In this paper, a fast algorithm for blind sparse source separation is proposed. Sources are estimated by means of minimizing (0-norm which is approximated using a predefined continuous and differentiable function. The proposed algorithm is easy to implement and runs fast. Then the algorithm is compared with several fast sparse reconstruction algorithms such as fast (1-norm minimization algorithm and OMP using synthetic data. Finally, we apply the proposed algorithm to underdetermined blind source separation using real world data. It is experimentally shown that the proposed algorithm runs faster than other algorithms, while acquiring almost the same (or better) quality.%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近似l0范数,采用最小化l0范数的方法实现对稀疏源信号的估计.该算法的特点是实现简单,速度快.采用人工生成的信号将算法与通过l1范数最小化的快速稀疏信号重构算法和OMP算法进行了比较.最后,将该算法用于实际信号的欠定盲源分离.仿真实验表明,算法在保证信号分离性能的前提下大幅度提高了算法的运行速度.
展开▼