S11 Parameter Calculation of Frequency Selective Surface Based on Deep Learning

Abstract

Abstract This paper proposes the FSS-CNN network model as a forward predictor, replacing the function of the Maxwell equation solver of commercial software. The predictor is different from the numerical optimization method, and the data-driven method based on machine learning (ML) can be expressed and generalize complex functions or data to discover unknown relationships between a large number of variables. In the frequency range of 2∼18GHz, the S11 parameter prediction of the corresponding metal pixel pattern can be easily realized by an accurate forward neural network model. The MSE reaches the level of less than 0.1 and the time consumption is less than 0.07s, which meets the requirements of fast, efficient and automatic calculation.