Abstract: A new two-step algorithm for underdetermined source separation is proposed. Mixing matrix is estimated using clustering methods. Sources are estimated using a fast sparse reconstructed algorithm which defines a continuous and differential function so as to approximate [?0]-norm. The new algorithm runs fast and is easily implemented. It is experimentally shown that the proposed algorithm runs faster than other two underdetermined source separation algorithms using fast minimization [?1]-norm and OMP methods, while acquiring almost the same quality.

Key words: underdetermined blind source separation, sparse component analysis, two-step method, [?0]-norm

摘要： 提出了一种基于两步法的欠定盲源分离新算法。在混合矩阵估计阶段，采用基于势函数的聚类方法，在源信号恢复阶段，提出一种快速的稀疏信号重构算法，通过定义一个连续可微函数来近似[?0]范数，使得[?0]范数可解。该算法的特点是实现简单、速度快。仿真实验表明，与现有的采用快速[?1]范数最小化和OMP算法的欠定盲源分离方法相比，提出的算法在保证分离性能的前提下大幅度提高了算法的运行速度。

关键词: 欠定盲源分离, 稀疏分量分析, 两步法, [?0]范数