| 1 | /* ======================================================================== *\
|
|---|
| 2 | ! $Name: not supported by cvs2svn $:$Id: MCalibrationChargeCalc.cc,v 1.170 2006-10-23 12:27:21 tbretz Exp $
|
|---|
| 3 | ! --------------------------------------------------------------------------
|
|---|
| 4 | !
|
|---|
| 5 | ! *
|
|---|
| 6 | ! * This file is part of MARS, the MAGIC Analysis and Reconstruction
|
|---|
| 7 | ! * Software. It is distributed to you in the hope that it can be a useful
|
|---|
| 8 | ! * and timesaving tool in analysing Data of imaging Cerenkov telescopes.
|
|---|
| 9 | ! * It is distributed WITHOUT ANY WARRANTY.
|
|---|
| 10 | ! *
|
|---|
| 11 | ! * Permission to use, copy, modify and distribute this software and its
|
|---|
| 12 | ! * documentation for any purpose is hereby granted without fee,
|
|---|
| 13 | ! * provided that the above copyright notice appear in all copies and
|
|---|
| 14 | ! * that both that copyright notice and this permission notice appear
|
|---|
| 15 | ! * in supporting documentation. It is provided "as is" without express
|
|---|
| 16 | ! * or implied warranty.
|
|---|
| 17 | ! *
|
|---|
| 18 | !
|
|---|
| 19 | !
|
|---|
| 20 | ! Author(s): Markus Gaug 02/2004 <mailto:markus@ifae.es>
|
|---|
| 21 | !
|
|---|
| 22 | ! Copyright: MAGIC Software Development, 2000-2004
|
|---|
| 23 | !
|
|---|
| 24 | !
|
|---|
| 25 | \* ======================================================================== */
|
|---|
| 26 | //////////////////////////////////////////////////////////////////////////////
|
|---|
| 27 | //
|
|---|
| 28 | // MCalibrationChargeCalc
|
|---|
| 29 | //
|
|---|
| 30 | // Task to calculate the calibration conversion factors and quantum efficiencies
|
|---|
| 31 | // from the fit results to the summed FADC slice distributions delivered by
|
|---|
| 32 | // MCalibrationChargeCam, MCalibrationChargePix, MCalibrationChargeBlindPix and
|
|---|
| 33 | // MCalibrationChargePINDiode, calculated and filled by MHCalibrationChargeCam,
|
|---|
| 34 | // MHCalibrationChargePix, MHCalibrationChargeBlindPix and MHCalibrationChargePINDiode.
|
|---|
| 35 | //
|
|---|
| 36 | // PreProcess(): Initialize pointers to MCalibrationChargeCam, MCalibrationChargeBlindPix
|
|---|
| 37 | // MCalibrationChargePINDiode and MCalibrationQECam
|
|---|
| 38 | //
|
|---|
| 39 | // Initialize pulser light wavelength
|
|---|
| 40 | //
|
|---|
| 41 | // ReInit(): MCalibrationCam::InitSize(NumPixels) is called from MGeomApply (which allocates
|
|---|
| 42 | // memory in a TClonesArray of type MCalibrationChargePix)
|
|---|
| 43 | // Initializes pointer to MBadPixelsCam
|
|---|
| 44 | //
|
|---|
| 45 | // Process(): Nothing to be done, histograms getting filled by MHCalibrationChargeCam
|
|---|
| 46 | //
|
|---|
| 47 | // PostProcess(): - FinalizePedestals()
|
|---|
| 48 | // - FinalizeCharges()
|
|---|
| 49 | // - FinalizeFFactorMethod()
|
|---|
| 50 | // - FinalizeBadPixels()
|
|---|
| 51 | // - FinalizeBlindCam()
|
|---|
| 52 | // - FinalizePINDiode()
|
|---|
| 53 | // - FinalizeFFactorQECam()
|
|---|
| 54 | // - FinalizeBlindPixelQECam()
|
|---|
| 55 | // - FinalizePINDiodeQECam()
|
|---|
| 56 | //
|
|---|
| 57 | // Input Containers:
|
|---|
| 58 | // MCalibrationChargeCam
|
|---|
| 59 | // MCalibrationChargeBlindPix
|
|---|
| 60 | // MCalibrationChargePINDiode
|
|---|
| 61 | // MCalibrationQECam
|
|---|
| 62 | // MPedestalCam
|
|---|
| 63 | // MBadPixelsCam
|
|---|
| 64 | // MGeomCam
|
|---|
| 65 | // MTime
|
|---|
| 66 | //
|
|---|
| 67 | // Output Containers:
|
|---|
| 68 | // MCalibrationChargeCam
|
|---|
| 69 | // MCalibrationChargeBlindPix
|
|---|
| 70 | // MCalibrationChargePINDiode
|
|---|
| 71 | // MCalibrationQECam
|
|---|
| 72 | // MBadPixelsCam
|
|---|
| 73 | //
|
|---|
| 74 | //
|
|---|
| 75 | // Preliminary description of the calibration in photons (email from 12/02/04)
|
|---|
| 76 | //
|
|---|
| 77 | // Why calibrating in photons:
|
|---|
| 78 | // ===========================
|
|---|
| 79 | //
|
|---|
| 80 | // At the Barcelona meeting in 2002, we decided to calibrate the camera in
|
|---|
| 81 | // photons. This for the following reasons:
|
|---|
| 82 | //
|
|---|
| 83 | // * The physical quantity arriving at the camera are photons. This is
|
|---|
| 84 | // the direct physical information from the air shower. The photons
|
|---|
| 85 | // have a flux and a spectrum.
|
|---|
| 86 | //
|
|---|
| 87 | // * The photon fluxes depend mostly on the shower energy (with
|
|---|
| 88 | // corrections deriving from the observation conditions), while the photon
|
|---|
| 89 | // spectra depend mostly on the observation conditions: zenith angle,
|
|---|
| 90 | // quality of the air, also the impact parameter of the shower.
|
|---|
| 91 | //
|
|---|
| 92 | // * The photomultiplier, in turn, has different response properties
|
|---|
| 93 | // (quantum efficiencies) for photons of different colour. (Moreover,
|
|---|
| 94 | // different pixels have slightly different quantum efficiencies).
|
|---|
| 95 | // The resulting number of photo-electrons is then amplified (linearly)
|
|---|
| 96 | // with respect to the photo-electron flux.
|
|---|
| 97 | //
|
|---|
| 98 | // * In the ideal case, one would like to disentagle the effects
|
|---|
| 99 | // of the observation conditions from the primary particle energy (which
|
|---|
| 100 | // one likes to measure). To do so, one needs:
|
|---|
| 101 | //
|
|---|
| 102 | // 1) A reliable calibration relating the FADC counts to the photo-electron
|
|---|
| 103 | // flux -> This is accomplished with the F-Factor method.
|
|---|
| 104 | //
|
|---|
| 105 | // 2) A reliable calibration of the wavelength-dependent quantum efficiency
|
|---|
| 106 | // -> This is accomplished with the combination of the three methods,
|
|---|
| 107 | // together with QE-measurements performed by David in order to do
|
|---|
| 108 | // the interpolation.
|
|---|
| 109 | //
|
|---|
| 110 | // 3) A reliable calibration of the observation conditions. This means:
|
|---|
| 111 | // - Tracing the atmospheric conditions -> LIDAR
|
|---|
| 112 | // - Tracing the observation zenith angle -> Drive System
|
|---|
| 113 | //
|
|---|
| 114 | // 4) Some knowlegde about the impact parameter:
|
|---|
| 115 | // - This is the only part which cannot be accomplished well with a
|
|---|
| 116 | // single telescope. We would thus need to convolute the spectrum
|
|---|
| 117 | // over the distribution of impact parameters.
|
|---|
| 118 | //
|
|---|
| 119 | //
|
|---|
| 120 | // How an ideal calibration would look like:
|
|---|
| 121 | // =========================================
|
|---|
| 122 | //
|
|---|
| 123 | // We know from the combined PIN-Diode and Blind-Pixel Method the response of
|
|---|
| 124 | // each pixel to well-measured light fluxes in three representative
|
|---|
| 125 | // wavelengths (green, blue, UV). We also know the response to these light
|
|---|
| 126 | // fluxes in photo-electrons. Thus, we can derive:
|
|---|
| 127 | //
|
|---|
| 128 | // - conversion factors to photo-electrons
|
|---|
| 129 | // - conversion factors to photons in three wavelengths.
|
|---|
| 130 | //
|
|---|
| 131 | // Together with David's measurements and some MC-simulation, we should be
|
|---|
| 132 | // able to derive tables for typical Cherenkov-photon spectra - convoluted
|
|---|
| 133 | // with the impact parameters and depending on the athmospheric conditions
|
|---|
| 134 | // and the zenith angle (the "outer parameters").
|
|---|
| 135 | //
|
|---|
| 136 | // From these tables we can create "calibration tables" containing some
|
|---|
| 137 | // effective quantum efficiency depending on these outer parameters and which
|
|---|
| 138 | // are different for each pixel.
|
|---|
| 139 | //
|
|---|
| 140 | // In an ideal MCalibrate, one would thus have to convert first the FADC
|
|---|
| 141 | // slices to Photo-electrons and then, depending on the outer parameters,
|
|---|
| 142 | // look up the effective quantum efficiency and get the mean number of
|
|---|
| 143 | // photons which is then used for the further analysis.
|
|---|
| 144 | //
|
|---|
| 145 | // How the (first) MAGIC calibration should look like:
|
|---|
| 146 | // ===================================================
|
|---|
| 147 | //
|
|---|
| 148 | // For the moment, we have only one reliable calibration method, although
|
|---|
| 149 | // with very large systematic errors. This is the F-Factor method. Knowing
|
|---|
| 150 | // that the light is uniform over the whole camera (which I would not at all
|
|---|
| 151 | // guarantee in the case of the CT1 pulser), one could in principle already
|
|---|
| 152 | // perform a relative calibration of the quantum efficiencies in the UV.
|
|---|
| 153 | // However, the spread in QE at UV is about 10-15% (according to the plot
|
|---|
| 154 | // that Abelardo sent around last time. The spread in photo-electrons is 15%
|
|---|
| 155 | // for the inner pixels, but much larger (40%) for the outer ones.
|
|---|
| 156 | //
|
|---|
| 157 | // I'm not sure if we can already say that we have measured the relative
|
|---|
| 158 | // difference in quantum efficiency for the inner pixels and produce a first
|
|---|
| 159 | // QE-table for each pixel. To so, I would rather check in other wavelengths
|
|---|
| 160 | // (which we can do in about one-two weeks when the optical transmission of
|
|---|
| 161 | // the calibration trigger is installed).
|
|---|
| 162 | //
|
|---|
| 163 | // Thus, for the moment being, I would join Thomas proposal to calibrate in
|
|---|
| 164 | // photo-electrons and apply one stupid average quantum efficiency for all
|
|---|
| 165 | // pixels. This keeping in mind that we will have much preciser information
|
|---|
| 166 | // in about one to two weeks.
|
|---|
| 167 | //
|
|---|
| 168 | //
|
|---|
| 169 | // What MCalibrate should calculate and what should be stored:
|
|---|
| 170 | // ===========================================================
|
|---|
| 171 | //
|
|---|
| 172 | // It is clear that in the end, MCerPhotEvt will store photons.
|
|---|
| 173 | // MCalibrationCam stores the conversionfactors to photo-electrons and also
|
|---|
| 174 | // some tables of how to apply the conversion to photons, given the outer
|
|---|
| 175 | // parameters. This is not yet implemented and not even discussed.
|
|---|
| 176 | //
|
|---|
| 177 | // To start, I would suggest that we define the "average quantum efficiency"
|
|---|
| 178 | // (maybe something like 25+-3%) and apply them equally to all
|
|---|
| 179 | // photo-electrons. Later, this average factor can be easily replaced by a
|
|---|
| 180 | // pixel-dependent factor and later by a (pixel-dependent) table.
|
|---|
| 181 | //
|
|---|
| 182 | //
|
|---|
| 183 | // ClassVersion 2:
|
|---|
| 184 | // - Float_t fPheErrLimit;
|
|---|
| 185 | // + Float_t fPheErrLowerLimit; // Lower limit acceptance nr. phe's w.r.t. area idx mean (in sigmas)
|
|---|
| 186 | // + Float_t fPheErrUpperLimit; // Upper limit acceptance nr. phe's w.r.t. area idx mean (in sigmas)
|
|---|
| 187 | //
|
|---|
| 188 | //
|
|---|
| 189 | // ClassVersion 3:
|
|---|
| 190 | // + Bool_t fUseExtractorRes; // Include extractor resolution in F-Factor method
|
|---|
| 191 | //
|
|---|
| 192 | // ClassVersion 4:
|
|---|
| 193 | // + Float_t fUnsuitablesLimit; // Limit for relative number of unsuitable pixels
|
|---|
| 194 | // + Float_t fUnreliablesLimit; // Limit for relative number of unreliable pixels
|
|---|
| 195 | // + Float_t fExternalNumPhes; // External mean number of photo-electrons set from outside
|
|---|
| 196 | // + Float_t fExternalNumPhesRelVar; // External rel. var. number of photo-electrons set from outside
|
|---|
| 197 | //
|
|---|
| 198 | // ClassVersion 5:
|
|---|
| 199 | // - TString fOutputPath
|
|---|
| 200 | // - TString fOutputFile
|
|---|
| 201 | //
|
|---|
| 202 | //////////////////////////////////////////////////////////////////////////////
|
|---|
| 203 | #include "MCalibrationChargeCalc.h"
|
|---|
| 204 |
|
|---|
| 205 | #include <TSystem.h>
|
|---|
| 206 | #include <TH1.h>
|
|---|
| 207 | #include <TF1.h>
|
|---|
| 208 |
|
|---|
| 209 | #include "MLog.h"
|
|---|
| 210 | #include "MLogManip.h"
|
|---|
| 211 |
|
|---|
| 212 | #include "MParList.h"
|
|---|
| 213 |
|
|---|
| 214 | #include "MCalibrationPattern.h"
|
|---|
| 215 |
|
|---|
| 216 | #include "MGeomCam.h"
|
|---|
| 217 | #include "MGeomPix.h"
|
|---|
| 218 | #include "MHCamera.h"
|
|---|
| 219 |
|
|---|
| 220 | #include "MPedestalCam.h"
|
|---|
| 221 | #include "MPedestalPix.h"
|
|---|
| 222 |
|
|---|
| 223 | #include "MCalibrationIntensityChargeCam.h"
|
|---|
| 224 | #include "MCalibrationIntensityQECam.h"
|
|---|
| 225 | #include "MCalibrationIntensityBlindCam.h"
|
|---|
| 226 |
|
|---|
| 227 | #include "MHCalibrationChargeCam.h"
|
|---|
| 228 | #include "MHCalibrationChargeBlindCam.h"
|
|---|
| 229 |
|
|---|
| 230 | #include "MCalibrationChargeCam.h"
|
|---|
| 231 | #include "MCalibrationChargePix.h"
|
|---|
| 232 | #include "MCalibrationChargePINDiode.h"
|
|---|
| 233 | #include "MCalibrationBlindPix.h"
|
|---|
| 234 | #include "MCalibrationBlindCam.h"
|
|---|
| 235 |
|
|---|
| 236 | #include "MExtractedSignalCam.h"
|
|---|
| 237 | #include "MExtractedSignalPix.h"
|
|---|
| 238 | #include "MExtractedSignalBlindPixel.h"
|
|---|
| 239 | #include "MExtractedSignalPINDiode.h"
|
|---|
| 240 |
|
|---|
| 241 | #include "MBadPixelsIntensityCam.h"
|
|---|
| 242 | #include "MBadPixelsCam.h"
|
|---|
| 243 |
|
|---|
| 244 | #include "MCalibrationQECam.h"
|
|---|
| 245 | #include "MCalibrationQEPix.h"
|
|---|
| 246 |
|
|---|
| 247 | ClassImp(MCalibrationChargeCalc);
|
|---|
| 248 |
|
|---|
| 249 | using namespace std;
|
|---|
| 250 |
|
|---|
| 251 | const Float_t MCalibrationChargeCalc::fgChargeLimit = 4.5;
|
|---|
| 252 | const Float_t MCalibrationChargeCalc::fgChargeErrLimit = 0.;
|
|---|
| 253 | const Float_t MCalibrationChargeCalc::fgChargeRelErrLimit = 1.;
|
|---|
| 254 | const Float_t MCalibrationChargeCalc::fgLambdaErrLimit = 0.2;
|
|---|
| 255 | const Float_t MCalibrationChargeCalc::fgLambdaCheckLimit = 0.5;
|
|---|
| 256 | const Float_t MCalibrationChargeCalc::fgPheErrLowerLimit = 6.0;
|
|---|
| 257 | const Float_t MCalibrationChargeCalc::fgPheErrUpperLimit = 5.5;
|
|---|
| 258 | const Float_t MCalibrationChargeCalc::fgFFactorErrLimit = 4.5;
|
|---|
| 259 | const Float_t MCalibrationChargeCalc::fgArrTimeRmsLimit = 3.5;
|
|---|
| 260 | const Float_t MCalibrationChargeCalc::fgUnsuitablesLimit = 0.1;
|
|---|
| 261 | const Float_t MCalibrationChargeCalc::fgUnreliablesLimit = 0.3;
|
|---|
| 262 | const TString MCalibrationChargeCalc::fgNamePedestalCam = "MPedestalCam";
|
|---|
| 263 |
|
|---|
| 264 |
|
|---|
| 265 | // --------------------------------------------------------------------------
|
|---|
| 266 | //
|
|---|
| 267 | // Default constructor.
|
|---|
| 268 | //
|
|---|
| 269 | // Sets the pointer to fQECam and fGeom to NULL
|
|---|
| 270 | //
|
|---|
| 271 | // Calls AddToBranchList for:
|
|---|
| 272 | // - MRawEvtData.fHiGainPixId
|
|---|
| 273 | // - MRawEvtData.fLoGainPixId
|
|---|
| 274 | // - MRawEvtData.fHiGainFadcSamples
|
|---|
| 275 | // - MRawEvtData.fLoGainFadcSamples
|
|---|
| 276 | //
|
|---|
| 277 | // Initializes:
|
|---|
| 278 | // - fArrTimeRmsLimit to fgArrTimeRmsLimit
|
|---|
| 279 | // - fChargeLimit to fgChargeLimit
|
|---|
| 280 | // - fChargeErrLimit to fgChargeErrLimit
|
|---|
| 281 | // - fChargeRelErrLimit to fgChargeRelErrLimit
|
|---|
| 282 | // - fFFactorErrLimit to fgFFactorErrLimit
|
|---|
| 283 | // - fLambdaCheckLimit to fgLambdaCheckLimit
|
|---|
| 284 | // - fLambdaErrLimit to fgLambdaErrLimit
|
|---|
| 285 | // - fNamePedestalCam to fgNamePedestalCam
|
|---|
| 286 | // - fPheErrLowerLimit to fgPheErrLowerLimit
|
|---|
| 287 | // - fPheErrUpperLimit to fgPheErrUpperLimit
|
|---|
| 288 | // - fPulserColor to MCalibrationCam::kCT1
|
|---|
| 289 | // - fOutputPath to "."
|
|---|
| 290 | // - fOutputFile to "ChargeCalibStat.txt"
|
|---|
| 291 | // - flag debug to kFALSE
|
|---|
| 292 | // - IsUseExtractorRes to kFALSE
|
|---|
| 293 | // - IsUseExternalNumPhes to kFALSE
|
|---|
| 294 | // - fExternalNumPhes to 0.
|
|---|
| 295 | // - fExternalNumPhesRelVar to 0.
|
|---|
| 296 | //
|
|---|
| 297 | // Sets all checks
|
|---|
| 298 | //
|
|---|
| 299 | // Calls:
|
|---|
| 300 | // - Clear()
|
|---|
| 301 | //
|
|---|
| 302 | MCalibrationChargeCalc::MCalibrationChargeCalc(const char *name, const char *title)
|
|---|
| 303 | : fGeom(NULL), fSignal(NULL), fCalibPattern(NULL), fExtractor(NULL)
|
|---|
| 304 | {
|
|---|
| 305 |
|
|---|
| 306 | fName = name ? name : "MCalibrationChargeCalc";
|
|---|
| 307 | fTitle = title ? title : "Task to calculate the calibration constants and MCalibrationCam ";
|
|---|
| 308 |
|
|---|
| 309 | AddToBranchList("MRawEvtData.fHiGainPixId");
|
|---|
| 310 | AddToBranchList("MRawEvtData.fLoGainPixId");
|
|---|
| 311 | AddToBranchList("MRawEvtData.fHiGainFadcSamples");
|
|---|
| 312 | AddToBranchList("MRawEvtData.fLoGainFadcSamples");
|
|---|
| 313 |
|
|---|
| 314 | SetArrTimeRmsLimit ();
|
|---|
| 315 | SetChargeLimit ();
|
|---|
| 316 | SetChargeErrLimit ();
|
|---|
| 317 | SetChargeRelErrLimit ();
|
|---|
| 318 | SetDebug ( kFALSE );
|
|---|
| 319 | SetFFactorErrLimit ();
|
|---|
| 320 | SetLambdaCheckLimit ();
|
|---|
| 321 | SetLambdaErrLimit ();
|
|---|
| 322 | SetNamePedestalCam ();
|
|---|
| 323 | SetPheErrLowerLimit ();
|
|---|
| 324 | SetPheErrUpperLimit ();
|
|---|
| 325 | SetUnsuitablesLimit ();
|
|---|
| 326 | SetUnreliablesLimit ();
|
|---|
| 327 | SetUseExtractorRes ();
|
|---|
| 328 | SetUseUnreliables ();
|
|---|
| 329 | SetUseExternalNumPhes(kFALSE);
|
|---|
| 330 |
|
|---|
| 331 | SetExternalNumPhes ();
|
|---|
| 332 | SetExternalNumPhesRelVar();
|
|---|
| 333 |
|
|---|
| 334 | SetCheckArrivalTimes ();
|
|---|
| 335 | SetCheckDeadPixels ();
|
|---|
| 336 | SetCheckDeviatingBehavior();
|
|---|
| 337 | SetCheckExtractionWindow ();
|
|---|
| 338 | SetCheckHistOverflow ();
|
|---|
| 339 | SetCheckOscillations ();
|
|---|
| 340 |
|
|---|
| 341 | Clear();
|
|---|
| 342 |
|
|---|
| 343 | }
|
|---|
| 344 |
|
|---|
| 345 | // --------------------------------------------------------------------------
|
|---|
| 346 | //
|
|---|
| 347 | // Sets:
|
|---|
| 348 | // - all variables to 0.,
|
|---|
| 349 | // - all flags to kFALSE
|
|---|
| 350 | // - all pointers to NULL
|
|---|
| 351 | // - the pulser colour to kNONE
|
|---|
| 352 | // - fBlindPixelFlags to 0
|
|---|
| 353 | // - fPINDiodeFlags to 0
|
|---|
| 354 | //
|
|---|
| 355 | void MCalibrationChargeCalc::Clear(const Option_t *o)
|
|---|
| 356 | {
|
|---|
| 357 |
|
|---|
| 358 | fNumHiGainSamples = 0.;
|
|---|
| 359 | fNumLoGainSamples = 0.;
|
|---|
| 360 | fSqrtHiGainSamples = 0.;
|
|---|
| 361 | fSqrtLoGainSamples = 0.;
|
|---|
| 362 | fNumInnerFFactorMethodUsed = 0;
|
|---|
| 363 |
|
|---|
| 364 | fNumProcessed = 0;
|
|---|
| 365 |
|
|---|
| 366 | fIntensBad = NULL;
|
|---|
| 367 | fBadPixels = NULL;
|
|---|
| 368 | fIntensCam = NULL;
|
|---|
| 369 | fCam = NULL;
|
|---|
| 370 | fHCam = NULL;
|
|---|
| 371 | fIntensQE = NULL;
|
|---|
| 372 | fQECam = NULL;
|
|---|
| 373 | fIntensBlind = NULL;
|
|---|
| 374 | fBlindCam = NULL;
|
|---|
| 375 | fHBlindCam = NULL;
|
|---|
| 376 | fPINDiode = NULL;
|
|---|
| 377 | fPedestals = NULL;
|
|---|
| 378 |
|
|---|
| 379 | SetPulserColor ( MCalibrationCam::kNONE );
|
|---|
| 380 |
|
|---|
| 381 | fStrength = 0.;
|
|---|
| 382 | fBlindPixelFlags.Set(0);
|
|---|
| 383 | fPINDiodeFlags .Set(0);
|
|---|
| 384 | fResultFlags .Set(0);
|
|---|
| 385 | }
|
|---|
| 386 |
|
|---|
| 387 |
|
|---|
| 388 | // -----------------------------------------------------------------------------------
|
|---|
| 389 | //
|
|---|
| 390 | // The following container are searched for and execution aborted if not in MParList:
|
|---|
| 391 | // - MPedestalCam
|
|---|
| 392 | // - MCalibrationPattern
|
|---|
| 393 | // - MExtractedSignalCam
|
|---|
| 394 | //
|
|---|
| 395 | Int_t MCalibrationChargeCalc::PreProcess(MParList *pList)
|
|---|
| 396 | {
|
|---|
| 397 |
|
|---|
| 398 | /*
|
|---|
| 399 | if (IsInterlaced())
|
|---|
| 400 | {
|
|---|
| 401 | fTrigPattern = (MTriggerPattern*)pList->FindObject("MTriggerPattern");
|
|---|
| 402 | if (!fTrigPattern)
|
|---|
| 403 | {
|
|---|
| 404 | *fLog << err << "MTriggerPattern not found... abort." << endl;
|
|---|
| 405 | return kFALSE;
|
|---|
| 406 | }
|
|---|
| 407 | }
|
|---|
| 408 | */
|
|---|
| 409 |
|
|---|
| 410 | fCalibPattern = (MCalibrationPattern*)pList->FindObject("MCalibrationPattern");
|
|---|
| 411 | if (!fCalibPattern)
|
|---|
| 412 | {
|
|---|
| 413 | *fLog << err << "MCalibrationPattern not found... abort." << endl;
|
|---|
| 414 | return kFALSE;
|
|---|
| 415 | }
|
|---|
| 416 |
|
|---|
| 417 | //
|
|---|
| 418 | // Containers that have to be there.
|
|---|
| 419 | //
|
|---|
| 420 | fSignal = (MExtractedSignalCam*)pList->FindObject("MExtractedSignalCam");
|
|---|
| 421 | if (!fSignal)
|
|---|
| 422 | {
|
|---|
| 423 | *fLog << err << "MExtractedSignalCam not found... aborting" << endl;
|
|---|
| 424 | return kFALSE;
|
|---|
| 425 | }
|
|---|
| 426 |
|
|---|
| 427 | if (fPedestals)
|
|---|
| 428 | return kTRUE;
|
|---|
| 429 |
|
|---|
| 430 | fPedestals = (MPedestalCam*)pList->FindObject(AddSerialNumber(fNamePedestalCam), "MPedestalCam");
|
|---|
| 431 | if (!fPedestals)
|
|---|
| 432 | {
|
|---|
| 433 | *fLog << err << fNamePedestalCam << " [MPedestalCam] not found... aborting" << endl;
|
|---|
| 434 | return kFALSE;
|
|---|
| 435 | }
|
|---|
| 436 |
|
|---|
| 437 | fPulserColor = MCalibrationCam::kNONE;
|
|---|
| 438 |
|
|---|
| 439 | return kTRUE;
|
|---|
| 440 | }
|
|---|
| 441 |
|
|---|
| 442 |
|
|---|
| 443 | // --------------------------------------------------------------------------
|
|---|
| 444 | //
|
|---|
| 445 | // Search for the following input containers and abort if not existing:
|
|---|
| 446 | // - MGeomCam
|
|---|
| 447 | // - MCalibrationIntensityChargeCam or MCalibrationChargeCam
|
|---|
| 448 | // - MCalibrationIntensityQECam or MCalibrationQECam
|
|---|
| 449 | // - MBadPixelsIntensityCam or MBadPixelsCam
|
|---|
| 450 | //
|
|---|
| 451 | // Search for the following input containers and give a warning if not existing:
|
|---|
| 452 | // - MCalibrationBlindPix
|
|---|
| 453 | // - MCalibrationChargePINDiode
|
|---|
| 454 | //
|
|---|
| 455 | // It retrieves the following variables from MCalibrationChargeCam:
|
|---|
| 456 | //
|
|---|
| 457 | // - fNumHiGainSamples
|
|---|
| 458 | // - fNumLoGainSamples
|
|---|
| 459 | //
|
|---|
| 460 | // It defines the PixId of every pixel in:
|
|---|
| 461 | //
|
|---|
| 462 | // - MCalibrationIntensityChargeCam
|
|---|
| 463 | // - MCalibrationChargeCam
|
|---|
| 464 | // - MCalibrationQECam
|
|---|
| 465 | //
|
|---|
| 466 | // It sets all pixels in excluded which have the flag fBadBixelsPix::IsBad() set in:
|
|---|
| 467 | //
|
|---|
| 468 | // - MCalibrationChargePix
|
|---|
| 469 | // - MCalibrationQEPix
|
|---|
| 470 | //
|
|---|
| 471 | // Sets the pulser colour and tests if it has not changed w.r.t. fPulserColor in:
|
|---|
| 472 | //
|
|---|
| 473 | // - MCalibrationChargeCam
|
|---|
| 474 | // - MCalibrationBlindPix (if existing)
|
|---|
| 475 | // - MCalibrationChargePINDiode (if existing)
|
|---|
| 476 | //
|
|---|
| 477 | Bool_t MCalibrationChargeCalc::ReInit(MParList *pList )
|
|---|
| 478 | {
|
|---|
| 479 |
|
|---|
| 480 | fGeom = (MGeomCam*)pList->FindObject("MGeomCam");
|
|---|
| 481 | if (!fGeom)
|
|---|
| 482 | {
|
|---|
| 483 | *fLog << err << "No MGeomCam found... aborting." << endl;
|
|---|
| 484 | return kFALSE;
|
|---|
| 485 | }
|
|---|
| 486 |
|
|---|
| 487 | fIntensCam = (MCalibrationIntensityChargeCam*)pList->FindObject(AddSerialNumber("MCalibrationIntensityChargeCam"));
|
|---|
| 488 | if (fIntensCam)
|
|---|
| 489 | *fLog << inf << "Found MCalibrationIntensityChargeCam... " << flush;
|
|---|
| 490 | else
|
|---|
| 491 | {
|
|---|
| 492 | fCam = (MCalibrationChargeCam*)pList->FindObject(AddSerialNumber("MCalibrationChargeCam"));
|
|---|
| 493 | if (!fCam)
|
|---|
| 494 | {
|
|---|
| 495 | *fLog << err << "Cannot find MCalibrationChargeCam ... abort." << endl;
|
|---|
| 496 | *fLog << "Maybe you forget to call an MFillH for the MHCalibrationChargeCam before..." << endl;
|
|---|
| 497 | return kFALSE;
|
|---|
| 498 | }
|
|---|
| 499 | }
|
|---|
| 500 |
|
|---|
| 501 | fHCam = (MHCalibrationChargeCam*)pList->FindObject(AddSerialNumber("MHCalibrationChargeCam"));
|
|---|
| 502 | if (!fHCam)
|
|---|
| 503 | {
|
|---|
| 504 | *fLog << err << "Cannot find MHCalibrationChargeCam ... abort." << endl;
|
|---|
| 505 | *fLog << "Maybe you forget to call an MFillH for the MHCalibrationChargeCam before..." << endl;
|
|---|
| 506 | return kFALSE;
|
|---|
| 507 | }
|
|---|
| 508 |
|
|---|
| 509 | fIntensQE = (MCalibrationIntensityQECam*)pList->FindObject(AddSerialNumber("MCalibrationIntensityQECam"));
|
|---|
| 510 | if (fIntensQE)
|
|---|
| 511 | *fLog << inf << "Found MCalibrationIntensityQECam... " << flush;
|
|---|
| 512 | else
|
|---|
| 513 | {
|
|---|
| 514 | fQECam = (MCalibrationQECam*)pList->FindObject(AddSerialNumber("MCalibrationQECam"));
|
|---|
| 515 | if (!fQECam)
|
|---|
| 516 | {
|
|---|
| 517 | *fLog << err << "Cannot find MCalibrationQECam ... abort." << endl;
|
|---|
| 518 | *fLog << "Maybe you forget to call an MFillH for the MHCalibrationQECam before..." << endl;
|
|---|
| 519 | return kFALSE;
|
|---|
| 520 | }
|
|---|
| 521 | }
|
|---|
| 522 |
|
|---|
| 523 | fIntensBlind = (MCalibrationIntensityBlindCam*)pList->FindObject(AddSerialNumber("MCalibrationIntensityBlindCam"));
|
|---|
| 524 | if (fIntensBlind)
|
|---|
| 525 | *fLog << inf << "Found MCalibrationIntensityBlindCam... " << flush;
|
|---|
| 526 | else
|
|---|
| 527 | {
|
|---|
| 528 | fBlindCam = (MCalibrationBlindCam*)pList->FindObject(AddSerialNumber("MCalibrationBlindCam"));
|
|---|
| 529 | if (!fBlindCam)
|
|---|
| 530 | *fLog << inf << "No MCalibrationBlindCam found... no Blind Pixel method!" << endl;
|
|---|
| 531 | }
|
|---|
| 532 |
|
|---|
| 533 | fHBlindCam = (MHCalibrationChargeBlindCam*)pList->FindObject(AddSerialNumber("MHCalibrationChargeBlindCam"));
|
|---|
| 534 | if (!fHBlindCam)
|
|---|
| 535 | *fLog << inf << "No MHCalibrationChargeBlindCam found... no Blind Pixel method!" << endl;
|
|---|
| 536 |
|
|---|
| 537 | fIntensBad = (MBadPixelsIntensityCam*)pList->FindObject(AddSerialNumber("MBadPixelsIntensityCam"));
|
|---|
| 538 | if (fIntensBad)
|
|---|
| 539 | *fLog << inf << "Found MBadPixelsIntensityCam... " << flush;
|
|---|
| 540 | else
|
|---|
| 541 | {
|
|---|
| 542 | fBadPixels = (MBadPixelsCam*)pList->FindObject(AddSerialNumber("MBadPixelsCam"));
|
|---|
| 543 | if (!fBadPixels)
|
|---|
| 544 | {
|
|---|
| 545 | *fLog << err << "Cannot find MBadPixelsCam ... abort." << endl;
|
|---|
| 546 | return kFALSE;
|
|---|
| 547 | }
|
|---|
| 548 | }
|
|---|
| 549 |
|
|---|
| 550 | //
|
|---|
| 551 | // Optional Containers
|
|---|
| 552 | //
|
|---|
| 553 | fPINDiode = (MCalibrationChargePINDiode*)pList->FindObject("MCalibrationChargePINDiode");
|
|---|
| 554 | if (!fPINDiode)
|
|---|
| 555 | *fLog << inf << "No MCalibrationChargePINDiode found... no PIN Diode method!" << endl;
|
|---|
| 556 |
|
|---|
| 557 | MCalibrationQECam *qecam = fIntensQE
|
|---|
| 558 | ? (MCalibrationQECam*) fIntensQE->GetCam() : fQECam;
|
|---|
| 559 | MCalibrationChargeCam *chargecam = fIntensCam
|
|---|
| 560 | ? (MCalibrationChargeCam*)fIntensCam->GetCam() : fCam;
|
|---|
| 561 | MBadPixelsCam *badcam = fIntensBad
|
|---|
| 562 | ? (MBadPixelsCam*) fIntensBad->GetCam() : fBadPixels;
|
|---|
| 563 |
|
|---|
| 564 | UInt_t npixels = fGeom->GetNumPixels();
|
|---|
| 565 |
|
|---|
| 566 | for (UInt_t i=0; i<npixels; i++)
|
|---|
| 567 | {
|
|---|
| 568 |
|
|---|
| 569 | MCalibrationChargePix &pix = (MCalibrationChargePix&)(*chargecam)[i];
|
|---|
| 570 | MCalibrationQEPix &pqe = (MCalibrationQEPix&) (*qecam) [i];
|
|---|
| 571 | MBadPixelsPix &bad = (*badcam) [i];
|
|---|
| 572 |
|
|---|
| 573 | if (bad.IsBad())
|
|---|
| 574 | {
|
|---|
| 575 | pix.SetExcluded();
|
|---|
| 576 | pqe.SetExcluded();
|
|---|
| 577 | continue;
|
|---|
| 578 | }
|
|---|
| 579 |
|
|---|
| 580 | if (IsDebug())
|
|---|
| 581 | pix.SetDebug();
|
|---|
| 582 | }
|
|---|
| 583 |
|
|---|
| 584 | fResultFlags.Set(fGeom->GetNumAreas());
|
|---|
| 585 |
|
|---|
| 586 | return kTRUE;
|
|---|
| 587 | }
|
|---|
| 588 |
|
|---|
| 589 | // ----------------------------------------------------------------------------------
|
|---|
| 590 | //
|
|---|
| 591 | // Set the correct colour to the charge containers
|
|---|
| 592 | //
|
|---|
| 593 | Int_t MCalibrationChargeCalc::Process()
|
|---|
| 594 | {
|
|---|
| 595 |
|
|---|
| 596 | const MCalibrationCam::PulserColor_t col = fCalibPattern->GetPulserColor();
|
|---|
| 597 | const Float_t strength = fCalibPattern->GetPulserStrength();
|
|---|
| 598 | const Float_t strdiff = TMath::Abs(strength-fStrength);
|
|---|
| 599 |
|
|---|
| 600 | if (col == fPulserColor && strdiff < 0.05 )
|
|---|
| 601 | {
|
|---|
| 602 | fNumProcessed++;
|
|---|
| 603 | return kTRUE;
|
|---|
| 604 | }
|
|---|
| 605 |
|
|---|
| 606 | if (col == MCalibrationCam::kNONE)
|
|---|
| 607 | return kTRUE;
|
|---|
| 608 |
|
|---|
| 609 |
|
|---|
| 610 | //
|
|---|
| 611 | // Now retrieve the colour and check if not various colours have been used
|
|---|
| 612 | //
|
|---|
| 613 | if (!fIntensCam)
|
|---|
| 614 | {
|
|---|
| 615 | if (fPulserColor != MCalibrationCam::kNONE)
|
|---|
| 616 | {
|
|---|
| 617 | *fLog << warn << "Multiple colours used simultaneously!" << flush;
|
|---|
| 618 | fHCam->Finalize();
|
|---|
| 619 | if (fHBlindCam)
|
|---|
| 620 | fHBlindCam->Finalize();
|
|---|
| 621 |
|
|---|
| 622 | Finalize();
|
|---|
| 623 |
|
|---|
| 624 | fHCam->ResetHists();
|
|---|
| 625 | if (fHBlindCam)
|
|---|
| 626 | fHBlindCam->ResetHists();
|
|---|
| 627 |
|
|---|
| 628 | *fLog << inf << "Starting next calibration... " << flush;
|
|---|
| 629 |
|
|---|
| 630 | fHCam->SetColor(col);
|
|---|
| 631 | if (fHBlindCam)
|
|---|
| 632 | fHBlindCam->SetColor(col);
|
|---|
| 633 |
|
|---|
| 634 | fCam->SetPulserColor(col);
|
|---|
| 635 | if (fBlindCam)
|
|---|
| 636 | fBlindCam->SetPulserColor(col);
|
|---|
| 637 | }
|
|---|
| 638 | }
|
|---|
| 639 |
|
|---|
| 640 | fPulserColor = col;
|
|---|
| 641 | fStrength = strength;
|
|---|
| 642 |
|
|---|
| 643 | *fLog << inf << "Found new colour ... " << flush;
|
|---|
| 644 |
|
|---|
| 645 | switch (col)
|
|---|
| 646 | {
|
|---|
| 647 | case MCalibrationCam::kGREEN: *fLog << "Green"; break;
|
|---|
| 648 | case MCalibrationCam::kBLUE: *fLog << "Blue"; break;
|
|---|
| 649 | case MCalibrationCam::kUV: *fLog << "UV"; break;
|
|---|
| 650 | case MCalibrationCam::kCT1: *fLog << "CT1"; break;
|
|---|
| 651 | default: break;
|
|---|
| 652 | }
|
|---|
| 653 |
|
|---|
| 654 | *fLog << inf << " with strength: " << strength << endl;
|
|---|
| 655 |
|
|---|
| 656 | fHCam->SetColor(col);
|
|---|
| 657 | if (fHBlindCam)
|
|---|
| 658 | fHBlindCam->SetColor(col);
|
|---|
| 659 |
|
|---|
| 660 | MCalibrationBlindCam *blindcam = fIntensBlind
|
|---|
| 661 | ? (MCalibrationBlindCam*)fIntensBlind->GetCam() : fBlindCam;
|
|---|
| 662 | MCalibrationChargeCam *chargecam = fIntensCam
|
|---|
| 663 | ? (MCalibrationChargeCam*)fIntensCam->GetCam() : fCam;
|
|---|
| 664 |
|
|---|
| 665 | chargecam->SetPulserColor(col);
|
|---|
| 666 |
|
|---|
| 667 | if (blindcam)
|
|---|
| 668 | blindcam->SetPulserColor(col);
|
|---|
| 669 | if (fPINDiode)
|
|---|
| 670 | fPINDiode->SetColor(col);
|
|---|
| 671 |
|
|---|
| 672 | fNumProcessed = 0;
|
|---|
| 673 |
|
|---|
| 674 | return kTRUE;
|
|---|
| 675 | }
|
|---|
| 676 |
|
|---|
| 677 | // -----------------------------------------------------------------------
|
|---|
| 678 | //
|
|---|
| 679 | // Return if number of executions is null.
|
|---|
| 680 | //
|
|---|
| 681 | Int_t MCalibrationChargeCalc::PostProcess()
|
|---|
| 682 | {
|
|---|
| 683 |
|
|---|
| 684 | if (GetNumExecutions() < 1)
|
|---|
| 685 | return kTRUE;
|
|---|
| 686 |
|
|---|
| 687 | if (fPulserColor == MCalibrationCam::kNONE)
|
|---|
| 688 | {
|
|---|
| 689 | *fLog << warn << "WARNING - No Pulse colour has been set or used at all..." << endl;
|
|---|
| 690 | return kTRUE;
|
|---|
| 691 | }
|
|---|
| 692 |
|
|---|
| 693 | if (fNumProcessed == 0)
|
|---|
| 694 | return kTRUE;
|
|---|
| 695 |
|
|---|
| 696 | *fLog << endl;
|
|---|
| 697 |
|
|---|
| 698 | return Finalize();
|
|---|
| 699 | }
|
|---|
| 700 |
|
|---|
| 701 | // -----------------------------------------------------------------------
|
|---|
| 702 | //
|
|---|
| 703 | // Return kTRUE if fPulserColor is kNONE
|
|---|
| 704 | //
|
|---|
| 705 | // First loop over pixels, average areas and sectors, call:
|
|---|
| 706 | // - FinalizePedestals()
|
|---|
| 707 | // - FinalizeCharges()
|
|---|
| 708 | // for every entry. Count number of valid pixels in loop and return kFALSE
|
|---|
| 709 | // if there are none (the "Michele check").
|
|---|
| 710 | //
|
|---|
| 711 | // Call FinalizeBadPixels()
|
|---|
| 712 | //
|
|---|
| 713 | // Call FinalizeFFactorMethod() (second and third loop over pixels and areas)
|
|---|
| 714 | //
|
|---|
| 715 | // Call FinalizeBlindCam()
|
|---|
| 716 | // Call FinalizePINDiode()
|
|---|
| 717 | //
|
|---|
| 718 | // Call FinalizeFFactorQECam() (fourth loop over pixels and areas)
|
|---|
| 719 | // Call FinalizeBlindPixelQECam() (fifth loop over pixels and areas)
|
|---|
| 720 | // Call FinalizePINDiodeQECam() (sixth loop over pixels and areas)
|
|---|
| 721 | //
|
|---|
| 722 | // Call FinalizeUnsuitablePixels()
|
|---|
| 723 | //
|
|---|
| 724 | // Call MParContainer::SetReadyToSave() for fIntensCam, fCam, fQECam, fBadPixels and
|
|---|
| 725 | // fBlindCam and fPINDiode if they exist
|
|---|
| 726 | //
|
|---|
| 727 | // Print out some statistics
|
|---|
| 728 | //
|
|---|
| 729 | Int_t MCalibrationChargeCalc::Finalize()
|
|---|
| 730 | {
|
|---|
| 731 | // The number of used slices are just a mean value
|
|---|
| 732 | // the real number might change from event to event.
|
|---|
| 733 | // (up to 50%!)
|
|---|
| 734 |
|
|---|
| 735 | fNumHiGainSamples = fSignal->GetNumUsedHiGainFADCSlices();
|
|---|
| 736 | fNumLoGainSamples = fSignal->GetNumUsedLoGainFADCSlices();
|
|---|
| 737 |
|
|---|
| 738 | fSqrtHiGainSamples = TMath::Sqrt(fNumHiGainSamples);
|
|---|
| 739 | fSqrtLoGainSamples = TMath::Sqrt(fNumLoGainSamples);
|
|---|
| 740 |
|
|---|
| 741 | if (fPINDiode)
|
|---|
| 742 | if (!fPINDiode->IsValid())
|
|---|
| 743 | {
|
|---|
| 744 | *fLog << warn << GetDescriptor()
|
|---|
| 745 | << ": MCalibrationChargePINDiode is declared not valid... no PIN Diode method! " << endl;
|
|---|
| 746 | fPINDiode = NULL;
|
|---|
| 747 | }
|
|---|
| 748 |
|
|---|
| 749 | MCalibrationBlindCam *blindcam = fIntensBlind ? (MCalibrationBlindCam*) fIntensBlind->GetCam() : fBlindCam;
|
|---|
| 750 | MCalibrationQECam *qecam = fIntensQE ? (MCalibrationQECam*) fIntensQE->GetCam() : fQECam;
|
|---|
| 751 | MCalibrationChargeCam *chargecam = fIntensCam ? (MCalibrationChargeCam*)fIntensCam->GetCam() : fCam;
|
|---|
| 752 | MBadPixelsCam *badcam = fIntensBad ? (MBadPixelsCam*) fIntensBad->GetCam() : fBadPixels;
|
|---|
| 753 |
|
|---|
| 754 | //
|
|---|
| 755 | // First loop over pixels, call FinalizePedestals and FinalizeCharges
|
|---|
| 756 | //
|
|---|
| 757 | Int_t nvalid = 0;
|
|---|
| 758 |
|
|---|
| 759 | for (Int_t pixid=0; pixid<fPedestals->GetSize(); pixid++)
|
|---|
| 760 | {
|
|---|
| 761 |
|
|---|
| 762 | MCalibrationChargePix &pix = (MCalibrationChargePix&)(*chargecam)[pixid];
|
|---|
| 763 | //
|
|---|
| 764 | // Check if the pixel has been excluded from the fits
|
|---|
| 765 | //
|
|---|
| 766 | if (pix.IsExcluded())
|
|---|
| 767 | continue;
|
|---|
| 768 |
|
|---|
| 769 | MPedestalPix &ped = (*fPedestals)[pixid];
|
|---|
| 770 | MBadPixelsPix &bad = (*badcam) [pixid];
|
|---|
| 771 |
|
|---|
| 772 | const Int_t aidx = (*fGeom)[pixid].GetAidx();
|
|---|
| 773 |
|
|---|
| 774 | FinalizePedestals(ped,pix,aidx);
|
|---|
| 775 |
|
|---|
| 776 | if (FinalizeCharges(pix,bad,"pixel "))
|
|---|
| 777 | nvalid++;
|
|---|
| 778 |
|
|---|
| 779 | FinalizeArrivalTimes(pix,bad,"pixel ");
|
|---|
| 780 | }
|
|---|
| 781 |
|
|---|
| 782 | *fLog << endl;
|
|---|
| 783 |
|
|---|
| 784 | //
|
|---|
| 785 | // The Michele check ...
|
|---|
| 786 | //
|
|---|
| 787 | if (nvalid == 0)
|
|---|
| 788 | {
|
|---|
| 789 | if (!fIntensCam)
|
|---|
| 790 | {
|
|---|
| 791 | *fLog << warn << GetDescriptor() << ": All pixels have non-valid calibration. "
|
|---|
| 792 | << "Did you forget to fill the histograms "
|
|---|
| 793 | << "(filling MHCalibrationChargeCam from MExtractedSignalCam using MFillH) ? " << endl;
|
|---|
| 794 | *fLog << warn << GetDescriptor() << ": Or, maybe, you have used a pedestal run "
|
|---|
| 795 | << "instead of a calibration run " << endl;
|
|---|
| 796 | return kFALSE;
|
|---|
| 797 | }
|
|---|
| 798 | }
|
|---|
| 799 |
|
|---|
| 800 | for (UInt_t aidx=0; aidx<fGeom->GetNumAreas(); aidx++)
|
|---|
| 801 | {
|
|---|
| 802 |
|
|---|
| 803 | const MPedestalPix &ped = fPedestals->GetAverageArea(aidx);
|
|---|
| 804 | MCalibrationChargePix &pix = (MCalibrationChargePix&)chargecam->GetAverageArea(aidx);
|
|---|
| 805 | const MArrayI &arr = fHCam->GetAverageAreaNum();
|
|---|
| 806 |
|
|---|
| 807 | FinalizePedestals(ped,pix,aidx);
|
|---|
| 808 |
|
|---|
| 809 | //
|
|---|
| 810 | // Correct for sqrt(number of valid pixels) in the pedestal RMS
|
|---|
| 811 | // (already done for calibration sigma in MHCalibrationCam::CalcAverageSigma()
|
|---|
| 812 | //
|
|---|
| 813 | pix.SetPedRMS(pix.GetPedRms()*TMath::Sqrt((Float_t)arr[aidx]),
|
|---|
| 814 | pix.GetPedRmsErr()*TMath::Sqrt((Float_t)arr[aidx]));
|
|---|
| 815 | pix.SetSigma (pix.GetSigma()/pix.GetFFactorFADC2Phe());
|
|---|
| 816 |
|
|---|
| 817 | FinalizeCharges(pix, chargecam->GetAverageBadArea(aidx),"area id");
|
|---|
| 818 | FinalizeArrivalTimes(pix, chargecam->GetAverageBadArea(aidx), "area id");
|
|---|
| 819 | }
|
|---|
| 820 |
|
|---|
| 821 | *fLog << endl;
|
|---|
| 822 |
|
|---|
| 823 | for (UInt_t sector=0; sector<fGeom->GetNumSectors(); sector++)
|
|---|
| 824 | {
|
|---|
| 825 |
|
|---|
| 826 | const MPedestalPix &ped = fPedestals->GetAverageSector(sector);
|
|---|
| 827 |
|
|---|
| 828 | MCalibrationChargePix &pix = (MCalibrationChargePix&)chargecam->GetAverageSector(sector);
|
|---|
| 829 | FinalizePedestals(ped,pix, 0);
|
|---|
| 830 | }
|
|---|
| 831 |
|
|---|
| 832 | *fLog << endl;
|
|---|
| 833 |
|
|---|
| 834 | //
|
|---|
| 835 | // Finalize Bad Pixels
|
|---|
| 836 | //
|
|---|
| 837 | FinalizeBadPixels();
|
|---|
| 838 |
|
|---|
| 839 | //
|
|---|
| 840 | // Finalize F-Factor method
|
|---|
| 841 | //
|
|---|
| 842 | if (FinalizeFFactorMethod())
|
|---|
| 843 | chargecam->SetFFactorMethodValid(kTRUE);
|
|---|
| 844 | else
|
|---|
| 845 | {
|
|---|
| 846 | *fLog << warn << "Could not calculate the photons flux from the F-Factor method " << endl;
|
|---|
| 847 | chargecam->SetFFactorMethodValid(kFALSE);
|
|---|
| 848 | if (!fIntensCam)
|
|---|
| 849 | return kFALSE;
|
|---|
| 850 | }
|
|---|
| 851 |
|
|---|
| 852 | *fLog << endl;
|
|---|
| 853 |
|
|---|
| 854 | //
|
|---|
| 855 | // Finalize Blind Pixel
|
|---|
| 856 | //
|
|---|
| 857 | qecam->SetBlindPixelMethodValid(FinalizeBlindCam());
|
|---|
| 858 |
|
|---|
| 859 | //
|
|---|
| 860 | // Finalize PIN Diode
|
|---|
| 861 | //
|
|---|
| 862 | qecam->SetBlindPixelMethodValid(FinalizePINDiode());
|
|---|
| 863 |
|
|---|
| 864 | //
|
|---|
| 865 | // Finalize QE Cam
|
|---|
| 866 | //
|
|---|
| 867 | FinalizeFFactorQECam();
|
|---|
| 868 | FinalizeBlindPixelQECam();
|
|---|
| 869 | FinalizePINDiodeQECam();
|
|---|
| 870 | FinalizeCombinedQECam();
|
|---|
| 871 |
|
|---|
| 872 | //
|
|---|
| 873 | // Re-direct the output to an ascii-file from now on:
|
|---|
| 874 | //
|
|---|
| 875 | *fLog << inf << endl;
|
|---|
| 876 | *fLog << GetDescriptor() << ": Fatal errors statistics:" << endl;
|
|---|
| 877 |
|
|---|
| 878 | PrintUncalibrated(MBadPixelsPix::kChargeIsPedestal,
|
|---|
| 879 | Form("%s%2.1f%s","Signal less than ",fChargeLimit," Pedestal RMS: "));
|
|---|
| 880 | PrintUncalibrated(MBadPixelsPix::kChargeRelErrNotValid,
|
|---|
| 881 | Form("%s%2.1f%s","Signal Error bigger than ",fChargeRelErrLimit," times Mean Signal: "));
|
|---|
| 882 | PrintUncalibrated(MBadPixelsPix::kLoGainSaturation,
|
|---|
| 883 | "Low Gain Saturation: ");
|
|---|
| 884 | PrintUncalibrated(MBadPixelsPix::kMeanTimeInFirstBin,
|
|---|
| 885 | Form("%s%2.1f%s","Mean Abs. Arr. Time in First ",1.," Bin(s): "));
|
|---|
| 886 | PrintUncalibrated(MBadPixelsPix::kMeanTimeInLast2Bins,
|
|---|
| 887 | Form("%s%2.1f%s","Mean Abs. Arr. Time in Last ",2.," Bin(s): "));
|
|---|
| 888 | PrintUncalibrated(MBadPixelsPix::kHiGainOverFlow,
|
|---|
| 889 | "Pixels with High Gain Overflow: ");
|
|---|
| 890 | PrintUncalibrated(MBadPixelsPix::kLoGainOverFlow,
|
|---|
| 891 | "Pixels with Low Gain Overflow : ");
|
|---|
| 892 | PrintUncalibrated(MBadPixelsPix::kFluctuatingArrivalTimes,
|
|---|
| 893 | "Fluctuating Pulse Arrival Times: ");
|
|---|
| 894 | PrintUncalibrated(MBadPixelsPix::kDeadPedestalRms,
|
|---|
| 895 | "Presumably dead from Pedestal Rms: ");
|
|---|
| 896 | PrintUncalibrated(MBadPixelsPix::kDeviatingNumPhes,
|
|---|
| 897 | "Deviating number of phes: ");
|
|---|
| 898 | PrintUncalibrated(MBadPixelsPix::kLoGainBlackout,
|
|---|
| 899 | "Too many blackout events in low gain: ");
|
|---|
| 900 | PrintUncalibrated(MBadPixelsPix::kPreviouslyExcluded,
|
|---|
| 901 | "Previously excluded: ");
|
|---|
| 902 |
|
|---|
| 903 | *fLog << inf << endl;
|
|---|
| 904 | *fLog << GetDescriptor() << ": Unreliable errors statistics:" << endl;
|
|---|
| 905 |
|
|---|
| 906 | PrintUncalibrated(MBadPixelsPix::kChargeSigmaNotValid,
|
|---|
| 907 | "Signal Sigma smaller than Pedestal RMS: ");
|
|---|
| 908 | PrintUncalibrated(MBadPixelsPix::kHiGainOscillating,
|
|---|
| 909 | "Changing Hi Gain signal over time: ");
|
|---|
| 910 | PrintUncalibrated(MBadPixelsPix::kLoGainOscillating,
|
|---|
| 911 | "Changing Lo Gain signal over time: ");
|
|---|
| 912 | PrintUncalibrated(MBadPixelsPix::kHiGainNotFitted,
|
|---|
| 913 | "Unsuccesful Gauss fit to the Hi Gain: ");
|
|---|
| 914 | PrintUncalibrated(MBadPixelsPix::kLoGainNotFitted,
|
|---|
| 915 | "Unsuccesful Gauss fit to the Lo Gain: ");
|
|---|
| 916 | PrintUncalibrated(MBadPixelsPix::kDeviatingFFactor,
|
|---|
| 917 | "Deviating F-Factor: ");
|
|---|
| 918 |
|
|---|
| 919 | chargecam->SetReadyToSave();
|
|---|
| 920 | qecam ->SetReadyToSave();
|
|---|
| 921 | badcam ->SetReadyToSave();
|
|---|
| 922 |
|
|---|
| 923 | if (blindcam)
|
|---|
| 924 | blindcam->SetReadyToSave();
|
|---|
| 925 | if (fPINDiode)
|
|---|
| 926 | fPINDiode->SetReadyToSave();
|
|---|
| 927 |
|
|---|
| 928 | //
|
|---|
| 929 | // Finalize calibration statistics
|
|---|
| 930 | //
|
|---|
| 931 | return FinalizeUnsuitablePixels();
|
|---|
| 932 | }
|
|---|
| 933 |
|
|---|
| 934 | // ----------------------------------------------------------------------------------
|
|---|
| 935 | //
|
|---|
| 936 | // Retrieves pedestal and pedestal RMS from MPedestalPix
|
|---|
| 937 | // Retrieves total entries from MPedestalCam
|
|---|
| 938 | // Sets pedestal*fNumHiGainSamples and pedestal*fNumLoGainSamples in MCalibrationChargePix
|
|---|
| 939 | // Sets pedRMS *fSqrtHiGainSamples and pedRMS *fSqrtLoGainSamples in MCalibrationChargePix
|
|---|
| 940 | //
|
|---|
| 941 | // If the flag MCalibrationPix::IsHiGainSaturation() is set, call also:
|
|---|
| 942 | // - MCalibrationChargePix::CalcLoGainPedestal()
|
|---|
| 943 | //
|
|---|
| 944 | void MCalibrationChargeCalc::FinalizePedestals(const MPedestalPix &ped, MCalibrationChargePix &cal, const Int_t aidx)
|
|---|
| 945 | {
|
|---|
| 946 |
|
|---|
| 947 | //
|
|---|
| 948 | // get the pedestals
|
|---|
| 949 | //
|
|---|
| 950 | const Float_t pedes = ped.GetPedestal();
|
|---|
| 951 | const Float_t prms = ped.GetPedestalRms();
|
|---|
| 952 | const Int_t num = fPedestals->GetTotalEntries();
|
|---|
| 953 |
|
|---|
| 954 | //
|
|---|
| 955 | // RMS error set by PedCalcFromLoGain, 0 in case MPedCalcPedRun was used.
|
|---|
| 956 | //
|
|---|
| 957 | const Float_t prmserr = num>0 ? prms/TMath::Sqrt(2.*num) : ped.GetPedestalRmsError();
|
|---|
| 958 |
|
|---|
| 959 | //
|
|---|
| 960 | // set them in the calibration camera
|
|---|
| 961 | //
|
|---|
| 962 | if (cal.IsHiGainSaturation())
|
|---|
| 963 | {
|
|---|
| 964 | cal.SetPedestal(pedes * fNumLoGainSamples,
|
|---|
| 965 | prms * fSqrtLoGainSamples,
|
|---|
| 966 | prmserr * fSqrtLoGainSamples);
|
|---|
| 967 | cal.CalcLoGainPedestal(fNumLoGainSamples);
|
|---|
| 968 | }
|
|---|
| 969 | else
|
|---|
| 970 | {
|
|---|
| 971 |
|
|---|
| 972 | cal.SetPedestal(pedes * fNumHiGainSamples,
|
|---|
| 973 | prms * fSqrtHiGainSamples,
|
|---|
| 974 | prmserr * fSqrtHiGainSamples);
|
|---|
| 975 | }
|
|---|
| 976 |
|
|---|
| 977 | }
|
|---|
| 978 |
|
|---|
| 979 | // ----------------------------------------------------------------------------------------------------
|
|---|
| 980 | //
|
|---|
| 981 | // Check fit results validity. Bad Pixels flags are set if:
|
|---|
| 982 | //
|
|---|
| 983 | // 1) Pixel has a mean smaller than fChargeLimit*PedRMS ( Flag: MBadPixelsPix::kChargeIsPedestal)
|
|---|
| 984 | // 2) Pixel has a mean error smaller than fChargeErrLimit ( Flag: MBadPixelsPix::kChargeErrNotValid)
|
|---|
| 985 | // 3) Pixel has mean smaller than fChargeRelVarLimit times its mean error
|
|---|
| 986 | // ( Flag: MBadPixelsPix::kChargeRelErrNotValid)
|
|---|
| 987 | // 4) Pixel has a sigma bigger than its Pedestal RMS ( Flag: MBadPixelsPix::kChargeSigmaNotValid )
|
|---|
| 988 | //
|
|---|
| 989 | // Further returns if flags: MBadPixelsPix::kUnsuitableRun is set
|
|---|
| 990 | //
|
|---|
| 991 | // Calls MCalibrationChargePix::CalcReducedSigma() and sets flag: MBadPixelsPix::kChargeIsPedestal
|
|---|
| 992 | // and returns kFALSE if not succesful.
|
|---|
| 993 | //
|
|---|
| 994 | // Calls MCalibrationChargePix::CalcFFactor() and sets flag: MBadPixelsPix::kDeviatingNumPhes)
|
|---|
| 995 | // and returns kFALSE if not succesful.
|
|---|
| 996 | //
|
|---|
| 997 | // Calls MCalibrationChargePix::CalcConvFFactor()and sets flag: MBadPixelsPix::kDeviatingNumPhes)
|
|---|
| 998 | // and returns kFALSE if not succesful.
|
|---|
| 999 | //
|
|---|
| 1000 | Bool_t MCalibrationChargeCalc::FinalizeCharges(MCalibrationChargePix &cal, MBadPixelsPix &bad, const char* what)
|
|---|
| 1001 | {
|
|---|
| 1002 |
|
|---|
| 1003 | if (bad.IsUnsuitable(MBadPixelsPix::kUnsuitableRun))
|
|---|
| 1004 | return kFALSE;
|
|---|
| 1005 |
|
|---|
| 1006 | const TString desc = Form("%7s%4d: ", what, cal.GetPixId());
|
|---|
| 1007 |
|
|---|
| 1008 | if (cal.GetMean() < fChargeLimit*cal.GetPedRms())
|
|---|
| 1009 | {
|
|---|
| 1010 | *fLog << warn << desc
|
|---|
| 1011 | << Form("Fitted Charge: %5.2f < %2.1f",cal.GetMean(),fChargeLimit)
|
|---|
| 1012 | << Form(" * Pedestal RMS %5.2f",cal.GetPedRms()) << endl;
|
|---|
| 1013 | bad.SetUncalibrated( MBadPixelsPix::kChargeIsPedestal);
|
|---|
| 1014 | }
|
|---|
| 1015 |
|
|---|
| 1016 | if (cal.GetMean() < fChargeRelErrLimit*cal.GetMeanErr())
|
|---|
| 1017 | {
|
|---|
| 1018 | *fLog << warn << desc
|
|---|
| 1019 | << Form("Fitted Charge: %4.2f < %2.1f",cal.GetMean(),fChargeRelErrLimit)
|
|---|
| 1020 | << Form(" * its error %4.2f",cal.GetMeanErr()) << endl;
|
|---|
| 1021 | bad.SetUncalibrated( MBadPixelsPix::kChargeRelErrNotValid );
|
|---|
| 1022 | }
|
|---|
| 1023 |
|
|---|
| 1024 | if (cal.GetSigma() < cal.GetPedRms())
|
|---|
| 1025 | {
|
|---|
| 1026 | *fLog << warn << desc << "Sigma of Fitted Charge: "
|
|---|
| 1027 | << Form("%6.2f <",cal.GetSigma()) << " Ped. RMS="
|
|---|
| 1028 | << Form("%5.2f", cal.GetPedRms()) << endl;
|
|---|
| 1029 | bad.SetUncalibrated( MBadPixelsPix::kChargeSigmaNotValid );
|
|---|
| 1030 | return kFALSE;
|
|---|
| 1031 | }
|
|---|
| 1032 |
|
|---|
| 1033 | if (!cal.CalcReducedSigma())
|
|---|
| 1034 | {
|
|---|
| 1035 | *fLog << warn << desc << "Could not calculate the reduced sigma" << endl;
|
|---|
| 1036 | bad.SetUncalibrated( MBadPixelsPix::kChargeSigmaNotValid );
|
|---|
| 1037 | return kFALSE;
|
|---|
| 1038 | }
|
|---|
| 1039 |
|
|---|
| 1040 | if (!cal.CalcFFactor())
|
|---|
| 1041 | {
|
|---|
| 1042 | *fLog << warn << desc << "Could not calculate the F-Factor"<< endl;
|
|---|
| 1043 | bad.SetUncalibrated(MBadPixelsPix::kDeviatingNumPhes);
|
|---|
| 1044 | bad.SetUnsuitable(MBadPixelsPix::kUnsuitableRun);
|
|---|
| 1045 | return kFALSE;
|
|---|
| 1046 | }
|
|---|
| 1047 |
|
|---|
| 1048 | if (cal.GetPheFFactorMethod() < 0.)
|
|---|
| 1049 | {
|
|---|
| 1050 | bad.SetUncalibrated(MBadPixelsPix::kDeviatingNumPhes);
|
|---|
| 1051 | bad.SetUnsuitable(MBadPixelsPix::kUnsuitableRun);
|
|---|
| 1052 | cal.SetFFactorMethodValid(kFALSE);
|
|---|
| 1053 | return kFALSE;
|
|---|
| 1054 | }
|
|---|
| 1055 |
|
|---|
| 1056 | if (!cal.CalcConvFFactor())
|
|---|
| 1057 | {
|
|---|
| 1058 | *fLog << warn << desc << "Could not calculate the Conv. FADC counts to Phes"<< endl;
|
|---|
| 1059 | bad.SetUncalibrated(MBadPixelsPix::kDeviatingNumPhes);
|
|---|
| 1060 | return kFALSE;
|
|---|
| 1061 | }
|
|---|
| 1062 |
|
|---|
| 1063 | if (!IsUseExtractorRes())
|
|---|
| 1064 | return kTRUE;
|
|---|
| 1065 |
|
|---|
| 1066 | if (!fExtractor)
|
|---|
| 1067 | {
|
|---|
| 1068 | *fLog << err << "Extractor resolution has been chosen, but not extractor is set. Cannot calibrate." << endl;
|
|---|
| 1069 | return kFALSE;
|
|---|
| 1070 | }
|
|---|
| 1071 |
|
|---|
| 1072 | const Float_t resinphes = cal.IsHiGainSaturation()
|
|---|
| 1073 | ? cal.GetPheFFactorMethod()*fExtractor->GetResolutionPerPheLoGain()
|
|---|
| 1074 | : cal.GetPheFFactorMethod()*fExtractor->GetResolutionPerPheHiGain();
|
|---|
| 1075 |
|
|---|
| 1076 | Float_t resinfadc = cal.IsHiGainSaturation()
|
|---|
| 1077 | ? resinphes/cal.GetMeanConvFADC2Phe()/cal.GetConversionHiLo()
|
|---|
| 1078 | : resinphes/cal.GetMeanConvFADC2Phe();
|
|---|
| 1079 |
|
|---|
| 1080 | if (resinfadc > 3.0*cal.GetPedRms() )
|
|---|
| 1081 | {
|
|---|
| 1082 | *fLog << warn << desc << "Extractor Resolution " << Form("%5.2f", resinfadc) << " bigger than 3 Pedestal RMS "
|
|---|
| 1083 | << Form("%4.2f", cal.GetPedRms()) << endl;
|
|---|
| 1084 | resinfadc = 3.0*cal.GetPedRms();
|
|---|
| 1085 | }
|
|---|
| 1086 |
|
|---|
| 1087 | if (!cal.CalcReducedSigma(resinfadc))
|
|---|
| 1088 | {
|
|---|
| 1089 | *fLog << warn << desc << "Could not calculate the reduced sigma" << endl;
|
|---|
| 1090 | bad.SetUncalibrated( MBadPixelsPix::kChargeSigmaNotValid );
|
|---|
| 1091 | return kFALSE;
|
|---|
| 1092 | }
|
|---|
| 1093 |
|
|---|
| 1094 | if (!cal.CalcFFactor())
|
|---|
| 1095 | {
|
|---|
| 1096 | *fLog << warn << desc << "Could not calculate the F-Factor" << endl;
|
|---|
| 1097 | bad.SetUncalibrated(MBadPixelsPix::kDeviatingNumPhes);
|
|---|
| 1098 | bad.SetUnsuitable(MBadPixelsPix::kUnsuitableRun);
|
|---|
| 1099 | return kFALSE;
|
|---|
| 1100 | }
|
|---|
| 1101 |
|
|---|
| 1102 | if (cal.GetPheFFactorMethod() < 0.)
|
|---|
| 1103 | {
|
|---|
| 1104 | bad.SetUncalibrated(MBadPixelsPix::kDeviatingNumPhes);
|
|---|
| 1105 | bad.SetUnsuitable(MBadPixelsPix::kUnsuitableRun);
|
|---|
| 1106 | cal.SetFFactorMethodValid(kFALSE);
|
|---|
| 1107 | return kFALSE;
|
|---|
| 1108 | }
|
|---|
| 1109 |
|
|---|
| 1110 | if (!cal.CalcConvFFactor())
|
|---|
| 1111 | {
|
|---|
| 1112 | *fLog << warn << desc << "Could not calculate the Conv. FADC counts to Phes" << endl;
|
|---|
| 1113 | bad.SetUncalibrated(MBadPixelsPix::kDeviatingNumPhes);
|
|---|
| 1114 | return kFALSE;
|
|---|
| 1115 | }
|
|---|
| 1116 |
|
|---|
| 1117 | return kTRUE;
|
|---|
| 1118 | }
|
|---|
| 1119 |
|
|---|
| 1120 | // -----------------------------------------------------------------------------------
|
|---|
| 1121 | //
|
|---|
| 1122 | // Test the arrival Times RMS of the pixel and set the bit
|
|---|
| 1123 | // - MBadPixelsPix::kFluctuatingArrivalTimes
|
|---|
| 1124 | //
|
|---|
| 1125 | void MCalibrationChargeCalc::FinalizeArrivalTimes(MCalibrationChargePix &cal, MBadPixelsPix &bad, const char* what)
|
|---|
| 1126 | {
|
|---|
| 1127 | if (bad.IsUnsuitable(MBadPixelsPix::kUnsuitableRun))
|
|---|
| 1128 | return;
|
|---|
| 1129 |
|
|---|
| 1130 | const TString desc = Form("%7s%4d: ", what, cal.GetPixId());
|
|---|
| 1131 |
|
|---|
| 1132 | if (cal.GetAbsTimeRms() > fArrTimeRmsLimit)
|
|---|
| 1133 | {
|
|---|
| 1134 | *fLog << warn;
|
|---|
| 1135 | *fLog << desc << "RMS of pulse arrival times: " << Form("%2.1f", cal.GetAbsTimeRms());
|
|---|
| 1136 | *fLog << " FADC sl. < " << Form("%2.1f", fArrTimeRmsLimit) << endl;
|
|---|
| 1137 | bad.SetUncalibrated( MBadPixelsPix::kFluctuatingArrivalTimes);
|
|---|
| 1138 | }
|
|---|
| 1139 | }
|
|---|
| 1140 |
|
|---|
| 1141 | // -----------------------------------------------------------------------------------
|
|---|
| 1142 | //
|
|---|
| 1143 | // Sets pixel to MBadPixelsPix::kUnsuitableRun, if one of the following flags is set:
|
|---|
| 1144 | // - MBadPixelsPix::kChargeIsPedestal
|
|---|
| 1145 | // - MBadPixelsPix::kChargeRelErrNotValid
|
|---|
| 1146 | // - MBadPixelsPix::kMeanTimeInFirstBin
|
|---|
| 1147 | // - MBadPixelsPix::kMeanTimeInLast2Bins
|
|---|
| 1148 | // - MBadPixelsPix::kDeviatingNumPhes
|
|---|
| 1149 | // - MBadPixelsPix::kHiGainOverFlow
|
|---|
| 1150 | // - MBadPixelsPix::kLoGainOverFlow
|
|---|
| 1151 | //
|
|---|
| 1152 | // - Call MCalibrationPix::SetExcluded() for the bad pixels
|
|---|
| 1153 | //
|
|---|
| 1154 | // Sets pixel to MBadPixelsPix::kUnreliableRun, if one of the following flags is set:
|
|---|
| 1155 | // - MBadPixelsPix::kChargeSigmaNotValid
|
|---|
| 1156 | //
|
|---|
| 1157 | void MCalibrationChargeCalc::FinalizeBadPixels()
|
|---|
| 1158 | {
|
|---|
| 1159 |
|
|---|
| 1160 | MBadPixelsCam *badcam = fIntensBad ? (MBadPixelsCam*)fIntensBad->GetCam() : fBadPixels;
|
|---|
| 1161 |
|
|---|
| 1162 | for (Int_t i=0; i<badcam->GetSize(); i++)
|
|---|
| 1163 | {
|
|---|
| 1164 |
|
|---|
| 1165 | MBadPixelsPix &bad = (*badcam)[i];
|
|---|
| 1166 |
|
|---|
| 1167 | if (IsCheckDeadPixels())
|
|---|
| 1168 | {
|
|---|
| 1169 | if (bad.IsUncalibrated( MBadPixelsPix::kChargeIsPedestal))
|
|---|
| 1170 | bad.SetUnsuitable( MBadPixelsPix::kUnsuitableRun );
|
|---|
| 1171 | }
|
|---|
| 1172 |
|
|---|
| 1173 | if (IsCheckExtractionWindow())
|
|---|
| 1174 | {
|
|---|
| 1175 | if (bad.IsUncalibrated( MBadPixelsPix::kMeanTimeInFirstBin ))
|
|---|
| 1176 | bad.SetUnsuitable( MBadPixelsPix::kUnsuitableRun );
|
|---|
| 1177 |
|
|---|
| 1178 | if (bad.IsUncalibrated( MBadPixelsPix::kMeanTimeInLast2Bins ))
|
|---|
| 1179 | bad.SetUnsuitable( MBadPixelsPix::kUnsuitableRun );
|
|---|
| 1180 | }
|
|---|
| 1181 |
|
|---|
| 1182 | if (IsCheckDeviatingBehavior())
|
|---|
| 1183 | {
|
|---|
| 1184 | if (bad.IsUncalibrated( MBadPixelsPix::kDeviatingNumPhes ))
|
|---|
| 1185 | bad.SetUnsuitable( MBadPixelsPix::kUnsuitableRun );
|
|---|
| 1186 | }
|
|---|
| 1187 |
|
|---|
| 1188 | if (IsCheckHistOverflow())
|
|---|
| 1189 | {
|
|---|
| 1190 | if (bad.IsUncalibrated( MBadPixelsPix::kHiGainOverFlow ))
|
|---|
| 1191 | bad.SetUnsuitable( MBadPixelsPix::kUnsuitableRun );
|
|---|
| 1192 |
|
|---|
| 1193 | if (bad.IsUncalibrated( MBadPixelsPix::kLoGainOverFlow ))
|
|---|
| 1194 | bad.SetUnsuitable( MBadPixelsPix::kUnsuitableRun );
|
|---|
| 1195 | }
|
|---|
| 1196 |
|
|---|
| 1197 | if (IsCheckArrivalTimes())
|
|---|
| 1198 | {
|
|---|
| 1199 | if (bad.IsUncalibrated( MBadPixelsPix::kFluctuatingArrivalTimes ))
|
|---|
| 1200 | bad.SetUnsuitable( MBadPixelsPix::kUnsuitableRun );
|
|---|
| 1201 | }
|
|---|
| 1202 |
|
|---|
| 1203 | if (bad.IsUncalibrated( MBadPixelsPix::kChargeSigmaNotValid ))
|
|---|
| 1204 | bad.SetUnsuitable( MBadPixelsPix::kUnreliableRun );
|
|---|
| 1205 | }
|
|---|
| 1206 | }
|
|---|
| 1207 |
|
|---|
| 1208 | // ------------------------------------------------------------------------
|
|---|
| 1209 | //
|
|---|
| 1210 | //
|
|---|
| 1211 | // First loop: Calculate a mean and mean RMS of photo-electrons per area index
|
|---|
| 1212 | // Include only pixels which are not MBadPixelsPix::kUnsuitableRun nor
|
|---|
| 1213 | // MBadPixelsPix::kChargeSigmaNotValid (see FinalizeBadPixels()) and set
|
|---|
| 1214 | // MCalibrationChargePix::SetFFactorMethodValid(kFALSE) in that case.
|
|---|
| 1215 | //
|
|---|
| 1216 | // Second loop: Get mean number of photo-electrons and its RMS including
|
|---|
| 1217 | // only pixels with flag MCalibrationChargePix::IsFFactorMethodValid()
|
|---|
| 1218 | // and further exclude those deviating by more than fPheErrLimit mean
|
|---|
| 1219 | // sigmas from the mean (obtained in first loop). Set
|
|---|
| 1220 | // MBadPixelsPix::kDeviatingNumPhes if excluded.
|
|---|
| 1221 | //
|
|---|
| 1222 | // For the suitable pixels with flag MBadPixelsPix::kChargeSigmaNotValid
|
|---|
| 1223 | // set the number of photo-electrons as the mean number of photo-electrons
|
|---|
| 1224 | // calculated in that area index.
|
|---|
| 1225 | //
|
|---|
| 1226 | // Set weighted mean and variance of photo-electrons per area index in:
|
|---|
| 1227 | // average area pixels of MCalibrationChargeCam (obtained from:
|
|---|
| 1228 | // MCalibrationChargeCam::GetAverageArea() )
|
|---|
| 1229 | //
|
|---|
| 1230 | // Set weighted mean and variance of photo-electrons per sector in:
|
|---|
| 1231 | // average sector pixels of MCalibrationChargeCam (obtained from:
|
|---|
| 1232 | // MCalibrationChargeCam::GetAverageSector() )
|
|---|
| 1233 | //
|
|---|
| 1234 | //
|
|---|
| 1235 | // Third loop: Set mean number of photo-electrons and its RMS in the pixels
|
|---|
| 1236 | // only excluded as: MBadPixelsPix::kChargeSigmaNotValid
|
|---|
| 1237 | //
|
|---|
| 1238 | Bool_t MCalibrationChargeCalc::FinalizeFFactorMethod()
|
|---|
| 1239 | {
|
|---|
| 1240 | MBadPixelsCam *badcam = fIntensBad
|
|---|
| 1241 | ? (MBadPixelsCam*) fIntensBad->GetCam() : fBadPixels;
|
|---|
| 1242 | MCalibrationChargeCam *chargecam = fIntensCam
|
|---|
| 1243 | ? (MCalibrationChargeCam*)fIntensCam->GetCam() : fCam;
|
|---|
| 1244 |
|
|---|
| 1245 | const Int_t npixels = fGeom->GetNumPixels();
|
|---|
| 1246 | const Int_t nareas = fGeom->GetNumAreas();
|
|---|
| 1247 | const Int_t nsectors = fGeom->GetNumSectors();
|
|---|
| 1248 |
|
|---|
| 1249 | TArrayF lowlim (nareas);
|
|---|
| 1250 | TArrayF upplim (nareas);
|
|---|
| 1251 | TArrayD areavars (nareas);
|
|---|
| 1252 | TArrayD areaweights (nareas);
|
|---|
| 1253 | TArrayD sectorweights (nsectors);
|
|---|
| 1254 | TArrayD areaphes (nareas);
|
|---|
| 1255 | TArrayD sectorphes (nsectors);
|
|---|
| 1256 | TArrayI numareavalid (nareas);
|
|---|
| 1257 | TArrayI numsectorvalid(nsectors);
|
|---|
| 1258 |
|
|---|
| 1259 | //
|
|---|
| 1260 | // First loop: Get mean number of photo-electrons and the RMS
|
|---|
| 1261 | // The loop is only to recognize later pixels with very deviating numbers
|
|---|
| 1262 | //
|
|---|
| 1263 | MHCamera camphes(*fGeom,"Camphes","Phes in Camera");
|
|---|
| 1264 |
|
|---|
| 1265 | for (Int_t i=0; i<npixels; i++)
|
|---|
| 1266 | {
|
|---|
| 1267 | MCalibrationChargePix &pix = (MCalibrationChargePix&)(*chargecam)[i];
|
|---|
| 1268 | MBadPixelsPix &bad = (*badcam)[i];
|
|---|
| 1269 |
|
|---|
| 1270 | if (!pix.IsFFactorMethodValid())
|
|---|
| 1271 | continue;
|
|---|
| 1272 |
|
|---|
| 1273 | if (bad.IsUnsuitable(MBadPixelsPix::kUnsuitableRun))
|
|---|
| 1274 | {
|
|---|
| 1275 | pix.SetFFactorMethodValid(kFALSE);
|
|---|
| 1276 | continue;
|
|---|
| 1277 | }
|
|---|
| 1278 |
|
|---|
| 1279 | if (bad.IsUncalibrated(MBadPixelsPix::kChargeSigmaNotValid))
|
|---|
| 1280 | continue;
|
|---|
| 1281 |
|
|---|
| 1282 | if (!IsUseUnreliables())
|
|---|
| 1283 | if (bad.IsUnsuitable(MBadPixelsPix::kUnreliableRun))
|
|---|
| 1284 | continue;
|
|---|
| 1285 |
|
|---|
| 1286 | const Float_t nphe = pix.GetPheFFactorMethod();
|
|---|
| 1287 | const Int_t aidx = (*fGeom)[i].GetAidx();
|
|---|
| 1288 | camphes.Fill(i,nphe);
|
|---|
| 1289 | camphes.SetUsed(i);
|
|---|
| 1290 | areaphes [aidx] += nphe;
|
|---|
| 1291 | areavars [aidx] += nphe*nphe;
|
|---|
| 1292 | numareavalid[aidx] ++;
|
|---|
| 1293 | }
|
|---|
| 1294 |
|
|---|
| 1295 | for (Int_t i=0; i<nareas; i++)
|
|---|
| 1296 | {
|
|---|
| 1297 | if (numareavalid[i] == 0)
|
|---|
| 1298 | {
|
|---|
| 1299 | *fLog << warn << GetDescriptor() << ": No pixels with valid number of photo-electrons found "
|
|---|
| 1300 | << "in area index: " << i << endl;
|
|---|
| 1301 | continue;
|
|---|
| 1302 | }
|
|---|
| 1303 |
|
|---|
| 1304 | if (numareavalid[i] == 1)
|
|---|
| 1305 | areavars[i] = 0.;
|
|---|
| 1306 | else
|
|---|
| 1307 | {
|
|---|
| 1308 | areavars[i] = (areavars[i] - areaphes[i]*areaphes[i]/numareavalid[i]) / (numareavalid[i]-1);
|
|---|
| 1309 | areaphes[i] = areaphes[i] / numareavalid[i];
|
|---|
| 1310 | }
|
|---|
| 1311 |
|
|---|
| 1312 | if (areavars[i] < 0.)
|
|---|
| 1313 | {
|
|---|
| 1314 | *fLog << warn << "Area " << setw(4) << i << ": No pixels with valid variance of photo-electrons found." << endl;
|
|---|
| 1315 | continue;
|
|---|
| 1316 | }
|
|---|
| 1317 |
|
|---|
| 1318 | lowlim[i] = areaphes[i] - fPheErrLowerLimit*TMath::Sqrt(areavars[i]);
|
|---|
| 1319 | upplim[i] = areaphes[i] + fPheErrUpperLimit*TMath::Sqrt(areavars[i]);
|
|---|
| 1320 |
|
|---|
| 1321 |
|
|---|
| 1322 | TH1D *hist = camphes.ProjectionS(TArrayI(),TArrayI(1,&i),"_py",100);
|
|---|
| 1323 | hist->Fit("gaus","Q0");
|
|---|
| 1324 | const Float_t mean = hist->GetFunction("gaus")->GetParameter(1);
|
|---|
| 1325 | const Float_t sigma = hist->GetFunction("gaus")->GetParameter(2);
|
|---|
| 1326 | const Int_t ndf = hist->GetFunction("gaus")->GetNDF();
|
|---|
| 1327 |
|
|---|
| 1328 | if (IsDebug())
|
|---|
| 1329 | hist->DrawClone();
|
|---|
| 1330 |
|
|---|
| 1331 | if (ndf < 5)
|
|---|
| 1332 | {
|
|---|
| 1333 | *fLog << warn << GetDescriptor() << ": Cannot use a Gauss fit to the number of photo-electrons " << endl;
|
|---|
| 1334 | *fLog << " in the camera with area index: " << i << endl;
|
|---|
| 1335 | *fLog << " Number of dof.: " << ndf << " is smaller than 5 " << endl;
|
|---|
| 1336 | *fLog << " Will use the simple mean and rms " << endl;
|
|---|
| 1337 | delete hist;
|
|---|
| 1338 | SetPheFitOK(i,kFALSE);
|
|---|
| 1339 | continue;
|
|---|
| 1340 | }
|
|---|
| 1341 |
|
|---|
| 1342 | const Double_t prob = hist->GetFunction("gaus")->GetProb();
|
|---|
| 1343 |
|
|---|
| 1344 | if (prob < 0.001)
|
|---|
| 1345 | {
|
|---|
| 1346 | *fLog << warn << GetDescriptor() << ": Cannot use a Gauss fit to the number of photo-electrons " << endl;
|
|---|
| 1347 | *fLog << " in the camera with area index: " << i << endl;
|
|---|
| 1348 | *fLog << " Fit probability " << prob << " is smaller than 0.001 " << endl;
|
|---|
| 1349 | *fLog << " Will use the simple mean and rms " << endl;
|
|---|
| 1350 | delete hist;
|
|---|
| 1351 | SetPheFitOK(i,kFALSE);
|
|---|
| 1352 | continue;
|
|---|
| 1353 | }
|
|---|
| 1354 |
|
|---|
| 1355 | if (mean < 0.)
|
|---|
| 1356 | {
|
|---|
| 1357 | *fLog << inf << "Area " << setw(4) << i << ": Fitted mean number of phe smaller 0." << endl;
|
|---|
| 1358 | *fLog << warn << "Area " << setw(4) << i << ": Will use the simple mean and rms " << endl;
|
|---|
| 1359 | SetPheFitOK(i,kFALSE);
|
|---|
| 1360 | delete hist;
|
|---|
| 1361 | continue;
|
|---|
| 1362 | }
|
|---|
| 1363 |
|
|---|
| 1364 | *fLog << inf << "Area " << setw(4) << i << ": Mean number of phes: "
|
|---|
| 1365 | << Form("%7.2f+-%6.2f",mean,sigma) << endl;
|
|---|
| 1366 |
|
|---|
| 1367 | lowlim[i] = mean - fPheErrLowerLimit*sigma;
|
|---|
| 1368 | upplim[i] = mean + fPheErrUpperLimit*sigma;
|
|---|
| 1369 |
|
|---|
| 1370 | delete hist;
|
|---|
| 1371 |
|
|---|
| 1372 | SetPheFitOK(i,kTRUE);
|
|---|
| 1373 | }
|
|---|
| 1374 |
|
|---|
| 1375 | *fLog << endl;
|
|---|
| 1376 |
|
|---|
| 1377 | numareavalid.Reset();
|
|---|
| 1378 | areaphes .Reset();
|
|---|
| 1379 | areavars .Reset();
|
|---|
| 1380 | //
|
|---|
| 1381 | // Second loop: Get mean number of photo-electrons and its RMS excluding
|
|---|
| 1382 | // pixels deviating by more than fPheErrLimit sigma.
|
|---|
| 1383 | // Set the conversion factor FADC counts to photo-electrons
|
|---|
| 1384 | //
|
|---|
| 1385 | for (Int_t i=0; i<npixels; i++)
|
|---|
| 1386 | {
|
|---|
| 1387 |
|
|---|
| 1388 | MCalibrationChargePix &pix = (MCalibrationChargePix&)(*chargecam)[i];
|
|---|
| 1389 |
|
|---|
| 1390 | if (!pix.IsFFactorMethodValid())
|
|---|
| 1391 | continue;
|
|---|
| 1392 |
|
|---|
| 1393 | MBadPixelsPix &bad = (*badcam)[i];
|
|---|
| 1394 |
|
|---|
| 1395 | if (bad.IsUncalibrated(MBadPixelsPix::kChargeSigmaNotValid))
|
|---|
| 1396 | continue;
|
|---|
| 1397 |
|
|---|
| 1398 | if (!IsUseUnreliables())
|
|---|
| 1399 | if (bad.IsUnsuitable(MBadPixelsPix::kUnreliableRun))
|
|---|
| 1400 | continue;
|
|---|
| 1401 |
|
|---|
| 1402 | const Float_t nvar = pix.GetPheFFactorMethodVar();
|
|---|
| 1403 | if (nvar <= 0.)
|
|---|
| 1404 | {
|
|---|
| 1405 | pix.SetFFactorMethodValid(kFALSE);
|
|---|
| 1406 | continue;
|
|---|
| 1407 | }
|
|---|
| 1408 |
|
|---|
| 1409 | const Int_t aidx = (*fGeom)[i].GetAidx();
|
|---|
| 1410 | const Int_t sector = (*fGeom)[i].GetSector();
|
|---|
| 1411 | const Float_t area = (*fGeom)[i].GetA();
|
|---|
| 1412 | const Float_t nphe = pix.GetPheFFactorMethod();
|
|---|
| 1413 |
|
|---|
| 1414 | if ( nphe < lowlim[aidx] || nphe > upplim[aidx] )
|
|---|
| 1415 | {
|
|---|
| 1416 | *fLog << warn << "Pixel " << setw(4) << i << ": Num of phe "
|
|---|
| 1417 | << Form("%7.2f out of +%3.1f-%3.1f sigma limit ",nphe,fPheErrUpperLimit,fPheErrLowerLimit)
|
|---|
| 1418 | << Form("[%7.2f,%7.2f]",lowlim[aidx],upplim[aidx]) << endl;
|
|---|
| 1419 | bad.SetUncalibrated( MBadPixelsPix::kDeviatingNumPhes );
|
|---|
| 1420 |
|
|---|
| 1421 | if (IsCheckDeviatingBehavior())
|
|---|
| 1422 | bad.SetUnsuitable(MBadPixelsPix::kUnsuitableRun);
|
|---|
| 1423 | continue;
|
|---|
| 1424 | }
|
|---|
| 1425 |
|
|---|
| 1426 | areaweights [aidx] += nphe*nphe;
|
|---|
| 1427 | areaphes [aidx] += nphe;
|
|---|
| 1428 | numareavalid [aidx] ++;
|
|---|
| 1429 |
|
|---|
| 1430 | if (aidx == 0)
|
|---|
| 1431 | fNumInnerFFactorMethodUsed++;
|
|---|
| 1432 |
|
|---|
| 1433 | sectorweights [sector] += nphe*nphe/area/area;
|
|---|
| 1434 | sectorphes [sector] += nphe/area;
|
|---|
| 1435 | numsectorvalid[sector] ++;
|
|---|
| 1436 | }
|
|---|
| 1437 |
|
|---|
| 1438 | *fLog << endl;
|
|---|
| 1439 |
|
|---|
| 1440 | for (Int_t aidx=0; aidx<nareas; aidx++)
|
|---|
| 1441 | {
|
|---|
| 1442 |
|
|---|
| 1443 | MCalibrationChargePix &apix = (MCalibrationChargePix&)chargecam->GetAverageArea(aidx);
|
|---|
| 1444 |
|
|---|
| 1445 | if (numareavalid[aidx] == 1)
|
|---|
| 1446 | areaweights[aidx] = 0.;
|
|---|
| 1447 | else if (numareavalid[aidx] == 0)
|
|---|
| 1448 | {
|
|---|
| 1449 | areaphes[aidx] = -1.;
|
|---|
| 1450 | areaweights[aidx] = -1.;
|
|---|
| 1451 | }
|
|---|
| 1452 | else
|
|---|
| 1453 | {
|
|---|
| 1454 | areaweights[aidx] = (areaweights[aidx] - areaphes[aidx]*areaphes[aidx]/numareavalid[aidx])
|
|---|
| 1455 | / (numareavalid[aidx]-1);
|
|---|
| 1456 | areaphes[aidx] /= numareavalid[aidx];
|
|---|
| 1457 | }
|
|---|
| 1458 |
|
|---|
| 1459 | if (areaweights[aidx] < 0. || areaphes[aidx] <= 0.)
|
|---|
| 1460 | {
|
|---|
| 1461 | *fLog << warn << "Area " << setw(4) << aidx << ": Mean number phes could not be calculated: "
|
|---|
| 1462 | << " Mean: " << areaphes[aidx]
|
|---|
| 1463 | << " Var: " << areaweights[aidx] << endl;
|
|---|
| 1464 | apix.SetFFactorMethodValid(kFALSE);
|
|---|
| 1465 | continue;
|
|---|
| 1466 | }
|
|---|
| 1467 |
|
|---|
| 1468 | *fLog << inf << "Area " << setw(4) << aidx << ": Average total phes: "
|
|---|
| 1469 | << Form("%7.2f +- %6.2f",areaphes[aidx],TMath::Sqrt(areaweights[aidx])) << endl;
|
|---|
| 1470 |
|
|---|
| 1471 | apix.SetPheFFactorMethod ( areaphes[aidx] );
|
|---|
| 1472 | apix.SetPheFFactorMethodVar( areaweights[aidx] / numareavalid[aidx] );
|
|---|
| 1473 | apix.SetFFactorMethodValid ( kTRUE );
|
|---|
| 1474 |
|
|---|
| 1475 | }
|
|---|
| 1476 |
|
|---|
| 1477 | *fLog << endl;
|
|---|
| 1478 |
|
|---|
| 1479 | for (Int_t sector=0; sector<nsectors; sector++)
|
|---|
| 1480 | {
|
|---|
| 1481 |
|
|---|
| 1482 | if (numsectorvalid[sector] == 1)
|
|---|
| 1483 | sectorweights[sector] = 0.;
|
|---|
| 1484 | else if (numsectorvalid[sector] == 0)
|
|---|
| 1485 | {
|
|---|
| 1486 | sectorphes[sector] = -1.;
|
|---|
| 1487 | sectorweights[sector] = -1.;
|
|---|
| 1488 | }
|
|---|
| 1489 | else
|
|---|
| 1490 | {
|
|---|
| 1491 | sectorweights[sector] = (sectorweights[sector]
|
|---|
| 1492 | - sectorphes[sector]*sectorphes[sector]/numsectorvalid[sector]
|
|---|
| 1493 | )
|
|---|
| 1494 | / (numsectorvalid[sector]-1.);
|
|---|
| 1495 | sectorphes[sector] /= numsectorvalid[sector];
|
|---|
| 1496 | }
|
|---|
| 1497 |
|
|---|
| 1498 | MCalibrationChargePix &spix = (MCalibrationChargePix&)chargecam->GetAverageSector(sector);
|
|---|
| 1499 |
|
|---|
| 1500 | if (sectorweights[sector] < 0. || sectorphes[sector] <= 0.)
|
|---|
| 1501 | {
|
|---|
| 1502 | *fLog << warn << "Sector " << setw(4) << sector
|
|---|
| 1503 | <<": Mean number phes/area could not be calculated:"
|
|---|
| 1504 | << " Mean: " << sectorphes[sector] << " Var: " << sectorweights[sector] << endl;
|
|---|
| 1505 | spix.SetFFactorMethodValid(kFALSE);
|
|---|
| 1506 | continue;
|
|---|
| 1507 | }
|
|---|
| 1508 |
|
|---|
| 1509 | *fLog << inf << "Sector " << setw(4) << sector
|
|---|
| 1510 | << ": Avg number phes/mm^2: "
|
|---|
| 1511 | << Form("%5.3f+-%4.3f",sectorphes[sector],TMath::Sqrt(sectorweights[sector]))
|
|---|
| 1512 | << endl;
|
|---|
| 1513 |
|
|---|
| 1514 | spix.SetPheFFactorMethod ( sectorphes[sector] );
|
|---|
| 1515 | spix.SetPheFFactorMethodVar( sectorweights[sector] / numsectorvalid[sector]);
|
|---|
| 1516 | spix.SetFFactorMethodValid ( kTRUE );
|
|---|
| 1517 |
|
|---|
| 1518 | }
|
|---|
| 1519 |
|
|---|
| 1520 | //
|
|---|
| 1521 | // Third loop: Set mean number of photo-electrons and its RMS in the pixels
|
|---|
| 1522 | // only excluded as: MBadPixelsPix::kChargeSigmaNotValid
|
|---|
| 1523 | //
|
|---|
| 1524 | for (Int_t i=0; i<npixels; i++)
|
|---|
| 1525 | {
|
|---|
| 1526 |
|
|---|
| 1527 | MCalibrationChargePix &pix = (MCalibrationChargePix&)(*chargecam)[i];
|
|---|
| 1528 | MBadPixelsPix &bad = (*badcam)[i];
|
|---|
| 1529 |
|
|---|
| 1530 | if (bad.IsUnsuitable(MBadPixelsPix::kUnsuitableRun))
|
|---|
| 1531 | continue;
|
|---|
| 1532 |
|
|---|
| 1533 | if (bad.IsUncalibrated(MBadPixelsPix::kChargeSigmaNotValid))
|
|---|
| 1534 | {
|
|---|
| 1535 | const Int_t aidx = (*fGeom)[i].GetAidx();
|
|---|
| 1536 | MCalibrationChargePix &apix = (MCalibrationChargePix&)chargecam->GetAverageArea(aidx);
|
|---|
| 1537 |
|
|---|
| 1538 | pix.SetPheFFactorMethod ( apix.GetPheFFactorMethod() );
|
|---|
| 1539 | pix.SetPheFFactorMethodVar( apix.GetPheFFactorMethodVar() );
|
|---|
| 1540 |
|
|---|
| 1541 | if (!pix.CalcConvFFactor())
|
|---|
| 1542 | {
|
|---|
| 1543 | *fLog << warn << GetDescriptor()
|
|---|
| 1544 | << "Pixel " << setw(4) << pix.GetPixId()
|
|---|
| 1545 | <<": Could not calculate the Conv. FADC counts to Phes"
|
|---|
| 1546 | << endl;
|
|---|
| 1547 | bad.SetUncalibrated( MBadPixelsPix::kDeviatingNumPhes );
|
|---|
| 1548 | if (IsCheckDeviatingBehavior())
|
|---|
| 1549 | bad.SetUnsuitable ( MBadPixelsPix::kUnsuitableRun );
|
|---|
| 1550 | }
|
|---|
| 1551 |
|
|---|
| 1552 | }
|
|---|
| 1553 | }
|
|---|
| 1554 |
|
|---|
| 1555 | return kTRUE;
|
|---|
| 1556 | }
|
|---|
| 1557 |
|
|---|
| 1558 |
|
|---|
| 1559 |
|
|---|
| 1560 | // ------------------------------------------------------------------------
|
|---|
| 1561 | //
|
|---|
| 1562 | // Returns kFALSE if pointer to MCalibrationBlindCam is NULL
|
|---|
| 1563 | //
|
|---|
| 1564 | // The check returns kFALSE if:
|
|---|
| 1565 | //
|
|---|
| 1566 | // 1) fLambda and fLambdaCheck are separated relatively to each other by more than fLambdaCheckLimit
|
|---|
| 1567 | // 2) BlindPixel has an fLambdaErr greater than fLambdaErrLimit
|
|---|
| 1568 | //
|
|---|
| 1569 | // Calls:
|
|---|
| 1570 | // - MCalibrationBlindPix::CalcFluxInsidePlexiglass()
|
|---|
| 1571 | //
|
|---|
| 1572 | Bool_t MCalibrationChargeCalc::FinalizeBlindCam()
|
|---|
| 1573 | {
|
|---|
| 1574 |
|
|---|
| 1575 | MCalibrationBlindCam *blindcam = fIntensBlind
|
|---|
| 1576 | ? (MCalibrationBlindCam*)fIntensBlind->GetCam() : fBlindCam;
|
|---|
| 1577 |
|
|---|
| 1578 | if (!blindcam)
|
|---|
| 1579 | return kFALSE;
|
|---|
| 1580 |
|
|---|
| 1581 | Int_t nvalid = 0;
|
|---|
| 1582 |
|
|---|
| 1583 | for (Int_t i=0; i<blindcam->GetSize(); i++)
|
|---|
| 1584 | {
|
|---|
| 1585 |
|
|---|
| 1586 | MCalibrationBlindPix &blindpix = (MCalibrationBlindPix&)(*blindcam)[i];
|
|---|
| 1587 |
|
|---|
| 1588 | if (!blindpix.IsValid())
|
|---|
| 1589 | continue;
|
|---|
| 1590 |
|
|---|
| 1591 | const Float_t lambda = blindpix.GetLambda();
|
|---|
| 1592 | const Float_t lambdaerr = blindpix.GetLambdaErr();
|
|---|
| 1593 | const Float_t lambdacheck = blindpix.GetLambdaCheck();
|
|---|
| 1594 |
|
|---|
| 1595 | if (2.*(lambdacheck-lambda)/(lambdacheck+lambda) > fLambdaCheckLimit)
|
|---|
| 1596 | {
|
|---|
| 1597 | *fLog << warn << "BlindPix " << i << ": Lambda="
|
|---|
| 1598 | << Form("%4.2f", lambda) << " and Lambda-Check="
|
|---|
| 1599 | << Form("%4.2f", lambdacheck) << " differ by more than "
|
|---|
| 1600 | << Form("%4.2f", fLambdaCheckLimit) << endl;
|
|---|
| 1601 | blindpix.SetValid(kFALSE);
|
|---|
| 1602 | continue;
|
|---|
| 1603 | }
|
|---|
| 1604 |
|
|---|
| 1605 | if (lambdaerr > fLambdaErrLimit)
|
|---|
| 1606 | {
|
|---|
| 1607 | *fLog << warn << "BlindPix " << i << ": Error of Fitted Lambda="
|
|---|
| 1608 | << Form("%4.2f", lambdaerr) << " is greater than "
|
|---|
| 1609 | << Form("%4.2f", fLambdaErrLimit) << endl;
|
|---|
| 1610 | blindpix.SetValid(kFALSE);
|
|---|
| 1611 | continue;
|
|---|
| 1612 | }
|
|---|
| 1613 |
|
|---|
| 1614 | if (!blindpix.CalcFluxInsidePlexiglass())
|
|---|
| 1615 | {
|
|---|
| 1616 | *fLog << warn << "Could not calculate the flux of photons from Blind Pixel Nr." << i << endl;
|
|---|
| 1617 | blindpix.SetValid(kFALSE);
|
|---|
| 1618 | continue;
|
|---|
| 1619 | }
|
|---|
| 1620 |
|
|---|
| 1621 | nvalid++;
|
|---|
| 1622 | }
|
|---|
| 1623 |
|
|---|
| 1624 | if (!nvalid)
|
|---|
| 1625 | return kFALSE;
|
|---|
| 1626 |
|
|---|
| 1627 | return kTRUE;
|
|---|
| 1628 | }
|
|---|
| 1629 |
|
|---|
| 1630 | // ------------------------------------------------------------------------
|
|---|
| 1631 | //
|
|---|
| 1632 | // Returns kFALSE if pointer to MCalibrationChargePINDiode is NULL
|
|---|
| 1633 | //
|
|---|
| 1634 | // The check returns kFALSE if:
|
|---|
| 1635 | //
|
|---|
| 1636 | // 1) PINDiode has a fitted charge smaller than fChargeLimit*PedRMS
|
|---|
| 1637 | // 2) PINDiode has a fit error smaller than fChargeErrLimit
|
|---|
| 1638 | // 3) PINDiode has a fitted charge smaller its fChargeRelErrLimit times its charge error
|
|---|
| 1639 | // 4) PINDiode has a charge sigma smaller than its Pedestal RMS
|
|---|
| 1640 | //
|
|---|
| 1641 | // Calls:
|
|---|
| 1642 | // - MCalibrationChargePINDiode::CalcFluxOutsidePlexiglass()
|
|---|
| 1643 | //
|
|---|
| 1644 | Bool_t MCalibrationChargeCalc::FinalizePINDiode()
|
|---|
| 1645 | {
|
|---|
| 1646 |
|
|---|
| 1647 | if (!fPINDiode)
|
|---|
| 1648 | return kFALSE;
|
|---|
| 1649 |
|
|---|
| 1650 | if (fPINDiode->GetMean() < fChargeLimit*fPINDiode->GetPedRms())
|
|---|
| 1651 | {
|
|---|
| 1652 | *fLog << warn << "PINDiode : Fitted Charge is smaller than "
|
|---|
| 1653 | << fChargeLimit << " Pedestal RMS." << endl;
|
|---|
| 1654 | return kFALSE;
|
|---|
| 1655 | }
|
|---|
| 1656 |
|
|---|
| 1657 | if (fPINDiode->GetMeanErr() < fChargeErrLimit)
|
|---|
| 1658 | {
|
|---|
| 1659 | *fLog << warn << "PINDiode : Error of Fitted Charge is smaller than "
|
|---|
| 1660 | << fChargeErrLimit << endl;
|
|---|
| 1661 | return kFALSE;
|
|---|
| 1662 | }
|
|---|
| 1663 |
|
|---|
| 1664 | if (fPINDiode->GetMean() < fChargeRelErrLimit*fPINDiode->GetMeanErr())
|
|---|
| 1665 | {
|
|---|
| 1666 | *fLog << warn << "PINDiode : Fitted Charge is smaller than "
|
|---|
| 1667 | << fChargeRelErrLimit << "* its error" << endl;
|
|---|
| 1668 | return kFALSE;
|
|---|
| 1669 | }
|
|---|
| 1670 |
|
|---|
| 1671 | if (fPINDiode->GetSigma() < fPINDiode->GetPedRms())
|
|---|
| 1672 | {
|
|---|
| 1673 | *fLog << warn << "PINDiode : Sigma of Fitted Charge smaller than Pedestal RMS" << endl;
|
|---|
| 1674 | return kFALSE;
|
|---|
| 1675 | }
|
|---|
| 1676 |
|
|---|
| 1677 |
|
|---|
| 1678 | if (!fPINDiode->CalcFluxOutsidePlexiglass())
|
|---|
| 1679 | {
|
|---|
| 1680 | *fLog << warn << "PINDiode : Could not calculate the flux of photons, "
|
|---|
| 1681 | << "will skip PIN Diode Calibration " << endl;
|
|---|
| 1682 | return kFALSE;
|
|---|
| 1683 | }
|
|---|
| 1684 |
|
|---|
| 1685 | return kTRUE;
|
|---|
| 1686 | }
|
|---|
| 1687 |
|
|---|
| 1688 | // ------------------------------------------------------------------------
|
|---|
| 1689 | //
|
|---|
| 1690 | // Calculate the average number of photons outside the plexiglass with the
|
|---|
| 1691 | // formula:
|
|---|
| 1692 | //
|
|---|
| 1693 | // av.Num.photons(area index) = av.Num.Phes(area index)
|
|---|
| 1694 | // / MCalibrationQEPix::GetDefaultQE(fPulserColor)
|
|---|
| 1695 | // / MCalibrationQEPix::GetPMTCollectionEff()
|
|---|
| 1696 | // / MCalibrationQEPix::GetLightGuidesEff(fPulserColor)
|
|---|
| 1697 | // / MCalibrationQECam::GetPlexiglassQE()
|
|---|
| 1698 | //
|
|---|
| 1699 | // Calculate the variance on the average number of photons assuming that the error on the
|
|---|
| 1700 | // Quantum efficiency is reduced by the number of used inner pixels, but the rest of the
|
|---|
| 1701 | // values keeps it ordinary error since it is systematic.
|
|---|
| 1702 | //
|
|---|
| 1703 | // Loop over pixels:
|
|---|
| 1704 | //
|
|---|
| 1705 | // - Continue, if not MCalibrationChargePix::IsFFactorMethodValid() and set:
|
|---|
| 1706 | // MCalibrationQEPix::SetFFactorMethodValid(kFALSE,fPulserColor)
|
|---|
| 1707 | //
|
|---|
| 1708 | // - Call MCalibrationChargePix::CalcMeanFFactor(av.Num.photons) and set:
|
|---|
| 1709 | // MCalibrationQEPix::SetFFactorMethodValid(kFALSE,fPulserColor) if not succesful
|
|---|
| 1710 | //
|
|---|
| 1711 | // - Calculate the quantum efficiency with the formula:
|
|---|
| 1712 | //
|
|---|
| 1713 | // QE = ( Num.Phes / av.Num.photons ) * MGeomCam::GetPixRatio()
|
|---|
| 1714 | //
|
|---|
| 1715 | // - Set QE in MCalibrationQEPix::SetQEFFactor ( QE, fPulserColor );
|
|---|
| 1716 | //
|
|---|
| 1717 | // - Set Variance of QE in MCalibrationQEPix::SetQEFFactorVar ( Variance, fPulserColor );
|
|---|
| 1718 | // - Set bit MCalibrationQEPix::SetFFactorMethodValid(kTRUE,fPulserColor)
|
|---|
| 1719 | //
|
|---|
| 1720 | // - Call MCalibrationQEPix::UpdateFFactorMethod()
|
|---|
| 1721 | //
|
|---|
| 1722 | void MCalibrationChargeCalc::FinalizeFFactorQECam()
|
|---|
| 1723 | {
|
|---|
| 1724 |
|
|---|
| 1725 | if (fNumInnerFFactorMethodUsed < 2)
|
|---|
| 1726 | {
|
|---|
| 1727 | *fLog << warn << GetDescriptor()
|
|---|
| 1728 | << ": Could not calculate F-Factor Method: Less than 2 inner pixels valid! " << endl;
|
|---|
| 1729 | return;
|
|---|
| 1730 | }
|
|---|
| 1731 |
|
|---|
| 1732 | MCalibrationQECam *qecam = fIntensQE
|
|---|
| 1733 | ? (MCalibrationQECam*) fIntensQE->GetCam() : fQECam;
|
|---|
| 1734 | MCalibrationChargeCam *chargecam = fIntensCam
|
|---|
| 1735 | ? (MCalibrationChargeCam*)fIntensCam->GetCam() : fCam;
|
|---|
| 1736 | MBadPixelsCam *badcam = fIntensBad
|
|---|
| 1737 | ? (MBadPixelsCam*) fIntensBad->GetCam() : fBadPixels;
|
|---|
| 1738 |
|
|---|
| 1739 | MCalibrationChargePix &avpix = (MCalibrationChargePix&)chargecam->GetAverageArea(0);
|
|---|
| 1740 | MCalibrationQEPix &qepix = (MCalibrationQEPix&) qecam->GetAverageArea(0);
|
|---|
| 1741 |
|
|---|
| 1742 | if (IsDebug())
|
|---|
| 1743 | *fLog << dbginf << "External Phes: " << fExternalNumPhes
|
|---|
| 1744 | << " Internal Phes: " << avpix.GetPheFFactorMethod() << endl;
|
|---|
| 1745 |
|
|---|
| 1746 | const Float_t avphotons = ( IsUseExternalNumPhes()
|
|---|
| 1747 | ? fExternalNumPhes
|
|---|
| 1748 | : avpix.GetPheFFactorMethod() )
|
|---|
| 1749 | / qepix.GetDefaultQE(fPulserColor)
|
|---|
| 1750 | / qepix.GetPMTCollectionEff()
|
|---|
| 1751 | / qepix.GetLightGuidesEff(fPulserColor)
|
|---|
| 1752 | / qecam->GetPlexiglassQE();
|
|---|
| 1753 |
|
|---|
| 1754 | const Float_t avphotrelvar = ( IsUseExternalNumPhes()
|
|---|
| 1755 | ? fExternalNumPhesRelVar
|
|---|
| 1756 | : avpix.GetPheFFactorMethodRelVar() )
|
|---|
| 1757 | + qepix.GetDefaultQERelVar(fPulserColor) / fNumInnerFFactorMethodUsed
|
|---|
| 1758 | + qepix.GetPMTCollectionEffRelVar()
|
|---|
| 1759 | + qepix.GetLightGuidesEffRelVar(fPulserColor)
|
|---|
| 1760 | + qecam->GetPlexiglassQERelVar();
|
|---|
| 1761 |
|
|---|
| 1762 | const UInt_t nareas = fGeom->GetNumAreas();
|
|---|
| 1763 |
|
|---|
| 1764 | //
|
|---|
| 1765 | // Set the results in the MCalibrationChargeCam
|
|---|
| 1766 | //
|
|---|
| 1767 | chargecam->SetNumPhotonsFFactorMethod (avphotons);
|
|---|
| 1768 |
|
|---|
| 1769 | if (avphotrelvar > 0.)
|
|---|
| 1770 | chargecam->SetNumPhotonsFFactorMethodErr(TMath::Sqrt( avphotrelvar * avphotons * avphotons));
|
|---|
| 1771 |
|
|---|
| 1772 | TArrayF lowlim (nareas);
|
|---|
| 1773 | TArrayF upplim (nareas);
|
|---|
| 1774 | TArrayD avffactorphotons (nareas);
|
|---|
| 1775 | TArrayD avffactorphotvar (nareas);
|
|---|
| 1776 | TArrayI numffactor (nareas);
|
|---|
| 1777 |
|
|---|
| 1778 | const UInt_t npixels = fGeom->GetNumPixels();
|
|---|
| 1779 |
|
|---|
| 1780 | MHCamera camffactor(*fGeom,"Camffactor","F-Factor in Camera");
|
|---|
| 1781 |
|
|---|
| 1782 | for (UInt_t i=0; i<npixels; i++)
|
|---|
| 1783 | {
|
|---|
| 1784 |
|
|---|
| 1785 | MCalibrationChargePix &pix = (MCalibrationChargePix&)(*chargecam)[i];
|
|---|
| 1786 | MCalibrationQEPix &qpix = (MCalibrationQEPix&) (*qecam) [i];
|
|---|
| 1787 | MBadPixelsPix &bad = (*badcam) [i];
|
|---|
| 1788 |
|
|---|
| 1789 | if (bad.IsUnsuitable(MBadPixelsPix::kUnsuitableRun))
|
|---|
| 1790 | continue;
|
|---|
| 1791 |
|
|---|
| 1792 | const Float_t photons = avphotons / fGeom->GetPixRatio(i);
|
|---|
| 1793 | const Float_t qe = pix.GetPheFFactorMethod() / photons ;
|
|---|
| 1794 |
|
|---|
| 1795 | const Float_t qerelvar = avphotrelvar + pix.GetPheFFactorMethodRelVar();
|
|---|
| 1796 |
|
|---|
| 1797 | qpix.SetQEFFactor ( qe , fPulserColor );
|
|---|
| 1798 | qpix.SetQEFFactorVar ( qerelvar*qe*qe, fPulserColor );
|
|---|
| 1799 | qpix.SetFFactorMethodValid( kTRUE , fPulserColor );
|
|---|
| 1800 |
|
|---|
| 1801 | if (!qpix.UpdateFFactorMethod( qecam->GetPlexiglassQE() ))
|
|---|
| 1802 | *fLog << warn << GetDescriptor()
|
|---|
| 1803 | << ": Cannot update Quantum efficiencies with the F-Factor Method" << endl;
|
|---|
| 1804 |
|
|---|
| 1805 | //
|
|---|
| 1806 | // The following pixels are those with deviating sigma, but otherwise OK,
|
|---|
| 1807 | // probably those with stars during the pedestal run, but not the cal. run.
|
|---|
| 1808 | //
|
|---|
| 1809 | if (!pix.CalcMeanFFactor( photons , avphotrelvar ))
|
|---|
| 1810 | {
|
|---|
| 1811 | bad.SetUncalibrated( MBadPixelsPix::kDeviatingFFactor );
|
|---|
| 1812 | if (IsCheckDeviatingBehavior())
|
|---|
| 1813 | bad.SetUnsuitable ( MBadPixelsPix::kUnreliableRun );
|
|---|
| 1814 | continue;
|
|---|
| 1815 | }
|
|---|
| 1816 |
|
|---|
| 1817 | const Int_t aidx = (*fGeom)[i].GetAidx();
|
|---|
| 1818 | const Float_t ffactor = pix.GetMeanFFactorFADC2Phot();
|
|---|
| 1819 |
|
|---|
| 1820 | camffactor.Fill(i,ffactor);
|
|---|
| 1821 | camffactor.SetUsed(i);
|
|---|
| 1822 |
|
|---|
| 1823 | avffactorphotons[aidx] += ffactor;
|
|---|
| 1824 | avffactorphotvar[aidx] += ffactor*ffactor;
|
|---|
| 1825 | numffactor[aidx]++;
|
|---|
| 1826 | }
|
|---|
| 1827 |
|
|---|
| 1828 | for (UInt_t i=0; i<nareas; i++)
|
|---|
| 1829 | {
|
|---|
| 1830 |
|
|---|
| 1831 | if (numffactor[i] == 0)
|
|---|
| 1832 | {
|
|---|
| 1833 | *fLog << warn << GetDescriptor() << ": No pixels with valid total F-Factor found "
|
|---|
| 1834 | << "in area index: " << i << endl;
|
|---|
| 1835 | continue;
|
|---|
| 1836 | }
|
|---|
| 1837 |
|
|---|
| 1838 | avffactorphotvar[i] = (avffactorphotvar[i] - avffactorphotons[i]*avffactorphotons[i]/numffactor[i])
|
|---|
| 1839 | / (numffactor[i]-1.);
|
|---|
| 1840 | avffactorphotons[i] = avffactorphotons[i] / numffactor[i];
|
|---|
| 1841 |
|
|---|
| 1842 | if (avffactorphotvar[i] < 0.)
|
|---|
| 1843 | {
|
|---|
| 1844 | *fLog << warn << GetDescriptor() << ": No pixels with valid variance of total F-Factor found "
|
|---|
| 1845 | << "in area index: " << i << endl;
|
|---|
| 1846 | continue;
|
|---|
| 1847 | }
|
|---|
| 1848 |
|
|---|
| 1849 | lowlim [i] = 1.; // Lowest known F-Factor of a PMT
|
|---|
| 1850 | upplim [i] = avffactorphotons[i] + fFFactorErrLimit*TMath::Sqrt(avffactorphotvar[i]);
|
|---|
| 1851 |
|
|---|
| 1852 | TArrayI area(1);
|
|---|
| 1853 | area[0] = i;
|
|---|
| 1854 |
|
|---|
| 1855 | TH1D *hist = camffactor.ProjectionS(TArrayI(),area,"_py",100);
|
|---|
| 1856 | hist->Fit("gaus","Q0");
|
|---|
| 1857 | const Float_t mean = hist->GetFunction("gaus")->GetParameter(1);
|
|---|
| 1858 | const Float_t sigma = hist->GetFunction("gaus")->GetParameter(2);
|
|---|
| 1859 | const Int_t ndf = hist->GetFunction("gaus")->GetNDF();
|
|---|
| 1860 |
|
|---|
| 1861 | if (IsDebug())
|
|---|
| 1862 | camffactor.DrawClone();
|
|---|
| 1863 |
|
|---|
| 1864 | if (ndf < 2)
|
|---|
| 1865 | {
|
|---|
| 1866 | *fLog << warn << GetDescriptor() << ": Cannot use a Gauss fit to the F-Factor " << endl;
|
|---|
| 1867 | *fLog << " in the camera with area index: " << i << endl;
|
|---|
| 1868 | *fLog << " Number of dof.: " << ndf << " is smaller than 2 " << endl;
|
|---|
| 1869 | *fLog << " Will use the simple mean and rms." << endl;
|
|---|
| 1870 | delete hist;
|
|---|
| 1871 | SetFFactorFitOK(i,kFALSE);
|
|---|
| 1872 | continue;
|
|---|
| 1873 | }
|
|---|
| 1874 |
|
|---|
| 1875 | const Double_t prob = hist->GetFunction("gaus")->GetProb();
|
|---|
| 1876 |
|
|---|
| 1877 | if (prob < 0.001)
|
|---|
| 1878 | {
|
|---|
| 1879 | *fLog << warn << GetDescriptor() << ": Cannot use a Gauss fit to the F-Factor " << endl;
|
|---|
| 1880 | *fLog << " in the camera with area index: " << i << endl;
|
|---|
| 1881 | *fLog << " Fit probability " << prob << " is smaller than 0.001 " << endl;
|
|---|
| 1882 | *fLog << " Will use the simple mean and rms." << endl;
|
|---|
| 1883 | delete hist;
|
|---|
| 1884 | SetFFactorFitOK(i,kFALSE);
|
|---|
| 1885 | continue;
|
|---|
| 1886 | }
|
|---|
| 1887 |
|
|---|
| 1888 | *fLog << inf << "Area " << setw(4) << i <<": Mean F-Factor :"
|
|---|
| 1889 | << Form("%4.2f+-%4.2f",mean,sigma) << endl;
|
|---|
| 1890 |
|
|---|
| 1891 | lowlim [i] = 1.;
|
|---|
| 1892 | upplim [i] = mean + fFFactorErrLimit*sigma;
|
|---|
| 1893 |
|
|---|
| 1894 | delete hist;
|
|---|
| 1895 |
|
|---|
| 1896 | SetFFactorFitOK(i,kTRUE);
|
|---|
| 1897 | }
|
|---|
| 1898 |
|
|---|
| 1899 | *fLog << endl;
|
|---|
| 1900 |
|
|---|
| 1901 | for (UInt_t i=0; i<npixels; i++)
|
|---|
| 1902 | {
|
|---|
| 1903 |
|
|---|
| 1904 | MCalibrationChargePix &pix = (MCalibrationChargePix&)(*chargecam)[i];
|
|---|
| 1905 | MBadPixelsPix &bad = (*badcam) [i];
|
|---|
| 1906 |
|
|---|
| 1907 | if (bad.IsUnsuitable(MBadPixelsPix::kUnsuitableRun))
|
|---|
| 1908 | continue;
|
|---|
| 1909 |
|
|---|
| 1910 | const Float_t ffactor = pix.GetMeanFFactorFADC2Phot();
|
|---|
| 1911 | const Int_t aidx = (*fGeom)[i].GetAidx();
|
|---|
| 1912 |
|
|---|
| 1913 | if ( ffactor < lowlim[aidx] || ffactor > upplim[aidx] )
|
|---|
| 1914 | {
|
|---|
| 1915 | *fLog << warn << "Pixel " << setw(4) << i<< ": Overall F-Factor "
|
|---|
| 1916 | << Form("%5.2f",ffactor) << " out of range ["
|
|---|
| 1917 | << Form("%5.2f,%5.2f",lowlim[aidx],upplim[aidx]) << "]" << endl;
|
|---|
| 1918 |
|
|---|
| 1919 | bad.SetUncalibrated( MBadPixelsPix::kDeviatingFFactor );
|
|---|
| 1920 | if (IsCheckDeviatingBehavior())
|
|---|
| 1921 | bad.SetUnsuitable ( MBadPixelsPix::kUnreliableRun );
|
|---|
| 1922 | }
|
|---|
| 1923 | }
|
|---|
| 1924 |
|
|---|
| 1925 | for (UInt_t i=0; i<npixels; i++)
|
|---|
| 1926 | {
|
|---|
| 1927 |
|
|---|
| 1928 | MCalibrationChargePix &pix = (MCalibrationChargePix&)(*chargecam)[i];
|
|---|
| 1929 | MCalibrationQEPix &qpix = (MCalibrationQEPix&) (*qecam) [i];
|
|---|
| 1930 | MBadPixelsPix &bad = (*badcam) [i];
|
|---|
| 1931 |
|
|---|
| 1932 | if (bad.IsUnsuitable(MBadPixelsPix::kUnsuitableRun))
|
|---|
| 1933 | {
|
|---|
| 1934 | qpix.SetFFactorMethodValid(kFALSE,fPulserColor);
|
|---|
| 1935 | pix.SetFFactorMethodValid(kFALSE);
|
|---|
| 1936 | continue;
|
|---|
| 1937 | }
|
|---|
| 1938 | }
|
|---|
| 1939 | }
|
|---|
| 1940 |
|
|---|
| 1941 |
|
|---|
| 1942 | // ------------------------------------------------------------------------
|
|---|
| 1943 | //
|
|---|
| 1944 | // Loop over pixels:
|
|---|
| 1945 | //
|
|---|
| 1946 | // - Continue, if not MCalibrationBlindPix::IsFluxInsidePlexiglassAvailable() and set:
|
|---|
| 1947 | // MCalibrationQEPix::SetBlindPixelMethodValid(kFALSE,fPulserColor)
|
|---|
| 1948 | //
|
|---|
| 1949 | // - Calculate the quantum efficiency with the formula:
|
|---|
| 1950 | //
|
|---|
| 1951 | // QE = Num.Phes / MCalibrationBlindPix::GetFluxInsidePlexiglass()
|
|---|
| 1952 | // / MGeomPix::GetA() * MCalibrationQECam::GetPlexiglassQE()
|
|---|
| 1953 | //
|
|---|
| 1954 | // - Set QE in MCalibrationQEPix::SetQEBlindPixel ( QE, fPulserColor );
|
|---|
| 1955 | // - Set Variance of QE in MCalibrationQEPix::SetQEBlindPixelVar ( Variance, fPulserColor );
|
|---|
| 1956 | // - Set bit MCalibrationQEPix::SetBlindPixelMethodValid(kTRUE,fPulserColor)
|
|---|
| 1957 | //
|
|---|
| 1958 | // - Call MCalibrationQEPix::UpdateBlindPixelMethod()
|
|---|
| 1959 | //
|
|---|
| 1960 | void MCalibrationChargeCalc::FinalizeBlindPixelQECam()
|
|---|
| 1961 | {
|
|---|
| 1962 |
|
|---|
| 1963 |
|
|---|
| 1964 | MCalibrationBlindCam *blindcam = fIntensBlind
|
|---|
| 1965 | ? (MCalibrationBlindCam*) fIntensBlind->GetCam(): fBlindCam;
|
|---|
| 1966 | MBadPixelsCam *badcam = fIntensBad
|
|---|
| 1967 | ? (MBadPixelsCam*) fIntensBad->GetCam() : fBadPixels;
|
|---|
| 1968 | MCalibrationQECam *qecam = fIntensQE
|
|---|
| 1969 | ? (MCalibrationQECam*) fIntensQE->GetCam() : fQECam;
|
|---|
| 1970 | MCalibrationChargeCam *chargecam = fIntensCam
|
|---|
| 1971 | ? (MCalibrationChargeCam*)fIntensCam->GetCam() : fCam;
|
|---|
| 1972 |
|
|---|
| 1973 | if (!blindcam)
|
|---|
| 1974 | return;
|
|---|
| 1975 |
|
|---|
| 1976 | //
|
|---|
| 1977 | // Set the results in the MCalibrationChargeCam
|
|---|
| 1978 | //
|
|---|
| 1979 | if (!blindcam || !(blindcam->IsFluxInsidePlexiglassAvailable()))
|
|---|
| 1980 | {
|
|---|
| 1981 |
|
|---|
| 1982 | const Float_t photons = blindcam->GetFluxInsidePlexiglass() * (*fGeom)[0].GetA()
|
|---|
| 1983 | / qecam->GetPlexiglassQE();
|
|---|
| 1984 | chargecam->SetNumPhotonsBlindPixelMethod(photons);
|
|---|
| 1985 |
|
|---|
| 1986 | const Float_t photrelvar = blindcam->GetFluxInsidePlexiglassRelVar()
|
|---|
| 1987 | + qecam->GetPlexiglassQERelVar();
|
|---|
| 1988 |
|
|---|
| 1989 | if (photrelvar > 0.)
|
|---|
| 1990 | chargecam->SetNumPhotonsBlindPixelMethodErr(TMath::Sqrt( photrelvar * photons * photons));
|
|---|
| 1991 | }
|
|---|
| 1992 |
|
|---|
| 1993 | //
|
|---|
| 1994 | // With the knowledge of the overall photon flux, calculate the
|
|---|
| 1995 | // quantum efficiencies after the Blind Pixel and PIN Diode method
|
|---|
| 1996 | //
|
|---|
| 1997 | const UInt_t npixels = fGeom->GetNumPixels();
|
|---|
| 1998 | for (UInt_t i=0; i<npixels; i++)
|
|---|
| 1999 | {
|
|---|
| 2000 |
|
|---|
| 2001 | MCalibrationQEPix &qepix = (MCalibrationQEPix&) (*qecam) [i];
|
|---|
| 2002 |
|
|---|
| 2003 | if (!blindcam || !(blindcam->IsFluxInsidePlexiglassAvailable()))
|
|---|
| 2004 | {
|
|---|
| 2005 | qepix.SetBlindPixelMethodValid(kFALSE, fPulserColor);
|
|---|
| 2006 | continue;
|
|---|
| 2007 | }
|
|---|
| 2008 |
|
|---|
| 2009 | MBadPixelsPix &bad = (*badcam) [i];
|
|---|
| 2010 |
|
|---|
| 2011 | if (bad.IsUnsuitable (MBadPixelsPix::kUnsuitableRun))
|
|---|
| 2012 | {
|
|---|
| 2013 | qepix.SetBlindPixelMethodValid(kFALSE, fPulserColor);
|
|---|
| 2014 | continue;
|
|---|
| 2015 | }
|
|---|
| 2016 |
|
|---|
| 2017 | MCalibrationChargePix &pix = (MCalibrationChargePix&)(*chargecam)[i];
|
|---|
| 2018 | MGeomPix &geo = (*fGeom) [i];
|
|---|
| 2019 |
|
|---|
| 2020 | const Float_t qe = pix.GetPheFFactorMethod()
|
|---|
| 2021 | / blindcam->GetFluxInsidePlexiglass()
|
|---|
| 2022 | / geo.GetA()
|
|---|
| 2023 | * qecam->GetPlexiglassQE();
|
|---|
| 2024 |
|
|---|
| 2025 | const Float_t qerelvar = blindcam->GetFluxInsidePlexiglassRelVar()
|
|---|
| 2026 | + qecam->GetPlexiglassQERelVar()
|
|---|
| 2027 | + pix.GetPheFFactorMethodRelVar();
|
|---|
| 2028 |
|
|---|
| 2029 | qepix.SetQEBlindPixel ( qe , fPulserColor );
|
|---|
| 2030 | qepix.SetQEBlindPixelVar ( qerelvar*qe*qe, fPulserColor );
|
|---|
| 2031 | qepix.SetBlindPixelMethodValid( kTRUE , fPulserColor );
|
|---|
| 2032 |
|
|---|
| 2033 | if (!qepix.UpdateBlindPixelMethod( qecam->GetPlexiglassQE()))
|
|---|
| 2034 | *fLog << warn << GetDescriptor()
|
|---|
| 2035 | << ": Cannot update Quantum efficiencies with the Blind Pixel Method" << endl;
|
|---|
| 2036 | }
|
|---|
| 2037 | }
|
|---|
| 2038 |
|
|---|
| 2039 | // ------------------------------------------------------------------------
|
|---|
| 2040 | //
|
|---|
| 2041 | // Loop over pixels:
|
|---|
| 2042 | //
|
|---|
| 2043 | // - Continue, if not MCalibrationChargePINDiode::IsFluxOutsidePlexiglassAvailable() and set:
|
|---|
| 2044 | // MCalibrationQEPix::SetPINDiodeMethodValid(kFALSE,fPulserColor)
|
|---|
| 2045 | //
|
|---|
| 2046 | // - Calculate the quantum efficiency with the formula:
|
|---|
| 2047 | //
|
|---|
| 2048 | // QE = Num.Phes / MCalibrationChargePINDiode::GetFluxOutsidePlexiglass() / MGeomPix::GetA()
|
|---|
| 2049 | //
|
|---|
| 2050 | // - Set QE in MCalibrationQEPix::SetQEPINDiode ( QE, fPulserColor );
|
|---|
| 2051 | // - Set Variance of QE in MCalibrationQEPix::SetQEPINDiodeVar ( Variance, fPulserColor );
|
|---|
| 2052 | // - Set bit MCalibrationQEPix::SetPINDiodeMethodValid(kTRUE,fPulserColor)
|
|---|
| 2053 | //
|
|---|
| 2054 | // - Call MCalibrationQEPix::UpdatePINDiodeMethod()
|
|---|
| 2055 | //
|
|---|
| 2056 | void MCalibrationChargeCalc::FinalizePINDiodeQECam()
|
|---|
| 2057 | {
|
|---|
| 2058 |
|
|---|
| 2059 | const UInt_t npixels = fGeom->GetNumPixels();
|
|---|
| 2060 |
|
|---|
| 2061 | MCalibrationQECam *qecam = fIntensQE
|
|---|
| 2062 | ? (MCalibrationQECam*) fIntensQE->GetCam() : fQECam;
|
|---|
| 2063 | MCalibrationChargeCam *chargecam = fIntensCam
|
|---|
| 2064 | ? (MCalibrationChargeCam*)fIntensCam->GetCam() : fCam;
|
|---|
| 2065 | MBadPixelsCam *badcam = fIntensBad
|
|---|
| 2066 | ? (MBadPixelsCam*) fIntensBad->GetCam() : fBadPixels;
|
|---|
| 2067 |
|
|---|
| 2068 | if (!fPINDiode)
|
|---|
| 2069 | return;
|
|---|
| 2070 |
|
|---|
| 2071 | //
|
|---|
| 2072 | // With the knowledge of the overall photon flux, calculate the
|
|---|
| 2073 | // quantum efficiencies after the PIN Diode method
|
|---|
| 2074 | //
|
|---|
| 2075 | for (UInt_t i=0; i<npixels; i++)
|
|---|
| 2076 | {
|
|---|
| 2077 |
|
|---|
| 2078 | MCalibrationQEPix &qepix = (MCalibrationQEPix&) (*qecam) [i];
|
|---|
| 2079 |
|
|---|
| 2080 | if (!fPINDiode)
|
|---|
| 2081 | {
|
|---|
| 2082 | qepix.SetPINDiodeMethodValid(kFALSE, fPulserColor);
|
|---|
| 2083 | continue;
|
|---|
| 2084 | }
|
|---|
| 2085 |
|
|---|
| 2086 | if (!fPINDiode->IsFluxOutsidePlexiglassAvailable())
|
|---|
| 2087 | {
|
|---|
| 2088 | qepix.SetPINDiodeMethodValid(kFALSE, fPulserColor);
|
|---|
| 2089 | continue;
|
|---|
| 2090 | }
|
|---|
| 2091 |
|
|---|
| 2092 | MBadPixelsPix &bad = (*badcam) [i];
|
|---|
| 2093 |
|
|---|
| 2094 | if (!bad.IsUnsuitable (MBadPixelsPix::kUnsuitableRun))
|
|---|
| 2095 | {
|
|---|
| 2096 | qepix.SetPINDiodeMethodValid(kFALSE, fPulserColor);
|
|---|
| 2097 | continue;
|
|---|
| 2098 | }
|
|---|
| 2099 |
|
|---|
| 2100 | MCalibrationChargePix &pix = (MCalibrationChargePix&)(*chargecam)[i];
|
|---|
| 2101 | MGeomPix &geo = (*fGeom) [i];
|
|---|
| 2102 |
|
|---|
| 2103 | const Float_t qe = pix.GetPheFFactorMethod()
|
|---|
| 2104 | / fPINDiode->GetFluxOutsidePlexiglass()
|
|---|
| 2105 | / geo.GetA();
|
|---|
| 2106 |
|
|---|
| 2107 | const Float_t qerelvar = fPINDiode->GetFluxOutsidePlexiglassRelVar() + pix.GetPheFFactorMethodRelVar();
|
|---|
| 2108 |
|
|---|
| 2109 | qepix.SetQEPINDiode ( qe , fPulserColor );
|
|---|
| 2110 | qepix.SetQEPINDiodeVar ( qerelvar*qe*qe, fPulserColor );
|
|---|
| 2111 | qepix.SetPINDiodeMethodValid( kTRUE , fPulserColor );
|
|---|
| 2112 |
|
|---|
| 2113 | if (!qepix.UpdatePINDiodeMethod())
|
|---|
| 2114 | *fLog << warn << GetDescriptor()
|
|---|
| 2115 | << ": Cannot update Quantum efficiencies with the PIN Diode Method" << endl;
|
|---|
| 2116 | }
|
|---|
| 2117 | }
|
|---|
| 2118 |
|
|---|
| 2119 | // ------------------------------------------------------------------------
|
|---|
| 2120 | //
|
|---|
| 2121 | // Loop over pixels:
|
|---|
| 2122 | //
|
|---|
| 2123 | // - Call MCalibrationQEPix::UpdateCombinedMethod()
|
|---|
| 2124 | //
|
|---|
| 2125 | void MCalibrationChargeCalc::FinalizeCombinedQECam()
|
|---|
| 2126 | {
|
|---|
| 2127 |
|
|---|
| 2128 | const UInt_t npixels = fGeom->GetNumPixels();
|
|---|
| 2129 |
|
|---|
| 2130 | MCalibrationQECam *qecam = fIntensQE
|
|---|
| 2131 | ? (MCalibrationQECam*) fIntensQE->GetCam() : fQECam;
|
|---|
| 2132 | MBadPixelsCam *badcam = fIntensBad
|
|---|
| 2133 | ? (MBadPixelsCam*) fIntensBad->GetCam() : fBadPixels;
|
|---|
| 2134 |
|
|---|
| 2135 | for (UInt_t i=0; i<npixels; i++)
|
|---|
| 2136 | {
|
|---|
| 2137 |
|
|---|
| 2138 | MCalibrationQEPix &qepix = (MCalibrationQEPix&) (*qecam) [i];
|
|---|
| 2139 | MBadPixelsPix &bad = (*badcam) [i];
|
|---|
| 2140 |
|
|---|
| 2141 | if (!bad.IsUnsuitable (MBadPixelsPix::kUnsuitableRun))
|
|---|
| 2142 | {
|
|---|
| 2143 | qepix.SetPINDiodeMethodValid(kFALSE, fPulserColor);
|
|---|
| 2144 | continue;
|
|---|
| 2145 | }
|
|---|
| 2146 |
|
|---|
| 2147 | qepix.UpdateCombinedMethod();
|
|---|
| 2148 | }
|
|---|
| 2149 | }
|
|---|
| 2150 |
|
|---|
| 2151 | // -----------------------------------------------------------------------------------------------
|
|---|
| 2152 | //
|
|---|
| 2153 | // - Print out statistics about BadPixels of type UnsuitableType_t
|
|---|
| 2154 | // - store numbers of bad pixels of each type in fCam or fIntensCam
|
|---|
| 2155 | //
|
|---|
| 2156 | Bool_t MCalibrationChargeCalc::FinalizeUnsuitablePixels()
|
|---|
| 2157 | {
|
|---|
| 2158 |
|
|---|
| 2159 | *fLog << inf << endl;
|
|---|
| 2160 | *fLog << GetDescriptor() << ": Charge Calibration status:" << endl;
|
|---|
| 2161 | *fLog << dec << setfill(' ');
|
|---|
| 2162 |
|
|---|
| 2163 | const Int_t nareas = fGeom->GetNumAreas();
|
|---|
| 2164 |
|
|---|
| 2165 | TArrayI suit(nareas);
|
|---|
| 2166 | TArrayI unsuit(nareas);
|
|---|
| 2167 | TArrayI unrel(nareas);
|
|---|
| 2168 |
|
|---|
| 2169 | const MBadPixelsCam *badcam = fIntensBad ? (MBadPixelsCam*)fIntensBad->GetCam() : fBadPixels;
|
|---|
| 2170 | MCalibrationChargeCam *chargecam = fIntensCam ? (MCalibrationChargeCam*)fIntensCam->GetCam() : fCam;
|
|---|
| 2171 |
|
|---|
| 2172 | Int_t unsuitcnt=0;
|
|---|
| 2173 | Int_t unrelcnt =0;
|
|---|
| 2174 |
|
|---|
| 2175 | // Count number of succesfully calibrated pixels
|
|---|
| 2176 | for (Int_t aidx=0; aidx<nareas; aidx++)
|
|---|
| 2177 | {
|
|---|
| 2178 | suit[aidx] = badcam->GetNumSuitable(MBadPixelsPix::kUnsuitableRun, fGeom, aidx);
|
|---|
| 2179 | unsuit[aidx] = badcam->GetNumUnsuitable(MBadPixelsPix::kUnsuitableRun, fGeom, aidx);
|
|---|
| 2180 | unrel[aidx] = badcam->GetNumUnsuitable(MBadPixelsPix::kUnreliableRun, fGeom, aidx);
|
|---|
| 2181 |
|
|---|
| 2182 | unsuitcnt += unsuit[aidx];
|
|---|
| 2183 | unrelcnt += unrel[aidx];
|
|---|
| 2184 |
|
|---|
| 2185 | chargecam->SetNumUnsuitable(unsuit[aidx], aidx);
|
|---|
| 2186 | chargecam->SetNumUnreliable(unrel[aidx], aidx);
|
|---|
| 2187 | }
|
|---|
| 2188 |
|
|---|
| 2189 | TArrayI counts(nareas);
|
|---|
| 2190 | for (Int_t i=0; i<chargecam->GetSize(); i++)
|
|---|
| 2191 | {
|
|---|
| 2192 | MCalibrationPix &pix = (*chargecam)[i];
|
|---|
| 2193 | if (pix.IsHiGainSaturation())
|
|---|
| 2194 | {
|
|---|
| 2195 | const Int_t aidx = (*fGeom)[i].GetAidx();
|
|---|
| 2196 | counts[aidx]++;
|
|---|
| 2197 | }
|
|---|
| 2198 | }
|
|---|
| 2199 |
|
|---|
| 2200 | if (fGeom->InheritsFrom("MGeomCamMagic"))
|
|---|
| 2201 | {
|
|---|
| 2202 | *fLog << " Successfully calibrated Pixels: Inner: "
|
|---|
| 2203 | << Form("%3i",suit[0]) << " Outer: " << Form("%3i",suit[1]) << endl;
|
|---|
| 2204 | *fLog << " Uncalibrated Pixels: Inner: "
|
|---|
| 2205 | << Form("%3i",unsuit[0]) << " Outer: " << Form("%3i",unsuit[1]) << endl;
|
|---|
| 2206 | *fLog << " Unreliable Pixels: Inner: "
|
|---|
| 2207 | << Form("%3i",unrel[0]) << " Outer: " << Form("%3i",unrel[1]) << endl;
|
|---|
| 2208 | *fLog << " High-gain saturated Pixels: Inner: "
|
|---|
| 2209 | << Form("%3i",counts[0]) << " Outer: " << Form("%3i",counts[1]) << endl;
|
|---|
| 2210 | *fLog << endl;
|
|---|
| 2211 | }
|
|---|
| 2212 |
|
|---|
| 2213 | if (unsuitcnt > fUnsuitablesLimit*fGeom->GetNumPixels())
|
|---|
| 2214 | {
|
|---|
| 2215 | *fLog << err << "Number of unsuitable pixels: " << 100.*unsuitcnt/fGeom->GetNumPixels()
|
|---|
| 2216 | << "% exceeds limit of " << fUnsuitablesLimit*100 << "%" << endl;
|
|---|
| 2217 | return kFALSE;
|
|---|
| 2218 | }
|
|---|
| 2219 |
|
|---|
| 2220 | if (unrelcnt > fUnreliablesLimit*fGeom->GetNumPixels())
|
|---|
| 2221 | {
|
|---|
| 2222 | *fLog << err << "Relative number of unreliable pixels: " << 100.*unrelcnt/fGeom->GetNumPixels()
|
|---|
| 2223 | << "% exceeds limit of " << fUnreliablesLimit*100 << "%" << endl;
|
|---|
| 2224 | return kFALSE;
|
|---|
| 2225 | }
|
|---|
| 2226 | return kTRUE;
|
|---|
| 2227 | }
|
|---|
| 2228 |
|
|---|
| 2229 | // -----------------------------------------------------------------------------------------------
|
|---|
| 2230 | //
|
|---|
| 2231 | // Print out statistics about BadPixels of type UncalibratedType_t
|
|---|
| 2232 | //
|
|---|
| 2233 | void MCalibrationChargeCalc::PrintUncalibrated(MBadPixelsPix::UncalibratedType_t typ, const char *text) const
|
|---|
| 2234 | {
|
|---|
| 2235 |
|
|---|
| 2236 | UInt_t countinner = 0;
|
|---|
| 2237 | UInt_t countouter = 0;
|
|---|
| 2238 |
|
|---|
| 2239 | MBadPixelsCam *badcam = fIntensBad
|
|---|
| 2240 | ? (MBadPixelsCam*)fIntensBad->GetCam() : fBadPixels;
|
|---|
| 2241 |
|
|---|
| 2242 | for (Int_t i=0; i<badcam->GetSize(); i++)
|
|---|
| 2243 | {
|
|---|
| 2244 | MBadPixelsPix &bad = (*badcam)[i];
|
|---|
| 2245 |
|
|---|
| 2246 | if (bad.IsUncalibrated(typ))
|
|---|
| 2247 | {
|
|---|
| 2248 | if (fGeom->GetPixRatio(i) == 1.)
|
|---|
| 2249 | countinner++;
|
|---|
| 2250 | else
|
|---|
| 2251 | countouter++;
|
|---|
| 2252 | }
|
|---|
| 2253 | }
|
|---|
| 2254 |
|
|---|
| 2255 | *fLog << " " << setw(7) << text << "Inner: " << Form("%3i",countinner)
|
|---|
| 2256 | << " Outer: " << Form("%3i", countouter) << endl;
|
|---|
| 2257 | }
|
|---|
| 2258 |
|
|---|
| 2259 | // --------------------------------------------------------------------------
|
|---|
| 2260 | //
|
|---|
| 2261 | // Read the environment for the following data members:
|
|---|
| 2262 | // - fChargeLimit
|
|---|
| 2263 | // - fChargeErrLimit
|
|---|
| 2264 | // - fChargeRelErrLimit
|
|---|
| 2265 | // - fDebug
|
|---|
| 2266 | // - fFFactorErrLimit
|
|---|
| 2267 | // - fLambdaErrLimit
|
|---|
| 2268 | // - fLambdaCheckErrLimit
|
|---|
| 2269 | // - fPheErrLimit
|
|---|
| 2270 | //
|
|---|
| 2271 | Int_t MCalibrationChargeCalc::ReadEnv(const TEnv &env, TString prefix, Bool_t print)
|
|---|
| 2272 | {
|
|---|
| 2273 |
|
|---|
| 2274 | Bool_t rc = kFALSE;
|
|---|
| 2275 | if (IsEnvDefined(env, prefix, "ChargeLimit", print))
|
|---|
| 2276 | {
|
|---|
| 2277 | SetChargeLimit(GetEnvValue(env, prefix, "ChargeLimit", fChargeLimit));
|
|---|
| 2278 | rc = kTRUE;
|
|---|
| 2279 | }
|
|---|
| 2280 | if (IsEnvDefined(env, prefix, "ChargeErrLimit", print))
|
|---|
| 2281 | {
|
|---|
| 2282 | SetChargeErrLimit(GetEnvValue(env, prefix, "ChargeErrLimit", fChargeErrLimit));
|
|---|
| 2283 | rc = kTRUE;
|
|---|
| 2284 | }
|
|---|
| 2285 | if (IsEnvDefined(env, prefix, "ChargeRelErrLimit", print))
|
|---|
| 2286 | {
|
|---|
| 2287 | SetChargeRelErrLimit(GetEnvValue(env, prefix, "ChargeRelErrLimit", fChargeRelErrLimit));
|
|---|
| 2288 | rc = kTRUE;
|
|---|
| 2289 | }
|
|---|
| 2290 | if (IsEnvDefined(env, prefix, "Debug", print))
|
|---|
| 2291 | {
|
|---|
| 2292 | SetDebug(GetEnvValue(env, prefix, "Debug", IsDebug()));
|
|---|
| 2293 | rc = kTRUE;
|
|---|
| 2294 | }
|
|---|
| 2295 | if (IsEnvDefined(env, prefix, "ArrTimeRmsLimit", print))
|
|---|
| 2296 | {
|
|---|
| 2297 | SetArrTimeRmsLimit(GetEnvValue(env, prefix, "ArrTimeRmsLimit", fArrTimeRmsLimit));
|
|---|
| 2298 | rc = kTRUE;
|
|---|
| 2299 | }
|
|---|
| 2300 | if (IsEnvDefined(env, prefix, "FFactorErrLimit", print))
|
|---|
| 2301 | {
|
|---|
| 2302 | SetFFactorErrLimit(GetEnvValue(env, prefix, "FFactorErrLimit", fFFactorErrLimit));
|
|---|
| 2303 | rc = kTRUE;
|
|---|
| 2304 | }
|
|---|
| 2305 | if (IsEnvDefined(env, prefix, "LambdaErrLimit", print))
|
|---|
| 2306 | {
|
|---|
| 2307 | SetLambdaErrLimit(GetEnvValue(env, prefix, "LambdaErrLimit", fLambdaErrLimit));
|
|---|
| 2308 | rc = kTRUE;
|
|---|
| 2309 | }
|
|---|
| 2310 | if (IsEnvDefined(env, prefix, "LambdaCheckLimit", print))
|
|---|
| 2311 | {
|
|---|
| 2312 | SetLambdaCheckLimit(GetEnvValue(env, prefix, "LambdaCheckLimit", fLambdaCheckLimit));
|
|---|
| 2313 | rc = kTRUE;
|
|---|
| 2314 | }
|
|---|
| 2315 | if (IsEnvDefined(env, prefix, "CheckDeadPixels", print))
|
|---|
| 2316 | {
|
|---|
| 2317 | SetCheckDeadPixels(GetEnvValue(env, prefix, "CheckDeadPixels", IsCheckDeadPixels()));
|
|---|
| 2318 | rc = kTRUE;
|
|---|
| 2319 | }
|
|---|
| 2320 | if (IsEnvDefined(env, prefix, "CheckDeviatingBehavior", print))
|
|---|
| 2321 | {
|
|---|
| 2322 | SetCheckDeviatingBehavior(GetEnvValue(env, prefix, "CheckDeviatingBehavior", IsCheckDeviatingBehavior()));
|
|---|
| 2323 | rc = kTRUE;
|
|---|
| 2324 | }
|
|---|
| 2325 | if (IsEnvDefined(env, prefix, "CheckExtractionWindow", print))
|
|---|
| 2326 | {
|
|---|
| 2327 | SetCheckExtractionWindow(GetEnvValue(env, prefix, "CheckExtractionWindow", IsCheckExtractionWindow()));
|
|---|
| 2328 | rc = kTRUE;
|
|---|
| 2329 | }
|
|---|
| 2330 | if (IsEnvDefined(env, prefix, "CheckHistOverflow", print))
|
|---|
| 2331 | {
|
|---|
| 2332 | SetCheckHistOverflow(GetEnvValue(env, prefix, "CheckHistOverflow", IsCheckHistOverflow()));
|
|---|
| 2333 | rc = kTRUE;
|
|---|
| 2334 | }
|
|---|
| 2335 | if (IsEnvDefined(env, prefix, "CheckOscillations", print))
|
|---|
| 2336 | {
|
|---|
| 2337 | SetCheckOscillations(GetEnvValue(env, prefix, "CheckOscillations", IsCheckOscillations()));
|
|---|
| 2338 | rc = kTRUE;
|
|---|
| 2339 | }
|
|---|
| 2340 | if (IsEnvDefined(env, prefix, "CheckArrivalTimes", print))
|
|---|
| 2341 | {
|
|---|
| 2342 | SetCheckArrivalTimes(GetEnvValue(env, prefix, "CheckArrivalTimes", IsCheckArrivalTimes()));
|
|---|
| 2343 | rc = kTRUE;
|
|---|
| 2344 | }
|
|---|
| 2345 | if (IsEnvDefined(env, prefix, "PheErrLowerLimit", print))
|
|---|
| 2346 | {
|
|---|
| 2347 | SetPheErrLowerLimit(GetEnvValue(env, prefix, "PheErrLowerLimit", fPheErrLowerLimit));
|
|---|
| 2348 | rc = kTRUE;
|
|---|
| 2349 | }
|
|---|
| 2350 | if (IsEnvDefined(env, prefix, "PheErrUpperLimit", print))
|
|---|
| 2351 | {
|
|---|
| 2352 | SetPheErrUpperLimit(GetEnvValue(env, prefix, "PheErrUpperLimit", fPheErrUpperLimit));
|
|---|
| 2353 | rc = kTRUE;
|
|---|
| 2354 | }
|
|---|
| 2355 | if (IsEnvDefined(env, prefix, "UseExtractorRes", print))
|
|---|
| 2356 | {
|
|---|
| 2357 | SetUseExtractorRes(GetEnvValue(env, prefix, "UseExtractorRes", IsUseExtractorRes()));
|
|---|
| 2358 | rc = kTRUE;
|
|---|
| 2359 | }
|
|---|
| 2360 | if (IsEnvDefined(env, prefix, "UseUnreliables", print))
|
|---|
| 2361 | {
|
|---|
| 2362 | SetUseUnreliables(GetEnvValue(env, prefix, "UseUnreliables", IsUseUnreliables()));
|
|---|
| 2363 | rc = kTRUE;
|
|---|
| 2364 | }
|
|---|
| 2365 |
|
|---|
| 2366 | if (IsEnvDefined(env, prefix, "UseExternalNumPhes", print))
|
|---|
| 2367 | {
|
|---|
| 2368 | SetUseExternalNumPhes(GetEnvValue(env, prefix, "UseExternalNumPhes", IsUseExternalNumPhes()));
|
|---|
| 2369 | rc = kTRUE;
|
|---|
| 2370 | }
|
|---|
| 2371 |
|
|---|
| 2372 | if (IsEnvDefined(env, prefix, "UnsuitablesLimit", print))
|
|---|
| 2373 | {
|
|---|
| 2374 | SetUnsuitablesLimit(GetEnvValue(env, prefix, "UnsuitablesLimit", fUnsuitablesLimit));
|
|---|
| 2375 | rc = kTRUE;
|
|---|
| 2376 | }
|
|---|
| 2377 |
|
|---|
| 2378 | if (IsEnvDefined(env, prefix, "UnreliablesLimit", print))
|
|---|
| 2379 | {
|
|---|
| 2380 | SetUnreliablesLimit(GetEnvValue(env, prefix, "UnreliablesLimit", fUnreliablesLimit));
|
|---|
| 2381 | rc = kTRUE;
|
|---|
| 2382 | }
|
|---|
| 2383 |
|
|---|
| 2384 |
|
|---|
| 2385 | return rc;
|
|---|
| 2386 | }
|
|---|