Reinforcement Learning based Optimal Dynamic Resource Allocation for RIS-aided MIMO Wireless Network with Hardware Limitations

This paper studies the optimal dynamic resource allocation problem for a Reconfigurable Intelligent Surface (RIS) aided MIMO wireless network with multi-user under uncertain time-varying wireless channels. Recently, RIS has been considered one of the most promising techniques to upgrade dynamic wireless network quality. However, the capacity of RIS has been restricted due to RIS hardware limitations and uncertainties from time-varying wireless channels. Therefore, a novel dynamic resource allocation technique needs to be developed that cannot only optimize the overall network quality, e.g. maximizing energy efficiency, minimizing power consumption, etc., but also consider the RIS hardware limitations and the uncertainty from the time-varying wireless channels. In this paper, a novel online data-enabled actor-critic-barrier reinforcement learning algorithm is developed and utilized along with neural networks (NNs) to learn the optimal transmit power control, RIS phase shift control policies under hardware limitations and wireless channel uncertainties. Eventually, numerical simulations are provided to demonstrate the effectiveness of the developed scheme.