Near-Field Phase Reconstruction and Far-Field Prediction with Phaseless Spherical Data Using Neural Networks

To address the challenge of accurately obtaining phase information in near-field antenna measurements, this paper proposes a method for near-field phase recovery and farfield prediction utilizing artificial neural networks. The phaseless near-field data is obtained from single spherical near-field scans of antennas. Phase information on the sampled nearfield sphere is reconstructed from the amplitude-only data by a trained convolutional neural network (CNN). Subsequently, the least squares method is employed to construct equivalent dipole sources according to the spherical equivalent dipole array method, ultimately enabling far-field calculations. Compared to existing approaches of using optimization algorithms for searching equivalent dipole sources or near-field phase distribution, the neural network-based method offers superior online efficiency and improved prediction accuracy.