Machine Learning-Based Design Scheme for Multifunctional Antenna Arrays With Reconfigurable Scattering Patterns

Abstract

A machine learning-based multifunctional antenna array design scheme (ML-MAADS), including a structure design module and a scattering beam steering module, is proposed for the multifunctional antenna array with reconfigurable scattering patterns in this work. The structure design module enables the rapid structure codesign of radiation and scattering for the reconfigurable array element, while the scattering beam steering module facilitates the near real-time beam steering for reconfigurable scattering patterns. To validate the ML-MAADS, a reconfigurable antenna element integrated with two positive-intrinsic-negative (p-i-n) diodes is designed using the structure design module, to construct a multifunctional antenna array. The antenna array serves three functions. First, it enables y-polarized ±60° radiation beam scanning, acting as a phased array; second, it exhibits low-scattering performance under x-polarized incident wave; third, it supports reconfigurable scattering pattern in the x-polarization, functioning as a programmable coding reflected metasurface. Benefiting from the structure design, the reconfigurable scattering characteristics do not affect the antenna’s radiation, thus enabling simultaneous operation in both radiation and scattering functions. Following the structure design, the scattering beam steering module is employed to achieve real-time beam steering (including single, dual, and vortex beams) for a reconfigurable scattering pattern, with the prediction accuracy exceeding 93%. The ML-MAADS provides designers with a promising comprehensive solution for complex multifunctional antenna synthesis in the future.