RIS-Aided AANETs: Security Maximization Relying on Unsupervised Projection-Based Neural Networks

The security aspects of aeronautical
ad-hoc
networks (AANET) relying on reflective intelligent surface (RIS) are considered. A projection-based deep neural network (DNN) is designed for maximizing the secrecy rate of the proposed RIS-aided AANET. While the multiple-layer architecture of the DNN enables learning the functional relationship between the target variables of the optimization problem and the ground-air channels, the projection method guarantees that the constraint of the optimization problem is not violated. Our design outperforms the state-of-the-art projected gradient descent algorithms and that the RIS is capable of enhancing the security.