【作者】 袁继敏；

【导师】 古天祥；

【作者基本信息】 电子科技大学 ， 测试计量技术及仪器， 2009， 博士

【摘要】 本文从原理上阐述基于混沌理论的检测技术的可行性，根据混沌检测模型和神经网络检测模型在检测应用中存在的诸多需要研究的问题，研究混沌检测模型、神经网络检测模型、以及二者结合的复合检测模型的原理，提出应用混沌原理和神经网络原理相结合的检测技术检测混沌背景中的信号的时域参数。这是一个很有应用前景的研究方向。论文逆向运用混沌测量的原理，突破现有的理论，探索新的检测原理方法，直接在混沌状态下构建检测微弱信号时域参数的混沌模型和神经网络模型的方法，更有效地提取信号参数。同时，针对时域测试系统的本身动态特性属于非线性，寻找数学描述模型困难而难于校正的问题，运用逆系统理论，构建神经网络逆系统，获取原系统的非线性动态特性方法。论文重点研究建模算法、模型结构和建模方法，力求拓宽混沌和神经网络理论测试技术在时域测试、电路动态参数和瞬态参数捕获方面的应用范围，提高检测精度。论文结合时域测试的典型仪器——数字示波器，将研究成果应用到数字示波器中。针对数字示波器捕获微弱触发信号能力差，不能测量微弱信号，采样经典的理论校准静态和动态参数等许多问题，论文着力研究增强其测量微弱信号和捕获微弱触发信号的能力；建立基于混沌理论校准模型；神经网络逆系统模型，创新静态、稳态和动态校准方法。在课题研究中，作者主要做了以下几个方面的工作：(1)对神经网络识别和检测信号进行分析和研究。研究了空间分割的竞争神经网络识别规则模拟信号类别的算法，提出了将竞争神经网络应用在数字示波器中识别规则信号类别，为选择内插算法提供依据。(2)对Elman时空网络结构、稳定性及应用研究。讨论了Elman时空网络的结构和学习方法，重点研究了应用Elman时空网络测量时域信号的有效性和问题。研究了改进型Elman时空网络的算法和稳定性，并通过仿真证明了改进网络时域测量的优点。把混沌和神经网络结合起来，为构建了新的时域测量模型奠定了基础。(3)研究混沌背景下的微弱周期信号的检测。用Duffing-Holmes方程构建混沌测量模型，检测微弱周期性信号的频率；利用二维Henon映射的混沌检测模型，检测微弱的触发信号，应用在数字示波器时基中。(4)基于混沌和神经网络的微弱瞬时信号的检测的研究。研究基于FP算法的前向网络的结构和设计方法，构建了基于混沌背景下的微弱瞬时信号测量模型，在混沌状态下直接检测混沌背景下的瞬时信号。(5)基于混沌和神经网络的微弱信号时域参数检测的研究。研究了混沌系统和神经网络检测模型和方法，在混沌状态下直接获取信号时域参数。同时，深入研究了时空神经网络的结构，获得基于混沌的神经网络的微弱时域信号检测模型的建模的依据。(6)DSO校准和“NTN”校正方法中kick-out脉冲研究。研究数字示波器电压测量准确度、时基误差的估计和上升时间的测量和校准方法，建立了静态参数的混沌校准的模型。提出运用神经网络逆系统的方法，解决数字示波器动态参数校准的新思路。深入研究“NTN”校正方法和kick-out脉冲，为宽带高速数字示波器校准提供理论基础。更多还原

【Abstract】 To solve the problems that have existed in the detection applications of chaos modeland neural network model,the feasibility of detecting technology based on chaos theoryis elaborated in terms of theory in this dissertation.The principles of chaos detectionmodel,neural network detection model and the composite detection model of theaforementioned are further studied.The time-domain parameters to detect the signals inthe chaotic background by adopting the detecting technology in the appliance of thechaos theory and neural network theory are presented.This study demonstrates positiveapplications by reversing the existing principles.By reversing the chaos detectionprinciple,a new method to extract signal parameters more effectively by building thechaos model and the neural network model to detect the weak-signal time-domainparameters in the chaotic state directly,has been explored.Simultaneously,because ofthe dynamic and non-linear characteristics of time-domain detection system,and thedifficulties in updating the mathematical model,a neural network inverse system hasbeen built to obtain the nonlinear dynamic characteristics of original system by usingthe inverse system theory.In essence,the modeling algorithm,the model’s structure andmodeling methods are studied in depth in order to widen the application scope of usingchaos and neural network theory in the time-domain detection and the circuit dynamicparameters to improve the detection accuracy.The research results are applied in digital oscilloscope,the typical time-domaindetection equipment.To solve the problems of incapability in detecting weak signalswith the digital oscilloscope for its poor performance in capturing weak trigger signalsand invalidation of updating the static and dynamic parameters with the classic theories,this dissertation are focused on enhancing the capability of DSO weak signal detectionand weak trigger signal capture.Calibration models based on chaos theory and inversesystem model based on neural network are constructed in this dissertation.Innovativestatic,steady-state and dynamic calibration methods are also recreated.In this research,the author has contributed in the following major areas: Ⅰ.Analysis and research on the neural network to identify and detect signals.Thealgorithms of space division competition neural networks to identify the regularanalog signal classifications are studied.The opinion which identifies the signaltype with the digital oscilloscope by competition neural network provides a basisfor the selection of interpolation algorithm.Ⅱ.Research on structure,stability and applied of Elman space-time network.Thestructure and learning methods of Elman time-space network,especially thevalidity of its application to detect time-domain signals are discussed.Thealgorithm and stability of improved Elman space-time network are studied and itsadvantages are also proved by emulation.A new time-domain detection model hasbeen constructed on the basis of chaos theory and neural network.Ⅲ.Research on the weak periodic signal detection in the chaotic background.Chaoticdetecting model is constructed by Duffing-Holmes equation to detect weakperiodic signal frequency;and the applications of two-dimension Henon map indetecting weak trigger signals and digital oscilloscope time-base are discussed.Ⅳ.Research on weak transient signals detecting based on chaos and neural networktechnology.The structure of front-faced network and design method based on FPalgorithm are studied.A detection model to detect weak transient signals directlyin the chaotic background is constructed.Ⅴ.Research on the weak signal time-domain parameters detection based on chaosand neural network technology.The detection model and method of chaoticsystem and neural network to obtain time-domain signal parameters directly in thechaotic state are researched.The structure of space-time neural network isfurther-studied,based on which the weak time-domain signal detection model byusing the chaotic neural network has been constructed.Ⅵ.Research on the kick-out pulse in DSO calibration and“NTN”correction methods.The accuracy,time-based error estimation and the rise-time detection andcalibration methods in the application of digital oscilloscope voltage are studied aswell as chaos calibration model of static parameters is constructed.New ideas toadopt the neural network inverse system to solve the DSO dynamic parameterscalibration are proposed.The“NTN”updating methods and kick-out pulse are studied to provide theoretical basis for the broadband high-speed digitaloscilloscope calibration.更多还原