DRL based joint A-PBF optimization with the scattered NLOS paths existence in RIS assisted MU-MISO systems

Reconfigurable Intelligent Surface (RIS) has emerged as one of the most influential technologies to support all the ambitious goals of 6G networks such as ultra-high data rates, energy efficient networks, seamless and wider coverage. Moreover, the RIS assisted network demonstrated its capability to address attenuation and signal blocking issues. However, the optimization of the active passive beamforming (A-PBF) process is still an open challenge, so in this paper, the joint optimization of transmit beamforming and RIS phase shifts with the integration of the scattered non line of sight (NLOS) paths besides the RIS-reflected paths is investigated. This assumption, unlike former works which consider the existence of reflected RIS links only, allows better resource usage and significantly more improvements in spectral and energy efficiency. Moreover, deep reinforcement learning framework is utilized to address the A-PBF optimization problem, leveraging its effective ability to deal with time varying wireless environments and NP-hard problems. In which, Deep Deterministic Policy Gradient (DDPG) model is employed for effectively addressing high-dimensional and continuous optimization challenge. Through the simulation results, the proposed approach proves its effectiveness comparable to traditional methods as it requires lower complexity, though it guarantees higher system performance. Furthermore, the results prove that the proposed approach increased the sum rate by up to 30% compared to the case of considering the existence of reflected RIS links only.