一种适用于信源数时变的自适应盲源分离算法

Adaptive blind source separation algorithm for time-varying number of sources

针对信源数动态变化情况下的盲源分离问题,首先采用一种基于交叉验证技术的估计方法用于估计时变的源数,然后推导了一种基于自然梯度和Frobenius范数相结合的自适应盲源分离算法,该算法不需要对源信号作任何约束性的假设,因此该算法适合于分离服从超高斯或亚高斯分布的信号。提出的算法通过了源数不变和源数动态变化2种方式实验的验证。

Aiming at the blind source separation problem of time-varying number of sources,a dynamic source number estimation method based on cross-validation technique is proposed.Then,an adaptive blind source separation algorithm based on natural gradient and Frobenius norm is deduced.The innovative blind separation algorithm does not require the assumption of any restrictions or constraints on source signals;therefore it is suitable for separating the sources obeying super-and sub-Gaussian distributions.At last,the effectiveness of the proposed method was verified in the simulation experiments for time-invariant and time-varying number of sources.