单通道盲源分离算法的研究

发布时间：2019-04-04 15:00

【摘要】：盲源分离(Blind Sources Separation,BSS)技术指的是在传输信道和源信号都未知的情况下,根据源信号的统计特性,仅仅利用观测信号分离出各个源信号的过程。多通道BSS算法已经在生物医学信号处理、阵列信号处理、移动通信和文本分析与处理等领域得到广泛应用。近几年来,单通道盲源分离问题逐渐成为信号处理领域的研究热点。单通道盲源分离(Single Channel Blind Source Separation,SCBSS)是欠定盲源分离问题的极端情况,它仅仅利用单路观测信号的特征信息,分离出多路源信号,解决起来十分困难。但是SCBSS又是许多实际系统中常见的问题,因此研究SCBSS算法具有重要的理论意义和应用价值。本文主要研究多通道BSS算法和SCBSS算法。首先,研究了FastICA算法的改进。针对源信号数较多时,原有的FastICA算法迭代次数较多和分离性能恶化的问题,提出了Pm-FastICA算法。对算法中的非线性函数进行Pade逼近,得到能够减少FastICA算法迭代次数的有理函数,提高了收敛速度和分离性能。仿真表明,Pm-Fast ICA算法性能优于FastICA算法,且随着源信号数目的增多,Pm-FastICA算法的性能优势将更明显。同时提出了一种利用有理多项式非线性函数的FastICA(简称N-FastICA)算法。仿真表明,N-FastICA算法性能优于Pm-FastICA算法和FastICA算法,且随着源信号数目的增多,N-Fast ICA算法的性能优势将更明显。其次,研究了基于小波包分解的SCBSS(WPT-ICA)算法。由于小波变换不能很好地表示包含大量细节信息,基于小波变换的SCBSS算法性能有待提高。对此,本文提出了一种基于小波包分解的SCBSS算法。对观测信号进行小波包分解,选择能量百分比较高的系数进行重构,将重构信号与观测信号构成多路信号,利用N-FastICA算法实现信号的盲源分离。仿真结果表明,基于小波包分解的SCBSS算法性能优于基于小波分解的SCBSS算法。然后,研究了基于经验模态分解(Empirical Mode Decomposition,EMD)的SCBSS算法。基于EMD的SCBSS算法存在模态混叠现象,导致分离性能恶化,甚至分离不完全。本文针对该问题,提出了一种基于EMD、主成分分析(Principal Component Analysis,PCA)和独立分量分析(Independent Component Analysis,ICA)的单通道盲源分离算法(简称EP-ICA算法)。该算法利用EMD得到本征模函数分量(intrinsic mode function,IMF)分量,针对出现模态混叠的IMF分量,利用信号的周期性构造其多路信号,利用ICA消除模态混叠,利用PCA和互相关性剔除多路信号中的虚假分量,并将剩余分量信号与观测信号构成新的多路信号,最后利用N-Fast ICA实现盲源分离。仿真结果表明EP-ICA算法优于已有的基于EMD的SCBSS算法。最后,研究了基于变分模式分解(Variational Mode Decomposition,VMD)的SCBSS算法。将VMD引入SCBSS算法中,提出了基于VMD的SCBSS(VMD-SCBSS)算法;同时将反馈机制应用于VMD方法中,提出了一种基于反馈VMD的SCBSS(VMDF-SCBSS)算法。仿真结果表明,VMD-SCBSS算法和VMDF-SCBSS算法的分离性能优于EP-ICA算法,VMDF-SCBSS算法具有与VMD-SCBSS算法相当的分离性能,但该算法无需预知源信号中心频率差值,能够自动确定源信号数,算法的运算复杂度低于VMD-SCBSS算法。
[Abstract]:Blind source separation (BSS) technology refers to the process of separating only the individual source signals from the observation signal in the case where both the transmission channel and the source signal are unknown. The multi-channel BSS algorithm has been widely used in the fields of biomedical signal processing, array signal processing, mobile communication and text analysis and processing. In recent years, the problem of single-channel blind source separation has become a hot topic in the field of signal processing. Single-channel blind source separation (SCBSS) is the extreme case of the problem of the separation of blind source. It only uses the characteristic information of single-channel observation signal to separate the multi-channel source signal, which is very difficult to solve. However, the SCBSS is a common problem in many practical systems, so it is of great theoretical and practical value to study the SCBSS algorithm. This paper mainly studies the multi-channel BSS algorithm and the SCBSS algorithm. First, the improvement of the FastICA algorithm is studied. In order to solve the problem of more iteration times and degradation of the original FastICA algorithm, a Pm-FastICA algorithm is proposed. Pade approximation to the non-linear function in the algorithm results in a rational function that can reduce the number of iterations of the FastICA algorithm, and the convergence speed and the separation performance are improved. The simulation results show that the performance of Pm-Fast ICA is better than that of the FastICA algorithm, and the performance advantage of the Pm-FastICA algorithm will be more obvious with the increase of the number of source signals. In this paper, a FastICA (N-FastICA) algorithm using the nonlinear function of rational polynomial is proposed. The simulation shows that the performance of the N-Fast ICA algorithm is better than that of the Pm-FastICA algorithm and the FastICA algorithm, and the performance advantage of the N-Fast ICA algorithm will be more obvious with the increase of the number of source signals. Secondly, we study the SCBSS (WPT-ICA) algorithm based on wavelet packet decomposition. The performance of the SCBSS algorithm based on wavelet transform is to be improved because of the fact that the wavelet transform does not well represent the large amount of detail information. In this paper, an SCBSS algorithm based on wavelet packet decomposition is proposed in this paper. And carrying out wavelet packet decomposition on the observation signal, selecting a coefficient with higher energy percentage to reconstruct, and combining the reconstructed signal and the observation signal to form a multi-channel signal, and utilizing the N-FastICA algorithm to realize the blind source separation of the signal. The simulation results show that the SCBSS algorithm based on wavelet packet decomposition is superior to the SCBSS algorithm based on wavelet decomposition. Then, the SCBSS algorithm based on the empirical mode decomposition (EMD) is studied. The existence of the mode aliasing in the SCBSS algorithm based on EMD leads to the deterioration of the separation performance and even the incomplete separation. In this paper, a single-channel blind source separation algorithm based on EMD, Principal Component Analysis (PCA) and Independent Component Analysis (ICA) is proposed. The method uses EMD to obtain an eigenmode function (IMF) component, And the residual component signal and the observation signal form a new multipath signal, and finally, the blind source separation is realized by using the N-Fast ICA. The simulation results show that the EP-ICA algorithm is superior to the existing EMD-based SCBSS algorithm. Finally, the SCBSS algorithm based on the variational mode decomposition (VMD) is studied. In this paper, the VMD-based SCBSS (VMD-SCBSS) algorithm is proposed, and a SCBSS (VMDF-SCBSS) algorithm based on the feedback VMD is proposed. The simulation results show that the separation performance of the VMD-SCBSS algorithm and the VMDF-SCBSS algorithm is better than that of the EP-ICA algorithm, and the VMDF-SCBSS algorithm has the same separation performance as the VMD-SCBSS algorithm, but the algorithm does not need to predict the source signal center frequency difference, can automatically determine the source signal number, and the operation complexity of the algorithm is lower than the VMD-SCBSS algorithm.
【学位授予单位】：杭州电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN911.7

【参考文献】