Deep-Learning-based Inverse Modelling with CMA-ES as Applied to the Design of A Wideband High-isolation Septum Polarizer

This paper presents a design method for electromagnetic (EM) devices based on inverse modelling using deep neural networks (DNNs) and evolutionary algorithms, which is suitable for generating new designs for complex EM devices with many design parameters and multiple objectives at a reduced computational cost. This method has been successfully applied to the design of a W-band septum polarizer, achieving a high isolation> 40 dB over 15.8% bandwidth.