Study Of Microwave Neural Network

With the increasingly development of wireless communication and microwave technology, we need speedy and accurate component model urgently. In traditional microwave CAD method, the pursuit of computation speed and accuracy has often been a trade-off process. Models with better accuracy usually come with more computation and fast models have often sacrificed accuracy. We need new modeling techniques that allow very fast model evaluation and do not sacrifice accuracy at the same time. On the other hand, neural network technology had become a powerful mathematical tool, especially for the complex nonlinear problems. It has many advantages such as: generalization, no prior knowledge needed, modeling process simple, and high accuracy, which make it become a very potential microwave CAD tool.Recently, massive research works had been done about microwave neural network at abroad. Many of them had become commercial software. In our country, only a few initiative studies had been endeavored on the field. There only had very little papers about microwave neural network. Moreover the research key point laid on a few problems, which was lack of systematism and generalization.In this paper, we systematically describe microwave neural network technology, as the follow list shows:(The classical neural networks, such as MLP, RBF, GRNN, wavelet network and GA, had been implemented, and several novel neural network and train algorithm had been proposed.(Several new modeling methods, such as metamorphosed neural network, source difference method and it's derivatives, hybrid neural network, had been presented. They not only can improve neural network performance, but also decrease training data and training time.(Strip line, coupling strip line, microstrip, coupling microstrip, slot and CPW models had been established with neural network technology. (Many passive and active component models of microwave circuit had been built with microwave neural network. It illuminated microwave neural network was an efficient and flexible modeling method.(Microwave neural network database and toolbox had been established base above works.(MMIC modeling and design works had been performed by using microwave neural network toolbox and various of modeling methods. The result shows microwave neural network technology has good prospects in MMIC design field.The result shows that microwave neural network is a simple and effective method. We trust that microwave neural network can be used to solve the most difficult problems of microwave CAD in the future.