Machine Learning-Based Inverse Design of Angle-Stable Frequency Selective Surfaces With Hexagonal Unit Cells

To design the frequency selective surface (FSS) with more degrees of freedom (DoFs) and incident angle stability, an improved machine learning-based inverse topological design model is proposed for the hexagonal FSS in this letter. In the proposed model, a novel modeling grid is employed to accurately describe the FSS unit cell with the discrete triangular pixel. This can significantly increase DoFs to explore more optimized solution space, and reduce the complexity and computational burden in the training process. To realize incident angle stability, the electromagnetic (EM) responses from six incident angles, from 0° to 75° with a 15° step, are incorporated into the design process. Two single-layer FSS numerical examples are provided to demonstrate the effectiveness of the proposed model, and the corresponding topological structures are fabricated and measured to validate the incident angle stability of the FSS structure predicted from the proposed model.