[Paper Review] Real-time Timepix3 data clustering, visualization and classification with a new Clusterer framework

L. Meduna, B. Bergmann+4 more|arXiv (Cornell University)|Oct 29, 2019

Particle Detector Development and Performance4 citations

TL;DR

This paper presents a real-time, multi-threaded C++ framework called Clusterer for processing Timepix3 detector data, enabling online clustering, visualization, and filtering of particle events using a quadtree data structure and dynamic plug-in filters. The system achieves low-latency event reconstruction from unordered pixel streams and reduces data volume by up to 90% through intelligent filtering, validated in CERN's ATLAS and MoEDAL experiments with live radiation monitoring and 3D track reconstruction capabilities.

ABSTRACT

With the next-generation Timepix3 hybrid pixel detector, new possibilities and challenges have arisen. The Timepix3 segments active sensor area of 1.98 $cm^2$ into a square matrix of 256 x 256 pixels. In each pixel, the Time of Arrival (ToA, with a time binning of 1.56 $ns$) and Time over Threshold (ToT, energy) are measured simultaneously in a data-driven, i.e. self-triggered, read-out scheme. This contribution presents a framework for data acquisition, real-time clustering, visualization, classification and data saving. All of these tasks can be performed online, directly from multiple readouts through UDP protocol. Clusters are reconstructed on a pixel-by-pixel decision from the stream of not-necessarily chronologically sorted pixel data. To achieve quick spatial pixel-to-cluster matching, non-trivial data structures (quadtree) are utilized. Furthermore, parallelism (i.e multi-threaded architecture) is used to further improve the performance of the framework. Such real-time clustering offers the advantages of online filtering and classification of events. Versatility of the software is ensured by supporting all major operating systems (macOS, Windows and Linux) with both graphical and command-line interfaces. The performance of the real-time clustering and applied filtration methods are demonstrated using data from the Timepix3 network installed in the ATLAS and MoEDAL experiments at CERN.

Motivation & Objective

• To enable real-time processing of high-rate, data-driven Timepix3 detector streams for radiation monitoring and event pre-selection.
• To address the challenge of reconstructing particle tracks from partially unordered pixel data due to hardware limitations in Timepix3.
• To reduce data storage demands by implementing online filtering based on cluster properties such as size and energy.
• To ensure cross-platform compatibility and support for both GUI and command-line interfaces for broad usability.
• To provide a modular, extensible software framework compatible with various Timepix3 readout systems, including Katherine and SPIDR.

Proposed method

• The framework uses a quadtree spatial data structure to accelerate pixel-to-cluster matching during real-time clustering.
• It implements a multi-threaded architecture to parallelize input parsing, clustering, and visualization tasks for high performance.
• A novel event-building algorithm processes incoming pixels in chronological order by dynamically merging clusters based on time and spatial proximity (8-connected neighbors).
• The system uses a time window of 2000 ns to define cluster membership, ensuring coherence of particle interactions.
• Filtering is applied via plug-in libraries that allow user-defined criteria for event selection, such as energy deposition and cluster size.
• The framework supports energy calibration and time-walk correction, enhancing data accuracy for physics analysis.

Experimental results

Research questions

• RQ1 How can particle events be reconstructed in real time from unordered Timepix3 pixel data streams?
• RQ2 What data structures and parallelization strategies enable low-latency clustering of high-rate detector data?
• RQ3 To what extent can online filtering reduce data volume without losing physics-relevant events?
• RQ4 How effective is the framework in reconstructing 3D particle tracks and identifying inelastic interactions?
• RQ5 Can the software be deployed reliably across multiple operating systems with both GUI and CLI support?

Key findings

• The framework successfully reconstructs particle tracks in real time from Timepix3 data streams, even with up to 200 µs of temporal disorder in pixel data.
• Using a 2000 ns time window and 8-connected spatial criteria, the system achieves accurate clustering of particle interactions with minimal latency.
• Online filtering reduced data volume by up to 90% in test beam campaigns, significantly decreasing disk storage requirements while preserving physics-relevant events.
• The system demonstrated real-time radiation field monitoring with live statistics, including cluster rate and energy spectrum, as shown in CERN test beam data.
• Skeletonization of selected clusters enabled the identification of five-prong fragmentation tracks from inelastic nuclear interactions, confirming the framework’s capability for detailed event analysis.
• The framework achieved full cross-platform compatibility on Windows, macOS, and Linux, with both GUI and CLI interfaces, ensuring broad usability in experimental physics.