Implementação de métodos esparsos para deconvolução on-line de sinais no calorímetro hadrônico do ATLAS

Abstract

The particle accelerator, known as the large hadron collider went into operation in 2008 and, since then, has been undergoing a process of constant updating, where parameters such as collision energy and the amount of particles per bunch are increased. These increasing have affected all detectors, including the detectors in the ATLAS experiment. In this experiment, the trigger system is where the effects can be felt most forcefully, because there is an increasing in the occurrence of adjacent collisions, producing the effect known as pile-up (signals overlapping), that make the process difficult to selecting events of interest. Therefore, deconvolution on-line methods are being proposed to deal with this problem. Most of these methods are based on finite impulse response filter, because they already have a structure that allows the on-line implementation. In contrast, large hadron collider operates with train of bunches, where a time sequence with collision and another collision-less time sequence are interleaved. This feature can be interpreted as a fixed size acquisition windows, allowing the usage of iterative deconvolution methods. Such methods tend to present better results than the methods currently in use, based on finite impulse response filter, because it is possible to use calorimetry specific knowledge and non-causal algorithms. For this purpose, this work proposes iterative methods of deconvolution, based on modern theories of sparse data representation, and their performance is compared with the performance of methods based on finite impulse response filter. Among the analyzed methods, the Separable Surrogate Functionals stands out, being thus proposed a form of implementation of this method in field programmable gate array, whose results indicate the possibility of implementation of several instances of this method in parallel, in a single field programmable gate array, allowing then the on-line processing required in the first level trigger of ATLAS.