Optimized next-neighbor image cleaning method for Cherenkov Telescopes
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
Optimized next-neighbor image cleaning method for Cherenkov Telescopes
M. S
HAYDUK , FOR THE
CTA C
ONSORTIUM . DESY Zeuthen [email protected]
Abstract:
In photo-sensor cameras of Cherenkov telescopes the light images from particle showers alwayscontain the background noise induced by photons of the night sky. An image cleaning procedure is needed toreduce the contribution of those noise photons in further analysis stages. The conventional topological next-neighbor method lacks reconstruction efficiency for low light content images and image peripheries with lowsignal levels. We present here a simple optimization of the traditional next-neighbor image cleaning method thatexploits the limited time duration of shower flashes and short time-difference between neighboring image pixels.This method reduces greatly the noise contribution by applying dynamical cuts in the parameter space formed bysignal amplitude and time-difference between neighboring pixels
Keywords: image cleaning, Cherenkov telescopes, light of the night sky.
Very high energy (VHE) ground-based g -ray astronomyhas rapidly developed over the last several decades. Withan introduction of the Imaging Atmospheric Cherenkovtelescopes technique, pioneered by the Whipple [1],[2] andHEGRA [3] collaborations it has reached a very produc-tive state with the currently operating experiments H.E.S.S.[5], MAGIC [4] and VERITAS [6]. The next generationground-based g -ray instrument - Cherenkov Telescope Ar-ray (CTA) [7] is aiming to reach an order of magnitudehigher sensitivity, compared to currently running facilities,extending the energy range at the same time. It will con-sist of about 60 prime-focus Cherenkov telescopes withdifferent size and will be extended with double-mirrorSchwarzschild-Coude telescopes to further improve angu-lar resolution and sensitivity over wider field of view.One of the key component of these instruments is theimaging camera that usually comprises a large number ofphotosensors ( ∼ ∼ The ultimate task of an image cleaning procedure is todetermine the maximal amount of pixels with the signalfrom the shower, keeping to a minimum the number ofpixels with noise signals. Thus, a maximum of informationinitially contained in the data will be provided for furtheranalysis.One should keep in mind that larger numbers of recon-structed pixels do not always result in better performanceof the gamma/hadron separation, since usually analysismethods are strongly optimized to traditional image pro-cessing and signal extraction methods. Therefore, to real-ize the full potential of more complete shower images anal-ysis methods should be widely revised and adopted, but wewould like to keep these issues outside of the scope of thepresent paper.In the traditional image cleaning method pixels withbackground signals are rejected by applying a charge cut(core threshold) on the pixel signal. Those pixels whichwere not rejected by the charge cut are assigned to be animage core candidate. Next, the topological condition ofhaving at least one neighbor is applied and image core can-didates without a neighbor are discarded. In the last step,the vicinity of the selected core pixels is revised againand the lower charge cut (boundary threshold) is appliedto complete the image with boundary pixels. The neces-sary threshold values scale with the root mean square ofthe background fluctuations. Here one should note that theshort time duration and fine time structure of shower im-ages is not taken into account, hence all noise signals in-duced within the readout window of the data acquisitionsystem are contributing to the background.A natural improvement of the traditional image cleaningmethod can be achieved by considering the additional im-age time structure information, so that for neighboring pix-els not only signal amplitudes, but also arrival time differ-ences are examined. In this fashion, the unnecessary largephase space of accepted noise can be significantly reducedand more closely match to the shower image phase space.Moreover, the next-neighbor groups of pixels with multi-plicities larger then two can be considered as units for im-age formation. These ideas were successfully implemented
CRC 2013 Template33 ND I NTERNATIONAL C OSMIC R AY C ONFERENCE , R
IO DE J ANEIRO
Charge, phe R a t e , H z LST MST SST
Figure 1 : Differential charge spectra in a single pixel forthree types of CTA telescopes under exposure to night skybackground light of typical intensities for extra-galacticobservations (Monte-Carlo simulations). The charge wasintegrated over 4 ns time slice and converted to photo-electrons (phe). LST, MST and SST acronyms stand forLarge-, Medium- and Small- Size Telescope accordingly.The transition region from noise charges, induced by lightof the night sky to the regime, dominated by photosensorafter-pulses is revealed at ∼ Q in the group and maximal timedifference D T .Meanwhile, the noise differential probability contour inthis Q - D T phase plane is obviously a smooth functionof threshold and coincidence time. This issue has becomeimportant with the demanding requirements of CTA, espe-cially considering the much larger energy range and cam-era field of view, the low light content images could havesignificantly broader time spread, compared to typical im-ages of low intensity in the MAGIC telescope data. , phes Q T , n s D -4 -3 -2 -1 Figure 2 : Parameter plane, formed by minimal charge Q and maximal coincidence time D T in the tested next-neighbor group. Contours of constant rate are shown forrate value of ∼ D T range of 1-10 ns.The natural way to adopt the next-neighbor cleaningalgorithm to the extended phase space of shower images isto use a dynamical cut in the group Q - D T parameter plane.The shape of this dynamical cut curve can be obtainedfrom noise properties of the individual pixel, assuming thatall pixels in the camera are the same in this term.The typical differential noise rate in one pixel is pre-sented in Fig. 1. Noise charges higher then ∼ ´ acc of the next-neighbor groups in the camera with multiplicity n can bederived from the individual pixel noise rate R pix ( Q ) as fol-lowing: ´ acc ( n , Q , D T ) = C n · D T n − · R pix ( Q ) n , (1)where Q - is the charge in one pixel, D T - is the time co-incidence window and C n - combinatorial factor, depend-ing on the group multiplicity and total number of the pix-els in the camera. The contour for the constant differentialrate ´ acc = const is described by the following equationfor the time coincidence D T : D T ( ´ acc [ Hz ] , Q ) = exp [ n − · ln ( ´ acc [ Hz ] C n · R pix ( Q ) n )] (2)This contour can serve as a dynamic cut in the Q − D T parameter plane, so that any next-neighbor group withparameter values, above this contour should be rejectedas a background group. The value of the differential rate CRC 2013 Template33 ND I NTERNATIONAL C OSMIC R AY C ONFERENCE , R
IO DE J ANEIRO
E, Log([GeV]) , q D i r ec t i on o ff se t Figure 3 : Reconstructed direction offset q = | q true − q rec | versus energy. The curve represents the dynamic cut, ap-plied in the analysis of g -ray events and corresponds to thetypical angular resolution for arrays like H.E.S.S and VER-ITAS. ´ acc [ Hz ] can be chosen, according to the desired prob-ability level of accidental images. Contours of constantaccidental rate for next-neighbor groups, calculated withequation 2 are shown in Fig.2. Thresholds for core andboundary pixel search from the traditional image cleaningmethod would appear in the plot as vertical lines at posi-tions of 10 and 5 phes correspondingly. The 2-nn groupthreshold curve shape is fully determined by the photo-sensor after-pulsing additional noise for coincidence timesabove ∼ The performance of the optimized next-neighbor imagecleaning method was studied with Monte-Carlo data forthe CTA experiment. The g -ray extensive air showers weresimulated with a power law energy spectrum dN / dE ∼ E − , using the SimTel package [11]. The telescope triggerswere simulated in great detail by the trigger simulationcode
TrigSim [12].The photoelectrons, induced by showers in imagingcameras were converted to Flash ADC traces and the cor-responding background light of the night sky with intrin-sic noise of photosensors were added to the camera data.From these simulated Flash ADC traces the charge andthe arrival time in every camera pixel were extracted andthe image cleaning procedure was applied. The shower di-rection reconstruction was performed with the VERITASanalysis code [13]. Since events with poor/failed directionreconstruction are usually useless for next analysis stepsthey were discarded by the dynamic direction offset cut,presented in Fig.3. Then the number of well-reconstructedevents in each energy bin was compared for two imagecleaning methods: traditional 10/5 phes and the optimizednext-neighbor image cleaning. The result of this compari-
E, Log([GeV]) N u m b e r o f eve n t s , c oun t s Dynamical cutsStandard 10/5 phe cuts
Figure 4 : Number of events with well-reconstructed direc-tion versus g -ray energy for LST telescopes. Monte-Carlo g -ray events were simulated with power law dN / dE ∼ E − a , a =
2. The data was processed with traditional im-age cleaning with thresholds of 10 and 5 phes for core andboundary pixels accordingly (circles) and the new imagecleaning, described in this paper (triangles). The gain innumber of events for optimized next-neighbor image clean-ing algorithm is very prominent for energies below 100GeV.son for CTA Large Size Telescopes is shown in Fig.4, re-vealing the remarkable gain for the new method in the en-ergy range of 10-100 GeV.
The suggested image cleaning method exploits the timestructure of the shower flash, but does not constrain thetotal time spread of the image, can be applied to eventswithin a wide energy range and keeps more image features,providing room for novel sophisticated g /hadron separa-tion methods. Acknowledgment:
The ICRC 2013 is funded by FAPERJ,CNPq, FAPESP, CAPES and IUPAP.
References ND I NTERNATIONAL C OSMIC R AY C ONFERENCE , R
IO DE J ANEIRO2013Cherenkov Telescope Array facility”, Nucl.Instrum.Meth.A617 (2010) 422-423[11] K.Bernl¨ohr, Astroparticle Physics 30 (2008) 149 ; K.Bernl¨ohr, AIP Conf. Proc., vol.1085, 2008, p.874.[12] R. Wischnewski et. al., ”Performance Study of a DigitalCamera Trigger for CTA”, Proc. of the 32nd InternationalCosmic Ray Conference, Beijing, 2011, pp 63,vol 9[13]