Global Optimization of a Dual-Band Coaxial Feed using Multidimensional Piecewise Regression-Kriging Interpolation

A new approach for global optimization is presented with application in antenna and microwave circuit design. This algorithm, implemented in the Branch & Bound Interpolating Surrogate Global Optimizer (B2ISGO), constructs piecewise multilinear-Kriging response surfaces that mimic basic electromagnetic responses, and samples a multi-variable design space using a branch and bound approach that performs simultaneous exploitation of local minima and exploration of promising designs. Optimization results are shown pertaining to two components of a dual-band coaxial antenna feed, namely the coaxial radiating aperture and a rectangular to coaxial waveguide transition. Convergence results show that B2ISGO compares favourably against Tomlab´s glbfast global optimizer, an implementation of Jones´ Direct algorithm.