Design of antipodal vivaldi antennas using kernel regression optimization

Abstract

This paper proposes the design of antipodal Vivaldi antennas using the kernel regression method. The kernel regression is applied for training a cost function model to predict the next sample with improved cost values, and the information of the predicted sample is employed to re-train the model. This process is repeated until the cost value converges to our design goal. The shapes of the tapered slot line and the extended ground are determined by adjusting coefficients of multi-Gaussian functions. The optimized antenna has an overall dimensions of 88.1 × 120 mm2 and shows an average reflection coefficient of -11.1 dB in the frequency band.