Graph Neural Network Enabled Resource Optimization for Tethered UAV Networks With Rotatable Directional Antennas

Rotatable directional antennas have recently emerged as a low-complexity alternative to conventional digital beamforming architectures for uncrewed aerial vehicle (UAV)-assisted wireless communication systems. In particular, tethered UAVs equipped with rotatable directional antennas can provide adaptive directional transmission while maintaining relatively stationary UAV locations, making them suitable for applications requiring stable aerial communication platforms and continuous backhaul connectivity. However, the joint optimization of antenna orientation, user association, and transmit-power allocation in multi-tethered-UAV communication systems is highly coupled and non-convex since the antenna directions directly influence the directional channel gains, which subsequently affect both the user-association and resource-allocation decisions. Moreover, conventional iterative optimization frameworks may incur high computational complexity and convergence delays in large-scale dynamic wireless networks. To address these challenges, this paper proposes a graph neural network based resource optimization framework for multi-tethered-UAV communication systems employing rotatable directional antennas. The considered UAV-user communication system is modeled as a bipartite graph, where the spatial interactions between UAVs and users are captured through iterative graph message passing operations. The proposed framework jointly learns antenna orientation, user association, and transmit-power allocation through end-to-end differentiable optimization. In addition, the corresponding differentiable training procedure and gradient propagation associated with the antenna-orientation variables, association coefficients, and power-allocation variables are derived. Simulation results demonstrate that the proposed framework outperforms fixed-antenna systems and proximal policy optimization based frameworks in terms of achievable sum rate and outage probability while maintaining efficient scalability with increasing numbers of users and UAVs.