Microwave And RF Device Modeling Based On Neural Network Technique

In order to improve the performance of the products, shorten the design cycle and reduce the cost, the designers have to rely on the efficient and accurate models with computer-aided design(CAD) techniques. In the field of microwave and radio frequency(RF) modeling, Conventional approaches to create or modify models are heavily based on slow trial-and-error processes, which can’t meet the requirements of the design cycle and the accuracy for the new device modeling. In recent years, artificial neural network(ANN) techniques have been recognized as useful alternatives to conventional approaches in microwave modeling. In this thesis, we propose three enhanced neural network modeling methods suitable for heterojunction bipolar transistors(HBTs), transistors with package and power amplifier with memory effects respectively.Firstly, the new modeling method suitable for HBT modifies the structure of the existing Neuro-Space Mapping(Neuro-SM). The voltage mapping network in the existing Neuro-SM is replaced by the current and voltage mapping network. The method using the hurst parameter(H-parameter) to present the small signals of HBT is proposed and the computational formula from the H-parameter of empirical models to the H-parameter of the modeled object is derived. The modeling results for HBT measurement data demonstrate that the proposed HBT model can reflect the DC and small signal characteristics accurately, and the proposed model can be built in the simulation software easily. Secondly, this thesis proposes a new modeling method for transistor with package. The packaged transistor is divided into three parts: the input package circuit, the nonlinear circuit and the output package circuit. These three parts are modeled respectively. In the proposed method, we derive the linear formulas of package circuit based on the reciprocity of the package circuit, and build the package model combined the linear formulas with neural network. In addition, the relationship between the S-parameter of the input package circuit, nonlinear circuit and output package circuit and the S-parameter of the whole packaged transistor is proposed. The integrated structure of the new transistor model and the new circuit of mapping network are presented in this thesis. The packaged transistor modeling examples demonstrate the correctness and effectiveness of the proposed modeling method. Finally, this thesis proposes a new recurrent neural network(RNN) modeling method suitable for the power amplifier with memory effects. In the proposed technique, we extract slow-changing signals from the inputs and outputs of the PA and use these signals as extra inputs of RNN model in order to effectively represent long term memory effects. Compared with the conventional RNN methods, the input current as one of the inputs is added into the proposed RNN model. Adding the input current can enhance the accuracy as well as the convergence of the model with less hidden neurons and less delays. The examples of modeling for power amplifiers with memory effects show that the proposed RNN models can reflect the short term memory effects and long term memory effects accurately.