AI-Driven Resource Allocation for Secrecy Rate Maximization in RIS-Assisted Cell-Free MIMO

In this paper, we develop a secure cell-free massive multiple-input multiple-output (CF-mMIMO) communication system assisted by a passive reconfigurable intelligent surface (RIS). In this model, multiple mobile unmanned aerial vehicle access points (UAV-APs) are tasked with communicating with multiple legitimate users (UEs), while multiple jammers are in charge of sending interference signals to UEs. This work proposes a unique framework to maximize the average sum secrecy rate (ASSR) by jointly optimizing the UAV’s trajectory, total transmit power at UAV-APs, and RIS phase shift. Ensemble global optimization (EGO) is implemented followed by deep learning (DL) techniques for optimization with low computational complexity. In particular, we train a deep neural network offline, to optimize the RIS phase shift, UAV-AP trajectory, and UAV-AP transmission powers online. A real-time optimization capability of the trained model is assessed online with low computational complexity. The numerical results show the performance analysis of the best neural network for real-time optimization. Furthermore, numerical results demonstrate that passive RIS technology and increased UAV-AP speed significantly enhance the performance of the cell-free communication system.