Efficient Electromagnetic Near-Field Scanning Using Physics-Informed Gaussian Process Regression

This paper proposes a novel approach combining prior physics-based Gaussian Process Regression (GPR) with Bayesian Optimization for efficient and accurate electromagnetic near-field scanning. By considering the measurement setup and properties of the Device Under Test (DUT), the proposed approach enables fast and accurate construction of the radiated fields’ surrogate models. The GPR kernel parameters are efficiently obtained independently of measured data-samples by fitting on the radiation pattern of suitable infinitesimal dipoles, thus ensuring consistency with physical behavior. Integrating this approach with Bayesian Optimization for adaptive sampling demonstrates improved accuracy and computational efficiency compared to traditional methods, while maintaining low overhead. This framework offers a powerful tool for efficient and accurate near-field scanning, potentially reducing the time and cost associated with the scanning process.