Incident-Angle-Insensitive Dual-Polarized Metacell Using Machine-Learning-Based Optimization

In this paper, an incident-angle-insensitive, dual-polarization metacell is proposed using a machine-Iearning-based optimization method. The metacell is based on a three-layer dual-polarized metacell with two substrates vertically and symmetrically integrated, which enhances transmission performance across a wide range of incident angles for both TE and TM waves. A seven-layer artificial neural network (ANN) model is employed to optimize the metacell dimensions and achieve the desired performance. The input data for the ANN consists of the target transmission responses and the incident angle. The neural network predicts the geometric parameters that meet the specified performance criteria. As a result, a dual-polarized metacell is optimized to exhibit incident-angle insensitivity up to 60° for both TE and TM incident waves.