源数估计对于独立分量分析算法的影响分析

发布时间：2018-01-27 13:37

  本文关键词： 信号处理 独立分量分析 源数估计 自然梯度法 单源点检测法 子空间表示法　出处：《电光与控制》2017年08期 　论文类型：期刊论文

【摘要】：研究了源数估计对于独立分量分析算法的影响。对于正定模型,当估计源数少于真实源数时,模型变为超定模型,采用自然梯度法开展仿真实验;当估计源数多于真实源数时,模型变为欠定模型,采用基于时频单源点检测的混合矩阵估计算法和子空间表示信号恢复算法开展仿真实验。实验结果表明,在满足一定信噪比的条件下,对于正定模型超定化问题,通常有数目等于估计源数的源信号能够成功分离;对于正定模型欠定化问题,通常所有源信号都能正确分离,只是分离信号中出现了1个或多个源信号的拷贝,可以通过检测分离信号的相关性,对拷贝信号进行剔除或合并,对分离效果无影响。研究结论对于独立分量分析算法的应用具有一定参考价值。
[Abstract]:The influence of source number estimation on independent component analysis (ICA) algorithm is studied. For the positive definite model, when the estimated number of sources is less than the real number of sources, the model becomes overdetermined model, and the natural gradient method is used to carry out the simulation experiment. When the estimated number of sources is more than the real number of sources, the model becomes an underdetermined model. The hybrid matrix estimation algorithm based on time-frequency single source point detection and the subspace representation signal recovery algorithm are used to carry out simulation experiments. Under the condition of satisfying certain signal-to-noise ratio (SNR), there is usually a source signal with a number equal to the estimated number of sources which can be separated successfully for the problem of over-determination of the positive definite model. For the problem of unfixed positive definite model, usually all the source signals can be separated correctly, but there are one or more copies of the source signals in the separated signals, which can detect the correlation of the separation signals. Eliminating or merging the copy signal has no effect on the separation effect. The research results have some reference value for the application of the independent component analysis (ICA) algorithm.
【作者单位】： 电子信息系统复杂电磁环境效应国家重点实验室;
【基金】：国家自然科学基金(61372040) CEMEE国家重点实验室开放课题(CEMEE2015Z0302B)
【分类号】：TN911.7
【正文快照】： 0引言盲源分离(Blind Source Separation，

本文编号：1468596