Microwave Passive Components Modeling Using Neural Network Techniques

Parametric models for Microwave passive components using neural networks techniques have been developed. Two novel techniques by combining neural networks techniques and rational-functions are proposed in this paper. One is the circuit based neural network; another is the knowledge based neural network. Both techniques utilize the space mapping technique and vector fitting algorithm. The neural networks are used to learn the relationship between geometrical parameters of passive components and coefficients (pole-residues) of rational function. In the circuit based neural network, a SPICE compatible equivalent circuit is combined to represent the behavior of passive components. In the knowledge based neural network, vector fitting algorithm is used to obtain pole-residues of rational function. Knowledge based neural network module is used to represent the Y-parameters of Microwave passive components. The trained model can represent an accurate, fast and passive behavior of Microwave components. Also, the model developed can be used to do optimization. Compared to EM models, the model developed can be much faster. Compared to vector fitting algorithm, the proposed technique can generate a parametric model and the model can be used to do optimization.