Fast Filter‐Antenna Design Method Accelerated With Physical‐Inspired AI‐Driven Forward Solver

ABSTRACT To meet various customised requirements for RF devices efficiently, a fast artificial intelligence (AI)‐driven design method is proposed using a physically inspired forward solver with coupling matrix via the artificial neural network (ANN). Establishing the space mapping from the coarse model to the fine model through ANN enables the rapid and accurate acquisition of a large number of S‐parameters of the filter under different geometric parameters. To improve the construction efficiency and performance of the forward solver, the mutual coupling matrix of the resonator ( M matrix) is extracted from the S‐parameters obtained as input of the surrogated model, which exhibits better robustness, convergence and training efficiency. Furthermore, by integrating with the AI‐driven forward solver, the performance of the RF devices could be efficiently optimised according to the requirements with the heuristic algorithms. Simulation, comparison and experimental results all demonstrate that the designed RF filters and filter antennas accordingly exhibit high efficiency and excellent prediction accuracy at various operating frequencies and bandwidths, thus proving the effectiveness and practicality of the proposed optimisation method.