Presentation Information
[B-5B-09]Radio Propagation Feature Aware High-Dimensional Bayesian Optimization for Base Station Planning with Combinatorial Structure
◎△Koki Kanzaki1, Koya Sato1 (1. The University of Electro-Communications)
Keywords:
Bayesian optimization,Gaussian process regression
Base station and antenna design that accounts for environment-dependent radio propagation is essential for improving wireless communication quality. However, jointly optimizing base station selection and antenna parameters leads to a high-dimensional problem with a combinatorial structure, making conventional Bayesian optimization difficult to apply efficiently. We propose a high-dimensional Bayesian optimization framework that exploits radio propagation features obtained from ray tracing. The proposed method maps the original design variables, including base station selection and antenna parameters, into a lower-dimensional continuous feature space constructed from candidate-site locations and propagation-aware features. Bayesian optimization with a Gaussian process surrogate model is then performed in this feature space to maximize the coverage area satisfying a target throughput. Numerical evaluations using NVIDIA Sionna RT demonstrate that the proposed method outperforms some base line methods such as genetic algorithms. The results confirm that proposed method improves the efficiency and performance of base station planning optimization.