基于平方根无迹卡尔曼滤波的混沌信号盲分离

Blind Source Separation of Chaotic Signals Based on Square Root Unscented Kalman Filters

针对现有卡尔曼盲分离算法在分离混沌信号时性能较差的问题,提出了基于平方根无迹卡尔曼滤波器(SRUKF)的混沌信号盲分离方法。该方法采用递推方式实现,在每一次递推中,首先将分离向量作为状态变量进行卡尔曼估计,然后将分离向量视为已知量,再次利用SRUKF重建源信号,从而得到源信号在最小均方误差意义下的优化估计。实验仿真表明,所提算法能够快速收敛,并且在噪声环境下估计误差比现有的卡尔曼盲分离方法明显减小。

The existing blind source separation methods in the framework of Kalman filtering suffers great per-formance degeneration when applying to chaotic mixtures with a low signal to noise ratio.To solve this prob-lem,a new two-step recursive approach based on square root unscented Kalman filters was proposed.In every recursion,the separation vector was firstly estimated through a square root unscented Kalman filter as its state varibles.Then in the second step,the chaotic souce was estimated again through a Kalman filter other than di-rectly computed by mutiplying the observations to the former obtained separation vector.Thereby,an optimal estimaion of the chaotic source was obtained.A simulation example was designed in comparison with existing Kalman based blind separation methods.Simulation results indicated that the proposed method performed better than existing unscented Kalman approaches.