Total: 1
In this work, we propose a method for efficient resource utilization of simultaneously transmitting and reflecting reconfigurable intelligent surface (STAR-RIS) elements to ensure fair and high data rates. We introduce a subsurface assignment variable that determines the number of STAR-RIS elements allocated to each user and maximizes the sum of the data rates by jointly optimizing the phase shifts of the STAR-RIS and the subsurface assignment variables using an appropriately tailored deep reinforcement learning (DRL) algorithm. The proposed DRL method is also compared with a Dinkelbach algorithm and the designed hybrid DRL approach. A penalty term is incorporated into the DRL model to enhance resource utilization by intelligently deactivating STAR-RIS elements when not required. The proposed DRL method can achieve fair and high data rates for static and mobile users while ensuring efficient resource utilization through extensive simulations. Using the proposed DRL method, up to 27% and 21% of STAR-RIS elements can be deactivated in static and mobile scenarios, respectively, without affecting performance.