Silicon Strip Digitization

Complete: 3

Explains the algorithms used to digitize the silicon strip tracker.

Introduction

"Digitization" refers to the process of simulating the electronics signals read out by the detector. The energy deposit from Geant4 is converted into charges in the silicon bulk, that are then migrated to the surface of the sensor, and shaped by the electronics. The various steps will be described in the following. To speed-up the process, different assumptions are made depending of the working mode of the electronics, we will therefore start by describing shortly these operation modes.

Operation modes of the electronics

The APV25 chip of the tracker front-end can operate in different modes. We are interested in two of them, called respectively peak and deconvolution mode.

  • In peak mode, the signal is a RC-CR curve, that gives a sensitivity - somewhat reduced after some time - over about 200ns.
  • In deconvolution mode, the signal is processed with a three-weights deconvolution algorithm to reduce the signal to 25ns. The signal is then approximately a Gaussian with sigma=12.5ns.
Deconvolution is the default operation mode for collisions.

The tracker FED (Front End Driver) can also be operated in various modes. Simulation handles the two most important cases: Virgin-Raw and Zero-Suppression.

  • in Virgin-Raw mode, one digi is produced per strip, disregarding the presence or not of signal. Each strip contains a reading corresponding to the pedestal (typically 250 ADC counts), to which one adds the noise (positive or negative) and sometimes the signal. One digi is the measured amplitude, and there are 4 (or 6) times 128 digis in each detunit.
  • in Zero-suppression, the FED does pedestal and common-noise subtraction, and then selects strips above a given threshold (typically 5*noise for isolated strips and 2*noise for two or more adjacent strips. One digi is then a pair amplitude/position, and there are n>=0 digis in each detunit.
Zero-suppression is the default operation mode for p-p collisions.

Simulation steps

The silicon strip simulation proceeds as follows. First, the behavior of charges in the silicon is simulated:

  • the energy deposit by Geant4 is distributed along the path of the particle in the silicon, split in N sub-deposits taking into account Landau fluctuations.
  • each individual sub-deposit is migrated to the surface of the sensor, taking into account the thermal diffusion and the Lorentz angle. The result is a charge on the surface, assigned with a width and a position.
  • the various charge contributions are merged per strip.

The resulting signal is processed to include the effects of the electronics:

  • the signal shaping is included, which reduces the signal of off-time particles.
  • in Virgin-Raw, a pedestal and noise is added on all strips. It gives directly the vector of (raw)digis that are sent to the local reconstruction (clustering).
  • in Zero-Suppression, noise tails above 2 sigmas are added to the signal strips only to for a set of digis.

Parameters and configuration

Usually, the digitizer is configured by including one of the two following configuration fragments:

  • SiStripDigi_APVModePeak_cff to work in peak mode (special)
  • SiStripDigi_APVModeDec_cff to work in deconvolution mode (normal)

The various parameters are described in the following table.

ParameterSorted ascending Default value Comment
AppliedVoltage 300 Applied voltage for operations. It controls the diffusion
APVpeakmode False APV mode. It is set to True in SiStripDigi_APVModePeak_cff.py
APVSaturationFromHIP True The APV is killed if an hadron produces HIP. The probability of this happening can be configured by the parameter below
APVSaturationProb 0.001 Probability of an hadron producing HIP
BaselineShift True APV baseline shift down is applied and proportional to the APV occupancy and total charged deposited
ChargeDistributionRMS 6.5e-10 Diffusion factor
chargeDivisionsPerStrip 10 Controls the number of sub-deposits generated
ChargeMobility 310 Charge mobility in silicon (for holes). It controls the diffusion
cmnRMStec 2.44 CMN RMS for TEC
cmnRMStib 5.92 CMN RMS for TIB
cmnRMStid 3.08 CMN RMS for TID
cmnRMStob 1.08 CMN RMS for TOB
CMS.LorentzAngle '' Db tag for Lorentz angle
CommonModeNoise True The Gaussian CMN is added for each APV. The RMS is specified by the parameters below
CosmicDelayShift 0.0 Delay to be added to the simHit time for synchronization. 0.0 for collisions, it has to be adapted for Cosmic MC
DeltaProductionCut 0.120425 Geant4 parameter used for the Landau fluctuation. Should be identical to the G4 Simulation part
DepletionVoltage 170 Depletion voltage of silicon sensors. 170 is a typical value from integration. It controls the diffusion
DigiModeList --- Outputs
electronPerAdcDec 217.0 Conversion factor (gain) for the electronics, when Deconvolution mode
electronPerAdcPeak 262.0 Conversion factor (gain) for the electronics, when Peak mode
FedAlgorithm 4 Zero Suppression algorithm. That corresponds to what is set in the firmware.
Gain '' Db tag for Gain
GeometryType 'idealForDigi' Geometry
GevPerElectron 3.61e-09 Conversion factor for ionization
Inefficiency 0.0 Artificial inefficiency added to the digitization.
LandauFluctuations True Activates the Landau fluctuation of the sub-deposits in the silicon
noDiffusion False flag to activate/deactivate diffusion
Noise True Enable/Disable noise simulation. NOTE : turning Noise ON/OFF will make a big change
NoiseSigmaThreshold 2.0 Cut on the tail for noise simulation in ZS mode. Noise below N*sigma will not be added.
PedestalsCondition 128 It is the pedestals offset introduced to avoid negative values on the FED. In the Db the pedestals are subtracted of 128
RealPedestals True The pedestals as stored in the Db + 128 are used. If false just a constant shift of 128 is applied
ROUList --- Inputs
SingleStripNoise True only if Noise = True. It uses the values of noise for each strip as stored in the Db
Temperature 273 Silicon temperature. It controls the diffusion
TOFCutForDeconvolution 50.0 Time-of-Flight cut in deconvolution mode. Deposits timed more than N ns away will not be converted to digis
TOFCutForPeak 100.0 Time-of-Flight cut in peak mode. Deposits timed more than N ns away will not be converted to digis
TrackerConfigurationFromDB False Use ConfDb
ZeroSuppression True Switch ZS on/off
Parameter for VR data (ZeroSuppression = False)

Conditions DB tags can be found here : https://twiki.cern.ch/twiki/bin/view/CMS/StripTrackerDBTagsForCalibrations#MC_production

In addition, there are some parameters controlling the cross-talk (capacitive coupling). There is one vector of doubles per geometry of sensor, each time once in deconvolution, once in peak mode. In the vector, the first number is the charge remaining on the leading strip, and you have then the induced signal on the first, second, third, ... adjacent strip(s). By default, signal is only induced on the first adjacent strip.

IMPORTANT: If the RawDigis generated in VR mode are used for reconstruction, it is mandatory that RealPedestals = True.
In order to access the real conditions to simulate realistic VR date, it is required to insert in the .cfg file the following lines:

process.poolDBESSource = cms.ESSource("PoolDBESSource",
 BlobStreamerName = cms.untracked.string('TBufferBlobStreamingService'),
 DBParameters = cms.PSet(
      messageLevel = cms.untracked.int32(2),
      authenticationPath = cms.untracked.string('/afs/cern.ch/cms/DB/conddb')
  ),
  timetype = cms.untracked.string('runnumber'),
  connect = cms.string('frontier://FrontierProd/CMS_COND_31X_STRIP'),
  toGet = cms.VPSet(cms.PSet(
      record = cms.string('SiStripNoisesRcd'),
      tag = cms.string('SiStripNoise_CRAFT09_DecMode_ForTrackerSim')
  ),
   cms.PSet(
      record = cms.string('SiStripPedestalsRcd'),
      tag = cms.string('SiStripPedestal_CRAFT09_DecMode_ForTrackerSim')
  ),
   cms.PSet(
      record = cms.string('SiStripFedCablingRcd'),
      tag = cms.string('SiStripFedCabling_CRAFT09_ForTrackerSim')
  ),
  cms.PSet(
      record = cms.string('SiStripBadChannelRcd'),
      tag = cms.string('SiStripBadChannelsFromO2O_CRAFT09_DecMode_ForTrackerSim')
  )
)
) 


process.es_prefer_my =cms.ESPrefer("PoolDBESSource","poolDBESSource")

New parameters in Peak and Deco modes from CRAFT09 data

Yuri compared CRAFT09 and MC in Peak and Deco mode for TIB and TOB, obtaining:

For TID and TEC, the most reliable values so far are those extracted by Burt, see slide 8 of this talk (ignore the TIB and TOB values, superseded by Yuri's values listed in the wikis).

Clarification: Burt used VR runs at 0 T, Yuri used ZS runs. There are reasons to believe that the VR runs at 0 T (peak: 111753, deco: 111664,111673) had some unaccounted feature that makes them not comparable with the bulk of the runs from CRAFT09, therefore we prefer Yuri's values when available; nevertheless, Burt's values are likely a better approximation than the old digitizer defaults (which were "educated guesses", unmodified since years).

These changes are effective since tag V05-15-05.

Plans for tuning with collision data

New proposals for realistic simulation, with the aim of having a good data/MC agreement for the usual observables at cluster level (Yuri's analysis) and at the same time a realistic description of saturation (critical for dE/dx), while keeping things simple from the db point of view (only one payload for noise and one for gain); putting together inputs by Andrea Venturi, Marco De Mattia, etc.

See also the SiStripTKSignalNormalization wiki, and subscribe to the cms-tk-signalnorm@cernNOSPAMPLEASE.ch mailing list if interested.

Situation at present (April 23, 2010):

Data to be used:

None of those currently available, as the timing is going to be re-optimized soon (after the technical stop of April 26-28). (By the way, this will lead to inconsistency between old and new data; hopefully, statistics will start to grow up much faster soon, and ignoring old data will not be dramatic for analyses aiming at Summer 2010 conferences.)

Parameters of interest:

  • electronsPerAdc: constant factor for all SST (different between Peak and Deco)
  • i: strip index
  • tickmark(i): gain correction as measured from tickmarks, currently equal for all i's in the same APV pair
  • g(i): from a gaussian with μ=1 and σ equal to 0 for DESIGN and MC conditions and to a tuned value for START conditions
  • ideal_noise(i): proportional to strip length
  • miscalibrated_noise(i): ideal_noise(i)/tickmark(i)

Approach Calib of signal in MC truth (digitizer) Calib used in reco (clusterizer) Noise added to signal in MC truth (digitizer) Noise used in Zero Suppression (digitizer) Noise used in reco (clusterizer)
old electronsPerAdc 1 or g(i) ideal_noise(i) ideal_noise(i) ideal_noise(i)
new electronsPerAdc*tickmark(i) 1/[g(i)*tickmark(i)] miscalibrated_noise(i)/tickmark(i) miscalibrated_noise(i) miscalibrated_noise(i)

Note: S/N must be invariant by calibration, at every step (simulation, zero suppression, clusterization). Digitizer code must be modified, see below.

Payloads in db:

Digitizer:

  • Gain = tickmark(i)
  • Noise = miscalibrated_noise(i); ideal noise is then derived from this and the gain inside the digitizer

Clusterizer:

  • Gain = 1/[g(i)*tickmark(i)]; one payload for the unsmeared (σ=0) and one for the smeared (σ from tuning) cases
  • Noise = miscalibrated_noise(i); exactly as for the digitizer

Sniplet for cfg:

process.load("CondCore.DBCommon.CondDBSetup_cfi")

process.SiStripNoises = cms.ESSource("PoolDBESSource",process.CondDBSetup,
                             connect = cms.string("sqlite_file:dbfile_noise.db"),
                             toGet = cms.VPSet(cms.PSet(record = cms.string("SiStripNoisesRcd"),
                                                        tag = cms.string("SiStripNoiseNormalizedWithGain"))))
process.es_prefer_SiStripNoises = cms.ESPrefer("PoolDBESSource","SiStripNoises")

process.SiStripGain = cms.ESSource("PoolDBESSource",process.CondDBSetup,
                           connect = cms.string('sqlite_file:dbfile_gainFromData.db'),
                           toGet = cms.VPSet(cms.PSet( record = cms.string('SiStripApvGainRcd'),
                                                       tag = cms.string('SiStripApvGain_default'))))
process.es_prefer_SiStripGain = cms.ESPrefer("PoolDBESSource", "SiStripGain")

Note: content of a sqlite file can be seen with the command

cmscond_list_iov -c sqlite_file:dbfile_gainFromData.db -a

Changes to the digitizer code:

SiStripDigitizerAlgorithm now (until V05-16-02):

float noiseRMS = noiseHandle->getNoise(RefStrip,detNoiseRange);
theSiNoiseAdder->addNoise(detAmpl,firstChannelWithSignal,lastChannelWithSignal,numStrips,noiseRMS*theElectronPerADC);
theSiZeroSuppress->suppress(theSiDigitalConverter->convert(detAmpl, gainHandle, detID), digis, detID,noiseHandle,thresholdHandle);
Here noiseHandle and gainHandle are the noise and gain payloads.

SiTrivialDigitalConverter now (until V05-16-02):

int adc = convert( (gainHandle->getStripGain(i, detGainRange))*(analogSignal[i]) );

The new proposal requires this code change (effective since V05-16-04):

float noiseRMS = noiseHandle->getNoise(RefStrip,detNoiseRange);
float gainValue = gainHandle->getStripGain(RefStrip, detGainRange);
theSiNoiseAdder->addNoise(detAmpl,firstChannelWithSignal,lastChannelWithSignal,numStrips,noiseRMS*theElectronPerADC/gainValue);
theSiZeroSuppress->suppress(theSiDigitalConverter->convert(detAmpl, gainHandle, detID), digis, detID,noiseHandle,thresholdHandle);

Tuning sequence:

  • electronsPerAdc: set to the average cluster charge, corrected by 3D angle and equalized by active tickness, of all SST (remember, tickmark corrections are applied everywhere)
  • σ of g(i), with μ fixed to 1, fitted by looking at cluster charge and S/N simultaneously
  • noise scaling
  • cross-talk parameters

Checks:

  • these variables (Yuri)
  • number of pure noise clusters (Yuri?); the test is passed if there is no visible change
  • number of saturating clusters, and fit of dE/dx vs P (Loic?)

To be considered later:

  • The model for noise could be improved, according to the measurements in the Craft08 paper.
  • Tuning of σ is calibration-dependent, in the sense that tickmark calibration and particle calibration would likely have different spreads. Hence we have gt(i) and gp(i).

Responsible: AndreaGiammanco

Topic attachments
I Attachment History Action Size Date Who Comment
PDFpdf Gain_10_01_27_TkSimMtg.pdf r1 manage 504.7 K 2010-01-28 - 15:40 AndreaGiammanco  
Unknown file formatdb dbfile.db r1 manage 10682.0 K 2010-04-27 - 14:07 MarcoDeMattia Noise payload normalized with gain: noise*gain
Unknown file formatdb dbfile_gainFromData.db r2 r1 manage 1008.0 K 2010-04-27 - 18:01 AndreaGiammanco gain from data (tickmarks)
Edit | Attach | Watch | Print version | History: r26 < r25 < r24 < r23 < r22 | Backlinks | Raw View | WYSIWYG | More topic actions
Topic revision: r26 - 2014-12-16 - JlAgram
 
    • Cern Search Icon Cern Search
    • TWiki Search Icon TWiki Search
    • Google Search Icon Google Search

    CMSPublic All webs login

This site is powered by the TWiki collaboration platform Powered by PerlCopyright & 2008-2021 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
or Ideas, requests, problems regarding TWiki? use Discourse or Send feedback