Neural Network Model with Broadband Prior Knowledge Neurons for Microstrip T-junction Structure

Abstract

This paper presents a novel knowledge-based neural network model for microstrip T-junction structure. The generalized transmission-line equation from Maxwellian circuits theory can be utilized to extract distributed equivalent circuit parameters for a microstrip T-junction structure. Since the generalized transmission-line equation are determined by dynamical numerical methods, the transmission equation will be dynamical rather TEM. Thus, one hand the extracted circuit parameters from dynamical numerical solutions will be dynamical rather quasi-static, and on the other hand the extracted circuit parameters will have broadband characteristic. The broadband characteristics will be very useful to be regarded as the prior knowledge in the novel knowledge-based neural network model proposed in this paper. The novel model will have broadband characteristics than conventional knowledge-based neural network, such as NNKBN model. The novel KBNN model is electromagnetically developed with a set of data that are produced by the MoM method. Through numerical experiments, many advantages have been shown in the novel KBNN model over the conventional multi-layer perceptron model and NNKBN model.