ND3$\text{D}^{3}$ QNet: Noisy dueling‐double‐deep q‐network for reconfigurable intelligent surfaces

Low‐cost passive reconfigurable intelligent surfaces that provide coverage expansion in wireless communication networks are facing challenges in phase shift optimization due to limitations in acquiring channel information for sub‐channels. To overcome this, a new approach that improves both the convergence speed and overall performance relative to the existing state‐of‐the‐art scheme is proposed. Experimental results in an environment with a Rician factor of 20 showed a 9.3% increase in the average sum data rate, and the probability of the sum data rate exceeding the threshold of 10 increased by 30%. The paper delves into the rising interest in Reconfigurable Intelligent Surface (RIS) technology as a cost‐effective means of extending coverage. It highlights the challenge of optimizing phase adjustments due to limited channel information and presents a novel reinforcement learning‐based approach to improve both convergence speed and performance. In experimental tests within a Rician factor 20 environment, the proposed scheme exhibits a 9.3% increase in the average sum data rate and a 30% enhancement in the likelihood of exceeding the 10 threshold for the sum data rate when compared to a state‐of‐the‐art scheme.