Real-time Pattern Recognition in the Central Tracking Detector of the Belle II Experiment

Sumitted to PubDB: 2016-12-22

Category: Phd Thesis, Visibility: Public

Tags: -

Authors Jakob Lettenbichler
Non-Belle II authors Rudolf Frühwirth
Date Jan. 1, 2016
Belle II Number BELLE2-PTHESIS-2016-008
Abstract The Belle II experiment is currently under construction in the KEK laboratory in Tsukuba, Japan. It will record events produced by the world’s only second-generation B factory, the SuperKEKB collider, an asymmetric e+e− storage ring with 3km circumference. The collider will operate at the collision energy corresponding to the mass of the Υ(4𝑆) and the Υ(5𝑆) resonances, i.e., 10.58 GeV/𝑐2 and 10.86 GeV/𝑐2, respectively. With an integrated luminosity of 8 × 10^35 cm−2s−1 it will produce about 40 times more events than its predecessor KEKB and therefore more than any other before. Neither the detector of the predecessor experiment Belle nor its reconstruction software would have been able to deal with the amount of events expected and their properties. Therefore the entire event reconstruction software was redesigned from scratch, while crucial parts of the Belle were improved and rebuilt to form the new Belle II detector. This thesis describes the VerteX Detector Track Finder (VXDTF) which has no predecessor in the old reconstruction software. Its task is to reconstruct tracks in the innermost tracking detector, the VXD, a silicon based detector consisting of two parts, the PXD with two layers of pixel sensors using DEPFET technology, and the SVD with four layers of double-sided silicon strip sensors. The potential amount of data recorded by the PXD with its 8 million pixels with 8-bit depth each, read out with about 50,000 readout frames per seconds, cannot be stored, unless a data reduction by a factor of about 10 can be performed. To this end, the VXDTF will be deployed to reconstruct tracks in the SVD in real time. These tracks are then extrapolated to the PXD where they de ne regions of interest, that are then read out. The focus of the VXDTF therefore lies on full event reconstruction in the SVD only, with an additional emphasis on low-momentum tracking down to a transverse momentum of 50 MeV/𝑐. Such low-momentum tracks can be reconstructed only by a track finder operating in the innermost tracking detector. Given the small number of sensor layers in the SVD, the VXDTF hast to deal with little redundancy, with the stochastic disturbances due to material e ects that are particularly strong for low-momentum tracks, and with the increased background level caused by the high luminosity of the collider. The VXDTF consists of several steps: • In a first step, combinatorics are reduced by discarding physically implausible hit combinations. An effective set of Filters with cuts trained on simulated data is stored in a container called SecMap. The filters reduce the hits and their accepted combinations, which are then stored as a Tree. • On this Tree a CA is applied, which is used for extracting track candidates, the TCs. • These TCs then are fitted by a fast fit, which estimates their quality, stored in a quality indicator, the QI. • If there are TCs that share one or more hits, a Hop field network is used to find a subset of non-overlapping TCs. The definition of the weights and thresholds of the neural network uses the QIs, so that the final subset contains the TCs with the highest quality. The final TCs produced by the VXDTF are then used for ROI finding on the PXD, as previously described. This thesis describes the detector Belle II, the machine SuperKEKB and their features; illustrates how typical events recorded by the SVD look like; describes the details of the VXDTF implementation; analyzes its performance on MC events of Υ(4𝑆) decays with realistic background; reports its performance at a combined beam test where PXD and SVD sensors were jointly tested in an electron beam at the DESY laboratory. Finally, it describes the next steps regarding the VXDTF and its successor, the VXDTF 2, which is also described in this thesis. The performance of a current version of the VXDTF is summarized here. The reconstruction efficiency of the VXDTF for Υ(4𝑆) events with full background included is about 87% for all MC tracks creating enough hits in the SVD to be able to reconstructed. For tracks with a transverse momentum of 100 MeV/𝑐 the efficiency is still above 80% and for 50 MeV/𝑐 its about 65%. The complete details can be found in this thesis.
Conference Vienna

Files