Physics-Aware Adaptive Bayesian Reconstructor for mmWave Sparse Planar Near-Field Measurement

Planar near-field measurement efficiency is often compromised in position-agnostic scenarios, such as testing encapsulated devices, where traditional sparse sampling methods may fail to locate the unknown effective aperture. In this context, this letter proposes a physics-aware Bayesian sparse measurement-and-reconstruction framework that jointly addresses adaptive acquisition and complex near field recovery. By integrating a physical back-projection mechanism with a data-driven active learning strategy, the proposed scheme achieves autonomous region-of-interest (ROI) focusing and high-fidelity reconstruction of the effective aperture. Specifically, a sparse set of pilot samples is first processed via virtual back-projection to extract a ROI weight map, effectively focusing on the physical source. Subsequently, an ROI-weighted Gaussian process model is constructed to adaptively guide the sampling agents toward high-energy and high-uncertainty regions and to enable accurate recovery of the full complex near field. Simulations of an offset 60-GHz 8×8 patch array and measurements of a 29-GHz 4×4 magneto-electric dipole array demonstrate that the proposed method achieves a sampling reduction of over 80% compared to the standard Nyquist sampling benchmark while maintaining comparable far-field reconstruction accuracy, validating its potential for accelerating measurements in position-agnostic testing scenarios.