1) DATA: 2008 BSMD Calibration

Information about the 2008 BSMD Calibration effort will be posted below as sub-pages.

Fig 1. BSMD-E 2D mapping of soft ID. (plot for reverse mapping is attached)

01) raw spectra

 Goals:

  1. verify pedestals loaded to DB are reasonable for 2008 pp data
  2. estimate stats needed to find slopes of individual strips for minB events

 Method:

  1. look at pedestal residua for individual strips, exclude caps 1 & 2, use only status==1
  2. fit gauss & compare with histo mean
  3. find integrals of individual strips, sum  over 20 ADC channels starting from ped+5*sig 

To obtain muDst w/o zero suppression I run privately the following production chain:

chain="DbV20080703 B2008a ITTF IAna ppOpt VFPPV l3onl emcDY2 fpd ftpc trgd ZDCvtx NosvtIT NossdIT analysis Corr4 OSpaceZ2 OGridLeak3D beamLine BEmcDebug"

Fig 1

Examples of single strip pedestal residua, based on ~80K minB events from days 47-65, 30 runs. (1223 is # of bins, ignore it).

Left is typical good spectrum, see Fig2.3. Middle is also reasonable, but peds is 8 channels wide vs. typical 4 channels.

The strip shown on the right plot is probably broken. 

Fig 2

Detailed view on first 500 strips. X=stripID for all plots.

  1. Y=mean of the gauss fit to pedestal residuum, in ADC channels, error=sigma of the gauss.
  2. Y=integral of over 20 ADC channels starting from ped+5*sig. 
  3. Raw spectra, Y=ADC-ped, exclude caps 1 & 2, use only status==1

 

Fig 3

Broader view of ... problems in BSMD-E plane. Note, status flag was taken in to account.

Top plot is sum of 30 runs from days 47-65, 80K events. Bottom plot is just 1 run, 3K events. You can't distinguish individual channels, but scatter plot works like a sum of channels, so it is clear the slopes are there, we need just more data.

 

Conclusions:

  1. DB peds for BSMDE look good on average.
  2. with 1M eve we will start to see  gains for individual strip relativewith ~20% error. Production will finish tomorrow.
  3. there are portions of SMDE masked out (empty area in fig 3.2, id=1000) - do why know what broke? Will it be fixed in 2009 run
  4. there are portions of SMDE not masked but dead (solid line in fig 3.2, id=1400) - worth to go after those
  5. there are portions of SMDE not masked with unstable (or wrong) pedestal, (fig 3.1 id=15000)
  6. for most channels there is one or more caps with different ped not accounted for in DB ( thin line below pedestal in fig 2.3)
  7. One gets a taste of gain variation from fig 2.2
  8. Question: what width of pedestal is tolerable. Fig 2.1 shows width as error bars. Should I kill channel ID=152?

 

02) relative BSMD-E gains from 1M dAu events

 Glimpse of relative calibration of BSMDE from 2008 d+Au data

 

Input: 1M dAu minb events from runs: 8335112+8336019

Method : fit slopes to individual strips, as discussed 01) raw spectra

 

Fig 1

Examples of raw pedestal corrected spectra for first 9 strips, 1M dAu events

Fig 2

Detailed view on first 500 strips. X=stripID for all plots.

  1. Y=mean of the gauss fit to pedestal residuum, in ADC channels, error=sigma of the gauss.
  2. Y=integral of over 20 ADC channels starting from ped+5*sig. Raw spectra,
  3. Y=gain defined as "-slope" from the exponential fit over ADC range  20-40  channels, errors from expo fit. Blue line is constant fit to gains.

Fig 3

BSMDE strips cover the whole barrel and eta-phi representation is better suited to present 18K strips in one plot.

  1. Mapping of BSMDE-softID (Z-axis) in to eta-phi space. Eta bin 0 is eta=-1.0, eta bin 299 is eta=+1.0. Phi bin 0 starts at 12:30 and advances as phi angle. 
  2. gains for majority of 18K BSMDE strips. White means no data or discarded by rudimentary QA of peds, yields or slope.

Fig 4

For reference spectra from 1M pp events from ~12 EmcCheck runs from days 47-51. It proves I did it and it was naive on my side to expect 1M pp events is enough.

Fig 5

More pp events spectra - lot of problems with DB content. 

 

03) more details , answering Will

 This page provides more details addressing some of Will's questions.

 


   2) fig 2: well, 500 chns is not a very "natural" unit, but I wonder
      what corresponds to 50 chns (e.g., the region of fluctuation
      250-300) ... I need to check my electronics readout diagrams
      again, or maybe folks more expert will comment

Fig 1.

Zoom-in of the god-to-bad region of BSMDE

Fig 2. 

'Good' strips belong to barrel module 2, crate 2, sitting at ~1 o'cloCk on the WEST

 

Fig 3. 

'BAD' strips belong also  to barrel module 2, crate 2, sitting at ~1 o'cloCk on the WEST

 

 

04) bad CAP 123

 Study of pedestal correlation for BSMDE

Goal: identify source of the band below main pedestals. 

Figs 1,2 show pedestals 'breathe' in correlated way for channels in the same crate, but this mode is decoupled between crates. It may be enough to use individual peds for all CAPS to reduce this correlation.

Fig3 shows CAP=123 has bi-modal pedestals. FYI, CAPS 124,125 were excluded because they also are different.

 

Based on Fig1 & 3 one could write an algo identifying event by event in which mode CAP=123 settled, but for now I'll discard CAP123 as well.

 All plots are made based on 500K d-Au events from the run 8336052.

 

Fig 0
Example of pedestal residua for BSMDE strips 1-500, after CAPS 124 and 125 were excluded.

Fig 1
Correlation between pedestal residua for neighbor strips. Strip 100 is used on all plots on the X-axis

Fig 2
Correlation between pedestal residua for strips in different crates. Strip 100 is used on all plots on the X-axis

Fig 3
Squared pedestal residua for strips [1,150] were summed for every event and plotted as function of CAP ID (Y-axis).

Those strips belong to the same module #1 . X-axis shoes SQRT(sum) for convenience. CAP=123 has double pedestal.

 

05) BSMDE saturation, dAu, 500K minB eve

 Input: 500K d-Au events from run 8336052,

Method : drop CAPS 123,124,125, subtract single ped for all other CAPS.

 

Fig 1 full resolution, only 6 modules , every module contains 150 strips.

Fig 2 All 18K strips (120 modules), every module contain only 6 bins, every bin is sum of 25 strips. 

 h->RebinX(25), h->SetMinimum(2), h->SetMaximum(1e5)

06) QAed relative gains BSMDE, 3M d-Au events , ver1.0

Version 1 of relative gains for BSMDE, d-AU 2008.

 

INPUT: 3M d-AU events from day ~336 of 2007.

Method: fit slopes to ADC =ped+30,ped+100.

The spectra,  fits of pedestal residuum, and slopes were QAed.

Results: slopes were found for 16,577 of 18,000 strips of BSMDE.

 

Fig1   Good spectrum for strip ID=1. X-axis ADC-ped, CAPs=123,124,124 excluded.

Fig2 TOP:  slopes distribution (Y-axis) vs. stripID within given module ( X-axis). Physical eta=0.0 is at X=0, eta=1.0 is at X=150.

BOTTOM: status tables with marked eta-phi location excluded 1423 strips of BSMDE by multi-stage QA of the spectra. Different colors denote various failed tests.

 

Fig3 Mapping of known BSMDE topology on chosen by us eta-phi 2D localization. Official stripID is shown in color.

 

 

07) QA method for SMD-E, slopes , ver1.1

Automatic QA of BSMDE minB spectra. 

Content

  1. example of good spectra (fig 0)
  2. QA cuts definition (table 1) + spectra (figs 1-7)
  3. Result:
    1. # of bad strips per module. BSMDE modules 10,31,68  are damaged above  50%+. (modules 16-30 served by crate 4 were not QAed). 
    2. eta-phi distributions of: slopes, slope error (fig 9) , pedestal, pedestal (fig 10) width _after_ QA
    3. sample of good and bad plots from every module, including modules 16-30 (PDF at the end)

The automatic  procedure doing QA of  spectra was set up in order to preserve only good looking spectra as shown in the fig 0 below. 

Fig 0   Good spectra for random strips in module=2. X-axis shows pedestal residua. It is shown to set a scale for the bad strips shown below. 

 

INPUT: 3M d-AU events from day ~336 of 2007.

All spectra were pedestals subtracted, using one value per strip, CAPS=123,124,125 were excluded. Below I'll use term 'ped' instead of more accurate pedestal residuum.

Method: fit slopes to ADC =ped+40,ped+90 or 5*sig(ped) if too low.

The spectra,  fits of pedestal residuum, and slopes were QAed.

QA method was set up as sequential series of cuts, upon failure later cuts were not checked.

Note, BSMD rate 4 had old resistors in day 366 of 2007 and was excluded from this analysis.
This reduces # of strips from 18,000 to 15,750 . 

 

Table 1 Definition of QA cuts
cut# cut code description # of discarded strips figure
1 1 at least 10,000 entries in the MPV bin 4 -
2  2  MPV position within +/-5 ADC channels  57 Fig 1
3 4  sig(ped) of gauss fit in [1.6,8] ADC ch 335 Fig 2
4 8  position of mean gauss within +/- 4 ADC  11 Fig 3
5 16  yield from [ped+40,ped+90] out of range 441 Fig 4
6 32  chi2/dof from slop fit in [0.6,2.5] 62 Fig 5
7 64  slopeError/slop >16% 4 Fig 6
8 128  slop within [-0.015, -0.05] 23 Fig 7
-  sum  out of processed 15,750 strips discarded  937 ==> 5.9%  

 

 

Fig 1 Example of strips failing QA cut #2, MPV position out of range , random strip selection

 

Fig 2a Distribution of width of pedestal vs. eta-bin

Fig 2b Example of strips failing QA cut #3, width of pedestal out of range , random strip selection

Fig 3a Distribution of pedestal position vs. eta-bin

Fig 3b Example of strips failing QA cut #4, pedestal position out of range , random strip selection

Fig 4a Distribution of yield from the slope fit range vs. eta-bin

Fig 4b Example of strips failing QA cut #5, yield from the slope fit range out of range , random strip selection

Fig 5a Distribution of chi2/DOF from the slope fit vs. eta-bin

Fig 5b Example of strips failing QA cut #6, chi2/DOF from the slope fit out of range , random strip selection

Fig 6a Distribution of err/slope vs. eta-bin

Fig 6b Example of strips failing QA cut #7, err/slope out of range , random strip selection

Fig 7a Distribution of slope vs. eta-bin

Fig 7b Example of strips failing QA cut #8, slope out of range , random strip selection

Results

Fig 8a Distribution of # of bad strips per module. 

BSMDE modules 10,31,68  are damaged above  50%+. Ymax was set to 150, i.e. to the # of eat strips per module. Modules 16-30 served by crate 4 were not QAed. 

Fig 8b 2D Distribution of # of bad strips indexed by eta & phi strip location. Z-scale denotes error code from the 2nd column from table 1.

Fig 9 2D Distribution of slope indexed by eta & phi strip location.
TOP: slopes. There is room for gain improvement in the offline analysis. At fixed eta (vertical line) there should be no color variation.
BOTTOM error of slope/slope.

Fig 10 2D Distribution of pedestal and pedestal width indexed by eta & phi strip location.
TOP: pedestal
BOTTOM: pedestal width.

08) SMD-E gain equalization , ver 1.1

Goal : predict BSMD-E relative gain corrections for every eta bin 

Method : find average slope per eta slice, fit gauss, determine average slope : avrSlope(iEta)

Gain correction formula is used only for extreme deviations:  

  • gainCorr_i = slope_i/avrSlope(iEta),  i =1,..,18,000
  • if( fabs(1-gainCorr)<0.15) gainCor=1.0  // do not correct

Fig 1 Example of 2 eta slices

 

Fig 2 TOP: Slope distribution vs. eta-bin, average marked by crosses

BOTTOM: predicted gain correction. Correction=1 for strips w/ undetermined gains. 

Fig 3 Predicted gain correction. Correction=1 for ~14K of 18K strips.

Fig 4 The same predicted gain correction vs. stripID.

09) QA of SMD-P slopes, ver1.1

Automatic QA of BSMD-P minB spectra. 

Content

  1. example of good spectra (fig 0)
  2. QA cuts definition (table 1) + spectra (figs 1-7)
  3. Result:
    1. # of bad strips per module. BSMD-P modules 1,4,59,75,85  are damaged above  50%+. (modules 16-30 served by crate 4 were not QAed). 
    2. eta-phi distributions of: slopes, slope error (fig 9) , pedestal, pedestal (fig 10) width _after_ QA
    3. sample of good and bad plots from every module, including modules 16-30 (PDF at the end)

The automatic  procedure doing QA of  spectra was set up in order to preserve only good looking spectra as shown in the fig 0 below. 

Fig 0   Good spectra for random strips in module=2. X-axis shows pedestal residua. It is shown to set a scale for the bad strips shown below. 

 

INPUT: 3M d-AU events from day ~336 of 2007.

All spectra were pedestals subtracted, using one value per strip, CAPS=123,124,125 were excluded. Below I'll use term 'ped' instead of more accurate pedestal residuum.

Method: fit slopes to ADC =ped+40,ped+90 or 5*sig(ped) if too low.

The spectra,  fits of pedestal residuum, and slopes were QAed.

QA method was set up as sequential series of cuts, upon failure later cuts were not checked.

Note, BSMD rate 4 had old resistors in day 366 of 2007 and was excluded from this analysis.
This reduces # of strips from 18,000 to 15,750 . 

 

Table 1 Definition of QA cuts
cut# cut code description # of discarded strips figure
1 1 at least 10,000 entries in the MPV bin 2 -
2  2  MPV position within +/-5 ADC channels  10 Fig 1
3 4  sig(ped) of gauss fit in [0.75,8] ADC ch 32 Fig 2
4 8  position of mean gauss within +/- 4 ADC  0 Fig 3
5 16  yield from [ped+40,ped+90] out of range 758 Fig 4
6 32  chi2/dof from slop fit in [0.55,2.5] 23 Fig 5
7 64  slopeError/slop >10% 1 Fig 6
8 128  slop within [-0.025, -0.055] 6 Fig 7
-  sum  out of processed 15,750 strips discarded  831 ==> 5.2%  

 

 

Fig 1 Example of strips failing QA cut #2, MPV position out of range , random strip selection

 

Fig 2a Distribution of width of pedestal vs. strip # inside the module. For the East side I cout strips as -1,-2, ...,-150.

Fig 2b Example of strips failing QA cut #3, width of pedestal out of range , random strip selection

Fig 3 Distribution of pedestal position vs. strip # inside the module 

Fig 4a Distribution of yield from the slope fit range vs. eta-bin

Fig 4b Example of strips failing QA cut #5, yield from the slope fit range out of range , random strip selection

Fig 5a Distribution of chi2/DOF from the slope fit vs. eta-bin

Fig 5b Example of strips failing QA cut #6, chi2/DOF from the slope fit out of range , random strip selection

Fig 6 Distribution of err/slope vs. eta-bin

Fig 7a Distribution of slope vs. eta-bin

Fig 7b Example of strips failing QA cut #8, slope out of range , random strip selection

Results

Fig 8a Distribution of # of bad strips per module. 

BSMD-P modules 1,4,59,75,85  are damaged above  50%+. Ymax was set to 150, i.e. to the # of eat strips per module. Modules 16-30 served by crate 4 were not QAed. 

Fig 8b 2D Distribution of # of bad strips indexed by eta & phi strip location. Z-scale denotes error code from the 2nd column from table 1.

Fig 9 2D Distribution of slope indexed by eta & phi strip location.
TOP: slopes.  At fixed eta (horizontal line) there should be no color variation. red=dead strips
BOTTOM error of slope/slope. white=dead strip

Fig 10 2D Distribution of pedestal and pedestal width indexed by eta & phi strip location.
TOP: pedestal. dead strip have 0 residuum.
BOTTOM: pedestal width. white marks dead strips

10) SMD-P gain equalization , ver 1.1

Goal : predict BSMD-P relative gain corrections for every eta bin 

Method : find average slope per eta slice, fit gauss, determine average slope : avrSlope(iEta)

Gain correction formula is used only for extreme deviations:  

  • gainCorr_i = slope_i/avrSlope(iEta),  i =1,..,18,000
  • if( fabs(1-gainCorr)<0.10) gainCor=1.0  // do not correct

Fig 1 Example of 2 eta slices

 

Fig 2 LEFT: Slope distribution vs. eta-bin, average marked by crosses

RIGHT: predicted gain correction. Correction=1 for strips w/ undetermined gains. 

Fig 3 Predicted gain correction. Correction=1 for ~14K of 18K strips.

Fig 4 The same predicted gain correction vs. stripID.

12) investigating status of P-strips

 
The  plot below merges what we found w/ Matt in BSMDP West with  what  Oleg told us so far about know broken Anode wires.
 
Plot show is 2D location of all 9K strips. It is half of this plot shown before.
In color are various failure modes detected by our software processing minB spectra.
One narrow bar denotes one strip.
 
Module # is printed on the right.
Anode wire range is printed on the top.
 
Black labels A1, A3, etc are from Olegs table of known broken Anode wires from year 2007.
 
Conclusions for day 336 of 2007:
a) we do not see good signal from broken anode wires on BSMD-P West.
 
b) we see more broken anode wires/problems BSMD-P West
  module36: Anode12
 m54: A12
 m24: completely dead
 m30: mostly dead
 m29: odd strips dead
 m23: most of the odd strips dead
 
c) BSMD-P East is terra incognita - no  external info about any problems there
 
Problems listed in b) ,c) seems to be new for day 336 of 2007
 
 

Fig 1. West BSMD-P

 

Fig 2. East BSMD-P

p>

 

 

13) ver 1.2 : SMD-E, -P, status & relative gains, no Crate4

Automatic BSMD-E, -P   relative calibration and status tables. 

The second pass through both SMDE, SMDP was performed, learning from previous mistakes.

Main changes:

  • relax most of the QA cuts
  • smooth raw spectra to reduce significance 1-2 channels wide  of spikes and deeps in raw spectra
  • extend slop fit range to ped +[30,100]
  • still do not use crate 4 - it has to many problems
  • reduce margin for gain correction - now 10%+ deviation is corrected.  

INPUT: 3M d-AU events from day ~336 of 2007.

All spectra were pedestals subtracted, using one value per strip, CAPS=123,124,125 were excluded. Below I'll use term 'ped' instead of more accurate pedestal residuum.

Method: fit slopes to ADC =ped+30,ped+100 or 5*sig(ped) if too low.

The spectra,  fits of pedestal residuum, and slopes were QAed.

Note, BSMD rate 4 had old resistors in day 366 of 2007 and was excluded from this analysis.
This reduces # of strips from 18,000 to 15,750 . 

 

Table 1 Definition of QA cuts, all plots (PDF1) 
cut# cut code description # of discarded E strips # of discarded P strips figure in PDF1
1 1 at least 10,000 entries in the MPV bin 4 ? -
2 2  sig(ped) of gauss fit <~13 ADC ch 13 11 Fig 1
3 4  position of mean gauss within +/- 4 ADC  10 8 Fig 2
4 8  yield from [ped+30,ped+100] out of range 513 766 Fig 3
5 16  chi2/dof<2.3 from slop fit  6 1 Fig 4
6 32  slopeError/slop >10% 5 0 Fig 5
7 64  slope in range 19 6 Fig 6
-  sum  out of processed 15,750 strips discarded  635 ==> 4.0%  789 ==> 5.0%  

 

 

Relative gain corrections for every eta bin 

Method : find average slope per eta slice, fit gauss, determine average slope : avrSlope(iEta)

Gain correction formula is used only for extreme deviations:  

  • gainCorr_i = slope_i/avrSlope(iEta),  i =1,..,18,000
  • if( fabs(1-gainCorr)<0.10) gainCor=1.0  // do not correct

PDF2 plots  shows: 

  • Fig 1  SMDE status table, distribution per module
  • Fig 2  SMDP status table, distribution per module
  • Fig 3  SMDE  gain corrections,  changed 3408 strips=22%
  • Fig 4  SMDP  gain corrections,  changed 2067 strips=13%

Summary of BSMDE,P status tales and gains , ver 1.2 

14 Eval of BSMDE status tables for pp 2008, day 49,50

Method:

 from _private_ production w/o zero BSMD suppression we look at pedestal residua for raw spectra for minb events.

chain="DbV20080703 B2008a ITTF IAna ppOpt VFPPV l3onl emcDY2 fpd ftpc  
trgd ZDCvtx NosvtIT NossdIT analysis Corr4 OSpaceZ2 OGridLeak3D  
beamLine BEmcDebug"

 

The only QA was to require  MPV of the spectrum is below 100, one run contains ~80K events.

Fig 1

Good spectra look like this:

 

Fig 2. Run 9049053, 80K events,

9,292 strips out of 15,750 tested were discarded by condition MPV>100 eve

TOP: MPV value from all strips. White means 0 (zero) counts. Crate 4 was not evaluated.
BOTTOM: status table: red=bad, white means MPV>100 events

Fig 3. Run 9050022, 80K events,

9,335 strips out of 15,750 tested were discarded by condition MPV>100 eve

TOP: MPV value from all strips. White means 0 (zero) counts
BOTTOM: status table: red=bad, white means MPV>100 events

Fig 4. Run 9050088, 80K events,

9,169 strips out of 15,750 tested were discarded by condition MPV>100 eve

TOP: MPV value from all strips. White means 0 (zero) counts
BOTTOM: status table: red=bad, white means MPV>100 events

15 stability of BSDM peds, day 47 is good

 Peds from run minB 17 were used as reference

Fig1 . pedestal residua for runs 17,29,31

Fig2 . pedestal residua for run 31, full P-plane

Fig3 . pedestal residua for run 31, full E-plane

Fig4 . pedestal residua for run 31, zoom in E-plane

Fig5 . pedestal residua for run 29, West, E-plane red, P-plane black, error=ped error

 

15a ped stability day 47, take 2

 On August 8, BSMD peds in the offline DB where corrected for day 47.

Runs minb 34 & 74 were used to determine and upload DB peds.

Below I evaluated pedestal residua for 2 runs : 37 & 70, both belonging to the same RHIC fill.

I have used 500 zero-bias events from runs 37 & 70, for the official production w/o zero suppression.

All strips for which  mTables->getStatus(iEP, id, statPed,"pedestal"); returns !=1 and all events using CAP123,124,125 were dropped.

Fig 1,2 show big picture: all 38,000 strips.

Fig 3 is zoom in on some small & big problems.

Fig 4 & 5 illustrates improvement if run-by-run pedestal is used. 

Fig 1, run=9047037

 

Fig 2, run=9047070

 

Fig 3, run=9047070, zoom in

 

Private peds were determine for 16 runs for day 47 and used appropriately. Below is sum of ped residua from all 16 runs, from zero bias events.

Fig 4, run=9047001,...,83 zoom in

 

Fig 5, run=9047001,...,83 full range

 

16) Time stability by fill of BSMD pedestals

I calculated the pedestals for every PP fill for 2008. This plot shows the pedestal per stripID and fill index. The Z-axis is the value of the pedestal. Only module 13 is shown here, but the full 2D histogram (and others) are in the attached root files.

 

17) Absolute gains , take1

 Goal: reco isolated gammas from bht0,1,2 -triggered events 

Method: identify isolated EM shower and match BSMD cluster energy to tower energy, as exercised earlier on 4) demonstration of absolute calib algo on single particle M-C 

INPUT events: 7,574 events triggered by barrel HT0,1,2 (id 220500 or 220510 or 220520) from run 9047029.

Cluster finder algo (sliding window, 1+3+1 strips),  smd cluster threshold set at 5 keV,  use only barrel West.

Tower cluster is defined as 3x3 patch centered on the tower pointed by the SMD peak.

Assumed BSMD calibration:  

  • ene(GeV)= (adc-ped)*1e-7, one constant for all barrel
  • pedestals, status tables hand tuned, some modules are disabled, but crate 4 is on

Results for ~3,8K barrel triggered events (half of 7,6K was not used)

Fig 1, Any  Eta-cluster

TOP: a) Cluster (Geant) energy;

b) Cluster RMS, peak at 0.5 is from low energy pair of isolated strips with almost equal energy

c) # of cluster per event, 

BOTTOM: X-axis is eta location, 20 bins span eta [-1,+1]. d) cluster ene vs. eta, e) cluster RMS vs. eta,

f) cluster yield vs. eta & phi, white bands are masked modules.

Fig 2, Any  Phi-cluster

see Fig 1 for details

 

Fig 3, Isolated EM shower

TOP: a) cluster loss on subsequent cuts, b) # of accepted EM cluster vs. eta location,

c) ADC distribution of 3x3 tower cluster centered at SMD cluster. In principle you should see there 3 edges from bht0, bht1, and bht2 trigger.

BOTTOM: X-axis is eta location, 20 bins span eta [-1,+1].d) Eta-cluster , e) phi-cluster energy, f) hit tower ADC .

 

Fig 4a,b, Calibration plots

TOP: BSMD Eta vs. Phi  as function of pseudorapidity.
BOTTOM: BSMD vs. BTOW as function of pseudorapidity.

2  eta location of 0.1, 0.5  of reco EM cluster  are shown in 3 panels (2x2)

1D plots are ratios of the respective 2D plots.

The  mean values of 1D fits are  relative gains of BSMDP/BSMDP and  BSMD/BTOW .

 

Fig 4c, Same as above, eta=0.9

 

18 Absolute gains, take 2

 Goal: reco isolated gammas from bht0,1,2 -triggered events 

Method: identify isolated EM shower and match BSMD cluster energy to tower energy, as exercised earlier on 4) demonstration of absolute calib algo on single particle M-C 

INPUT events: 100K events triggered by barrel HT0,1,2 (id 220500 or 220510 or 220520) from day 47, runs 1..83

Cluster finder algo (sliding window, 1+4+1 strips),  smd cluster threshold set at 10 keV,  use only barrel West, BSMD CR=4 masked out.

Tower cluster is defined as 3x3 patch centered on the tower pointed by the SMD peak, must contain 90% of energy from 5x5 cluster.

Default pedestals from offline DB used.

Assumed BSMD calibration: see table 1 column J+K  

Results for ~25K barrel triggered events (7/8 of 100K was not used)

 Fig 1 is above

Fig 2, Eta strips, any cluster

Fig 3 Phi strips, any cluster

Fig 4  isolated cluster (different sort). Plot c has huge peak at 0 - X-axis is chopped. Similar but smaller peak is in fig d. Magenta are event with bht0 and bht2 trigger.

Fig 5  isolated cluster :

Left: eta & phi plane coincidence--> works,

Right: eta & phi & tower 3x3>150 fials for modules 30-60, I have mapping problem?? 

Fig 6  Example of Eta vs. Phi  and SMD vs. Tower calibrations for eta bins 0.15, 0.5, and 0.85.

19) Absolute BSMD Calibration, table ver2.0, Isolated Gamma Algo description

 BSMD calibration algo has been developed based on M-C response of BSMD & towers to single gammas.

Executive summary:

The purpose of BSMD absolute calibration summarized at this drupal page is to reconstruct integrated energy deposit (dE) in BSMD based on measured ADC.
By integrated dE in BSMD I mean sum over few strips forming EM cluster, no matter what is the cluster shape.
This calibration method accounts for the varying absorber in front of BSMD and between eta & phi planes.
This calibration will NOT help in reconstruction:  
- full energy of EM particle which gets absorbed in BEMC ( shower development after BSMD layer does not matter for this calibration).
- partial energy of hadrons passing or showering in BEMC
- correct for the incident angle of the particle passing through detector
- saturation of BSMD readout. I only state up to which  loss (DE) the formula used in reconstruction:
      dE/GeV= (rawAdc-ped) * C0 * (1 +/- C1etaBin)
- determine sampling fraction (SF) of BSMD with high accuracy 

Below you will find brief description of the algo, side by side comparison of selected plots for M-C and real data, finally PDF with many more plots.

Proposed absolute calibration coefficients are show in table 2.

 

Part 1

Description of algorithm finding isolated gammas in the Barrel.

Input events

  • M-C :  single gamma per event, 6 GeV ET, flat in eta, phi covers 3 barrel modules 12,13,14, geometry=y2006
  • DATA: BHT0,1,2 triggered pp 2008 events from day 47, total 100K, individual triggers: 44K, 33K, 40K, respectively
    events were privately produced w/ zero suppression.

Raw data processing based on muDst

  • M-C : BSMD - take geant energy deposit  ~100 keV range, towers - take ADC*0.493 to have nominal calibration of 4070 ADC=60 GeV ET.
  • Data : BSMD - use private  pedestals & status tables for day 47, use custom calibration
    use BSMD calibration dE/GeV= (rawAdc-ped) * C0 * (1 +/- C1etaBin), where '+' is for Phi-plane and '-' for eta plane, see table below
    skip strips 4 sigma above ped or with energy below 1keV, strip to strip relative gains NOT used , data from CAPs=123,124,125 were not used
     towers- take ADC as is, no offline gains correction.

Cluster finder algo (seed is sliding fixed window), tuned on M-C gamma events

  • work with 150 Eta-strips per module or 900 Phi-strips at fixed eta
  • all strips are marked as 'unused'
  1. sum  dE in fixed window of 4 unused strips, snap at location which maximizes the energy
  2. if sum below 10 keV STOP searching for clusters in this band
  3. add energy from one strip on each side, mark all 1+4+1 strips as 'used'
  4. compute energy weighted  cluster position and RMS
  5. goto 1

This cluster finder process full Barrel West, more details about clustering is in one cluster topology , definition of 'barrel cell' 

Isolated EM shower has been selected as follows, tuned on gamma events,

  • select isolated eta-cluster in every segment of 15 eta strips.
  • require cluster center is at least 3 strips away from edges  of this segment (defined by eta values of 0.0, 0.1, 0.2,....0.9, 1.0)
  • require there is only one phi-cluster in the same 0.1x0.1 eta.phi cell
    • require phi-cluster center is at least 3 strips from the edges
  • find hit tower matching to the cross of eta & phi cluster
  • sum tower energy from 3x3 patch centered on hit tower
    • require 3x3 tower  ADC sum >150 ADC (equivalent to 2.2 GeV ET, EM)
  • sum tower energy from 5x5 patch centered on hit tower
    • require 3x3 sum/ 5x5 sum >0.9  
  • require RMS of Phi & Eta-cluster  is above 0.2 strips

 Below is listing of all cuts used by this algo:

  useDbPed=1; // 0= use my private peds
  par_skipSpecCAP=1; // 0 means use all BSMD caps
  par_strWinLen=4; (3)  // length of integration window, total  1+4+1, in strips
  par_strEneThr=1.e-6;  (0.5e-6) // GeV, energy threshold for strip to start cluster search
  par_cluEneThr=10.0e-6; (2.0e-6) // GeV, energy threshold for cluster in window
  par_kSigPed=4.; (3) // ADC threshold
  par_isoRms=0.2;  (0.11) // minimal smd 1D cluster RMS 
  par_isoMinT3x3adc=150; //cut off for low tower response
  par_isoTowerEneR=0.9; //  ratio of 3x3/5x4 cluster
(in red are adjusted values for MIP or ET=1GeV cluster selection)

 Table 1 Tower cluster cut defines energy of isolated gammas.

3x3 tower ET (GeV), trigger used MIP, BHT0,1,2 1.0, BHT0,1,2

3.4, BHT0

4.7, BHT1

5.5, BHT2

7, BHT2
3x3 tower ADC sum range 15-30 ADC 50-75 ADC 170-250 ADC 250-300 ADC 300-380 ADC 400-500 ADC
3x3 energy &  RMS (GeV) @ eta=[0.1,0.2] 0.34 +/- 0.06 0.92 +/- 0.11 3.1 +/- 0.3 4.1 +/- 0.2 5.1 +/- 0.3 6.6 +/- 0.4
3x3 energy &  RMS (GeV) @ eta=[0.4,0.5] 0.37 +/- 0.07 1.0 +/- 0.11 3.4 +/- 0.4 4.6 +/- 0.3 5.6 +/- 0.4 7.3 +/- 0.5
3x3 energy &  RMS (GeV) @ eta=[0.8,0.9] 0.47 +/- 0.09 1.3 +/- 0.16 4.3 +/- 0.4 5.7 +/- 0.3 7.1 +/- 0.5 9.3 +/- 0.6

  

Table 2 shows assumed calibration.   

Contains relative calibration  of eta vs. phi plane, different for M-C vs. data,
and single absolute DATA normalization of the ratio of SMD (Eta+Phi) cluster energy vs. 3x3 tower cluster at eta=0.5 .

Table 3 shows what comes from data & M-C analysis using calibration from table 2.

 

Part 2

Side by side comparison of M-C and real data. 

Fig 2.1 BSMD "Any cluster" properties

TOP : RMS vs. energy, only Eta-plane shown, Phi-plane looks similar
BOTTOM: eta -phi distribution of found clusters. Left is M-C - only 3 modules were 'populated'. Right is data, white bands are masked modules or whole BSMD crate 4

 

 

Fig 2.2  Crucial  cuts after coincidence & isolation was required for a pair BSMD Eta & Phi clusters 

TOP :  3x3 tower energy (black), hit-tower energy (green) , if 3x3 energy below 150 ADC cluster is discarded
BOTTOM: eta dependence of 3x3 cluster energy. M-C has 'funny' calibration - there is no reason for U-shape, Y-value at eta=0.5 is correct by construction.

Fig 2.3  None-essential cuts, left by inertia 

TOP :  ratio of 3x3 tower energy to 5x5 tower energy ,  rejected if below 0.9 
BOTTOM:  RMS of Eta & Phi cluster must be above 0.2, to exclude single strip clusters

Part 3

Examples of relative response of BSMD Eta vs. Phi AFTER  calibration above is applied. 

I'm showing examples for 3 eta slices of 0.15, 0.55, 0.85 -  plots for all eta bins  are available as PDF, posted in Table 2 at the end.

The red vertical line marks the target calibration, first 2 columns are aligned by definition, 3rd column is independent measurement confirming calibration for data holds for ~40% lower gamma energy.

Fig 3.1 Phi-cluster vs. Eta cluster for eta range [0.1,0.2]. M-C on the left, data in the middle, right.

Fig 3.2 Phi-cluster vs. Eta cluster for eta range [0.4,0.5]. M-C on the left, data in the middle, right.

Fig 3.3 Phi-cluster vs. Eta cluster for eta range [0.8,0.9]. M-C on the left, data in the middle, right.

Fig 3.4 Phi-cluster vs. Eta cluster for eta range [0.9,1.0]. M-C on the left, data in remaining columns.

 

Part 4

Absolute response of BSMD (Eta + Phi) vs. 3x3 tower cluster,  AFTER  calibration above is applied. 

I'm showing eta slices [0.4,0.5] used to set absolute scale. The red vertical line marks the target calibration, first 2 columns are aligned by definition, 3rd column is independent measurement  for gammas with ~40% lower --> BSMD response is NOT proportional to gamma energy.

Fig 4.1 Phi-cluster vs. Eta cluster for eta range [0.4,0.5]. Only data are shown.

Fig 4.2 Absolute BSMD calibration for eta range [0.0,0.1] (top) and eta range [0.1,0.2] (bottom) . Only data are shown.
Left: Y-axis is BSMD(E+P) cluster energy, y-error is error of the mean; X-axis 3x3 tower cluster energy, x-error is RMS of distribution . Fit (magenta thick) is using only to 4 middle points - I trust them more. The MIP point is too high due to necessary SMD cluster threshold, the 7GeV point has very low stat. There is no artificial point at 0,0. Dashed line is extrapolation of the fit.
Right: only slope param (P1) from the left is used to compute full BSMD Phi & Eta-plane calibration using formulas: 

slope P1_Eta=P1/2./(1-C1[xCell])/C0
slope P1_Phi=P1/2./(1+C1[xCell])/C0
Using C1[xCell],C0 from table 2.

Fig 4.3 Absolute BSMD calibration for eta range [0.2,0.3] (top) and eta range [0.3,0.4] (bottom) . Only data are shown, description as above.

Fig 4.4 Absolute BSMD calibration for eta range [0.4,0.5] (top) and eta range [0.5,0.6] (bottom) . Only data are shown, description as above.

Fig 4.5 Absolute BSMD calibration for eta range [0.6,0.7] (top) and eta range [0.7,0.8] (bottom) . Only data are shown, description as above.

Fig 4.6 Absolute BSMD calibration for eta range [0.8,0.9] (top) and eta range [0.9,0.95] (bottom) . Only data are shown, description as above.

I'm showing the last eta bin because it is completely different - I do not understand it at all. It was different on all plots above - just reporting here.

Fig 4.7 Expected BSMD gain dependence on HV, from  Oleg document The 2008 working HV=1430 V (same for eta & phi planes) - in the middle of the measured gain curve.

 Part 5

Possible extensions of this algo.

  1. cover also East barrel (for the cross check)
  2. include vertex correction in projecting SMD cluster to tower (perhaps)
  3. study energy resolution of eta & phi plane - now I just compensated relative gains but the total BSMD energy is simply sum of both planes
  4. last eta bin [0.9,1.0] is completely different, e.g. there is no MIP peak in the 2D fig 2.2 - BTOW gain (HV) is factor 2 or more way too high in this 2 bins 
    Justification:     Inspect right plot on figures 4.2,...,4.6, in particular note at what gamma energy the blue line reaches ADC of 1000 counts. Look at this pattern vs. eta bin. On the last plot it should happen at gamma energy of ~5 GeV but in reality it is at ~10 GeV.    
  5. crate 4 (unmodified) would have different gains - excluded in this analysis
  6. Speculation: those multiple peaks in raws BSMD spectra (seen by others) could be correlated with BHT0,1,2 triggers
  7. Scott suggestion: more detailed study of BSMD saturation could use BSMD cluster location for fiducial cut forcing gamma to be in the tower center and use just the hit tower. This needs more stats. This analysis uses 1 day of data and ends up with just ~100 entries per energy point.
  8. non-linear BSMD response does not mean we can reco cluster position with accuracy better than 1 strip.

 

 

Fig 5.1 BSMD cluster energy vs. eta of the cluster.

 

 

Fig 5.2 hit tower to 3x3 cluster energy for accepted clusters. DATA, trigger BHT2, gamma ET~5.5 GeV.

 

 

Fig 5.3 hit tower to 3x3 cluster energy for accepted clusters. M-C, single gamma ET=6 GeV, flat in eta .

 

20 BSMD saturation

 The isolated BSMD cluster algo allows to select different range of tower energy cluster as shown in Fig.1

Fig 1. Tower energy spectrum, marked range [1.2,1.8] GeV.

In the analysis 5 energy tower slices were selected: MIP, 1.5 GeV, and around BHT0,1,2 thresholds.

Plots below show example of calibrated BSMD (eta+phi) cluster energy vs. tower cluster energy. (I added point at zero with error as for next point to constrain the fit) 

Fig 2. BSMD vs. tower energy for eta of 0.15, 0.55, and 0.85.

I'm concern we are beyond the middle of BSMD dynamic range for ~6 GeV (energy) gammas at eta 0.5. Also one may argue we se already saturation.

If we want BSMD to work up to 40 GeV ET we need to think a lot how to accomplish that.

Below is dump of one event contributing to the last dot on the middle plot. It always help me to think if I see real raw event.

BSMDE
  i=526, smdE-id=6085 rawADC=87.0 ped=71.4  adc=15.6 ene/keV=0.9
  i=527, smdE-id=6086 rawADC=427.0 ped=65.0  adc=362.0 ene/keV=20.0
  i=528, smdE-id=6087 rawADC=814.0 ped=71.8 adc=742.2 ene/keV=41.0
  i=529, smdE-id=6088 rawADC=92.0 ped=66.4  adc=25.6 ene/keV=1.4


BSMDP
  i=422, smdP-id=6086 rawADC=204.0 ped=99.3  adc=104.7 ene/keV=7.8
  i=423, smdP-id=6096 rawADC=375.0 ped=98.5  adc=276.5 ene/keV=20.3
  i=424, smdP-id=6106 rawADC=692.0 ped=100.1  adc=591.9 ene/keV=47.3


2D cluster
bsmdE CL  meanId=6086 rms=0.80 ene/keV=66.80 inTw 1632.or.1612 
bsmdP CL  meanId=6106 rms=0.68 ene/keV=75.45 inTw 1631.or.1632 

BTOW
  id=1631 rawADC=43.0 ene=0.2 ped=30.0,  adc=13.0
  id=1632 rawADC=401.0 ene=6.9 ped=32.4  adc=368.6
  id=1633 rawADC=43.0 ene=0.1 ped=35.5   adc=7.5

 gotTwId=1632
 gotTwAdc=368.6
tow3x3 sum=405.7 ADC
3x3Tene=7.3GeV