Design of low-RCS metasurface with out-of-band transmissionvia deep learning

To address the pressing need for out‑of‑band high transmission performance and radar cross section (RCS) reduction in antenna‑radome systems for stealth communication, this study proposes a deep‑learning‑based multifunctional metasurface design approach. The proposed structure combines a pixel‑type metasurface with a band‑pass frequency selective surface (FSS) to achieve simultaneous RCS reduction and narrow‑band high‑efficiency transmission. For the efficient design of a high‑degree‑of‑freedom pixel‑type metasurface that satisfies the beam‑scattering mechanism, a self‑attention conditional Wasserstein generative adversarial network (SA‑C‑WGAN) was constructed and integrated with a forward‑prediction network based on ResNet18. This framework enabled rapid and accurate generation of 16 × 16 pixelated topologies along with phase‑response prediction across the 1–20 GHz band. Based on this method, a metasurface array exhibiting low RCS in the 12.5–18.5 GHz band and high transmission in the 6.5–7 GHz band was designed and fabricated. Experimental results show that the prototype achieves a transmission rate better than –1 dB within 6.56–7.81 GHz and an RCS reduction exceeding 10 dB within 12.34–19.25 GHz, verifying the effectiveness of the proposed design method