An Efficient Trust-Region Algorithm for Wideband Antenna Optimization

Simulation-driven design closure is a necessary step in the design of modern antenna structures. It is typically realized using parameter sweeping. Application of numerical optimization routines brings better results but it may be computationally expensive due to numerous electromagnetic (EM) analyses involved in the process. Reduction of the optimization cost can be achieved using various methods, e.g., surrogate-assisted techniques, reduction of the problem dimensionality, etc. Yet, either the main task or its sub-problems (e.g., surrogate model optimization) have to be solved directly. Here, an improved trust-region algorithm working with numerical derivatives is proposed that permits considerable reduction of the number of EM simulations in the optimization process. Our approach utilizes a smart management of antenna response Jacobian updates derived from relative changes of geometry parameters during subsequent iterations. Our methodology is demonstrated using a benchmark set of wideband antennas. Comparison with a conventional trust-region routine is provided along with a comprehensive analysis of algorithm performance involving multiple starting points. The impact of control parameters is also investigated. The average computational savings across the benchmark set are around 50 percent with acceptable degradation of the solution quality.