Deconvolução free-running aplicada à estimação online de energia na calorimetria do ATLAS

Abstract

This thesis proposes signal processing algorithms to deconvolve the energy information in ATLAS calorimeters. Modern calorimeters have a pulse response that is longer than the acquisition rate of events and can generate the effect signal overlapping on your readout electronics. The high event rate environment foreseen to occur in the next few years in the largest particle collider in the world, the LHC, increases the likelihood of adjacent collisions occurring in its detector experiments. In ATLAS, the first-level online event selection operates uninterrupted and must provide the energy estimation for each collision. The pile-up effect may cause several problems in energy estimation, and some solutions which relies on the standard optimal filter algorithm treat the effect only as additive noise. In contrast, deconvolution techniques may compensate for the pileup effect and may be independent of parameters for unsupervised techniques. Two deconvolution algorithms to operate in this environment are proposed. The first one is supervised and it uses a FIR filter followed by the Shirinkage function, based on deconvolution by sparse representations. The second one, called Bussgang-Sparse, is unsupervised, and relies on the Bussgang algorithm associated with the sparsity criterion. The results show that deconvolution based on sparse representations offer better noise attenuation performance than FIR filter deconvolution based on the inverse pulse response of the calorimeter. For the second unsupervised technique, the results demonstrated convergence to deconvolution techniques based on FIR filters.