Surrogate Based Optimization of Wideband Reflector Feed Antennas

This paper presents a surrogate based optimization strategy for the rapid design of reflector antenna feeds. A multi-fidelity optimization scheme is employed, where the fast coarse model of the aperture efficiency is based on an interpolant of the Characteristic Basis Function Pattern (CBFP) coefficients. The training points of the interpolation model are selected by an expected improvement sequential design strategy, with the minimization goal function in this case, the negative aperture efficiency, penalized by the probability that the reflection coefficient magnitude exceeds a pre-defined limit. A Kriging model of the reflection coefficient magnitude is constructed from all available training data, which includes statistical information of the apparent model confidence over the full design space. An example of a quad-ridge flared horn design is presented, highlighting the efficiency of the suggested method.