Joint UAV 3D Trajectory and Resource Allocation for Integrated LEO Satellite and Multi-UAV-Enabled Marine IoT Networks: A Federated Multi-Agent Deep Reinforcement Learning Approach

The marine IoT (MIoT) has experienced widespread adoption in the maritime industry; however, its communication infrastructure faces significant challenges due to a lack of terrestrial networks and the dynamic nature of oceanic environments. In response, low-Earth orbit (LEO) satellites have been suggested as an alternative solution. However, LEO satellites also face limitations, including coverage gaps due to the power constraints of MIoT devices (MIoTDs) and the curvature of the Earth, which impede continuous connectivity. To address these challenges, integrating uncrewed aerial vehicles (UAVs) as aerial base stations between LEO satellites and MIoTDs has gained attention as an effective solution to enhance communication. Managing UAV mobility and ensuring reliable communication requires autonomous 3-D trajectory optimization and dynamic resource allocation. This article presents an integrated LEO and multi-UAV-enabled MIoT network designed to maximize communication coverage, system throughput, and fairness. To achieve this, we formulate a joint UAV trajectories and resource allocation (UTCRA) optimization problem. To address the UTCRA problem, we propose a two-stage learning approach: 1) Gaussian mixture models (GMMs)-based clustering is initially applied to dynamically cluster MIoTDs under different UAVs, ensuring a balanced distribution of user loads among UAVs. 2) A federated multiagent deep deterministic policy gradient (FL-MADDPG) algorithm is employed to continuously optimize UTCRA, thereby enhancing communication efficiency. Simulation results demonstrate that FL-MADDPG with GMM significantly enhances communication coverage, average throughput, and fairness. It outperforms benchmark algorithms across all evaluated metrics, highlighting its strong potential as a method for next-generation MIoT networks.