Cognition-Driven Optimization Technique for Broadband Metasurface Antenna Based on Feature-Assisted Neural Network Surrogate

This letter proposes a cognition-driven optimization technique based on feature-assisted artificial neural network (ANN) surrogate for broadband metasurface antenna design. In this work, we introduce the cognition-driven optimization technique into antenna design. The derivatives of gain with respect to elevation and azimuth angle in the neighborhood of the desired radiation direction are served as feature parameters to enhance the antenna's directivity. The proposed feature-assisted surrogate model comprises the S-parameter-based ANN model, the Gain-based ANN model and the Feature-Derivative-based ANN model. With the assistance of feature parameters, the technique achieves superior metasurface antenna performance by simultaneously optimizing S-parameter and gain. Additionally, a trust region algorithm is incorporated to accelerate optimization convergence. Compared to the ANN-based method without feature assistance and the trust region framework algorithm of CST Studio Suite, the proposed technique obtains the optimal solution with fewer EM simulations. For demonstration, a broadband metasurface antenna is optimized, fabricated and measured. The close agreement between measurement and simulation results validates the effectiveness of the proposed optimization technique.