Design And Implement Of Radio Frequency Devices Modeling And Simulation Software Based On Neural Networks

Radio frequency circuits contain a lot of passive elements which play very important roles in radio frequency circuits. How to provide accurate model for the circuit simulation is a basic and critical context of CAD tools. The main work of this paper is associated with the design and implement of radio frequency devices modeling and simulation software.In this paper, common used neural network algorithms are analyzed and compared firstly. The neural network is considered as a high accurate modeling method with less computation and thus has the special advantages compared with traditional modeling method. The Levenberg-Marquardt algorithm which has the best performance is introduced into the modeling of radio frequency devices, and its convergence has been improved. In order to raise the training speed, weights and threshold selection algorithms from Nguyen and Widrow has been improved. The transmission space of the neuron in every layer of the neural network is designed to cover the whole input space of the neural network. In the implement of LM algorithm, the mathematic and coding methods are combined to further optimize algorithm and improve the computing speed. The improved algorithm has been tested on convergence, and the result shows the new algorithm can achieve better performance. A universal radio frequenc y devices modeling and simulation software has been developed based on the above techniques. The spiral inductor is used to do the simulation and test, and the simulation results show that this software can achieve ideal predicting precision and can provide accurate model for the radio frequency circuits.