Filter approximation by RBF-NN and segmentation method

In this paper, radial basis function neural networks (RBF-NN) are used to approximate microwave filters. The method used is based on the segmentation of the structure in small units and the approximation of their scattering parameters. The approximated transmission matrices of each unit are multiplied to reproduce the whole filter response. This method is applied to a genetic algorithm in order to obtain a 13 sections microwave step filter, with a specified response. The RBF-NN response of the resulting filter is compared with its full-wave method of moments analysis showing a considerable save of computation time and increased accuracy in the results.