知识型神经网络的射频/微波器件建模方法(英文)

Knowledge-Based Neural Networks for Modeling of Radio-Frequency/Microwave Components

综述了用神经网络对射频与微波电子器件建模的方法,利用知识型的概念将射频或微波电子的等效电路和经验公式与神经网络有机的结合起来。将相关知识附加到神经网络模型的优势进行了论证.通过与无附加知识的传统多层感知器对比,证明了知识型神经网络的可行性。介绍了几种现有电路模型与神经网络结合的方法,如差分法、先于知识输入法及空间映射神经网络法。举例介绍了微波传输线和高电子迁移率晶体管的建模方法及过程,证明了基于知识的神经网络是各种微波器件建模的一种有效方法。

An overview of advanced neural network methods for modeling radio-frequency （RF） and microwave electronic devices is presented. Knowledge-based engineering concept is utilized where the knowledge of RF/microwave electronics in the form of equivalent circuits and empirical formulas is combined with neural networks. Advantages of adding knowledge on the performance of the neural models in terms of generalization ability versus different sizes of training data through a knowledge based neural network （KBNN） technique are demonstrated and examples of comparisons with conventional MLP （without any knowledge-base） are given. Several methods of combining existing circuit models with neural networks, including the source difference method, the prior knowledge input method, and the space-mapped neural models, are also introduced. Application examples on modeling microwave transmission line and high electron mobility transistor （HEMT） device demonstrate that KBNN is an efficient approach for modeling various types of microwave devices.