AI-Assisted Robust Design of Ku-Band Dual-Polarized AiP Antennas Considering Manufacturing Tolerance
Abstract
In this paper, a robust intelligent optimization method for Ku-band dual-polarized antenna-in-package (AiP) antennas is proposed. To address the high computational costs, extended design cycles, and sensitivity to fabrication errors in traditional full-wave electromagnetic optimization, an artificial neural network (ANN) surrogate model combined with particle swarm optimization (PSO) is employed for the rapid and robust dual-polarized antenna design. The ANN is employed to characterize the nonlinear relationship between the geometric parameters of the antenna and its scattering parameters (S-parameters). The ANN model is trained using datasets obtained from high frequency structure simulator (HFSS) simulations. Subsequently, fabrication tolerance robustness is introduced into the PSO optimization process. Monte Carlo analysis verifies that the proposed method enhances performance stability, achieving a manufacturing yield of over 90% under a ±2 mil fabrication error. Based on this validated framework, two dual-polarized patch antennas with different bandwidth characteristics are designed. Simulation results demonstrate that the optimized antennas achieve broadband impedance matching, port isolation exceeding 25dB, and high polarization purity with cross-polarization levels greater than 30dB.