Bayesian Optimization-Based Maximum Radiation Point Measurement System for Millimeter-Wave Array Antennas

In this study, a data-driven variable importance-based Bayesian optimization system was developed for the maximum radiation point measurement of UMFUS devices. To address the inefficiencies of conventional omnidirectional scanning, variable importance analysis through multiple methodologies based on empirical data and adaptive acquisition function transitions was applied. Performance verification using a ground truth dataset of 8,043 measurement points from a 4-axis positioner demonstrated that variable importance-based Bayesian optimization achieved 0.28 dB performance improvement and 52 % stability enhancement over pure Bayesian optimization. With 150 measurements achieving a target performance of 99.3 %, the system showed an efficiency improvement of approximately 30× over an exhaustive search. When combined with the early termination convergence detection system, the average convergence at 99 iterations achieved a 50.5 % time saving. The developed system provides a practical solution that significantly reduces measurement time and cost while satisfying international regulatory standards in RF certification environments.