Presentation Information
[ABT-1-02]High-Dimensional Bayesian Optimization for Automated Base Station Planning
〇Koya Sato1 (1. The University of Electro-Communications)
Keywords:
Bayesian optimization,Base station planning,Ray tracing,Radio propagation
To ensure high-speed and highly reliable wireless communications, we must carefully design system parameters such as base station placement and transmit power. However, before deployment, it is difficult to formulate a closed-form objective function for communication quality that accounts for complex radio propagation. We therefore need efficient parameter search methods based on repeated simulations, especially when high-fidelity simulations such as ray tracing are computationally expensive.
In this talk, we discuss Bayesian optimization (BO) and its application to wireless base station design. BO efficiently searches for optimal inputs under a limited evaluation budget by modeling the relationship between design parameters and the objective function, for example using Gaussian process regression (GPR), and sequentially selecting promising parameters.
Since radio propagation loss can be modeled as a Gaussian process over space, GPR can be used to model objectives such as area-averaged throughput. This enables efficient base station design using BO. We also discuss the need to reduce input dimensionality and exploit problem structure, because BO generally becomes less efficient as the number of design variables increases.
In this talk, we discuss Bayesian optimization (BO) and its application to wireless base station design. BO efficiently searches for optimal inputs under a limited evaluation budget by modeling the relationship between design parameters and the objective function, for example using Gaussian process regression (GPR), and sequentially selecting promising parameters.
Since radio propagation loss can be modeled as a Gaussian process over space, GPR can be used to model objectives such as area-averaged throughput. This enables efficient base station design using BO. We also discuss the need to reduce input dimensionality and exploit problem structure, because BO generally becomes less efficient as the number of design variables increases.