Cosmic Ray Rejection and Readout Efficiency for La

arXiv:astro-ph/0005486v124May2000AcceptedbyPASP:19May2000CosmicRayRejectionandReadoutEfficiencyforLarge-AreaArraysD.J.Fixsen1,J.D.Offenberg1,R.J.Hanisch2,J.C.Mather3,M.A.Nieto-Santisteban2,R.Sengupta1,andH.S.Stockman2Subjectheadings:CosmicRay,datacompression,NGSTABSTRACTWepresentanalgorithmtooptimallyprocessuniformlysampledarrayimagedataobtainedwithanondestructivereadout.Thealgorithmdiscardsfullwells,removescosmicray(particle)hitsandotherglitches,andmakesanearlyoptimumestimateofthesignaloneachpixel.Thealgorithmalsocompressesthedata.Thecomputerrequirementsaremodest,andtheresultsarerobust.TheresultsareshownandcomparedtoresultsofFowlersampledandprocesseddata.Non-idealdetectorperformancemayrequiresomeadditionalcode,butthisisnotexpectedtocostmuchprocessingtime.Knowntypesofdetectorfaultsareaddressed.1.IntroductionTheoptimalrejectionofcosmicrayglitchesfromastronomicalimagesisofcriticalimportanceforlarge-areapixelizeddetectorsinspace.Thedetectors(CCDs,etc)aregenerallystableandrepeatable,sotheycanbecarefullycalibrated.Theyareoftensensitivetocosmicraysandotherradiation.Thesignalsfromcosmicrayscanbethelargestcontamination.ButthecontaminationisfarfromGaussian;ittendstobedominatedby“glitches”whichhavealargeeffectononeorafewpixelsforashortduration.Finding,limiting,andrejectingaffecteddataisaproblemcommontomanyobservationanddatareductionstrategies.Cosmicraysaffectground-baseddetectorsaswell,buttoamuchsmallerextent.Thisispartlybecausetheatmosphereandmagneticfieldactasashield,eliminatingmostofthecosmicrays,andpartlybecauseatmosphericemissionandscatteringarevariableandlimittheextenttowhichitispossibletouncoverandunderstandothersystematicerrors.1RaytheonITSS2SpaceTelescopeScienceInstitute3NASAGoddardSpaceFlightCenter–2–ThisstudywasfocusedontheNextGenerationSpaceTelescope(Stockmanetal.1998);however,manyoftheresultsareapplicableoverawiderangeofobservatories.AlthoughwespecificallyconsiderIRdetectors,theresultscanbeappliedtovisiblelightdetectorsoranydetectorswherebothreadoutnoiseandPoissonstatisticsarepresent.2.NoiseEstimationTheultimatelimitonthesignal-to-noiseratioisthePoissonvariationinthephotonarrivalrate.Forlargenumbersofphotons,N,andlowphotonoccupationnumbers,thevarianceisapproximatelyequaltothenumberofphotons,sothesignal-to-noiseratiois√Nifothersourcesofnoisearenegligible.Forafixedtelescope,thisisproportionalto√AZξT=√STwhereAistheareaofthetelescope,Zisthesourcestrength,ξistheefficiencyofthetelescopeanddetector,andTistheobservingtime.Othersourcesofnoisecanreducethesignal-to-noiseratio.TheratioofthePoissonvariancetotherealvarianceisameasureoftheefficiencyofthereadoutscheme.Varianceisusedratherthansigma,sotheefficiencyrelatesdirectlytotime.Inwhatfollows,wewillusethisefficiencytocomparevariousstrategies.InIRdetectors,thephotonscannotyetbereliablycountedindividually.Thedetectorandreadoutelectronicsaddnoise(Fansonetal.1998,Tianetal.1996).Thiscanbesummarizedasthereadoutnoise,R,expressedasanumberofphotonsorelectrons.Typicalvaluesrangefromafewto∼40eforwell-designedsystems.Anondestructivereadoutallowstheuseofmultiplesamplesonthesamepixelandelectronstoreducethereadouterror.However,theadditionalsamplesmakeamorecomplexsystemwithmoreelectronicsandcomputationalcomplexity.Theprocessofsamplingthedetectorcanaddelectronicnoisedirectly,orheatthedetector,causingadditionalnoise.Withthecurrentstateoftheartdetectors,thenoisecanbereducedbyincreasingthenumberofsamples.Thereareoftensourcesofnoisewithapowerspectrumthatscalesinverselywithfrequency.Althoughthisnoiseispoorlyunderstood,itisoftenpresent,andinmanydetectorsitbecomesthedominantsourceofnoiseatverylowfrequencies.Thissortofnoiseincreaseswithtimejustlikethephotonnoisefromasource,soitcanbetreatedlikeabackground(orforeground)contamination.Thereadoutnoisedecreaseswiththenumberofsamples,andthesignal(andPoissonvariance)increaseswithtime.Thusafterasufficienttimeandnumberofsamples,thereadoutnoisewillbeinsignificantrelativetothephotonnoise,andtheidealefficiencyofǫ=1willbeapproached.However,therearetwoobstaclestolongintegrationtimes.Atsomepoint,thewellsofthedetectorfillup,andthedetectorsbecomenonlinearandultimatelyinsensitivetoadditionalphotons.Also,particularlyinspacebaseddetectors,cosmicraysstrikethedetectoratarateof5-30cm−2sec−1(Tribble1995,Barth&Isaacs1999)(muchhigherinradiationbeltsorduringsolarflares)andaddalargenoisesignal.–3–3.SamplingStrategyThesamplingstrategyisanintegralpartofthelargerobservationstrategy.Thebeststrategydepends,ofcourse,onthegoaloftheobservation.Possiblegoalsincludefindingsources,surveyingaregionofthesky,identifyingsources,examiningasourceforvariation,ormappingdiffuseemission.Hereweconcentrateondetectingdimsourcesandassumethatbandlimitingfiltersareusedforpurposesofidentificationandanalysis.Ifonlyafixednumberofsamples,n,areavailable,thereisanadvantageofusingthematthebeginningandtheendoftheintegration(Fowlersampling,Fowler&Gatley1990,andFowleretal.1996).Thisapproachgivesthelongesteffectiveintegrationtime,andthevariancefromthereadoutnoiseisreducedby4/n,ifthesamplenoiseisuncorrelated.Halfofthesamplesareusedatth