A Study Of Neural Network Techniques In Microwave Circuit CAD

Researching the applications of neural network techniques in the area of microwave circuits CAD and optimization design is one of the front edge issues in nonlinear microwave electronics domains. The emergence of neural network provide effective tools for a system with highly nonlinear input-output relations. A neural network model for a device/circuit can be developed by learning and abstracting from measured-simulated microwave data, through a process called training. Once be trained, the neural network model can be used during microwave design to provide instant answers to any pre-assigned task. The trained neural network model can be used to solve a variety of problems emerged in RF/microwave circuit design, such as in microwave circuit CAD, the established model structure can be used to characterize the nonlinear behavior of microwave circuits. And in microwave circuit design, the model can be used for the design of coplanar waveguides, transistors, transmission line, filters and amplifiers, etc. Besides, after in microwave circuit optimization, a neural network model can be used to optimize circuit parameters and match the microwave impedances, etc.Under this background, with the support of the project entitled "Modeling of Wireless Nonlinear Channel and Analysis of Its Characteristics" of of UESTC Youth Funddation, this dissertation is intended to develop some techniques for application of neural networks in the fields of nonlinear microwave circuit modeling and optimization. The main contributions of the dissertation are as following:1. Modeling time-domain nonlinear characteristics of microwave circuits by using the neural network techniques, then analyzing the time-domain transient and steady states of a power-amplifier and in collarboration with the HP-ADS. The results of computer simulation show that the output of the neural network model output is concidede with the output of a real nonlinear power-amplifier.2. The nonlinear input-output characteristics of a GaAs MESFET is modeled by the use of Volterra series in the frequency domain. Based on this frequency domain model, after an MLP network training, the input-output characteristics of the device as well as the Volterra series Kernal function, which is hard to be determined by traditional approaches, is obtained with satisfactory accuracy.3. A wavelet neural network training method is used for the optimization of a microwave circuit. This technique is used for the simulation and optimizationof the low-pass filter parameter. Simulation results indicate that this training algorithm is able to give solutions with high quality and remarkably reduced computation time.