Application of Artificial Intelligence in Modeling Polarization Convertor Unit-cell

In this letter, a data-driven learning algorithm is presented to reduce the highly time-consuming and computationally expensive iterative progression of design using full-wave (FW)simulators. It approximates the nonlinear relationship between the electromagnetic design and associated desired performance. A polarization convertor unit-cell with n number of diagonal dipoles as a patch is considered and the effect of geometrical parameters on the required performance and frequency range is investigated using convolutional neural network (CNN). The developed CNN has the capability of fast prediction of geometrical parameters for any user-defined frequency ranges between 10 GHz to 50 GHz of desired characteristics. As a proof-of-concept, FW simulation results are presented to demonstrate the efficiency of the proposed CNN.