PXDGainCalibrationAlgorithm.cc

/**************************************************************************
 * basf2 (Belle II Analysis Software Framework)                           *
 * Author: The Belle II Collaboration                                     *
 *                                                                        *
 * See git log for contributors and copyright holders.                    *
 * This file is licensed under LGPL-3.0, see LICENSE.md.                  *
 **************************************************************************/
 
#include <pxd/calibration/PXDGainCalibrationAlgorithm.h>
#include <pxd/dbobjects/PXDGainMapPar.h>
#include <pxd/dbobjects/PXDClusterChargeMapPar.h>
 
#include <string>
#include <algorithm>
#include <set>
 
#include <sstream>
#include <iostream>
 
#include <boost/format.hpp>
 
//ROOT
#include <TRandom.h>
#include <TH1.h>
#include <TF1.h>
 
using namespace std;
using boost::format;
using namespace Belle2;
 
 
// Anonymous namespace for data objects used by PXDGainCalibrationAlgorithm class
namespace {
 
  int m_signal;
  int m_run;
  int m_exp;
 
  void getNumberOfBins(const std::shared_ptr<TH1I>& histo_ptr, unsigned short& nBinsU, unsigned short& nBinsV)
  {
    set<unsigned short> uBinSet;
    set<unsigned short> vBinSet;
 
    // Loop over all bins of input histo
    for (auto histoBin = 1; histoBin <= histo_ptr->GetXaxis()->GetNbins(); histoBin++) {
      // The bin label contains the vxdid, uBin and vBin
      string label = histo_ptr->GetXaxis()->GetBinLabel(histoBin);
 
      // Parse label string format to read sensorID, uBin and vBin
      istringstream  stream(label);
      string token;
      getline(stream, token, '_');
      getline(stream, token, '_');
      unsigned short uBin = std::stoi(token);
 
      getline(stream, token, '_');
      unsigned short vBin = std::stoi(token);
 
      uBinSet.insert(uBin);
      vBinSet.insert(vBin);
    }
 
    if (uBinSet.empty() || vBinSet.empty()) {
      B2FATAL("Not able to determine the grid size. Something is wrong with collected data.");
    } else {
      nBinsU = *uBinSet.rbegin() + 1;
      nBinsV = *vBinSet.rbegin() + 1;
    }
  }
 
  unsigned short getNumberOfSensors(const std::shared_ptr<TH1I>& histo_ptr)
  {
    set<unsigned short> sensorSet;
 
    // Loop over all bins of input histo
    for (auto histoBin = 1; histoBin <= histo_ptr->GetXaxis()->GetNbins(); histoBin++) {
      // The bin label contains the vxdid, uBin and vBin
      string label = histo_ptr->GetXaxis()->GetBinLabel(histoBin);
 
      // Parse label string format to read sensorID, uBin and vBin
      istringstream  stream(label);
      string token;
      getline(stream, token, '_');
      VxdID sensorID(token);
      sensorSet.insert(sensorID.getID());
    }
    return sensorSet.size();
  }
 
}
 
 
PXDGainCalibrationAlgorithm::PXDGainCalibrationAlgorithm():
  CalibrationAlgorithm("PXDClusterChargeCollector"),
 
  minClusters(1000), noiseSigma(0.6), safetyFactor(2.0), forceContinue(false), strategy(0)
{
  setDescription(
    " -------------------------- PXDGainCalibrationAlgorithm ---------------------------------\n"
    "                                                                                         \n"
    "  Algorithm for estimating pxd gains (conversion factor from charge to ADU)              \n"
    " ----------------------------------------------------------------------------------------\n"
  );
}
 
 
 
 
CalibrationAlgorithm::EResult PXDGainCalibrationAlgorithm::calibrate()
{
 
  // Get counter histograms
  auto cluster_counter = getObjectPtr<TH1I>("PXDClusterCounter");
 
  // Extract number of sensors from counter histograms
  auto nSensors = getNumberOfSensors(cluster_counter);
 
  // Extract the number of grid bins from counter histograms
  unsigned short nBinsU = 0;
  unsigned short nBinsV = 0;
  getNumberOfBins(cluster_counter, nBinsU, nBinsV);
 
  // Check that we have collected enough Data
  if (cluster_counter->GetEntries() < int(safetyFactor * minClusters * nSensors * nBinsU * nBinsV)) {
    if (not forceContinue) {
      B2WARNING("Not enough Data: Only " <<  cluster_counter->GetEntries() << " hits were collected but " << int(
                  safetyFactor * minClusters*
                  nSensors * nBinsU * nBinsV) << " needed!");
      return c_NotEnoughData;
    } else {
      B2WARNING("Continue despite low statistics: Only " <<  cluster_counter->GetEntries() << " hits were collected but" << int(
                  safetyFactor * minClusters*
                  nSensors * nBinsU * nBinsV) << " would be desirable!");
    }
  }
 
  B2INFO("Start calibration using a " << nBinsU << "x" << nBinsV << " grid per sensor.");
  B2INFO("Number of collected clusters is " << cluster_counter->GetEntries());
 
  // This is the PXD gain correction payload for conditions DB
  PXDGainMapPar* gainMapPar = new PXDGainMapPar(nBinsU, nBinsV);
  set<VxdID> pxdSensors;
 
  // Loop over all bins of input histo
  for (auto histoBin = 1; histoBin <= cluster_counter->GetXaxis()->GetNbins(); histoBin++) {
    // The bin label contains the vxdid, uBin and vBin
    string label = cluster_counter->GetXaxis()->GetBinLabel(histoBin);
 
    // Parse label string format to read sensorID, uBin and vBin
    istringstream  stream(label);
    string token;
    getline(stream, token, '_');
    VxdID sensorID(token);
 
    getline(stream, token, '_');
    unsigned short uBin = std::stoi(token);
 
    getline(stream, token, '_');
    unsigned short vBin = std::stoi(token);
 
    // Read back the counters for number of collected clusters
    int numberOfHits = cluster_counter->GetBinContent(histoBin);
 
    // Only perform fitting, when enough data is available
    if (numberOfHits >= minClusters) {
 
      // Estimate the gain on a certain part of PXD
      auto gain = EstimateGain(sensorID, uBin, vBin);
 
      // Store the gain
      gainMapPar->setContent(sensorID.getID(), uBin, vBin, gain);
    } else {
      B2WARNING(label << ": Number of mc hits too small for fitting (" << numberOfHits << " < " << minClusters <<
                "). Use default gain.");
      gainMapPar->setContent(sensorID.getID(), uBin, vBin, 1.0);
    }
    pxdSensors.insert(sensorID);
  }
 
  // Post processing of gain map. It is possible that the gain
  // computation failed on some parts. Here, we replace default
  // values (1.0) by local averages of neighboring sensor parts.
 
  for (const auto& sensorID : pxdSensors) {
    for (unsigned short vBin = 0; vBin < nBinsV; ++vBin) {
      float meanGain = 0;
      unsigned short nGood = 0;
      unsigned short nBad = 0;
      for (unsigned short uBin = 0; uBin < nBinsU; ++uBin) {
        auto gain = gainMapPar->getContent(sensorID.getID(), uBin, vBin);
        // Filter default gains
        if (gain != 1.0) {
          nGood += 1;
          meanGain += gain;
        } else {
          nBad += 1;
        }
      }
      B2RESULT("Gain calibration on sensor=" << sensorID << " and vBin=" << vBin << " was successful on " << nGood << "/" << nBinsU <<
               " uBins.");
 
      // Check if we can repair bad calibrations with a local avarage
      if (nGood > 0 && nBad > 0) {
        meanGain /= nGood;
        for (unsigned short uBin = 0; uBin < nBinsU; ++uBin) {
          auto gain = gainMapPar->getContent(sensorID.getID(), uBin, vBin);
          if (gain == 1.0) {
            gainMapPar->setContent(sensorID.getID(), uBin, vBin, meanGain);
            B2RESULT("Gain calibration on sensor=" << sensorID << ", vBin=" << vBin << " uBin " << uBin << ": Replace default gain wih average "
                     << meanGain);
          }
        }
      }
    }
  }
 
 
  // Save the gain map to database. Note that this will set the database object name to the same as the collector but you
  // are free to change it.
  saveCalibration(gainMapPar, "PXDGainMapPar");
 
  B2INFO("PXD Gain Calibration Successful");
  return c_OK;
}
 
 
double PXDGainCalibrationAlgorithm::EstimateGain(VxdID sensorID, unsigned short uBin, unsigned short vBin)
{
  // Construct a tree name for requested part of PXD
  auto layerNumber = sensorID.getLayerNumber();
  auto ladderNumber = sensorID.getLadderNumber();
  auto sensorNumber = sensorID.getSensorNumber();
  const string treename = str(format("tree_%1%_%2%_%3%_%4%_%5%") % layerNumber % ladderNumber % sensorNumber % uBin % vBin);
  // Vector with cluster signals from collected mc
  vector<double> mc_signals;
 
  auto tree_MC = getObjectPtr<TTree>(treename);
  tree_MC->SetBranchAddress("signal", &m_signal);
 
  // Loop over tree_MC
  const auto nEntries_MC = tree_MC->GetEntries();
  for (int i = 0; i < nEntries_MC; ++i) {
    tree_MC->GetEntry(i);
 
    double noise = gRandom->Gaus(0.0, noiseSigma);
    mc_signals.push_back(m_signal + noise);
  }
 
  double dataMedian = GetChargeMedianFromDB(sensorID, uBin, vBin);
  double mcMedian = -1;
  // check if dataMedian makes sense
  if (dataMedian <= 0.0) {
    B2WARNING("Retrieved negative charge median from DB for sensor=" << sensorID << " uBin=" << uBin << " vBin=" << vBin <<
              ". Set gain to default value (=1.0) as well.");
    return 1.0;
  }
  if (strategy == 0) mcMedian = CalculateMedian(mc_signals);
  else if (strategy == 1) mcMedian = FitLandau(mc_signals);
  else {
    B2FATAL("strategy unavailable, use 0 for medians or 1 for landau fit!");
  }
 
  // check if mcMedian makes sense
  if (mcMedian <= 0.0) {
    B2WARNING("Retrieved negative charge median from DB for sensor=" << sensorID << " uBin=" << uBin << " vBin=" << vBin <<
              ". Set gain to default value (=1.0) as well.");
    return 1.0;
  }
 
 
  double gain =  dataMedian / mcMedian;
  double gainFromDB = GetCurrentGainFromDB(sensorID, uBin, vBin);
  B2DEBUG(10, "Gain from db used in PXDDigitizer is " << gainFromDB);
  B2DEBUG(10, "New gain correction derived is " << gain);
  B2DEBUG(10, "The total gain we should return is " << gain * gainFromDB);
 
  return gain * gainFromDB;
}
 
double PXDGainCalibrationAlgorithm::GetChargeMedianFromDB(VxdID sensorID, unsigned short uBin, unsigned short vBin)
{
  // Read back db payloads
  PXDClusterChargeMapPar* chargeMapPtr = 0;
  PXDGainMapPar* gainMapPtr = 0;
 
  auto dbtree = getObjectPtr<TTree>("dbtree");
  dbtree->SetBranchAddress("run", &m_run);
  dbtree->SetBranchAddress("exp", &m_exp);
  dbtree->SetBranchAddress("chargeMap", &chargeMapPtr);
  dbtree->SetBranchAddress("gainMap", &gainMapPtr);
 
  // Compute running average of charge medians from db
  double sum = 0;
  int counter = 0;
 
  // Loop over dbtree
  const auto nEntries = dbtree->GetEntries();
  for (int i = 0; i < nEntries; ++i) {
    dbtree->GetEntry(i);
    sum += chargeMapPtr->getContent(sensorID.getID(), uBin, vBin);
    counter += 1;
  }
  delete chargeMapPtr;
  chargeMapPtr = 0;
  delete gainMapPtr;
  gainMapPtr = 0;
 
  return sum / counter;
}
 
double PXDGainCalibrationAlgorithm::GetCurrentGainFromDB(VxdID sensorID, unsigned short uBin, unsigned short vBin)
{
  // Read back db payloads
  PXDClusterChargeMapPar* chargeMapPtr = 0;
  PXDGainMapPar* gainMapPtr = 0;
 
  auto dbtree = getObjectPtr<TTree>("dbtree");
  dbtree->SetBranchAddress("run", &m_run);
  dbtree->SetBranchAddress("exp", &m_exp);
  dbtree->SetBranchAddress("chargeMap", &chargeMapPtr);
  dbtree->SetBranchAddress("gainMap", &gainMapPtr);
 
  // Compute running average of gains from db
  double sum = 0;
  int counter = 0;
 
  // Loop over dbtree
  const auto nEntries = dbtree->GetEntries();
  for (int i = 0; i < nEntries; ++i) {
    dbtree->GetEntry(i);
    sum += gainMapPtr->getContent(sensorID.getID(), uBin, vBin);
    counter += 1;
  }
  delete chargeMapPtr;
  chargeMapPtr = 0;
  delete gainMapPtr;
  gainMapPtr = 0;
 
  return sum / counter;
}
 
 
double PXDGainCalibrationAlgorithm::CalculateMedian(vector<double>& signals)
{
  auto size = signals.size();
 
  if (size == 0) {
    return 0.0;  // Undefined, really.
  } else {
    sort(signals.begin(), signals.end());
    if (size % 2 == 0) {
      return (signals[size / 2 - 1] + signals[size / 2]) / 2;
    } else {
      return signals[size / 2];
    }
  }
}
 
double PXDGainCalibrationAlgorithm::FitLandau(vector<double>& signals)
{
  auto size = signals.size();
  if (size == 0) return 0.0; // Undefined, really.
 
  // get max and min values of signal vector
  int max = *max_element(signals.begin(), signals.end());
  int min = *min_element(signals.begin(), signals.end());
 
  // create histogram to hold signals and fill it
  TH1D* hist_signals = new TH1D("", "", max - min, min, max);
  for (auto it = signals.begin(); it != signals.end(); ++it) {
    hist_signals->Fill(*it);
  }
 
  // create fit function
  TF1* landau = new TF1("landau", "TMath::Landau(x,[0],[1])*[2]", min, max);
  landau->SetParNames("MPV", "sigma", "scale");
  landau->SetParameters(100, 1, 1000);
 
  // do fit and get results
  Int_t status = hist_signals->Fit("landau", "Lq", "", 0, 350);
  double MPV = landau->GetParameter("MPV");
 
  // clean up
  delete hist_signals;
  delete landau;
 
  // check fit status
  if (status == 0) return MPV;
  else {
    B2WARNING("Fit failed!. using default value.");
    return 0.0;
  }
}