Deep Learning-Enabled Secrecy Performance Analysis of UAV-Aided Reconfigurable Intelligent Surfaces With Non-Orthogonal Multiple Access

Abstract

This paper aims to examine the secrecy performance of a two-user non-orthogonal multiple access network consisting of a reflective intelligent surface (RIS) aided by an unmanned aerial vehicle to assist the weaker user in the presence of an eavesdropper. Also, a jammer whose signal is identifiable to the authorized users is employed to weaken the eavesdropper's attempts to intercept the signal intended to the weaker user. Analytical expressions of the secrecy outage probability (SOP) are derived to assess the secrecy performance of the underlying system. Furthermore, an asymptotic SOP analysis is performed to determine the influence of critical parameters on the overall system performance. Subsequently, the impact of using different signal strengths, power allocation coefficients, and a number of RIS elements on the system performance is investigated. However, it is noted that the analytical framework adopted in this work to obtain the above SOP expressions remains complex and challenging, especially in such dynamic environments. To tackle this problem, a deep neural network model is proposed for SOP prediction with fast execution in such dynamic environments. Monte Carlo simulations are provided to verify the tightness of the derived mathematical formulations. In addition, numerical results reveal the efficacy of the proposed deep learning framework in predicting the SOP performance. © 2025 IEEE.