微波电子器件与系统建模优化先进方法研究

项目名称： 微波电子器件与系统建模优化先进方法研究

项目编号： No.61271067

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 无线电电子学、电信技术

项目作者： 张齐军

作者单位： 天津大学

项目金额： 80万元

中文摘要： 本课题的目标是研究新的高频电子器件建模方法，用于射频/微波通信系统的开发。在高频情况下，器件/系统的电磁行为建模和优化是至关重要的。随着电磁场几何结构复杂性的增加，电磁建模和优化任务因为庞大的计算量给设计者提出了严峻的挑战。特别是在基于高成品率的设计中，高精度要求和众多物理/几何参数变化的情况下，需要对电磁场进行反复的计算，进而加剧了设计的难度。针对这些技术问题，课题组在已经处于微波器件神经网络建模研究的国际前沿基础上，将开展如下创新研究：一是基于电磁响应和电磁敏感度的神经网络模型训练算法，二是采用基于敏感度训练的知识型神经网络模型自动生成算法，三是采用状态空间估计和动态神经网络状态空间训练的非线性器件行为建模方法，四是新型用于状态空间动态神经网络模型的自动产生算法。本研究的目标是为快速和准确地设计新一代高频器件和系统，建立系统规范的神经网络模型生成方法。

中文关键词： 建模；电磁敏感度；状态空间；自动建模；微波器件

英文摘要： The objective of this research is to develop new methodologies for modeling of high-frequency electronic components and packages in wireless and wireline RF/Microwave communication systems. At high signal frequency, modeling and optimization of the electromagnetic behavior in devices /systems are vital. This task becomes computationally prohibitive as the complexity of the electromagnetic structure increase and optimization is applied. For yield-driven design with random variations in physical/geometrical parameters in the system, the computational difficulty is further compounded. The proposed research will address these challenges. Built on top of our advances in neural network technologies for microwave design, this research will open several new frontiers in the area: a new category of training algorithms using both electromagnetic response and electromagnetic sensitivity, a new category of behavioral modeling of nonlinear devices using state-space estimation and state-space training of dynamic neural models, new formulations of automated model generation algorithms for knowledge based neural networks with sensitivity based training, and new automated model generation algorithms for training state-space dynamic neural networks. The long term direction is towards a unified methodology for developing neural

英文关键词： modeling；electromagnetic sensitivity；state space；automated modeling；microwave components