基于知识模糊逻辑电路非线性行为宏模型-knowledge-based fuzzy logic circuit nonlinear behavior macro model.docxVIP

摘要器件与各类功能电路的模型是进行集成电路与系统设计的前提与基础，是 EDA 技术中重 要的研究课题。宏模型是表示一个单元或功能电路外部特性的一个简化等效模型，是集成电 路设计自动化的基础，但长期以来，对一般非线性电路的宏模型构造缺乏有效、建立在严格 电路与系统理论基础上的方法。本文首先论述了几种比较典型的基于系统辨识的电路宏模型建模方法，包括 Volterra 级 数模型、Hammerstein 模型、NARX 模型、神经网络模型等，简要说明了各种方法的原理与 特点。本文重点研究基于模糊逻辑的电路宏模型建模方法。模糊逻辑的优点是可自然地表达 出问题的经验知识，模型是可解释、有物理意义的、易于理解的。文章在以往工作的基础上，主要开展了基于先验知识的电路模糊逻辑建模方法的研究， 提出了基于电路结构分解的模糊逻辑建模方法与基于状态方程的模糊逻辑建模方法，并将这 两种方法应用于运算放大器与数字驱动器宏模型的构建。两种方法都利用了电路内部的结构 特点与工作特性等先验知识。根据电路的结构知识确定模型整体结构；根据电路的工作特性 确定模糊逻辑模型的规则（包括条数、前提与结论函数的形式）及初始参数。这就大大简化 了模型构造过程，使一开始构造出来的初始模型就有较好的逼近精度，后续模型参数的学习 训练只需经过少量的迭代，相对于传统的系统辨识方法具有很大的优势。文中对每类电路都 给出了实际建模例子，通过对模型预测结果和原始电路仿真结果的对比分析，说明了提出的 建模方法的有效性。关键词：模糊逻辑、行为宏模型、数字驱动器、运算放大器、结构分解、状态方程ABSTRACTThe models of devices and various functional circuits are the premise and basis of the integrated circuit and system design.It is also an important research topic in the EDA technology. The macro model is a simplified equivalent model describing external characteristics of a unit or a functional circuit, which is the basis for automation of integrated circuit design. But macro model structure of the general nonlinear circuit is short of effective methods based on strict circuit and system theory for a long time.This paper firstly discusses several typical circuit macro modeling method based on system identification, which contains Volterra series model, Hammerstein model, NARX model, and neural network model, etc. And a brief description of the principles and characteristics of these methods has been given. The paper focuses on the circuit macro modeling method based on fuzzy logic system. The advantage of fuzzy logic approach is a natural expression of experience and knowledge of the problem, and these models can explain the physical sense, easy-to-understand.This paper mainly researches the study of fuzzy logic modeling method based on prior knowledge of the circuit, comes up with the fuzzy logic modeling method based on the circuit structure decomposition and the fuzzy logic modeling method ba