Sum Rate Enhancement using Machine Learning for Semi-Self Sensing Hybrid RIS-Enabled ISAC in THz Bands

Abstract

This paper proposes a novel semi-self sensing hybrid reconfigurable intelligent surface (SS-HRIS) in terahertz (THz) bands, where the RIS is equipped with reflecting elements divided between passive and active elements in addition to sensing elements. SS-HRIS along with integrated sensing and communications (ISAC) can help to mitigate the multipath attenuation that is abundant in THz bands. In our proposed scheme, sensors are configured at the SS-HRIS to receive the radar echo signal from a target. A joint base station (BS) beamforming and HRIS precoding matrix optimization problem is proposed to maximize the sum rate of communication users while maintaining satisfactory sensing performance measured by the Cramer-Rao bound (CRB) for estimating the direction of angles of arrival (AoA) of the echo signal and thermal noise at the target. The CRB expression is first derived and the sum rate maximization problem is formulated subject to communication and sensing performance constraints. To solve the complex non-convex optimization problem, deep deterministic policy gradient (DDPG)-based deep reinforcement learning (DRL) algorithm is proposed, where the reward function, the action space and the state space are modeled. Simulation results show that the proposed DDPG-based DRL algorithm converges well and achieves better performance than several baselines, such as the soft actor-critic (SAC), proximal policy optimization (PPO), greedy algorithm and random BS beamforming and HRIS precoding matrix schemes. Moreover, it demonstrates that adopting HRIS significantly enhances the achievable sum rate compared to passive RIS and random BS beamforming and HRIS precoding matrix schemes.