DATCON: FPGA-based data reduction for the Belle II Pixel Detector
DATCON: FPGA-based data reduction for the Belle II Pixel Detector
Dokument öffnen (4.4MB)Art der Hochschulschrift
DissertationPrüfungsdatum
02.03.2026Datum der Veröffentlichung
16.06.2026Erstgutachter
Bernlochner, FlorianZweitgutachter
Dingfelder, JochenMetadaten
Zur Langanzeige
Zitierbare Links 
Inhalt
DATCON, the Data Acquisition Tracking and Concentrator Online Node, is a system of 15 FPGAs at the Belle II detector at KEK in Tsukuba, Japan. As part of the data acquisition (DAQ) chain, its task is to reduce the data volume produced by the Pixel Detector (PXD), while preserving the information needed for precise vertex reconstruction.
The PXD produces up to 20 GB/s of data. This rate has to be reduced by a factor of ~10 before the data can be transmitted and stored. DATCON performs a fast online track reconstruction using hit information from the surrounding Silicon Vertex Detector (SVD) to define regions of interest (ROIs) on the PXD sensors. Only PXD pixels inside these ROIs are retained.
To meet strict latency requirements, DATCON processes the SVD strip data using a Hough transformation based track reconstruction algorithm in two separate 2D projections: the xy-plane (transverse to the magnetic field) and the rz-plane (along the magnetic field). The resulting two-dimensional track candidates are extrapolated onto the PXD layers and combined to form ROIs.
This combination step was the main limitation of the previous architecture of DATCON. The independent combination of 2D track candidates generated an excessive number of fake ROIs, which saturated the output bandwidth, rendering the system insufficient for standalone data reduction during the physics runs of Belle II.
This thesis presents a redesigned tracking architecture that addresses this shortcoming. The new system encodes the three-dimensional information of SVD hits in geometric segments, which are then carried over to the combination step. By evaluating the shared segments, the system can correlate track candidates from the two projections. Requiring a minimum number of matching segments rejects unphysical combinations and suppresses fake ROIs. With the segment veto, DATCON achieves the necessary data reduction factor (DRF) while maintaining a high hit finding efficiency (HFE).
The PXD produces up to 20 GB/s of data. This rate has to be reduced by a factor of ~10 before the data can be transmitted and stored. DATCON performs a fast online track reconstruction using hit information from the surrounding Silicon Vertex Detector (SVD) to define regions of interest (ROIs) on the PXD sensors. Only PXD pixels inside these ROIs are retained.
To meet strict latency requirements, DATCON processes the SVD strip data using a Hough transformation based track reconstruction algorithm in two separate 2D projections: the xy-plane (transverse to the magnetic field) and the rz-plane (along the magnetic field). The resulting two-dimensional track candidates are extrapolated onto the PXD layers and combined to form ROIs.
This combination step was the main limitation of the previous architecture of DATCON. The independent combination of 2D track candidates generated an excessive number of fake ROIs, which saturated the output bandwidth, rendering the system insufficient for standalone data reduction during the physics runs of Belle II.
This thesis presents a redesigned tracking architecture that addresses this shortcoming. The new system encodes the three-dimensional information of SVD hits in geometric segments, which are then carried over to the combination step. By evaluating the shared segments, the system can correlate track candidates from the two projections. Requiring a minimum number of matching segments rejects unphysical combinations and suppresses fake ROIs. With the segment veto, DATCON achieves the necessary data reduction factor (DRF) while maintaining a high hit finding efficiency (HFE).
Klassifikation (DDC)
530 PhysikFarkas, Ralf: DATCON: FPGA-based data reduction for the Belle II Pixel Detector. - Bonn, 2026. - Dissertation, Rheinische Friedrich-Wilhelms-Universität Bonn.
Online-Ausgabe in bonndoc: https://nbn-resolving.org/urn:nbn:de:hbz:5-90569
Online-Ausgabe in bonndoc: https://nbn-resolving.org/urn:nbn:de:hbz:5-90569
@phdthesis{handle:20.500.11811/14209,
urn: https://nbn-resolving.org/urn:nbn:de:hbz:5-90569,
doi: https://doi.org/10.48565/bonndoc-885,
author = {{Ralf Farkas}},
title = {DATCON: FPGA-based data reduction for the Belle II Pixel Detector},
school = {Rheinische Friedrich-Wilhelms-Universität Bonn},
year = 2026,
month = jun,
note = {DATCON, the Data Acquisition Tracking and Concentrator Online Node, is a system of 15 FPGAs at the Belle II detector at KEK in Tsukuba, Japan. As part of the data acquisition (DAQ) chain, its task is to reduce the data volume produced by the Pixel Detector (PXD), while preserving the information needed for precise vertex reconstruction.
The PXD produces up to 20 GB/s of data. This rate has to be reduced by a factor of ~10 before the data can be transmitted and stored. DATCON performs a fast online track reconstruction using hit information from the surrounding Silicon Vertex Detector (SVD) to define regions of interest (ROIs) on the PXD sensors. Only PXD pixels inside these ROIs are retained.
To meet strict latency requirements, DATCON processes the SVD strip data using a Hough transformation based track reconstruction algorithm in two separate 2D projections: the xy-plane (transverse to the magnetic field) and the rz-plane (along the magnetic field). The resulting two-dimensional track candidates are extrapolated onto the PXD layers and combined to form ROIs.
This combination step was the main limitation of the previous architecture of DATCON. The independent combination of 2D track candidates generated an excessive number of fake ROIs, which saturated the output bandwidth, rendering the system insufficient for standalone data reduction during the physics runs of Belle II.
This thesis presents a redesigned tracking architecture that addresses this shortcoming. The new system encodes the three-dimensional information of SVD hits in geometric segments, which are then carried over to the combination step. By evaluating the shared segments, the system can correlate track candidates from the two projections. Requiring a minimum number of matching segments rejects unphysical combinations and suppresses fake ROIs. With the segment veto, DATCON achieves the necessary data reduction factor (DRF) while maintaining a high hit finding efficiency (HFE).},
url = {https://hdl.handle.net/20.500.11811/14209}
}
urn: https://nbn-resolving.org/urn:nbn:de:hbz:5-90569,
doi: https://doi.org/10.48565/bonndoc-885,
author = {{Ralf Farkas}},
title = {DATCON: FPGA-based data reduction for the Belle II Pixel Detector},
school = {Rheinische Friedrich-Wilhelms-Universität Bonn},
year = 2026,
month = jun,
note = {DATCON, the Data Acquisition Tracking and Concentrator Online Node, is a system of 15 FPGAs at the Belle II detector at KEK in Tsukuba, Japan. As part of the data acquisition (DAQ) chain, its task is to reduce the data volume produced by the Pixel Detector (PXD), while preserving the information needed for precise vertex reconstruction.
The PXD produces up to 20 GB/s of data. This rate has to be reduced by a factor of ~10 before the data can be transmitted and stored. DATCON performs a fast online track reconstruction using hit information from the surrounding Silicon Vertex Detector (SVD) to define regions of interest (ROIs) on the PXD sensors. Only PXD pixels inside these ROIs are retained.
To meet strict latency requirements, DATCON processes the SVD strip data using a Hough transformation based track reconstruction algorithm in two separate 2D projections: the xy-plane (transverse to the magnetic field) and the rz-plane (along the magnetic field). The resulting two-dimensional track candidates are extrapolated onto the PXD layers and combined to form ROIs.
This combination step was the main limitation of the previous architecture of DATCON. The independent combination of 2D track candidates generated an excessive number of fake ROIs, which saturated the output bandwidth, rendering the system insufficient for standalone data reduction during the physics runs of Belle II.
This thesis presents a redesigned tracking architecture that addresses this shortcoming. The new system encodes the three-dimensional information of SVD hits in geometric segments, which are then carried over to the combination step. By evaluating the shared segments, the system can correlate track candidates from the two projections. Requiring a minimum number of matching segments rejects unphysical combinations and suppresses fake ROIs. With the segment veto, DATCON achieves the necessary data reduction factor (DRF) while maintaining a high hit finding efficiency (HFE).},
url = {https://hdl.handle.net/20.500.11811/14209}
}
Die folgenden Nutzungsbestimmungen sind mit dieser Ressource verbunden:
