Transcript Fast reconstruction of tracks in the inner tracker of - Indico

Fast reconstruction of tracks
in the inner tracker
of the CBM experiment
Ivan Kisel
(for the CBM Collaboration)
KIP
Kirchhoff Institute of Physics
University of Heidelberg, Germany
CBM
CHEP 2004
Interlaken, Switzerland, 30.09.04
Outline
• CBM Experiment at GSI
• Cellular Automaton (CA) Method
• CBM Track Finding based on CA
• Efficiency and Timing
• Our Experience (HERA-B and LHCb)
• Summary
30 September 2004, CHEP 04
Ivan Kisel, KIP, Uni-Heidelberg
2
Compressed Baryonic Matter (CBM) Experiment at GSI
 Radiation hard Silicon pixel/strip detectors in a magnetic dipole field
 Electron detectors: RICH & TRD & ECAL: pion suppression up to 105
 Hadron identification: RPC, RICH
 Measurement of photons, p0, h: electromagnetic calorimeter (ECAL)
 High speed data acquisition and trigger system
30 September 2004, CHEP 04
Ivan Kisel, KIP, Uni-Heidelberg
3
CBM Reconstruction Algorithms
TRD, ECAL
• Conformal Mapping
• Hough Transform
• Cellular Automaton
RICH
S5
S4
S3
S2
S1
M2
M1
J/y
 107 Au+Au reactions/sec with high track multiplicity (700 – 1000)
 determination of displaced vertices with high resolution ( 30 m)
 identification of electrons and hadrons
30 September 2004, CHEP 04
Ivan Kisel, KIP, Uni-Heidelberg
4
Cellular Automaton Method
1.
2.
3.
4.
Define :
NIM A329 (1993) 262
NIM A387 (1997) 433
NIM A489 (2002) 389
NIM A490 (2002) 546
•CELLS
•NEIGHBORS
•RULES
•EVOLUTION
 Being essentially local and parallel
cellular automata avoid exhaustive
combinatorial searches, even when
implemented on conventional computers.
 Since cellular automata operate with
highly structured information (for
instance sets of tracklets connecting space
points), the amount of data to be
processed in the course of the track search
is significantly reduced.
 Further reduction of information to
be processed is achieved by smart
definition of neighborhood.
.
 Usually cellular automata employ a
very simple track model which leads to
utmost computational simplicity and a
fast algorithm.
.
0
.
-> TRACKLETS
-> TRACK MODEL
-> BEST TRACK CANDIDATE
-> CONSECUTIVE OR PARALLEL
1
2
3
4
5
Create tracklets
Collect tracks
30 September 2004, CHEP 04
Ivan Kisel, KIP, Uni-Heidelberg
5
CBM Track Finding
Parabola
Main Program
Event Loop
Reconstruction Part
RECONSTRUCTION
•Fetch MC data
•Copy to local arrays and sort
•Create tracklets
•Link tracklets
•Create track candidates
•Select tracks
MC Truth -> NO
Performance Part
PERFORMANCE
•Evaluation of efficiencies
•Evaluation of resolutions
•Histogramming
•Timing
•Statistics
•Event display
Straight line
MC Truth -> YES
30 September 2004, CHEP 04
Ivan Kisel, KIP, Uni-Heidelberg
6
CBM Cellular Automaton Tracking Efficiency
RECO STATISTICS 100 events
Refprim efficiency
98.36 46562
Refset efficiency
94.85 49250
Allset efficiency
90.09 64860
Extra efficiency
77.79 15610
Clone probability
0.11
74
Ghost probability
5.18 3358
Reco MC tracks/event 648
Timing/event
175 ms
ALL MC TRACKS
RECONSTRUCTABLE TRACKS
Number of hits >= 3
REFERENCE TRACKS
Momentum > 1 GeV
FPGA
co-processor
98%
CA – INTRINSICALLY LOCAL
AND PARALLEL
TIMING (ms)
Fetch ROOT MC data
63.3
Copy to local arrays and sort 12.4
Create and link tracklets
115.7
Create track candidates
53.5
Select tracks
2.6
CPU
2%
30 September 2004, CHEP 04
Ivan Kisel, KIP, Uni-Heidelberg
7
Our Experience: HERA-B Pattern Tracking
Cellular
Automaton
Kalman
Filter
Hough
Transform
Accuracy
Efficiency
Time per
Event, sec
RANGER
~3x
CATS
~300 tracks/event
NIM A489 (2002) 389; NIM A490 (2002) 546; I. Gorbounov, Ph.D. Thesis, Uni-Siegen, 2004
30 September 2004, CHEP 04
Ivan Kisel, KIP, Uni-Heidelberg
8
Our Experience: LHCb Level-1 Trigger
1.
2.
3.
Mean: 4.8 ms
 time (ms)
• Cellular Automaton algorithm
• FPGA co-processor at 50 MHz
• 8 processing units running in parallel
=> 15 s !
Expect a factor 7—8 in CPU power in 2007
(PASTA report)
=> we are already within 1 ms !
FPGA (CA)
 Events
 Events
CPU (CA)
Tracking efficiency 97—99%
Primary vertex resolution 46 m
Timing 4.8 ms
Mean: 15 s
 time (s)
LHCb Trigger TDR, CERN-LHCC-2003-031; LHCb Note 2003-064; K. Giapoutzis, Diploma Thesis, Uni-Heidelberg, 2002
30 September 2004, CHEP 04
Ivan Kisel, KIP, Uni-Heidelberg
9
Summary
• Fast and efficient track finder based on the cellular automaton method
• Locality suitable for inhomogeneous magnetic field
• Possible implementation in hardware to accelerate the combinatorial part
30 September 2004, CHEP 04
Ivan Kisel, KIP, Uni-Heidelberg
10