一种基于多层神经网络的盲信号分离算法

基于多层神经网络,提出一种盲信号分离算法.该算法不对信号的密度模型做任何假设,通过多层神经网络估计任意信号的概率密度函数,并由此估计信号的评价函数.同其他方法相比,该方法不仅具有更好的分离性能,而且收敛速度较快.该方法可直接应用于所有以非线性函数代替评价函数的盲信号分离算法.实验验证了方法的有效性.

Blind Source Separation Algorithm Using Multilayer Networks

An algorithm for blind source separation based on multilayer networks is proposed.Using a multilayer network density estimation technique,the algorithm may estimate the unknown probability density functions and its derivative of the score functions of the source signals and the algorithm is truly blind to the particular underlying distribution of the mixed signals.The new algorithm not only outperforms other methods,but also the approach has a better convergence property.The method can be applied to all the blind source separation algorithms where the score function is obtained by a nonlinear function.Simulation results show good performances of the proposed algorithm on both demixing and convergence to the desired solutions.