Analysis Of Electromagnetic Stop Band Structures In Coplanar Waveguide Using Artificial Neural Networks

Rapid development of microwave techniques leads to emergence of a series of new circuits and devices. Electromagnetic stop band structure is one of these structures. Due to excellent stop band and slow wave characteristics, they are applied to filters, resonators, directional couplers, diplexers and antennas.Traditional analysis methods are full wave analysis methods, such as method of moment, finite element method. These methods give very accurate results. However, they are too time-consuming to be applied to fast analysis and design. Meanwhile, artificial neural network method attracts more and more of our interest. And its application in microwave circuit design has been broadly discussed. Based on former researchers' works, artificial neural network method is introduced to the analysis of electromagnetic stop band structure to fulfill fast analysis in this dissertation.Three types of artificial neural networks, namely radial basis function network, feedforward network, stochastic network, and their application in the analysis of electromagnetic stop band structure in coplanar waveguide have been discussed.Radial basis function network constructs many basis functions, according to the distances between the unknown style and the known styles, to generate the function value of the unknown style. In this dissertation, characteristic impedance and effective permittivity of two well-known electromagnetic stop band structures have been fitted by this method.Feedforward networks use back propagation algorithm to train a multi-layer network. After training, the multi-layer network can fit the function in the data space very well. The scattering parameters of two proposed electromagnetic stop band filters have been fitted by this method.Stochastic network can employ simulated annealing method to optimize parameters of equivalent circuits. Equivalent circuits of two proposed electromagnetic stop band filters have been developed. And their parameters have been optimized using this method.All above analysis results have been verified by experiments.