中文摘要：

分析了已有源信号数目未知的盲信号分离算法中分离矩阵的动态变化，找到了分离矩阵范数趋于无穷大而导致算法发散的原因：分离矩阵在某一等价类中的冗余移动．利用正交投影方法消去了自然梯度中引起冗余移动的冗余分量，推导出源信号数目未知的投影自然梯度算法．新算法具有与自然梯度算法相同的收敛速度，而且克服了已有算法不能稳定收敛的缺点．仿真验证了新算法的分离性能和收敛稳定性．

英文摘要：

An analysis of the demixing matrix＇s dynamic changing in available blind source separation algorithms with an unknown number of sources is made and the conclusion that it is the redundant movement of demixing matrix in some equivalent class which cases the divergence is drawn. By canceling the redundant component of the natural gradient in an orthogonal projection approach, a new algorithm is developed, which overcomes the weak point of being unable to converge stably. Finally, simulation proves the capacity to perform the blind source separation with an unknown number of sources and the convergent stability of the new algorithm.