时频域多分量信号分析识别研究

发布时间：2018-05-28 16:00

本文选题：时频域多分量信号 + 时频分析　；参考：《电子科技大学》2016年博士论文

【摘要】：在过去的几十年里,随着电磁设备研发的多样性及应用的广泛性,采集到的信号形式与种类也越来越复杂,分布越来越密集,在同一信道中存在多个信号的情况变得越来越普遍,这种情况为后续的信号处理研究增加了一些新的问题。由于广泛的应用背景和巨大的现实意义,尽管时频域多分量信号的分析识别问题是一个极端欠定的病态问题,但仍然得到了很多研究人员的青睐和投入。尽管在研究过程中,研究人员针对生电、机械振动、语音等自然信号和雷达、通信等人工信号的分析识别问题,分别定义了确定性信号模型、非平稳信号模型、非线性时间序列模型和状态空间模型等等各种模型,提出了各种时频变换方法、矩阵分解方法和参数估计重构方法。但归根结底,所有这些研究都是针对特定的应用场景的,而在更为复杂的情况下或者更为广泛的应用背景下常常存在失效的问题。针对这些问题,本文在前人现有研究成果的基础上,对时频域多分量信号的分析识别问题进行了更加深入地研究和探讨,提出了一些新的思路和算法,进而扩展了这个方向上的研究内容,具体包括:1.针对在多个线性调频信号的全盲分析识别中对较弱分量信号的检测失效问题,本文将谱峭度的盲检测技术和分数阶傅里叶变换算法相结合,在现有研究的基础上提出了基于分数阶傅里叶谱峭度的变换新算法,并将其应用于较低信噪比下对多个非等功率分量信号的检测估计之中。该算法首先讨论了分数阶傅里叶变换的圆特性,然后将谱峭度的盲检测技术引入分数阶傅里叶域,定义了分数阶傅里叶谱峭度的概念并推导了它的一些特性,进而将这些特性用于多个线性调频信号的盲的分析识别中。最后,理论分析和仿真结果都验证了所提算法在较低信噪比下对较弱分量信号的检测性能优于其它算法;2.针对在多个非线性调频信号的全盲分析识别中对多个相近分量信号的检测失效问题,将分数阶傅里叶谱峭度和短时分数阶傅里叶变换算法相结合,在现有研究的基础上提出了基于分数阶傅里叶谱峭度的自适应短时分数阶傅里叶变换新算法,并将其用于多个相近非线性调频信号的检测估计之中。该算法首先推导了高斯旋转窗的谱密度和非圆特性,并将分数阶傅里叶谱峭度的特性用于修正后的最优旋转高斯窗的参数选择,从而定义了基于分数阶傅里叶谱峭度的自适应短时分数阶傅里叶变换新算法;在此基础上,将基于分数阶傅里叶谱峭度的时频分割算法用于时频变换后对多个非线性调频信号的盲检测提取之中。最后,理论分析和仿真结果验证了所提算法的时频聚焦性优于其它算法,因此可以识别非常接近的多个非线性调频信号;3.针对全盲条件下多个非平稳非线性时间序列的分析识别问题,将高阶统计技术和非平稳分割技术、相空间重构技术以及单通道独立分量分析算法相结合,在现有研究的基础上提出了基于高阶单通道独立分量分析的变换新算法,并将其应用于多个非平稳非线性时间序列的全盲分离识别之中。该算法主要分为三个步骤:首先通过应用基于高阶启发式的非平稳检测和分割算法成功得到高阶平稳的非线性时间子序列;然后通过选取合适的重构参数将分割后的非线性时间子序列有效重构为多维轨迹矩阵,并应用基于高阶奇异值分解的坐标变换方法将该矩阵转换成伪多通道的瞬时线性混合模型;最后应用盲源分离方法将其中感兴趣的信号分量分离并提取出来。理论分析和仿真显示都验证了所提算法不仅可以有效分离多个非平稳非线性时间序列,而且对噪声和重构参数的鲁棒性也优于传统的单通道独立分量分析算法;4.针对数字通信中多个相位调制信号的共信道盲分析识别问题,在现有研究的基础上提出了基于高阶奇异值分解的盲源分离新算法,并将其应用于共信道码速率不同的多个数字相位调制信号的分离估计之中。该算法首先通过过采样和矩阵重排将多个数字相位调制信号的盲分析识别问题转化成相位变化的多个周期信号的盲分离问题,然后采用高阶奇异值分解算法估计各个相位调制信号的码元波形;最后应用盲源分离方法直接估计出各个相位调制信号的符号序列。仿真显示所提算法在一定程度上可以解决共信道多个数字相位调制信号的盲分离问题,且对噪声干扰和非等功率影响具有一定的鲁棒性。
[Abstract]:In the past few decades, with the diversity of the development of electromagnetic devices and the wide range of applications, the forms and types of signals collected are becoming more and more complex and more and more dense. The presence of multiple signals in the same channel becomes more and more common. This situation has added some new problems to the follow-up signal processing research. Although the analysis and recognition of multi component signals in time and frequency domain is an extremely ill defined and ill conditioned problem, it still gets the favor and input of many researchers. Although in the process of research, researchers are aiming at natural signals such as electricity, machinery vibration, voice and other natural signals and radar, communication and so on. In the analysis and recognition of the signal, the deterministic signal model, the non-stationary signal model, the nonlinear time series model and the state space model are defined respectively. Various time-frequency transformation methods, matrix decomposition methods and parameter estimation reconstruction methods are proposed. All these studies are aimed at specific applications. In the context of more complex situations or more extensive application background, there are often failures. In this paper, based on the existing research results of previous studies, this paper makes a further research and Discussion on the problem of analysis and recognition of multi component signals in time-frequency domain, and puts forward some new ideas and algorithms. The research contents of this direction are expanded, including: 1. for the detection and failure of weak component signals in all blind analysis and recognition of multiple linear frequency modulation signals, this paper combines the blind detection technique of spectral kurtosis with the fractional Fu Liye transform algorithm, and proposes a fractional Fu Liye based on the existing research. A new algorithm for spectral kurtosis is applied to the detection and estimation of multiple non equal power component signals at low signal to noise ratio. The algorithm first discusses the circular characteristics of fractional Fourier transform, and then introduces the blind detection technique of spectral kurtosis to the fractional Fourier domain, and defines the concept of fractional Fourier spectral kurtosis. Some of its characteristics are applied to the blind analysis and recognition of multiple linear frequency modulation signals. Finally, both theoretical analysis and simulation results demonstrate that the proposed algorithm is superior to other algorithms for weak component signals under low signal to noise ratio; 2. With the combination of fractional Fourier spectrum kurtosis and short-time fractional Fourier transform algorithm, a new adaptive short time fractional Fourier transform algorithm based on fractional Fourier spectral kurtosis is proposed on the basis of the existing research, and it is applied to multiple similar Nonlinear FM signals. The algorithm first derives the spectral density and non circular characteristics of the Gauss rotating window, and uses the characteristic of the fractional Fourier spectral kurtosis to select the parameters of the modified optimal rotating Gauss window, and then defines a new adaptive short time fractional Fourier transform algorithm based on the fractional Fourier spectral kurtosis. The time frequency segmentation algorithm based on fractional Fourier spectral kurtosis is used to extract the blind detection of multiple nonlinear FM signals after the time frequency transformation. Finally, the theoretical analysis and simulation results show that the time frequency focusing of the proposed algorithm is better than other algorithms, so it can identify very close multiple nonlinear FM signals; 3. With the combination of high order statistical technology and non-stationary segmentation, phase space reconstruction and single channel independent component analysis, a new algorithm based on high order single channel independent component analysis is proposed on the basis of the existing research, and the new algorithm is applied to the problem of the analysis and recognition of multiple nonstationary nonlinear time series. The algorithm is divided into three steps: first, a high order stationary nonlinear time sequence is successfully obtained by using a non-stationary detection and segmentation algorithm based on high order heuristic, and then the nonlinear time sequence of the segmented nonlinear time sequence is obtained by selecting the appropriate reconfiguration parameters. The column is effectively reconstructed into a multidimensional trajectory matrix, and the matrix transformation method based on high order singular value decomposition is used to convert the matrix into a pseudo multi channel instantaneous linear mixed model. Finally, the blind source separation method is used to separate and extract the signal components of interest. Both theoretical analysis and simulation display verify that the proposed algorithm is not only effective. It can effectively separate multiple non-stationary nonlinear time series, and it is more robust to noise and reconstruction parameters than traditional single channel independent component analysis algorithm. 4. for the blind analysis and recognition problem of multiple phase modulation signals in digital communication, the blind algorithm based on high order singular value decomposition is proposed on the basis of the existing research. A new source separation algorithm is applied to the separation estimation of multiple digital phase modulation signals with different cochannel code rates. First, through over sampling and matrix rearrangement, the blind analysis recognition problem of multiple digital phase modulation signals is transformed into a blind separation problem of multiple phase signals with phase change, and then the higher order is adopted. The singular value decomposition algorithm estimates the symbol waveforms of each phase modulation signal. At last, the blind source separation method is used to directly estimate the symbol sequence of each phase modulation signal. The simulation shows that the proposed algorithm can solve the blind separation problem of multiple digital phase modulation signals of common channel to a certain extent, and the noise interference and non equal power can be used. The influence has a certain robustness.
【学位授予单位】：电子科技大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TN911.6

【参考文献】