Reconfigurable reflective metasurface reinforced by optimizing mutual coupling based on a deep neural network

Abstract

Mutual coupling between different states of active unit cells is a challenging factor that has not been considered in the design of reconfigurable transmissive or reflective metasurface antennas. In this work, we propose a gain-predicting deep neural network (GPDNN) that predicts the radiation patterns of a reconfigurable reflective metasurface (RRM) composed of a 12-by-12 one-bit active unit cell array and is used to search for the best combination of unit cell on-or-off states for beam forming. First, the GPDNN is trained to accurately predict the radiation pattern based on the combination of unit cells. Second, it is merged with a search algorithm that retrieves the best on-or-off states near the boundary of the two states determined using the conventional beam-forming calculation method. As proof of concept, the proposed scheme is employed to find the highest realized gain in five directions: (θ, φ) = (0°, 0°), (−60°, 0°), (60°, 0°), (−60°, 90°), and (60°, 90°). The proposed deep neural network–based search algorithm takes 3.27 × 10−7 seconds per design, which is considerably faster than that based on full-wave simulation (1.5 h per design). The accuracy of the proposed method is verified by comparing the predicted results with those of the full-wave simulation. Finally, the best combination of on-or-off states for each beam-forming case is experimentally verified by measuring the radiation pattern. Compared with the conventional design, the maximum gain increases up to 0.771 dB at (θ, φ) = (−60°, 0°), and the side lobe levels decrease substantially in the other cases.