基于决策图贝叶斯的盲源信号分离算法

Blind source separation based on Bayesian optimization algorithm with decision graphs

从混合观测数据向量中恢复不可观测的各个源信号是阵列处理和数据分析的一个典型问题。提出了一种基于决策图贝叶斯的盲源信号分离算法,该算法利用决策图贝叶斯优化算法代替JADE算法中的联合对角化操作,通过构造和学习网络来替代传统遗传算法中的交叉重组和变异等遗传算子,避免了对大量控制参数和遗传算子的人工选择和重要构造块的破坏。仿真结果表明,提出的算法比JADE算法和基于遗传算法的盲源信号分离方法均具有更高的分离精度。

Recovering the unobserved source signals from their mixtures is a typical problem in array processing and data analysis.In this paper,a blind source separation algorithm using Bayesian optimization algorithm with decision graphs is proposed,which uses Bayesian optimization algorithm with decision graphs instead of the joint diagonalization operation in JADE to improve the accurateness of the solutions.The suggested algorithm replaces some genetic operators such as crossover and mutation in traditional genetic algorithms by building and learning Bayesian networks,which avoids setting a lot of parameters manually and destroying some important building blocks.The analysis and simulations suggest that the algorithm has a higher separation accuracy than JADE algorithm and blind source separation based on GA.