LEARNNG-BASED RIS ELEMENTS PARTITIONING FOR ISAC USING MULTI-ARMED BANDITS

Abstract

Integrated Sensing and Communications (ISAC)is oneoftheemergingparadigms for future generation wireless communications which enables communication and sens ing to be performed simultaneously by making use of common spectrum and hardware infrastructure. Performance of ISAC can further be boosted through the usage of RISs that manipulate wireless channels intelligently. The objective of this thesis is to optimize partition of RIS into communication and sensing sub-functions with the aid of reinforcement learning algorithm with Gaussian TS and UCB approaches applied to RIS-assisted ISAC systems under fading conditions. A realistic 4-hop monostatic sensing channel model is considered along with Nakagami fading based wireless propagation. The communication and sensing objectives are jointly optimized using a geometric mean reward formulation that ensures balanced ISAC performance. The proposed algorithm examines different partition setups of the RIS and deter mines the best partitioning policy through a series of interactions within the wireless environment. It is shown that through extensive simulations that the Gaussian Thomp son Sampling approach performs significantly better than UCB in terms of cumulative regret, convergence time, average reward, and stability. The results achieved confirm the applicability of reinforcement learning in online RIS partition optimization in upcoming ISAC-based wireless communication systems.