A Simulation-Based AI-Assisted Framework for Beam Steering in Reconfigurable Metasurface Antennas

Millimeter-wave communication systems require highly directive and dynamically steerable antennas to mitigate the severe propagation losses and blockage effects. Reconfigurable metasurface antennas have emerged as a promising solution for adaptive beam control. However, conventional beam synthesis methods rely on computationally intensive full-wave electromagnetic optimization or exhaustive parameter sweeps, limiting their suitability for rapid reconfiguration and large configuration spaces. This paper presents a simulation-based artificial intelligence-assisted framework for efficient beam steering in reconfigurable metasurface antennas. The proposed approach formulates beam synthesis as an inverse control problem, in which a supervised learning model is trained on full-wave electromagnetic simulation data to directly infer metasurface configuration states corresponding to the desired beam directions. Once trained, the model enabled fast and direct prediction without repeated electromagnetic optimization. The numerical results demonstrate accurate beam steering over a ±60° angular range, achieving mean beam-pointing errors below 2.0°, while maintaining stable gain and controlled sidelobe levels across multiple operating frequencies. Furthermore, the proposed framework achieves a reduction of more than two orders of magnitude in the online inference latency compared with conventional full-wave optimization. These results demonstrate the effectiveness of learning-assisted inverse synthesis for reducing the computational burden of metasurface beam control and establishing a scalable, simulation-driven foundation for future intelligent and experimentally realizable electromagnetic surface systems.