Performance Evaluation of Deep Q Networks for Hybrid Reconfigurable Intelligent Surface in 6G Networks

Abstract

The emergence of 6G wireless communication in-troduces a new era of connectivity demands, marked by high data rates and varying network conditions. To address these challenges, we propose HRISDQN, a framework that combines Hybrid Reconfigurable Intelligent Surfaces (HRIS) with Deep Q-Network (DQN)-based reinforcement learning. HRISDQN represents a significant advancement in optimising communication in 6G networks, enabling transformative improvements. In our work, we compare HRISDQN with conventional Semi Definite Relaxation (SDR), Maximum Ratio Transmission (MRT), and Minimum Mean Square Error (MMSE) as traditional beam-forming techniques. We demonstrate HRISDQN's adaptability to dynamic scenarios through extensive simulations and evaluations, including varying Signal-to-Noise Ratios (SNR) and changing user densities. Our results show that HRISDQN consistently outperforms its counterparts; HRISDQN's resource allocation capability ensures 40% better fairness, lower delay by 80%, and three times higher spectral efficiency, even in high-density user environments. The designed HRISDQN excels under diverse SNR conditions, providing robust and reliable connectivity. HRIS-DQN's exceptional performance holds great promise for the future of 6G communication. HRISDQN offers ultra-efficient, low-latency, and adaptive communication networks for augmented reality and autonomous vehicles using HRIS and DQN.