Two-Dimensional Gaussian Process Regression Interpolation for Spherical Near-Field Measurements

This study presents a novel electromagnetic spherical near-field (NF) interpolation method based on Gaussian Process Regression (GPR) to address challenges associated with under-sampling in NF data acquisition. The proposed approach mitigates errors in far-field (FF) radiation pattern predictions caused by insufficient NF sampling. The NF-to-FF transformation employs spherical wave expansion with double numerical integration. A seven-patch array antenna was simulated in CST Microwave Studio, with NF sampling angles exceeding the limits of the spherical sampling theorem, resulting in a 25.65% under-sampling rate. Results demonstrate that GPR effectively reconstructs NF data, achieving FF radiation patterns closely matching those obtained directly from CST Microwave Studio. The absolute error is reduced to under 10% in the ZY plane and approximately 3% in the ZX plane. By enabling accurate FF radiation pattern predictions and reducing NF data acquisition time, the GPR-based method offers a robust and efficient solution for spherical NF measurements.