基于扩展Infomax算法的变步长在线盲分离

由于扩展的Infomax算法需要一定的样本长度来估计信号的峭度,所以常采用离线批处理的方式,但这种方法不能处理混合矩阵发生变化的情况。通过判断系统混合矩阵是否改变,改进了在线估计峭度的模型,同时为解决在线算法中收敛速度和稳态误差的矛盾,提出了一种新的步长更新算法,该算法根据信号的分离状态与峭度曲线收敛程度的关系,采用峭度方差为参数来控制步长的变化,使得步长的选择与分离状态相结合,减小了稳态误差,仿真结果证实了该算法的有效性。

Variable Step-size Algorithm for On-line Blind Source Separation Based on Extended Infomax

The Extended Infomax algorithm usually separates the signals off-line,because it requires adequate samples to estimate the kurtosis of the outputs. However,the batch processing will fail when the mixed channel changes. An improved on-line method to estimate the kurtosis was introduced,which was according to the detection of mixed channel matrix,mean while,a novel variable step-size algorithm was proposed to solve the tradeoff between convergence speed and steady-state error. This step-size update regulation was controlled by the kurtosis covariance of outputs,as reason was that the fluctuation of the kurtosis could describe the state of separation. This adaptive algorithm decreased the steady-state error efficiently. The simulations have verified its validity.