基于贝叶斯网络理论的时变混合盲源分离算法

以状态空间模型作为信道的变化模型，研究了时变混合情况下非平稳信号的盲分离问题。首先将隐马尔可夫模型（HMM）和混合高斯（MOG）模型结合起来对具有动态结构和复杂分布的非平稳源信号进行建模，然后运用贝叶斯网络理论处理信道时变情况下独立成分分析（ICA）模型中各变量和参数之间的关系，提出了一种基于贝叶斯推断的可同时完成混合矩阵盲估计及源信号盲分离的算法，通过采用逼近方法有效地减小了算法计算量。计算机仿真试验证明本文算法的有效性。

Bayesian Network Theory Based Blind Source Separation from Time-Varying Mixture

By combining HMM with MOG to model the dynamic and complex-distribution nonstationary signal, and processing relationship between hidden variables and parameters of ICA model with Bayesian network theory, an algorithm based on Bayesian inferring for the blind source separation from time-varying mixture is proposed. The approximation method is utilized to reduce the calculation amount. The simulation experiment shows the efficiency of the method.