应用JADE盲分离算法分离统计相关源

在一些如故障诊断等复杂的应用系统中，不相关源与相关源往往以相互混合方式同时存在于传感观测中。由于相关源不满足盲源离理论有关源的统计独立性前提假设，因此现有的盲源分离算法如基于矩阵联合对角化的算法等，无法从混有相关源的传感观测中准确分离源信号(不相关源信号和相关源信号)。本文在基于矩阵联合对角化的盲分离算法的基础上，提出通过对其源估计进行修正的相关源分离算法。理论分析和仿真结果表明：修正后的基于矩阵联合对角化的盲分离算法，能有效地分离包含相关源的混合观测信号。

Application of JADE to Separation of Statistically Correlated Sources

Uncorrelated sources and correlated sour ces often occur in a complex system simu ltaneously, which are mixed up in the me asurements sensor. Algorithms for blind source sparation, such as the one based on joint diagonalization of Eigen Matri ces, are not able to separate correlated sources and uncorrelated sources from t he mixed observations correctly because the requirement of statistical independe nce can not be satisfied. Joint approxim ation diagonalization of eigenmatrices ( JADE) based blind sources algorithm is m odified for separating correlated source from mixed observation in this paper. R esults of the simulation show that the m odified algorithm can separate the corre lated sources and uncorrelated sources f rom the mixed observations effectively.