A Data-Driven Pattern-to-Code Mapping for Digital Metasurface

This paper presents a method for mapping pattern masks to coding sequences using a convolutional neural network (CNN). To assess the feasibility of this approach, we generated a dataset based on a numerically simulated 15 × 15, 2-bit spatially coded digital metasurface operating at 100 GHz. The CNN model was trained to predict the coding sequence required to produce a specific pattern mask. The mask was generated by assigning a value of ‘1’ if the radiation exceeded one-tenth of the maximum value, and ‘0’ otherwise. The model performance was evaluated by the error, defined as the difference in magnitude between the actual and predicted patterns.