Implementação online de representação esparsa para reconstrução de energia no calorímetro hadrônico do ATLAS

Abstract

In High Energy Physics experiments, digital signal reconstruction is a fundamental step for the accurate estimation of the energy of particles produced in collisions. From an Electrical Engineering perspective, this problem can be formulated as a digital signal processing task involving filtering, deconvolution, and parameter estimation in highly noisy environments. In the ATLAS experiment at the Large Hadron Collider (LHC), the hadronic calorimeter TileCal employs dedicated electronics and digital filtering techniques in the trigger system to perform real-time event selection. With the accelerator upgrade to the HL-LHC (High Luminosity Large Hadron Collider), the significant increase in the rate of simultaneous events intensifies the pile-up effect, degrading the performance of classical reconstruction methods based on linear filtering. This scenario motivates the investigation of new signal processing approaches capable of operating under strict latency and computational complexity constraints. Sparse signal representation has shown promising results in this context, as it exploits the intrinsic structure of the signals to improve estimation performance in high-interference environments. However, its practical application in real-time systems is limited by the use of iterative algorithms. This work proposes a non-iterative adaptation of these methods, based on a deconvolution filter followed by a feed-forward neural network, which applies a shrinkage function with an adaptive threshold defined on a sample-by-sample basis. The proposed method was compared with techniques reported in the literature and demonstrated superior performance in both low- and high-occupancy scenarios, achieving reductions of up to 54% in reconstruction error. The approach maintains low computational complexity and features a structure compatible with real-time processing, making it suitable for implementation in embedded digital systems, such as the TileCal trigger system.