Accelerated Parameter Tuning of Antenna Structures by Means of Response Features and Principal Directions

The popularity of numerical optimization has been steady on the rise in the design of modern antenna systems. Resorting to mathematical rigorous parameter tuning methods is a matter of practical necessity as interactive techniques (e.g., parameter sweeping) that are no longer adequate when handling several performance figures over multidimensional parameter spaces. The most common design scenarios involve local tuning since decent initial designs are often rendered at the early phases of the design process. Notwithstanding, antenna optimization is usually executed using full-wave electromagnetic (EM) simulation tools, inevitably requiring a considerable amount of computational resources. This article introduces a novel technique for expedited gradient-based antenna optimization with numerical derivatives. The two major acceleration mechanisms, both of which exploit the problem-specific knowledge carried by the antenna characteristics, include the response feature methodology and sparse sensitivity updates restricted to selected principal vectors. The former permits flattening the landscape of the objective function, whereas the latter effectively reduces the problem dimensionality to dimensions that have the most significant effect on antenna characteristics, which are established using an automated decision-making procedure. Our methodology is applied to optimize three antenna structures. Comparisons with several benchmark procedures indicate that the relevance of the adopted mechanisms and considerable computational savings that can be achieved. The average savings amount to 50% over conventional trust-region (TR) gradient search, and slightly lower, yet still significant, over accelerated versions thereof.