Expedited Design Optimization of Antenna Input Characteristics Using Trust-Region Search with Adaptive Jacobian Updates

摘要

Reducing the number of electromagnetic (EM) simulations in the course of design closure procedures is a key to make them more practical for antenna designers. This applies to both direct and surrogate-assisted optimization algorithms. In either case, local methods such as gradient search are ultimately applied to find an improved design at either (low-or high-fidelity) level of antenna evaluation. In this paper, a simple yet robust technique to reduce the computational cost of gradient-based optimization process is proposed. Our methodology involves selective application of Jacobian updating formulas controlled by the alignment between the recent design relocation and coordinate system basis vectors, with the acceptance threshold made dependent on the algorithm convergence status. The algorithm is demonstrated to provide considerable speedup of up to 63 percent as compared to the reference trust-region-based procedure. The results are consistent across a three-antenna benchmark set with the robustness confirmed by means of statistics over multiple runs with random initial designs.