Modeling Of Microwave Passive Components Based On Fuzzy Logic

Circuits and devices'model is the basis of analysis and design. With the development of the Integrated Circuit, the feature size of devices are shrinking down, the scale of the circuits are becoming larger and larger, and the operating frequency/speed are increasing tremendously, the function and performance of the circuits are becoming more and more complex. Using the traditional model of circuits and devices to circuit design and simulation can no longer meet the needs of complex large-scale Integrated Circuit design. New modeling technology is the most important subject in current circuit and system theory.In recent years, because of the emergence of intelligent information processing technology, such as neural networks and fuzzy logic, circuit and device modeling have begun to use such methods. Compared with the traditional numerical methods, neural networks and fuzzy logic system can be better-expressed and processing multi-dimensional, complex non-linear relations. They are possible to achieve better results in versatile nature, approximation accuracy and efficiency.This paper studies modeling of microwave passive components based on fuzzy logic. Some approximate empirical formula can be used for model passive components in RF and microwave monolithic integrated circuit while the frequency is low. However, with increasing frequency, they are need to full-wave electromagnetic numerical simulation, so it is necessary to establish the relationship model between the physical and structural parameters of these passive components and the electromagnetic properties.Based on fuzzy logic modeling have two steps. First, determining the initial system by the method of subtractive clustering. Second, learning the system. The consequent parameters of a TSK fuzzy logic system can always be determined by the premise parameters using least squares method, then the optimizing learning algorithm is given which depends on the premise parameters only. This reduces the dimension of the problem, it helps to reduce the computational complexity and improve convergence. The gradient descendent optimizing of this problem is equivalent to the hybrid learning algorithm with one BP iteration with respect to the premise parameters and the least squares solution to the consequent parameters. The Quasi-Newton optimizing of the problem leads to an efficient Newton-type hybrid-learning algorithm. In order to achieve this modeling method, using VC++ compiled a fuzzy logic modeling tool.Adopting the above algorithm, this paper has set up the relationship model between the S parameters and the physical parameters for several microstrip line discontinuities and coplanar waveguide components of RF / microwave circuits and given experiment results. At the same time with the ANFIS and neural network modeling results are compared, indicating the merits of using Quasi-Newton hybrid learning algorithm for modeling. The results of these examples show that the modeling method is effective.