中文摘要：

针对盲源分离的初始化问题，提出一种盲源分离初始化方法。该方法首先对接收到的混合信号进行小波稀疏分解，然后选取稀疏性最好的分解系数组，并在其星图中通过聚类方法寻求聚轴来估计混合矩阵。最后，采用这一混合矩阵估计值对FastICA算法进行初始化。仿真实验表明，该初始化方法能避免盲源分离算法收敛时陷入局部最小，加快算法收敛，同时使盲源分离算法的分离精度提高10～26dB。

英文摘要：

An initialization method for blind source separation was proposed. By using wavelet transform, the received mixed signals were firstly decomposed into a low frequency subband and a set of multisacle high frequency subbands with different sparsity. Then the best sparse subbands were chosen to detect the cluster centers in their seatter plot, and get the estimation of the mixing matrix. Lastly, the blind source separation was initialized by the estimated mixing matrix. Simulation results show that the blind source separation algorithm using the proposed initialization method can obtain a faster convergence rate, and a separation precision improvement about 10 to 16 dB in PSNR.