**************************************************************************/
/* $Id$ */
-/* History of cvs commits:
- *
- * $Log$
- * Revision 1.10 2007/12/06 13:58:11 hristov
- * Additional pritection. Do not delete the mapping, it is owned by another class
- *
- * Revision 1.9 2007/12/06 02:19:51 jklay
- * incorporated fitting procedure from testbeam analysis into AliRoot
- *
- * Revision 1.8 2007/12/05 02:30:51 jklay
- * modification to read Altro mappings into AliEMCALRecParam and pass to AliEMCALRawUtils from AliEMCALReconstructor; add option to AliEMCALRawUtils to set old RCU format (for testbeam) or not
- *
- * Revision 1.7 2007/11/14 15:51:46 gustavo
- * Take out few unnecessary prints
- *
- * Revision 1.6 2007/11/01 01:23:51 mvl
- * Removed call to SetOldRCUFormat, which is only needed for testbeam data
- *
- * Revision 1.5 2007/11/01 01:20:33 mvl
- * Further improvement of peak finding; more robust fit
- *
- * Revision 1.4 2007/10/31 17:15:24 mvl
- * Fixed bug in raw data unpacking; Added pedestal to signal fit; Added logic to deal with high/low gain
- *
- * Revision 1.3 2007/09/27 08:36:46 mvl
- * More robust setting of fit range in FitRawSignal (P. Hristov)
- *
- * Revision 1.2 2007/09/03 20:55:35 jklay
- * EMCAL e-by-e reconstruction methods from Cvetan
- *
- * Revision 1.1 2007/03/17 19:56:38 mvl
- * Moved signal shape routines from AliEMCAL to separate class AliEMCALRawUtils to streamline raw data reconstruction code.
- * */
+//_________________________________________________________________________
+// Utility Class for handling Raw data
+// Does all transitions from Digits to Raw and vice versa,
+// for simu and reconstruction
+//
+// Note: the current version is still simplified. Only
+// one raw signal per digit is generated; either high-gain or low-gain
+// Need to add concurrent high and low-gain info in the future
+// No pedestal is added to the raw signal.
//*-- Author: Marco van Leeuwen (LBL)
+
#include "AliEMCALRawUtils.h"
#include "TF1.h"
#include "TGraph.h"
-#include "TSystem.h"
+#include <TRandom.h>
+class TSystem;
-#include "AliLog.h"
+class AliLog;
#include "AliRun.h"
#include "AliRunLoader.h"
-#include "AliCaloAltroMapping.h"
+class AliCaloAltroMapping;
#include "AliAltroBuffer.h"
#include "AliRawReader.h"
-#include "AliCaloRawStream.h"
+#include "AliCaloRawStreamV3.h"
#include "AliDAQ.h"
#include "AliEMCALRecParam.h"
#include "AliEMCALLoader.h"
#include "AliEMCALGeometry.h"
-#include "AliEMCALDigitizer.h"
+class AliEMCALDigitizer;
#include "AliEMCALDigit.h"
#include "AliEMCAL.h"
ClassImp(AliEMCALRawUtils)
// Signal shape parameters
- Double_t AliEMCALRawUtils::fgTimeBinWidth = 100E-9 ; // each sample is 100 ns
+Int_t AliEMCALRawUtils::fgTimeBins = 256; // number of time bins for EMCAL
+Double_t AliEMCALRawUtils::fgTimeBinWidth = 100E-9 ; // each sample is 100 ns
Double_t AliEMCALRawUtils::fgTimeTrigger = 1.5E-6 ; // 15 time bins ~ 1.5 musec
// some digitization constants
Int_t AliEMCALRawUtils::fgThreshold = 1;
Int_t AliEMCALRawUtils::fgDDLPerSuperModule = 2; // 2 ddls per SuperModule
-Int_t AliEMCALRawUtils::fgPedestalValue = 32; // pedestal value for digits2raw
-Double_t AliEMCALRawUtils::fgFEENoise = 3.; // 3 ADC channels of noise (sampled)
+Int_t AliEMCALRawUtils::fgPedestalValue = 32; // pedestal value for digits2raw
+Double_t AliEMCALRawUtils::fgFEENoise = 3.; // 3 ADC channels of noise (sampled)
AliEMCALRawUtils::AliEMCALRawUtils()
: fHighLowGainFactor(0.), fOrder(0), fTau(0.), fNoiseThreshold(0),
//To make sure we match with the geometry in a simulation file,
//let's try to get it first. If not, take the default geometry
- AliRunLoader *rl = AliRunLoader::GetRunLoader();
+ AliRunLoader *rl = AliRunLoader::Instance();
if(!rl) AliError("Cannot find RunLoader!");
if (rl->GetAliRun() && rl->GetAliRun()->GetDetector("EMCAL")) {
fGeom = dynamic_cast<AliEMCAL*>(rl->GetAliRun()->GetDetector("EMCAL"))->GetGeometry();
//____________________________________________________________________________
AliEMCALRawUtils::~AliEMCALRawUtils() {
+ //dtor
}
{
// convert digits of the current event to raw data
- AliRunLoader *rl = AliRunLoader::GetRunLoader();
+ AliRunLoader *rl = AliRunLoader::Instance();
AliEMCALLoader *loader = dynamic_cast<AliEMCALLoader*>(rl->GetDetectorLoader("EMCAL"));
// get the digits
for (Int_t i=0; i < nDDL; i++)
buffers[i] = 0;
- Int_t adcValuesLow[fgkTimeBins];
- Int_t adcValuesHigh[fgkTimeBins];
+ TArrayI adcValuesLow(fgTimeBins);
+ TArrayI adcValuesHigh(fgTimeBins);
// loop over digits (assume ordered digits)
for (Int_t iDigit = 0; iDigit < digits->GetEntries(); iDigit++) {
buffers[iDDL]->WriteTrailer(3, ieta, iphi, nSM); // trailer
// calculate the time response function
} else {
- Bool_t lowgain = RawSampledResponse(digit->GetTimeR(), digit->GetAmp(), adcValuesHigh, adcValuesLow) ;
+ Bool_t lowgain = RawSampledResponse(digit->GetTimeR(), digit->GetAmp(), adcValuesHigh.GetArray(), adcValuesLow.GetArray()) ;
if (lowgain)
- buffers[iDDL]->WriteChannel(ieta, iphi, 0, GetRawFormatTimeBins(), adcValuesLow, fgThreshold);
+ buffers[iDDL]->WriteChannel(ieta, iphi, 0, GetRawFormatTimeBins(), adcValuesLow.GetArray(), fgThreshold);
else
- buffers[iDDL]->WriteChannel(ieta,iphi, 1, GetRawFormatTimeBins(), adcValuesHigh, fgThreshold);
+ buffers[iDDL]->WriteChannel(ieta,iphi, 1, GetRawFormatTimeBins(), adcValuesHigh.GetArray(), fgThreshold);
}
}
return;
}
- AliCaloRawStream in(reader,"EMCAL",fMapping);
+ AliCaloRawStreamV3 in(reader,"EMCAL",fMapping);
// Select EMCAL DDL's;
- reader->Select("EMCAL");
+ reader->Select("EMCAL",0,43);
//Updated fitting routine from 2007 beam test takes into account
//possibility of two peaks in data and selects first one for fitting
Int_t id = -1;
Float_t time = 0. ;
Float_t amp = 0. ;
+ Int_t i = 0;
+ Int_t startBin = 0;
//Graph to hold data we will fit (should be converted to an array
//later to speed up processing
TGraph * gSig = new TGraph(GetRawFormatTimeBins());
- Int_t readOk = 1;
Int_t lowGain = 0;
+ Int_t caloFlag = 0; // low, high gain, or TRU, or LED ref.
- while (readOk && in.GetModule() < 0)
- readOk = in.Next(); // Go to first digit
-
- Int_t col = 0;
- Int_t row = 0;
-
- while (readOk) {
-
- id = fGeom->GetAbsCellIdFromCellIndexes(in.GetModule(), in.GetRow(), in.GetColumn()) ;
- lowGain = in.IsLowGain();
- Int_t maxTime = in.GetTime(); // timebins come in reverse order
- if (maxTime < 0 || maxTime >= GetRawFormatTimeBins()) {
- AliWarning(Form("Invalid time bin %d",maxTime));
- maxTime = GetRawFormatTimeBins();
- }
- gSig->Set(maxTime+1);
- // There is some kind of zero-suppression in the raw data,
- // so set up the TGraph in advance
- for (Int_t i=0; i < maxTime; i++) {
- gSig->SetPoint(i, i , 0);
- }
-
- Int_t iTime = 0;
- do {
- if (in.GetTime() >= gSig->GetN()) {
- AliWarning("Too many time bins");
- gSig->Set(in.GetTime());
- }
- col = in.GetColumn();
- row = in.GetRow();
+ // start loop over input stream
+ while (in.NextDDL()) {
+ while (in.NextChannel()) {
- gSig->SetPoint(in.GetTime(), in.GetTime(), in.GetSignal()) ;
+ //Check if the signal is high or low gain and then do the fit,
+ //if it is from TRU do not fit
+ caloFlag = in.GetCaloFlag();
+ if (caloFlag != 0 && caloFlag != 1) continue;
+
+ // There can be zero-suppression in the raw data,
+ // so set up the TGraph in advance
+ for (i=0; i < GetRawFormatTimeBins(); i++) {
+ gSig->SetPoint(i, i , 0);
+ }
+
+ Int_t maxTime = 0;
+ Int_t min = 0x3ff; // init to 10-bit max
+ Int_t max = 0; // init to 10-bit min
+ int nsamples = 0;
+ while (in.NextBunch()) {
+ const UShort_t *sig = in.GetSignals();
+ startBin = in.GetStartTimeBin();
+
+ if (((UInt_t) maxTime) < in.GetStartTimeBin()) {
+ maxTime = in.GetStartTimeBin(); // timebins come in reverse order
+ }
+
+ if (maxTime < 0 || maxTime >= GetRawFormatTimeBins()) {
+ AliWarning(Form("Invalid time bin %d",maxTime));
+ maxTime = GetRawFormatTimeBins();
+ }
+ nsamples += in.GetBunchLength();
+ for (i = 0; i < in.GetBunchLength(); i++) {
+ time = startBin--;
+ gSig->SetPoint(time, time, (Double_t) sig[i]) ;
+ if (max < sig[i]) max= sig[i];
+ if (min > sig[i]) min = sig[i];
+ }
+ } // loop over bunches
+
+ if (nsamples > 0) { // this check is needed for when we have zero-supp. on, but not sparse readout
- if (in.GetTime() > maxTime)
- maxTime = in.GetTime();
- iTime++;
- } while ((readOk = in.Next()) && !in.IsNewHWAddress());
+ id = fGeom->GetAbsCellIdFromCellIndexes(in.GetModule(), in.GetRow(), in.GetColumn()) ;
+ lowGain = in.IsLowGain();
- FitRaw(gSig, signalF, amp, time) ;
+ gSig->Set(maxTime+1);
+ if ( (max - min) > fNoiseThreshold) FitRaw(gSig, signalF, amp, time) ;
- if (amp > 0 && amp < 2000) { //check both high and low end of
- //result, 2000 is somewhat arbitrary
- AliDebug(2,Form("id %d lowGain %d amp %g", id, lowGain, amp));
-
- AddDigit(digitsArr, id, lowGain, (Int_t)amp, time);
- }
+ //if (caloFlag == 0 || caloFlag == 1) { // low gain or high gain
+ if (amp > 0 && amp < 2000) { //check both high and low end of
+ //result, 2000 is somewhat arbitrary - not nice with magic numbers in the code..
+ AliDebug(2,Form("id %d lowGain %d amp %g", id, lowGain, amp));
- // Reset graph
- for (Int_t index = 0; index < gSig->GetN(); index++) {
- gSig->SetPoint(index, index, 0) ;
- }
- // Reset starting parameters for fit function
- signalF->SetParameters(10.,0.,fTau,fOrder,5.); //reset all defaults just to be safe
-
- }; // EMCAL entries loop
+ AddDigit(digitsArr, id, lowGain, (Int_t)amp, time);
+ }
+
+ //}
+
+ // Reset graph
+ for (Int_t index = 0; index < gSig->GetN(); index++) {
+ gSig->SetPoint(index, index, 0) ;
+ }
+ // Reset starting parameters for fit function
+ signalF->SetParameters(10.,0.,fTau,fOrder,5.); //reset all defaults just to be safe
+
+ } // nsamples>0 check, some data found for this channel; not only trailer/header
+ } // end while over channel
+ } //end while over DDL's, of input stream
delete signalF ;
delete gSig;
}
//____________________________________________________________________________
-void AliEMCALRawUtils::FitRaw(TGraph * gSig, TF1* signalF, Float_t & amp, Float_t & time)
+void AliEMCALRawUtils::FitRaw(TGraph * gSig, TF1* signalF, Float_t & amp, Float_t & time) const
{
// Fits the raw signal time distribution; from AliEMCALGetter
-
amp = time = 0. ;
Double_t ped = 0;
Int_t nPed = 0;
ped = 10; // put some small value as first guess
}
- Int_t max_found = 0;
- Int_t i_max = 0;
+
+ Int_t maxFound = 0;
+ Int_t iMax = 0;
Float_t max = -1;
- Float_t max_fit = gSig->GetN();
- Float_t min_after_sig = 9999;
- Int_t tmin_after_sig = gSig->GetN();
- Int_t n_ped_after_sig = 0;
- Int_t plateau_width = 0;
- Int_t plateau_start = 9999;
- Float_t Cut = 0.3;
+ Float_t maxFit = gSig->GetN();
+ Float_t minAfterSig = 9999;
+ Int_t tminAfterSig = gSig->GetN();
+ Int_t nPedAfterSig = 0;
+ Int_t plateauWidth = 0;
+ Int_t plateauStart = 9999;
+ Float_t cut = 0.3;
for (Int_t i=fNPedSamples; i < gSig->GetN(); i++) {
Double_t ttime, signal;
- gSig->GetPoint(i, ttime, signal) ;
- if (!max_found && signal > max) {
- i_max = i;
+ gSig->GetPoint(i, ttime, signal) ;
+ if (!maxFound && signal > max) {
+ iMax = i;
max = signal;
}
else if ( max > ped + fNoiseThreshold ) {
- max_found = 1;
- min_after_sig = signal;
- tmin_after_sig = i;
+ maxFound = 1;
+ minAfterSig = signal;
+ tminAfterSig = i;
}
- if (max_found) {
- if ( signal < min_after_sig) {
- min_after_sig = signal;
- tmin_after_sig = i;
+ if (maxFound) {
+ if ( signal < minAfterSig) {
+ minAfterSig = signal;
+ tminAfterSig = i;
}
- if (i > tmin_after_sig + 5) { // Two close peaks; end fit at minimum
- max_fit = tmin_after_sig;
+ if (i > tminAfterSig + 5) { // Two close peaks; end fit at minimum
+ maxFit = tminAfterSig;
break;
}
- if ( signal < Cut*max){ //stop fit at 30% amplitude(avoid the pulse shape falling edge)
- max_fit = i;
+ if ( signal < cut*max){ //stop fit at 30% amplitude(avoid the pulse shape falling edge)
+ maxFit = i;
break;
}
if ( signal < ped + fNoiseThreshold)
- n_ped_after_sig++;
- if (n_ped_after_sig >= 5) { // include 5 pedestal bins after peak
- max_fit = i;
+ nPedAfterSig++;
+ if (nPedAfterSig >= 5) { // include 5 pedestal bins after peak
+ maxFit = i;
break;
}
}
//Add check on plateau
if (signal >= fgkRawSignalOverflow - fNoiseThreshold) {
- if(plateau_width == 0) plateau_start = i;
- plateau_width++;
+ if(plateauWidth == 0) plateauStart = i;
+ plateauWidth++;
}
}
- if(plateau_width > 0) {
- for(int j = 0; j < plateau_width; j++) {
+ if(plateauWidth > 0) {
+ for(int j = 0; j < plateauWidth; j++) {
//Note, have to remove the same point N times because after each
//remove, the positions of all subsequent points have shifted down
- gSig->RemovePoint(plateau_start);
+ gSig->RemovePoint(plateauStart);
}
}
if ( max - ped > fNoiseThreshold ) { // else its noise
AliDebug(2,Form("Fitting max %d ped %d", max, ped));
- signalF->SetRange(0,max_fit);
+ signalF->SetRange(0,maxFit);
if(max-ped > 50)
signalF->SetParLimits(2,1,3);
signalF->SetParameter(4, ped) ;
- signalF->SetParameter(1, i_max);
+ signalF->SetParameter(1, iMax);
signalF->SetParameter(0, max);
gSig->Fit(signalF, "QROW"); // Note option 'W': equal errors on all points
//
Double_t signal ;
Double_t tau =par[2];
- Double_t N =par[3];
+ Double_t n =par[3];
Double_t ped = par[4];
Double_t xx = ( x[0] - par[1] + tau ) / tau ;
if (xx <= 0)
signal = ped ;
else {
- signal = ped + par[0] * TMath::Power(xx , N) * TMath::Exp(N * (1 - xx )) ;
+ signal = ped + par[0] * TMath::Power(xx , n) * TMath::Exp(n * (1 - xx )) ;
}
return signal ;
}
//Double_t noise = gRandom->Gaus(0.,fgFEENoise);
//signal = sqrt(signal*signal + noise*noise);
+ // March 17,09 for fast fit simulations by Alexei Pavlinov.
+ // Get from PHOS analysis. In some sense it is open questions.
+ Double_t noise = gRandom->Gaus(0.,fgFEENoise);
+ signal += noise;
+
adcH[iTime] = static_cast<Int_t>(signal + 0.5) ;
if ( adcH[iTime] > fgkRawSignalOverflow ){ // larger than 10 bits
adcH[iTime] = fgkRawSignalOverflow ;