Enhancing Deep Learning-Based Antenna Array Coefficient Reconstruction through Optimized Data Management

Abstract

Complex array coefficients are essential parameters for designing planar antennas with desired radiation patterns. However, reconstructing these coefficients from nearfield phaseless measurements is a challenging inverse problem that requires a large amount of data and a robust optimization algorithm. In this paper, we propose a novel method that combines an optimal non-uniform sampling strategy and a U-Net model, a convolutional neural network (CNN) that has shown high accuracy in image segmentation, to achieve this purpose. In this scheme, the most informative measurements are collected and discards the redundant ones, reducing the data size and enhancing the data quality. Furthermore, we exploit a U-Net model to learn the mapping from the measurements to the array coefficients, avoiding the difficulties of solving a non-linear inverse problem. We demonstrate the effectiveness of our method through numerical simulations, showing that our method can accurately reconstruct the array coefficients with a reduced number of measurements and outperform existing classic methods.