A Dual-Path Generative Adversarial Network-based inverse design method for broadband RCS reduction metasurface element patterns

To address the problem of low efficiency in the traditional trial-and-error method, as well as provide a method for the inverse design of broadband radar-cross-section reduction (RCSR) coding metasurfaces elements patterns, the dual path network (DPN)-Generative adversarial network (GAN) is proposed. To enable the network to learn the complex mapping relationship between cross-polarized reflectance and anisotropic metasurface element patterns, the DPN-GAN loss function is improved by the fusion of Wasserstein distance and classification loss, so that the network can converge stably during the training process, and finally realize the inverse design of metasurface elements with wide operating band. The generator and discriminator consist of the proposed DPN models, thus enhancing the ability of the network to learn data features. Simulation results show that the mean square error (MSE) was 0.02584 of 300 sets of test data, which proves the high prediction accuracy of the network. Finally, the measurement results show that the broadband of RCSR less than -10 dB is 12.8-28.0 GHz for the coding metasurface sample, which proves the network can inverse design the element patterns for the coding metasurface with broadband RCSR performance.