针对传统基于随机初始化的变分贝叶斯独立分量分析(VBICA)方法的不足,即随机初始化导致经过不同的学习得到的分离结果存在差异性,提出了一种基于主分量分析(PCA)的变分贝叶斯独立分量分析的盲源分离方法,在提出的方法中,利用PCA来初始化模型参数。并与传统的变分贝叶斯独立分量分析方法进行对比。仿真结果验证了该方法的有效性,提出的方法不仅比传统的VBICA方法取得了更好的分离性能,并且保持分离结果的稳定性,克服了传统的VBICA方法的不足。
For the deficiencies of the traditional variational Bayesian independent component analysis (VBI- CA),i. e. the differences resulting from the random initialization in the separation results from different learnings. A blind source separation method based on principal component analysis (PCA) and VBICA was proposed,where PCA was used to initialize the model parameters. The proposed method was compared with the traditional VBICA method. The simulation results verified the effectiveness of this method,which was superior to the traditional VBICA in the separation performance and the stability of the separation resuits. It also overcame the deficiencies of the traditional VBICA.