The Influence of Varying Pulse Shapes on the Energy Reconstruction of ATLAS Liquid Argon Calorimeter Signals using Convolutional Neural Networks

From 2026 to 2028 the large particle accelerator LHC at CERN will get an upgrade. The number of simultaneous proton-proton-collisions will be increased to make rare processes or even new physics discoverable. Increasing the amount of simultaneous collisions will be a big challenge for the signal processing and triggers. The so called Optimal Filter, which is currently used at the ATLAS Liquid Argon calorimeter for the energy reconstruction, will reach its limits. For a reliable energy reconstruction alternative algorithms are needed. A possible alternative is the usage of artificial neural networks. In this thesis the energy reconstruction using convolutional neural networks is investigated. Especially approaches of resource-saving optimizations of previously developed networks are presented. It is being studied how time shifted pulses influence the energy reconstruction and how these negative influences can be reduced. Additionally, convolutional neural networks which reconstruct the hit times are presented.