Reinforcement Learning System of UAV for Antenna Beam Localization

Abstract

Along with the growth of satellite communication industry, the demands and benefits to perform satellite terminal antenna evaluation are increasing. UAV based in-situ measurement can increase the efficiency of the measurement procedure. Main beam localization is a necessary procedure to execute the antenna evaluation test. To accelerate the process of finding the antenna beam centre, this paper develop a meta-reinforcement learning based algorithm. The developed algorithm is compared with other methods and it showed the best performance in terms of accuracy, robustness and travelling efficiency not only in the simulated radiation pattern environment but also in the empirically obtained radiation pattern.

Autonomous Drone-Based Antenna Radiation Pattern Characterization

“…Overall, it can be concluded that the EI acquisition function with restricted sampling is the most optimal configuration for peak detection with respect to lowest localization error using Bayesian optimization. Recent novel approach is proposed in [9], where the authors used recurrent neural networks for antenna man beam localization. However, this approach requires training the algorithm based on a specific antenna's radiation pattern and does not provide a more generic solution to this problem.…”