基于多个时间点联合解相关的卷积盲源分离

Convolutive blind source separation based on multiple times and decorrelation

实际信号的混合均为卷积混合,且信号是非平稳的。盲源分离的目标就是找到一组分离滤波器,使得源信号的估计信号互相统计独立。结合信号的非平稳性,利用二阶解相关原理,文章阐明了一种在频域实现卷积混合的盲源分离算法,并且考虑了噪声对分离性能的影响。为了避免频点排列次序的不确定性,利用了多阶段盲源分离思想。利用该算法,对两路混合的实录水声信号进行盲分离,得到了两路源信号的估计信号,通过对估计信号的分析,利用信噪比提高率这一标准,验证了该算法的有效性。该算法收敛速度快,精度高,可用于浅海环境下实录水声混合信号的盲分离。

Mixed signals in practice can be viewed as sums of differently convolved sources, and the signals are non-stationary. The task of blind source separation is to obtain a set of separation filters and make the estimated signals of sources statistically independent. This paper discusses a convolutional blind source separation algorithm based on second-order decorrelation, taking into account non-stationarity of signals. Influence of noise on the quality of separation is considered as well. To avoid inconsistency of frequency bin permutation, a multi-resolution approach to blind source separation is studied. The algorithm is used to separate real acoustic signals successfully. Experimental results are presented and separation performance analyzed. Validity of the algorithm is shown by the improvement of SNR. The algorithm converges rapidly and has high precision. It can be used to separate actual signals recorded in shallow sea.