Chengshuo Zhang | arXiv Science

arXiv2026

CTPX1: A Highly Integrated and High-Throughput Data-Driven Camera Based on Timepix4

The upgrade of the China Spallation Neutron Source (CSNS-II) will raise the proton beam power to 500 kW. Consequently, the existing Timepix3-based detector systems, limited to a count rate of 80 Mhits/s, will encounter severe saturation challenges. To address the demand of the Energy-Resolved Neutron Imaging instrument (ERNI) for next-generation higher count-rate electronics, this paper presents CTPX1, a high-performance data-driven camera system based on the Timepix4 ASIC. The system adopts a compact modular architecture, integrating readout electronics, a precision high-voltage bias unit, and a TEC temperature control subsystem. To fully exploit the readout potential of the Timepix4 ASIC's 16 high-speed serial links, this paper proposes a two-stage parallel processing architecture. This architecture achieves real-time data aggregation with a total bandwidth of up to 81.92 Gbps. Over a continuous 12-hour operation period, temperature fluctuations were kept within 0.1 °C while the high-voltage output noise remained below 1 mV. High-flux X-ray testing indicates that the system achieves a peak event readout rate of 1.17 Ghits/s, approaching the limit of the configured link speed. In-beam neutron verification at CSNS confirms that the camera successfully resolves fine spatial structures, achieving an imaging performance consistent with the 55 μm pixel pitch of the sensor. Furthermore, the clear observation of spectral features in the Time of flight (TOF) spectrum of a γ-Fe sample validates the system's good time resolution. This camera effectively addresses the data readout saturation challenges, validates the feasibility of Timepix4 technology for neutron imaging, and provides a viable solution for next-generation high-performance neutron imaging instruments.

physics.ins-det↗

arXiv2025

Personalized Federated Dictionary Learning for Modeling Heterogeneity in Multi-site fMRI Data

Data privacy constraints pose significant challenges for large-scale neuroimaging analysis, especially in multi-site functional magnetic resonance imaging (fMRI) studies, where site-specific heterogeneity leads to non-independent and identically distributed (non-IID) data. These factors hinder the development of generalizable models. To address these challenges, we propose Personalized Federated Dictionary Learning (PFedDL), a novel federated learning framework that enables collaborative modeling across sites without sharing raw data. PFedDL performs independent dictionary learning at each site, decomposing each site-specific dictionary into a shared global component and a personalized local component. The global atoms are updated via federated aggregation to promote cross-site consistency, while the local atoms are refined independently to capture site-specific variability, thereby enhancing downstream analysis. Experiments on the ABIDE dataset demonstrate that PFedDL outperforms existing methods in accuracy and robustness across non-IID datasets.

cs.LG↗