Inverse GRIN Lens Design Using Artificial Neural Network and Geometrical Optics

This paper proposes an efficient artificial neural network for inverse design of GRIN lens antennas. A fast ray tracing method based on Geometrical Optics (GO) is implemented to numerically generate sufficient datasets for training. A simple artificial neural network is built for inverse design. A demonstration shows such a network can realize good mean absolute percentage error (MAPE) between the desired and predicted geometrical variables. The training and testing MAPE are 3.42% and 4.38%, respectively. Radiation patterns of a test lens and its prediction are presented to show the GO method and network accuracy.