Online-Learning-Based Multi-RIS-Aided Wireless Systems

The evolution of software-defined radios and reconfigurable intelligent surfaces (RIS) has enabled on-the-fly control and reconfigurability at the physical layer parameters and radio propagation environment. In multi-RIS-aided communication, the RIS block, comprising a certain number of RIS elements from one or more RIS, is selected to achieve high throughput reliable communication between transmitter and receiver. However, selecting an RIS block when there are multiple RIS and receivers is not trivial due to the large number of candidate blocks. In this article, a novel multiarmed bandit (MAB) framework, which can learn and select the optimal RIS block using focused exploration, is proposed. We provide the theoretical regret bound for the proposed algorithm and demonstrate the gain in performance over existing state-of-the-art statistical and MAB approaches via detailed simulation results in terms of rate, ergodic capacity, outage probability, energy efficiency, and received SNR. The proposed algorithm offers 33%–85% lower latency than existing MAB algorithms on various edge platforms. Furthermore, the gain in performance and latency improves with the increase in the number and size of the RIS in the network.