Research Article
Secrecy Capacity Maximization in UAV-Assisted RSMA Systems with Mobile Users Using Multi-agent Deep Reinforcement Learning
@INPROCEEDINGS{10.1007/978-3-032-44094-5_10, author={Zhaoyi Feng and Shengpeng Liu and Chunjia Tang and Ali A. Nasir and Zhichao Sheng and Xue-Xia Yang}, title={Secrecy Capacity Maximization in UAV-Assisted RSMA Systems with Mobile Users Using Multi-agent Deep Reinforcement Learning}, proceedings={Industrial Networks and Intelligent Systems. 12th EAI International Conference, INISCOM 2026, Da Nang, Vietnam, February 26--27, 2026, Proceedings}, proceedings_a={INISCOM}, year={2026}, month={11}, keywords={Multi-Agent DRL UAV Rate-Splitting Multiple Access Secrecy Capacity Social Force Model}, doi={10.1007/978-3-032-44094-5_10} }- Zhaoyi Feng
Shengpeng Liu
Chunjia Tang
Ali A. Nasir
Zhichao Sheng
Xue-Xia Yang
Year: 2026
Secrecy Capacity Maximization in UAV-Assisted RSMA Systems with Mobile Users Using Multi-agent Deep Reinforcement Learning
INISCOM
Springer
DOI: 10.1007/978-3-032-44094-5_10
Abstract
Unmanned aerial vehicles (UAVs) are pivotal in shaping the vision of sixth-generation mobile communication due to their exceptional flexibility and mobility. Rate-splitting multiple access (RSMA) stands out as a highly efficient technology among the next generation multiple access technologies. In this treatise, we investigate a multi-UAV-assisted RSMA secure transmission system. Meanwhile, considering the influence of obstacles, user mobility adheres to the social force model. The problem of joint optimization of multi-UAV trajectory, common secrecy capacity vector, and precoding matrix is formulated to maximize the secrecy capacity. To effectively tackle this nonconvex issue, we propose an action embedding integrated multi-agent twin delayed deep deterministic policy gradient (AcE-MATD3) method. This method incorporates an action embedding approach based on policy decomposition to overcome the limitation of the tanh activation function in directly producing constrained actions. Simulation results verified the superiority of our proposed RSMA enhanced multi-UAV scheme over multi-agent deep deterministic policy gradient based on action embedding (AcE-MADDPG).