Detecção de Sinal em Calorimetria de Altas Energias Utilizando Redes Neurais Artificiais

Abstract

Signal superposition is a recurrent problem when information is transmitted through a communication channel at a rate higher than its response time. Modern particle colliders, operating with collision rates of tens of millions of subatomic particles per second, exhibit signal superposition in their front-end electronics, which hinders the selection of events of interest by the detector’s trigger systems. Current deconvolution methods model the front-end electronics as a linear transmission channel whose transfer function can be compensated using linear channel equalization techniques, allowing the cancellation of intersymbol interference and proper interpretation of previously superimposed data. Optimal approximations of this equalizer are typically implemented using FIR filters. However, this linear compensation approach does not account for intrinsic nonlinearities present in sensor outputs. In this work, instead of using linear channel equalization, a neural network (NN) classifier is employed to perform the detection of the signal of interest. Simulations are used to compare the detection efficiency of both methods under different event rate scenarios.