UAV-Assisted Wireless-Powered Secure Communications: Integration of Optimization and Deep Learning

This paper presents a novel framework that combines an optimization-based approach with a deep learning (DL)-based approach to devise a cooperative strategy for unmanned aerial vehicles (UAVs) in wireless-based secure communications by leveraging the strengths of both approaches and addressing their respective limitations. We first formulate a joint optimization problem to maximize the minimum achievable secrecy rate while guaranteeing the minimum harvested energy requirement for each ground node by optimizing scheduling, transmit power, and trajectory. To address the difficulty of solving the formulated non-convex mixed-integer nonlinear programming problem, optimization techniques, such as continuous convex approximation and the block coordinate descent algorithm, are used to efficiently find feasible solutions. To tackle the challenges posed by the high computational complexity and initialization sensitivity of the optimization-based approach, we also propose an unsupervised learning-based deep neural network (DNN) structure with a specialized loss function tailored to our goals that allows the DNN to effectively approximate the optimal strategies for UAVs. Finally, we design a pioneering method that integrates the strengths of the two aforementioned approaches, in which the output of the trained DNN serves as the initial values of the optimization variables, and subsequently, optimization techniques are applied to fine-tune these optimization variables, leading to further performance improvements. Through intensive simulations, we confirm that the integrated scheme provides superior performance without constraint violation compared to the DL-based scheme, while guaranteeing faster convergence than the optimization-based scheme.