Efficient Beam Tracking for Resource-Constrained mmWave UAV Communications
Abstract
Intelligent beam tracking is one of the key techniques of millimeter wave (mmWave) unmanned aerial vehicle (UAV) communications. However, UAVs typically face severe constraints in power supply and computational resources, hindering the deployment of complex neural networks on them. To address this issue, an extremely lightweight contextual bandit-based beam tracking algorithm is presented in this letter. The proposed algorithm exploits the independence of the UAV’s flight behavior from its position and heading direction, thereby compressing the neural network to just a few hundred neurons. Simulation results show that the proposed algorithm can achieve $100~\mu s$ -level inference and second-level convergence on a cost-sensitive Cortex-A7 processor while maintaining high performance.