Deep Learning Enabled Inverse Design and Optimization of the Frequency Selective Surface (FSS)

In this paper, a neural network model is proposed to design and optimize the frequency selective surface (FSS) structure. The training set and testing set of several classic structures are obtained by High Frequency Structure Simulator (HFSS) simulations. Using neural network, a targeted transmission coefficient curve (v. s. frequency) is firstly classified to select the most suitable type of classical structure as an initial template for design, and then further modification is carried out based on this template to match the targeted transmission coefficient in a more precise way. Subsequently, Bayesian optimization is used to adjust the hyper-parameters of the network. Through the proposed model, we can successfully design the FSS structural parameters corresponding to the required transmission coefficient $(S_{21})$. Compared to the traditional FSS design method using neural network, this proposed method greatly improves the accuracy of inverse design and can be used to design FSS structures corresponding to more types of S
21
curves.