论文摘要
非平稳信号的分析和处理已经引起了大量的时频方法和算法的发展,而本论文的目的是希望为该领域的研究提供一些新的方法。具体一点来讲,论文工作主要针对以下两方面的研究内容:(1)改进非平稳信号分析的表象工具;(2)检验信号相对于某一观察范围的平稳性。就非平稳光谱评估而言,现存的时频工具仍然需要得到进一步的改善,尤其是对于啁啾信号嵌入在噪声中的情况。因此,论文第一部分的工作主要讨论解决的就是该问题,最终的目标是一方面能够加强啁啾成分的局域性,另一方面又能降低噪声背景的统计波动水平。技术的关键在于结合“时频再分配技术”和“多窗技术”。根据这一思想,我们提出了两种不同的方法:第一个方法是基于对使用不同窗口产生的再分配评估求和,主要针对的是非平稳光谱评估;而第二个方法是基于对同样的(不同窗口产生的再分配)评估求差,主要是为了进一步加强啁啾信号。方法的技术原理在文中具体阐述,基于厄密多函数的执行被讨论论证,并且论文也提供了一些典型的例子(包括欧拉盘问题)来支持方法的有效性。再说平稳性,本论文第二部分的工作在于打破经典的平稳性定义,开发具有实际运作意义的平稳性检验。经典的平稳性是指信号的统计属性随时间严格不变,而本论文中的广义平稳性则是相对于某个观察范围而言的,并且同时适用于统计和确定性的情况。该平稳性检验基于局域时频特性与全局时频特性的比较,原创性地使用了由原始被测信号产生的平稳的“替代品”来定义平稳性的零假设。基于这个框架,两种不同的方法被提出:第一个方法是适当地选择局域谱和全局谱的”距离”度量,构建一个来源于替代品属性分布的参数模型,从而定义平稳性零假设;而第二个方法是从时频平面上提取信号特征量,交由一类支持向量机执行,而把替代品的这一特征量作为训练集。检验的原理和具体方法在文中详细阐述,并且对一些典型的相对于不同观察范围下平稳性情况不同的信号(包括语音信号)进行了测试。
论文目录
Abstract摘要Ⅰ.Introduction1 Introduction to Stationarity and Nonstationarity1.1 Classical Definition of Stationarity1.2 Generalized Notion of Stationarity1.3 Issues in Representation of Nonstationary Signals2 Introduction to Time-Frequency Method2.1 Fourier Transform2.2 Characteristics of Time-Frequency Representation2.3 Short-Time Fourier Transform2.4 Wigner-Ville Distribution2.5 Wigner-Ville Spectrum3 Reassignment3.1 Spectrogram3.2 Reassigned Spectrogram3.3 Some Historical Comments4 Multitapering4.1 Power Spectrum Density4.2 Welch Method of Averaged Periodograms4.3 Thomson's Method of Multitapering4.4 Extensions of MultitaperingⅡ.Multitaper Time-Frequency Reassignment1 Multitaper Time-Frequency Reassignment for Nonstationary Spectrum Estimation1.1 Principle and Implementation1.1.1 Wedding multitapering with reassignment1.1.2 Choice of tapers1.1.3 Effectiveness measure by Renyi Entropy1.2 Performance Evaluation:Error Measure1.3 Examples2 Variations for Chirp Enhancement2.1 Principle and Implementation2.1.1 Differences between estimates based on different tapers2.1.2 Final estimate combined by "sums" and "differences"2.2 Performance Evaluations:Contrast Measure2.3 Examples3 Application:Euler's Disk3.1 Time-Frequency Estimation3.2 Parameter Estimation of Chirp by Hough Transform3.3 Performance Evaluation on Synthetic Model3.4 Test on Experimental Signal4 ConclusionⅢ.Test of Stationarity with Surrogates1 Time-Frequency Framework:Comparison of Local vs.Global Features2 Surrogates:Stationary Reference2.1 Surrogates in(Non)stationarity Test2.2 Surrogates in(Nonlinear)Physics3 Stationarity Test-Parametric Model Approach3.1 Comparison of Local vs.Global Time-Frequency Spectra3.2 Choice of Distance3.2.1 Test signals3.2.2 Probability laws distances3.2.3 Spectral distances3.2.4 Combined distance3.3 Structure of the Test3.4 Null Hypothesis of Stationarity3.4.1 Distribution based on surrogates3.4.2 Gamma model?3.4.3 How many surrogates needed?3.4.4 Threshold based on a statistical significance3.4.5 Verification of null hypothesis3.5 Index and Scale of Nonstationarity3.6 Examples4 Stationarity Test-One-Class Support Vector Machine Approach4.1 One-Class Support Vector Machines4.1.1 Machine learning4.1.2 Support Vector machines4.1.3 One-class Support Vector machines4.2 Comparison of Local vs.Global Time-Frequency Spectra4.3 Examples5 Application:Speech6 ConclusionⅣ.Conclusions and PerspectivesⅤ.Appendix1 Alternative model2 Choice of Thresholds3 Variations of Test4 PublicationsBibliographyAcknowledgment
相关论文文献
标签:时频论文; 非平稳光谱论文; 啁啾论文; 多窗论文; 再分配论文; 平稳性检验论文; 支持向量机论文;