具有非奇异约束的线性卷积混合信号盲分离联合对角化方法

联合对角化能够成功解决盲分离问题, 但在求解时会得到非期望的奇异解, 从而无法完全分离出源信号. 鉴于此, 提出一种用于线性卷积混合盲分离的联合对角化方法, 将卷积混合模型变换为瞬时模型, 并对变换后的模型应用联合对角化求取分离矩阵. 在求解过程中, 引入约束条件对解的范围进行限定, 避免了奇异解的出现. 仿真结果表明, 所提出的方法能够成功实现卷积混合信号盲分离.

Joint diagonalization method with non-singular constraints for blind separation of linear convolutive mixtures

Joint diagonalization can solve the problem of the blind source separation(BSS) approach. However, the method may converge to some unexpected singular solutions, thus the separation process fails in the end. Therefore, a blind source separation method based on the joint diagonalization approach for the linear convolutive mixing model is proposed. In the proposed method, the convolutive mixing model is firstly transformed to the instantaneous mixing model. Then a joint diagonalization method is applied on the transformed model in order to compute out the separating matrix. Meanwhile, in the process of diagonalization, a constraint condition is introduced for the limitation of the class of the separating matrices such that the singular solutions are avoided. Simulation results show that the proposed method can realize BSS of convolutive mixture signals successfully.