研究在线盲信号分离问题,先提出一种递归最小二乘(RLS)由化算法,然后与一种基于自然梯度的RLS信号分离算法相结合,并经合理近似,得到一种新的RLS盲信号分离算法,它不需要对观测数据进行白化预处理,RLS由化算法和RLS信号分离算法的平衡点分析表明,这两种算法具有所期望的收敛性。理论证明了提出的RLS盲信号分离算法具有等变化性和分离矩阵的非奇异性这两个关键性能,仿真实验验证了新算法的有效性。 In this paper, the problem of online blind signal separation is studied. First, a recursive least square (RLS) algorithm is proposed, and then a RLS signal separation algorithm based on natural gradient is proposed. A rational RLS blind signal It does not need to whiten the observed data. The equilibrium analysis of RLS algorithm and RLS signal separation algorithm shows that these two algorithms have the expected convergence. The theory proves that the proposed RLS blind signal separation algorithm has such two key properties as equal variability and non-singularity of separation matrix, and the simulation results verify the effectiveness of the new algorithm.