Tracking: From Raw Data to Analysis

1. Tracking: From Raw Data to Analysis Matthew Herndon, May 2006 University of Wisconsin CDF Silicon Detector Workshop • Raw Data and Calibration • Clusters Hits, Resolution • Tracking • Physics • Conclusion BEACH 04 J. Piedra 1

2. Raw Data Bit 0 Bit 1 Bit ff, 7f Dup Chan 33 types Rick Snider, Matt Herndon, Lester Miller Cannot abandon nearest neighbour readout • Organized by HDI, Chip and Strips • Interpreted by bank unpacker • Bit errors common from optical readout • Chip ID errors, Strip ID errors • Dropped readout • Errors in termination characters • Improved version: NN(nearest neighbour) bank unpacker • Chip ID errors and termination characters corrected by understanding readout order • Nearest neighbour readout mode allows correction of strip IDs from context • Bits 0-1, lowest significance bits correctable: higher bits correctible if lower bits look correct • Large amount of data corrected • Efficiency improved by a few %, number of bad clusters reduced 2

3. Calibration Jason Nielsen SICHIPPED, SICHIPON, SISTRIPDH PROD_PHYS_SVX Small residual pedestal: Dynamic Pedestal Subtraction(DPS) • Operational requirements • Calibration needed on short time scale(24 hours): Before Production calibration jobs • Production calibration jobs include beamline determination using tracking • Calibration • Calibrate per strip pedestals, noise, and identify bad strips(hot or dead) • Provide short time scale detector performance evaluation along with SVXMon • Procedure • Take SVXCAL run with DPS-on and DPS-off • Integrate calibration run information with dead strip list and create final tables • Other calibration runs: bias scans/gain scans performed by experts • Calibration benefits • Improves cluster resolution and identifies clusters with potentially bad resolution • Used in tracking efficiency measurements and realistic detector simulation SiExpected and bad strips M. Herndon 3

4. Clustering Rick Snider, Matt Herndon Doug Glenzinski • Clustering algorithm • Thresholds: 1 strip 4, 2 strips 3 and 2, 3 strips all 2 • Loose clustering requirements for high efficiency • Clusters • Typically two strips: ~1/2 each • Charge varies between types/sides of sensors • Online Thresholds • Needs to be efficient for Landau peak of ~25ADC and RMS of order 5 counts • 7-10 ADC conservative • >12 ADC aggressive 4

5. The Integration Gate Matt Herndon, David Stuart R-phi,SAS, Z • Detector needs to be timed in • Integration window set relative to beam crossing time • Done for each type of sensor • Found that optimum window was different for different types of sensors • No optimum choice: May have changed with integrated radiation M. Herndon 5

6. Cluster Resolution Aaron Dominguez • Measuring intrinsic resolution • Constrain track to hits on two layers around layer of interest • Calculate intersection point using hit positions and transverse momentum of track • Measure resolution using likelihood fit • Found resolution had a considerable tail • Dropped entries outside of 2.5 to measure core • Track Fit • Tails still a problem for the track fit • Toy MC with % in tails reproduces misestimates of resolution seen in track fit • Double Gaussian model for track fit would be too time consuming M. Herndon 6

7. Alignment Ray Culbertson ISL: Juan Pablo Fernandez L00: Ray, Aart Heijboer • Alignment Methodology • Start from survey results: extremely important as a starting point, crosscheck or constraint. Compare to survey at every step • First step global alignment SVXII, ISL, L00 - compare COT and silicon beamlines • Fix tracks to layer 5 and beamline and align inner layers • Align r-phi, SAS and then Z • Ailgn for bows and small rotations • Constrain/realign using overlap regions • ISL and L00 • Separate process starting with OI tracks after SVXII alignment M. Herndon 7

8. Tracking Overview • Uses a general outside-in methodology • Start where evidence of tracks is most clear • Highest angular resolution portions of detector • Outer layers of COT and silicon • Algorithms • Form track seeds in outer layers • Helix from outer hits and beamline or vertex • Look for consistent hits moving inward • Refit and repeat after adding each hit • This method used for COT, OI(COT to silicon outside-in) and silicon only algorithms M. Herndon 8

9. OI Tracking Efficiency Matt Herndon Very low and went down with pT and time! • Measurement of the efficiency of adding silicon hits to COT tracks • Use J/ψ→ μ+μ- data (2001-2002 data) (2001-2002 data) • Problem is with COT alignment • Pointing very precise at high pt unless the alignment is not perfect • εSVX = 74.5 ± 0.3(stat) ± 2.2(sys)%(2001-2003 data) • Average efficiency for adding silicon to two tracks: In silicon fid. M. Herndon 9

10. New OI Efficiency old avg (88%) w/ same data (2003) The Silicon Group Matt Herndon +2.5% and flat - improved code +6% εSVX improved code and new quality cuts • Improved silicon pattern recognition code and detector performance • Hold search window open to at least 1mm until first silicon hit found εSVX = 74.5% → 88.5% 2001-2003 → 2001-2005 data M. Herndon 10

11. Silicon Only Tracking 80% TDR  Stephanie Menzemer, Thorsten Schiedle, Matt Herndon • Silicon Standalone Tracking: SiSA • Algorithm • Use two outer 3D point(r-phi + SAS) and vertex to find initial helix. Gives helix and constraint in z. • ISL cooling problems, wire bond oscillation, AVDD2 chip failures, beam incidents effect the stereo side disproportionately • SiSA Performance is suboptimal M. Herndon 11

12. Silicon Forward Tracking Z->ee MC 45% 85% 70% • Also new forward COT segment based algorithm • Combination will be very powerful Matt Herndon, Thorsten Schiedle Antonio Boveia, David Stuart Realistic MC predicts efficiency well • New forward silicon algorithms • Use 1 outer 3D point(r-phi + SAS) and vertex to find initial helix. No constraint • Much higher potential performance M. Herndon 12

13. L3 Tracking Dynamic Pedestal Sub(DPS) very effective Matt Herndon • Algorithm to confirm L2 SVT decisions • Reduces output from L3 • Often a limiting factor - Consumer Server Logger • Requirements - Fast • Solution • Only perform OI tracking • Apply no calibration (calibration people were unhappy) • Don’t read unnecessary code: Reading L00 or forward ISL • Profile all code and optimize anything that seems slow • Level 3 tracking • Executes in 1/5 the time • Loose a few % in efficiency and resolution M. Herndon 13

14. L00 Matt Herndon, Aart Heijboer Auke Pieter Colijn, Chris Hill, David Stuart Tim Nelson • Makes a huge difference in B physics • Data not usable as is • Large coherent noise: Continuous pattern across all strips on a sensor • Induced by silicon readout: Different event by event • Coupling between cables and shields and due to space constraint • Pedestal fit • Event by event fit to find the pedestal distribution • Ignoring sharp peaks: real clusters • Effectively finds clusters • CPU intensive: can’t be used in L3 • Clusters after pedestal fit good for physics • Using tight quality cuts, but probably not necessary • Special track fit routine and alignment 14

15. Physics Impact The B Group 3D 2D Requires Z • Z strips and L00 • Bhh • Two tracks always meet in 2D. 3D needed to constrain vertex • Bs  • 3D pointing to vertex • Use of L00 M. Herndon 15

16. Bs Mixing Adding L00 Improvement in sensitivity at high ms comes from improvement in vertex resolution The BsMixing Group Sensitivity at 25ps-1 |Vtd| / |Vts| = 0.208 +0.008 -0.007 • L00/forward tracking • Bs mixing • L00 improves proper time resolution • Sensitivity dependent on the exponential of the square of t • 30% muon tags from forward tracking - most powerful tag M. Herndon 16

17. Conclusions A/A (17.25 ps-1) = 3.5 • Highly successful silicon tracker • All aspects of the system making a difference in physics analysis • SVXII gives high efficiency tracking and good vertexing for all analysis • ISL forward tracking: Seeded tracking used for Z and electroweak physics standard forward tracking for B physics • Z strips reduce background in B physics two track modes • L00 and SVT critical for Bs mixing • Many things could still be improved • Use of L00 • Forward tracking • Tracking at high Luminosity • Faster code • See Rainer and Mircea M. Herndon 17