Data-Driven Learning Based Secrecy Rate Optimization for RIS-Assisted Secure Wireless Networks: A Multi-Armed Bandit Approach

In this paper, the secrecy rate optimization problem for Reconfigurable Intelligent Surface (RIS) assisted secure wireless networks is investigated. However, to fully stimulate the potential of RIS, a novel algorithm is needed to optimize the security of RIS-assisted wireless networks. Specifically, a secrecy optimization for multiple RIS-assisted wireless networks with uncertain environments as well as eavesdropping is given first. Then, to solve the formulated optimal secrecy problem, the Multi-Armed Bandit (MAB) has been adopted and further integrated with a deep reinforcement learning algorithm along with Deep Deterministic Policy Gradients (DDPG) technique. This developed algorithm can effectively enhance the secrecy of RIS-assisted wireless networks, i.e. minimizing secrecy outage probability, etc., not only through selecting appropriate RIS, but also by optimizing transmit power and phase shift matrix in selected RIS. Eventually, the effectiveness of the developed algorithm has been demonstrated through a series of numerical simulations.