CA+KF Track Reconstruction in the STS S. Gorbunov and I. Kisel GSI/KIP/LIT CBM Collaboration Meeting Dresden, September 26, 2007
26 September 2007, DresdenIvan Kisel, KIP, Uni-Heidelberg2/12 CBM Note on SIMDized Kalman Filter Track Fit Implemented: in CbmL1CATracker Different CPU architectures The Kalman filter based track fit works with single precision floating point variables, and we are now not far from implementing it in integers in order to port it later to FPGA.
26 September 2007, DresdenIvan Kisel, KIP, Uni-Heidelberg3/12 Cellular Automaton Pseudocode 1 Create tracklets 2 Collect tracks
26 September 2007, DresdenIvan Kisel, KIP, Uni-Heidelberg4/12 Reconstruction Time vs. Number of MC Tracks (2) all tracks (1) all quasi-primary tracks (0) fast quasi-primary tracks
26 September 2007, DresdenIvan Kisel, KIP, Uni-Heidelberg5/12 Structure and Data cbmroot/L1 L1Algo L1Geometry L1Event (L1Strips, L1Hits) L1Tracks Strips: float vStripValues[NStrips]; // strip coordinates (32b) unsigned char vStripFlags [NStrips]; // strip iStation (6b) + used (1b) + used_by_dublets (1b)Hits: struct L1StsHit { unsigned short int f, b; // front (16b) and back (16b) strip indices }; L1StsHit L1StsHit vHits[NHits]; unsigned short int vRecoHits [NRecoHits]; // hit index (16b) unsigned char vRecoTracks [NRecoTracks]; // N hits on track (8b) class L1Triplet{ unsigned short int w0; // left hit (16b) unsigned short int w1; // first neighbour (16b) or middle hit (16b) unsigned short int w2; // N neighbours (16b) or right hit (16b) unsigned char b0; // chi2 (5b) + level (3b) unsigned char b1; // qp (8b) unsigned char b2; // qp error (8b) } Input: Output: Internal: L1Algo A standalone L1Algo module 300 kB About 300 kB per central event
26 September 2007, DresdenIvan Kisel, KIP, Uni-Heidelberg6/12 Event Display
26 September 2007, DresdenIvan Kisel, KIP, Uni-Heidelberg7/12 CA Track Finder Efficiency MBias events Central events Efficiency, % Track category Efficiency, % ( )Reference set (>1 GeV/c) ( ) ( )All set (>=4 hits, >100 MeV/c) ( ) ( )Extra set (<1 GeV/c) ( ) 0.4Clone Ghost MC tracks/event found633 Standard geometry: 2M2P4S Reconstructed + (Damaged + Good)
26 September 2007, DresdenIvan Kisel, KIP, Uni-Heidelberg8/12 Low Momentum Tracks 1. In general, efficiency calculation is based on similarity between parameters of generated and reconstructed tracks. 2. The simplest efficiency calculation is based on association of hits used for track fitting. 3. In the region of low momentum tracks it can be based on association of hits within the track road because of large multiple scattering and high hit density. Therefore, ghost in (2) can here contribute to (1). Central events
26 September 2007, DresdenIvan Kisel, KIP, Uni-Heidelberg9/12 Detector Inefficiency Tracking is gathering of 1/2D measurements into 5D tracks, here combinatorics Tracking is gathering of 1/2D measurements into 5D tracks, here combinatorics Therefore, tracking is split into two parts: local (1) and global (2) Therefore, tracking is split into two parts: local (1) and global (2) In the local part a gap between 1/2D and 5D is filled with triplets In the local part a gap between 1/2D and 5D is filled with triplets If there is no triplet in the local step, no track in the global step If there is no triplet in the local step, no track in the global step Therefore, short tracks are weak against detector inefficiency Therefore, short tracks are weak against detector inefficiency Tracks interesting for physics are usually long (long vs. short tracks) Tracks interesting for physics are usually long (long vs. short tracks) Specialized extra tracking step (usually indicates weakness of the detector) Specialized extra tracking step (usually indicates weakness of the detector) Increase acceptance ( ) keeping N ch constant (longer or chained strips) Increase acceptance ( ) keeping N ch constant (longer or chained strips) Double stations – 4x-, 3x-, 2x-strip space points (no inefficiency, no dead zones) Double stations – 4x-, 3x-, 2x-strip space points (no inefficiency, no dead zones)
26 September 2007, DresdenIvan Kisel, KIP, Uni-Heidelberg10/12 CA Track Finder Efficiency MBias events Central events Standard geometry: 2M2P4S Efficiency, % Track category Efficiency, % ( )Reference set (>1 GeV/c) ( ) ( )All set (>=4 hits, >100 MeV/c) ( ) ( )Extra set (<1 GeV/c) ( ) 0.3Clone Ghost < % MC tracks/event found 617 < % Reconstructed + (Damaged + Good) Detector efficiency (MC points) 98%
26 September 2007, DresdenIvan Kisel, KIP, Uni-Heidelberg11/12 CBM Note on Reconstruction of Decayed Particles Implemented: in CBM as CbmKFParticle in ALICE as AliKFParticle D0D0D0D0 D0D0D0D0 x, y, z, p x, p y, p z, E, m, L, c K-K-K-K- ++++ In addition to vertices, which are now production and decay points, all physical parameters of decayed particles together with the corresponding errors are provided by the package.
26 September 2007, DresdenIvan Kisel, KIP, Uni-Heidelberg12/12 Summary and Plans SIMDized CA and KF algorithms are released SIMDized CA and KF algorithms are released Low momentum tracks down to 100 MeV/c are found by default Low momentum tracks down to 100 MeV/c are found by default CA track finder works with inefficient detectors CA track finder works with inefficient detectors Need direct access to strips (fixed geometry at the level of modules) Further analysis of robustness of the CA track finder