Signal Processing in PHAT P' Decowski, A' Olszewski, H' Pernegger, P' Sarin

1
Signal Processing in PHAT(P. Decowski, A.
Olszewski, H. Pernegger, P. Sarin)

• Outline
• Requirement for the signal calculation
• Layout of the measurement chain and contributions
  to the measured ADC value
• Overview of the signal processing in PHAT
• Definition of the data format
• How to obtain pedestal and noise values
• How to common-mode correct and zero-suppress the
  data
• Calibration, HitArrays and Hits
• First test results using this chain
• Open questions

2
What are the aims of the signal processing

• TWO quite different detector with different aims
• The Octagon measures multiplicity by
• summation of calibrated but non-zero-suppressed
  signals in a given detector area
• application of a threshold and digital counting
  of pads with signals above a threshold
• This requires to minimizes systematic errors in
  the calculation
• from large correlated common mode shifts
• calibration errors
• pile-up
• Statistical fluctuation from the Landau
  distribution and intrinsic noise are less
  important than in the spectrometer

3

• The Spectrometer want to reconstruct tracks and
  identify particles
• clearly identify pads with hits with high
  efficiency for straight and inclined tracks
• minimize the number of hits not associated with
  tracks for an easier pattern recognition
• understand (and minimize) the width of the
  measured signal distribution for good pion-kaon
  separation
• This requires for the tracking to
• apply signal threshold as low as possible
  (inclined tracks!) -gt reduce the calculated noise
  value to the intrinsic noise
• filter out defect channels with strong
  fluctuations, which fake signals
• and for the particle ID to minimize the signal
  distribution width. Irreducible is the energy
  straggling and intrinsic noise but reducible
  components are
• common mode shift
• gain fluctuations and variation in the detector
  thickness
• systematic differences in the signal response
  from type 1 to type 5 detectors

4
Layout of the signal measurement can
contributions to the measured ADC value
Preamp
OutBuffer
ADC
InputAmp
shaper
FEC
Module
SignalNoiseCMPedestal
(SignalNoiseCMPedestal)gain
SignalNoiseCM

• The measured ADC(I)S(I)Ped(I)CM(k) Ichannel
  k chip
• the particle signal intrinsic noise (S)
• the channel pedestal (Ped) which corresponds to
  the channels DC levels in the chip
• the common mode shift (CM) for a group of
  channels

5
How large are this contributions in our module
tests?

• The tests are done on final modules but without
  FEC or final power supplies
• The Signal and Noise
• from a 90Sr source (close to MIP energy
  deposition)
• peak S 23mV with a peak variation of /- 1.5mV
• The rms noise is 1.2mV to 1.6mV depending on
  sensor type and quality (analog readout
  contribution is 0.4mV)
• The Common Mode Noise
• for most modules rms common mode noise is 1.5mV
  to 2mV
• few sensors (approx. 5) have large common mode
  noise (4 to 10mV rms)
• The noise level of defect channels (approx 1)
  is varies between 3 and 12mV rms. Some channels
  fake signals.
• The Pedestal Variation
• per chip 200mV pp, I.e 10MIPs
• per string 400mV pp , I.e. 20MIPs

6
Overview of the signal processing steps in PHAT

• The aim for ZSS is to unfold Ped and CM
  contributions and find hits
• the code in PHAT and the Mercury system needs to
  be consistent
• Precise calibration and unfold systematic effects
  (hit merging, detector thickness, track
  inclination) are NOT part of the ZSS
• The algorithm used at the moment is the one
  provided by Pradeep and implemented by Andrzej.

Initialization


Raw n0
PedPre Process
Pedestal Process
Noise Process
CM Correct
Zero Suppress
Make HitArray
Make Hits
Do PR tracking
Calibrate
Raw s0
Event Loop
7
Raw data format for Non-Zero Suppressed Data

• One basic block per FEC, one 16bit word per
  channel with 12 bit integer for the ADC value
  20 words trailer
• all channels are written without any processing

8
Raw data format for Zero Suppressed Data

• One basic block per FEC, one 24bit word per
  channel with a hit12 bit integer for the signal
  value (LSB1ADC) trailer with CM correction for
  each chip
• calculation in the ZSS are based on 32float
• only channels with signals larger than a
  threshold are written.

9
Special Events

• The present data format do not contain any
  information on pedestal and noise used for zero
  suppression
• The present assumption is that
• noise and pedestal values will be preprocessed
  offline
• they are loaded to the Mercury system which uses
  them for ZS
• they are written to the data stream for all
  channels once at the beginning and at the end of
  the run
• We need to define a format for these special
  events
• The algorithms for offline preprocessing exist
  and are tested

10
The Initalization phase

• Get start values for pedestal and noise
• Structure is based on FEC and String units
• TPhFECDetectorPedPreProcess
• calculated the average ADC value for each channel
  for N events
• output is a preliminary pedestal value/channel
• TPhFECDetectorPedProcess
• calculated the average ADC value for each channel
  for N events for all channels where ADC-Pedlt 0.5
  MIPADC ( this removes eventual signals)
• output is a mean ADC value (PedMean) and its rms
  (PedRMS)
• there is no common mode correction
• TPhFECDetectorNoiseProcess
• calculated the rms for SADC-Ped-CM for all
  channels where ADC-Pedlt nPedRMS ( this removes
  signals)
• output is a rms(S)Noise/channel

11
The Event Loop

• Corrects common mode shifts and zero-suppresses
  empty channels
• Each function is called once/event and processes
  all FECs
• TPhFECDetectorCMNProcess
• calculated the average ADC-Ped value for one chip
  for all signals with abs(ADC-PedMean)lt nPedRMS
• checks that a minimum of 5 channels are used in
  the calculation
• stores the CMNMean for each chip
• TPhFECDetectorZeroSuppress
• calculated the signal for each channel as
  SADC-PedMean-CMNMean
• suppressed all channels with abs(S(I))ltnNoise(I)
• channels on chips where CM correction failed are
  not suppressed
• Stores S in TPhRawHitFECs0 (as integer...!) with
  LSB1ADC count

12
Signal Processing after Zero-Suppression

• Get the detector geometry, our sensor layout and
  the FEC assignment to sensors all processing is
  based on physical sensors
• TPhDetectorMakeHitArrays
• takes the ZS data (string/channels) and fills it
  to the sensors as signal on pads row/column
• stores the data in HitArray containers
• TPhDetectorCalibrate
• converts the measured ADC values to deposited
  charge using a pre-measured calibration curve .
  Charge is stored in units of 0.1fC
• The present calibration curve is a 2nd order
  polynom plus a cutoff parameter
• TPhDetectorMakeHitsOutOfHitArray
• converts charge to energy loss (in GeV)
• stores all pads with signalgt0 as hits with their
  geometrical coordinates x,y,z
• NOW you merge hits and start the TrackSeedFinder
  and reconstruct the tracks with Inkyus code

13
First Results of tests with real data

• Data are from the october beam test at AGS
• ADC are 12bit with similar dynamic range as the
  FEC
• data are written in the PHAT non-zero-suppressed
  data format
• we used 4 type 1 module (MOD10001,0002,003,0006)
• What were the first tests
• The pedestal for one string
• The noise before and after common mode correction
• The signal distribution for zero-suppressed data
  on a nice and a less nice module
• The Hit-signal after calibration
• The final Landau distribution in keV after hit
  merging and track reconstruction

14
The calculated pedestal

• Output of TPhFECDetectorPedProcess

15
The noise before and after CM correction

• Output of TPhFECDetectorPedProcess and Noise
  Process

16
Zero-Suppressed signals on a good module

• Output of TPhFECDetectorZeroSuppress for
  MOD10003 (2 noisy channels/sensor )

17
After making the calibration...

• Output of TPhDetectorCalibrate for MOD10003

18
After making hits we reconstruct the first tracks
in 4 planes

• Use 4 plane for a TPhTrackSeedFinder for a planes
  NA,NB,NC,ND

19
Landau distribution in keV and reconstructed
tracks

• Signal for hits after hit-merging, which are
  associated with the track in planes NA,NB,NC,ND

20
X-Y hit distribution for reconstructed tracks

• We had one dead chip (unfortunately)

21
Open questions

• Definition of raw data format for loading and
  writing noise and pedestal information
• Procedure for offline processing of pedestal,
  noise, gains
• Proper definition of the signal value for optimal
  processing (how to treat negative signals, what
  is the LSB)
• Consistent variable definition for signal, ADC,
  pedestal, noise, CMN
• Test of the common mode calculation
• and...