergy thresholds were obtained for the trigger implemen-tations with the faster pulses, especially for LST part ofthe array. Fort the MST and SST the solutions with slowerpulses had tendency to possess ∼
15% gain in trigger effi-ciency for energies above ∼
500 GeV.The possible basic shapes of camera regions, associatedwith the trigger patches are presented in Fig.1. In our no-tation, the one hardware unit is the cluster of 7 pixels. Thetrigger geometrical patches formed by one, two and threesuch clusters, called ”Singlets”, ”Doublets” and ”Triplets”correspondingly. The data from these patches, overlappingas shown in Fig.1, is examined by the trigger logic to is-sue the patch trigger. The final camera trigger is the logi-cal ”OR” of all patch triggers. The optimal patch area andshape can differ for different trigger concepts and the de-voted simulations are ongoing.The trigger threshold of the Cherenkov telescope is de-termined by the intensity of the night sky background light,which can dramatically vary during observations, depend-ing on the telescope pointing position and the moon bright-ness. Moreover, the intrinsic noise of photosensors, so-called after-pulses, contributes to the camera noise rate andaffect the telescope performance. Therefore, these aspects
Charge, phe R a t e , H z LST (nominal) MST (nominal) SST (nominal) LST (NSBx4.5) MST (NSBx4.5) SST (NSBx4.5)
Figure 2 : Differential charge spectra in a single pixel fortwo levels of the night sky background light intensities:”nominal” denotes the typical intensity level for extra-galactic observations, and ”NSBx4.5” - galactic/moonlightobservation. The ”LST”, ”MST” and ”SST” labels standfor Large-, Medium- and Small- Size Telescope accord-ingly. The single photoelectron pulses are gaussian withFWHM=2.6 ns. For high charges (above 6 phe for nominaland 9 phe for high intensity) the noise spectrum is fully de-termined by the photosensor after-pulses. Rate values satu-rate at 2 · Hz due to the 50 ns gate-width of the counter.should be studied as well and the corresponding perfor-mances of all trigger concepts should be investigated. TheMonte-Carlo simulation for the individual pixel noise spec-tra, induced by the light of the night sky with nominal and4.5 times higher intensity are presented in Fig.2.The accidental camera trigger rates for all consideredtrigger concepts with different intrinsic scenarios areshown in Fig.3. The camera trigger threshold is definedas the threshold value, corresponding to 10kHz rate. Fol-lowing the notation of trigger patches defined in Fig.1 thelabels ”SumSingl”, ”SumDoubl” and ”SumTripl” standfor the
Analogue Sum Trigger approach with the intrin-sic scenario of summing the analogue signals from the de-scribed above trigger Singlets (1 cluster, 7 pixels), Dou-blets (2 clusters, 14 pixels) and Triplets (3 clusters, 21pixels). The single photoelectron pulse is a gaussian withFWHM=2.6 ns. Similarly, the ”DigitalSingl”, ”Digital-Doubl” and ”DigitalTripl” labels denote the digital sumover the same patches in
Digital Trigger concept. In orderto match the 250MHz sampling rate of the Flash-ADC theinitial analogue pulse has a special shape, that after FPGAprocessing is roughly equivalent to the gaussian pulse withFWHM ≈
10 ns (for details, see [9]).Trigger schemes listed in lower legends in Fig.3 arethe fast (analogue pulse FWHM=2.6 ns)
Majority Trig-ger algorithms, designated as following: ”ScSglMaj3” -any 3 pixels out of 7 - pixel cluster should have a signalabove a reference threshold within ∼ Comparison of Different Trigger and Readout Approaches for Cameras in the CTA Project”33 ND I NTERNATIONAL C OSMIC R AY C ONFERENCE , R
IO DE J ANEIRO
Threshold, phes A cc i d e n t a l R a t e , H z MST, NSB=104.0 MHzSumSinglSumDoublSumTriplDigitalSinglDigitalDoublDigitalTriplScSglMaj3IC2nnMaj3IC2nnMaj4TriplMaj5TriplMaj7
Threshold, phes A cc i d e n t a l R a t e , H z MST, NSB=467.0 MHzSumSinglSumDoublSumTriplDigitalSinglDigitalDoublDigitalTriplScSglMaj3IC2nnMaj3IC2nnMaj4TriplMaj5TriplMaj7
Figure 3 : Accidental MST camera rates as a function of discriminator threshold for different flavors of
Analogue Sum (labels with prefix ”Sum”),
Majority (ScSgMaj3, IC2nnMaj3, IC2nnMaj4) and
Digital (labels with prefix ”Digital”)trigger approaches, for standard dark sky NSB rate (left plot) and for 4.5 times larger rate (right plot). For
Majority schemes the threshold value correspond to the individual pixel discriminator reference amplitude, while for the
AnalogueSum and
Digital approaches the threshold value is for the trigger patch discriminator. The two-threshold
Binary trigger approach can be obtained from the majority triggers ”IC2nnMaj3” and ”IC2nnMaj4” as the logical ”OR”.fancy ”IC2nnMaj3” and ”IC2nnMaj4” labels denote themajority trigger algorithms with slightly modified logic.The condition of short coincidence time is only requiredfor pairs of neighboring pixels, in contrast to the conven-tional majority scheme, where this condition is examinedfor all triggered pixels. Simultaneous implementation ofthese two triggers in the FPGA can serve as the cameratwo-threshold
Binary Trigger . One of the key characteristics of the readout system is thethroughput analog bandwidth. Relevantly to the scope ofCherenkov Telescopes, it essentially determines the mini-mal possible noise contribution to the recorded signal [10].We compare here readout system approaches differingbasically by the analog bandwidth settings: the high band-width possesses fast gaussian analogue pulses with 2.6ns FWHM, digitized with 1GHz Flash-ADC and the lowbandwidth approach with 250MHz Flash-ADC samplingrate and slower analogue pulses with FWHM=10.4 ns. Forall of these options events were recorded with wide 50 nsdefault readout window, dynamically extended up to 100ns for events with time duration longer than the defaultwindow. For short integration windows, the reduction ofthe readout window will allow to lower image cleaningthresholds, since the signal position search range is con-siquentely shrinks, but potentially can lead to the loss ofsignal for high-energy events with the large time spread.The optimization of readout algorithms in this respect isongoing [11].
As an input for our simulations we used the data, pro-duced by simtelarray program [12]. This package includes full detector simulations [7] and provides arrival times ofphotoelectrons (p.e.) from showers at the telescope focalplanes, relevant for our studies. Next, in the framework of trigsim we simulated the electronic response of the pho-tomultiplier tube (PMT) response by convolving the p.e.times with the individual single p.e. pulses from the CTAPMT candidate. The pulses have been widened and sam-pled according to the bandwidth and the sampling rate ofthe studied front-end electronics designs.The simulated data have been analyzed using evndisplay software package for the analysis of simulations and databy arrays of imaging Cherenkov telescopes [13], originallydeveloped for VERITAS experiment, and further extendedto AGIS and CTA arrays [14, 15]. We implemented con-version from sim telarray [12]
EVENTIO format to evndis-play format within trigsim , in order to analyze all triggerpatterns. At the conversion stage, the NSB contribution toFlash-ADC traces is added to all pixels.The simulated FADC trace integration, data calibrationand the selection of triggered events is being done within evndisplay , using trigger thresholds values obtained with trigsim . For the subsequent image cleaning, the novel im-age cleaning procedure [16] has been implemented. Thenthe second-moment parameterization of the recorded im-ages and stereo reconstruction of shower geometry (direc-tion and impact parameter) is performed. The gamma -rayenergy is estimated using beforehand trained lookup tables.At the next step we discard events with poor/failed direc-tion reconstruction by applying the dynamic direction off-set cut which corresponds to the typical angular resolutioncurve for arrays like H.E.S.S. and VERITAS.
For the comparison of the trigger and readout approacheswe selected following camera electronic chains: the
Comparison of Different Trigger and Readout Approaches for Cameras in the CTA Project”33 ND I NTERNATIONAL C OSMIC R AY C ONFERENCE , R
IO DE J ANEIRO
E, Log([GeV]) / N R a t i o , N NSBx1NSBx4.5 (LowBW) evts
N (highBW) evts
NLST:
E, Log([GeV]) / N R a t i o , N NSBx1NSBx4.5 (LowBW) evts
N (highBW) evts
NMST:
Figure 4 : Number of reconstructed events after direction cut for the high bandwidth camera approach, normalized to thenumber of reconstructed events for the low bandwidth solution.
Left:
LST part of array ”E”. Nominal and high NSBconditions are labelled with circles and triangles correspondingly.
Right:
MST part of array ”E”. Significant gain in thenumber of events in the low energy range for arrays equipped with high BW cameras is revealed.high bandwidth chain (denoted in Fig.4 as ”high BW”),equipped with
Analogue Sum trigger with fast 2.6 ns gaus-sian pulses and 1GHz Flash-ADC readout system and thelow bandwidth chain (”low BW”), comprising
AnalogueTrigger with slower pulses of 10.4 ns and correspondinglyslower 250MHz readout. It was demonstrated in our previ-ous studies [8], that the trigger scenario of direct summa-tion of pulses, irrelevant to their amplitudes, leads to lowerenergy thresholds (for Doublets), compared to the conven-tional majority schemes. Thus we selected this scenariofor both ”high BW” and ”low BW” approaches.After the analysis steps described in the previous sec-tion, the number of well-reconstructed events for both cam-era solutions is examined. The number of events for highBW approach, normalized to the number of events for lowBW system is shown on Fig. 4. The ratio curves for thenominal and the high (4.5 times brighter) light of the nightsky level are depicted with circles and triangles accord-ingly. As it can be seen, the high BW design providesslightly more events for further analysis up to the 1 TeVenergies, gaining more prominently in the near-thresholdenergy region. For the LST subarray the gain of exploit-ing the high BW camera solution is substantial for all rel-evant energies (below 100 GeV). Moreover, the gain forthe high BW approach increases for the conditions of thebright night sky background and extends to the whole en-ergy range .
Lowering energy threshold and improving sensitivity oflarge and medium CTA telescope size sections will bebeneficial for the long-term monitoring of AGNs, the de-tectability of distant AGNs and gamma-ray bursts, pulsarsand for the variability studies. The R & D work with novelphotosensors and high bandwidth front-end camera elec-tronics performed by the CTA consortium [6, 7] is verypromising for improving the array performances. We ex-tensively studied the impact of various trigger schemes andthe rapidity of the front-end electronics on CTA gamma ac-ceptances. The study on flux sensitivity and other key per-formances are ongoing.The observed lowering of energy threshold and improve- ment of gamma acceptances for high bandwidth designsis moderate. However, there is a room for improvementon the analysis side (e.g. using pixel-by-pixel image tem-plates [17] and / or DISP-like [18] methods etc.), whichwould further clarify the obtained performance differencebetween two approaches. Next steps of our work will incor-porate some of these advanced image analysis methods, aswell the next round of CTA simulations, which account forupdated parameters of CTA candidate photosensors, front-end electronics and other detector components.
Acknowledgment:
References [1] Weekes T C et. al., 1989 Astrophys. J. 342 379[2] O.C. Allkofer et. al., in: Proc. of the 19th InternationalCosmic Ray Conference, La Jolla, vol. 3, 1985, p. 418.[3] [4] [5] http://veritas.sao.arizona.edu/ [6] ”Design Concepts for The Cherenkov Telescope Array”,The CTA Consortium, (2010), arXiv 1008.3703[7] K. Bernl¨ohr, et al. (for the CTA Consortium), AstroparticlePhysics 43 (2013) 171-188[8] R. Wischnewski et. al., in: Proc. of the 32nd InternationalCosmic Ray Conference, Beijing, 2011, pp 63,vol 9[9] G. P¨uhlhofer et. al., in: Proc. of the 332th InternationalCosmic Ray Conference, Beijing, 2011, pp 138, vol 9.[10] S. Vorobiov et al., NIM A,Volume 695, 11 December 2012,Pages 394-397[11] C.L.Naumann et al., NIM A, Volume 695, 11 December2012, Pages 44-51[12] K.Bernl¨ohr, Astroparticle Physics 30 (2008) 149 ; K.Bernl¨ohr, AIP Conf. Proc., vol.1085, 2008, p.874.[13] https://wiki-zeuthen.desy.de/CTA/Eventdisplay%20Softwarehttps://wiki-zeuthen.desy.de/CTA/Eventdisplay%20Software