A Spatial Inverse Design Method (SIDM) Based on Machine Learning for Frequency-Selective-Surface (FSS) Structures

To efficiently and conveniently realize the design of frequency-selective-surface (FSS) structures with many degrees of freedoms (DoFs), a spatial inverse design method (SIDM) based on machine learning technology is proposed. The proposed SIDM takes advantages of the inverse modeling and topological design to spatially design for FSS. Different from simple parametric or topological modeling, which only involves one type of variable, i.e., binary or continuous variables, the proposed SIDM contains both binary and continuous variables to flexibly model FSS with less cost. Meanwhile, multilayer perceptron (MLP)-mixer is employed to capture the feature of both kinds of variables to realize the mapping relationship from the EM response to the corresponding FSS structure. Three numerical examples of single-layer FSS, FSS absorber, and multilayer FSS are employed to verify the effectiveness of the proposed SIDM. It indicates that compared with genetic algorithm (GA) with full-wave simulation and forward machine learning model, the proposed SIDM has higher accuracy and efficiency. Meanwhile, the fabricated topological structures are also measured to verify the performance of designed FSSs obtained from SIDM.