An Efficient Spherical Near-Field Sparse Sampling Method Based on Neural Network for Antenna Far-Field Prediction

A novel method for antenna far-field (FF) pattern reconstruction using sparse spherical near-field (SNF) measurements is presented. A conditional generative adversarial network (CGAN) is employed to significantly reduce the sampling density while maintaining high accuracy in FF prediction. The near-field (NF) data sampling is close to one-ninth of the Nyquist rate, converted to the initial FF patterns by fast Fourier transform (FFT). These initial patterns are utilized to train the neural network, which eventually reconstructs FF patterns closely approximating those obtained from full Nyquist-sampled measurements. This method avoids the use of superresolution networks, which can introduce artifacts and struggle with phase reconstruction in complex electromagnetic environments. The simulation and measurement results demonstrate substantial reduction in measurement time and data acquisition while preserving accuracy, offering an efficient solution for antenna characterizations.