U-Net Based Near-to-Far-Field Transformation Under Limited Near-Field Measurement Area

초록

Accurate far-field radiation pattern prediction is essential for evaluating large antennas used in satellite communication, radar systems, and mobile networks, where performance verification under realistic deployment conditions is critical. However, direct far-field measurement is often impractical due to the large required distances and spatial constraints of measurement environments. Conventional NF2FF transformations demand dense and extensive near-field sampling, significantly increasing both cost and measurement time. In this study, we propose a Residual U-Net-based deep learning model that directly predicts the 2D far-field pattern from complex-valued near-field measurements. The proposed model demonstrates robust and consistent performance even under limited measurement areas, offering a computationally efficient and practical alternative to traditional physics-based NF2FF methods. © 2025 IEEE.

키워드