A High-Quality Data Acquisition Method for Machine-Learning-Based Design and Analysis of Electromagnetic Structures

Abstract

Electromagnetic (EM) structures play a significant role in wireless communication, radar detection, medical imaging, and so on. Machine learning (ML) has been increasingly applied to facilitate the design and analysis of EM structures. Data acquisition is a major bottleneck. Conventional methods blindly sweep geometric parameters on a uniform grid and acquire corresponding responses via simulation. Acquired data have unstable quality due to inconsistent informativeness of responses, leading to a low ratio of model performance to data amount. This article proposes a high-quality data acquisition method to increase the ratio of model performance to data amount. It anticipates and generates high-quality data by analyzing the distribution of existing data iteratively. Comparative analysis of four implementations proves that the proposed method reduces the required data amount by around 40% for the same model performance and hence saves around 40% simulation and computing resources. The proposed method benefits ML applications of metasurfaces, antennas, and many other microwave structures.