基于线性预测的固定点盲源分离算法

Fixed Point Blind Source Separation Algorithm Based on Linear Prediction

将可预测性作为盲源分离的代价函数,可以提取具有不同时序结构的源信号。但这种算法的批处理形式收敛速度较慢,本文针对这种算法进行了改进,根据Kuhn-Tucker条件,提出了基于线性预测的固定点盲源分离算法。与A.K.Barros等提出的批处理算法相比,基于线性预测的固定点批处理算法具有更快的收敛速度。本文提出的固定点批处理算法具有运算量低、无需设置步进长度的优点。实际采集的语音数据试验验证了该算法的有效性。

The source signal with different temporal structures can be extracted by the cost function of predictability in the blind source separation. But the batch version of the algorithm slowly converges. A improved point blind source separation algorithm is proposed to accelerate the convergence rate of the algorithm under the Kuhn-Tucker condition based on the same cost function. Compared with the algorithm proposed by A. K. Barros and A. Cichocki et al, the proposed algorithm converges faster. Moreover, the algorithm has the advantages of no need of step size and low computation. And the experiment on real data also shows the validity of the proposed algorithm.