基于帕斯维尔定理的频域积分盲源分离算法

Frequency-domain blind separation of convolutive mixtures based on Parseval's theorem

为了克服卷积混合信号盲源分离双最小均方(Double least mean squeres,Double-LMS)算法在分离滤波器过长时计算量过大的问题,借助于傅里叶变换理论中的帕斯维尔定理,将其转化为频域积分算法。频域积分算法可以利用快速傅里叶变换实现,具有较高的计算效率,可以克服当分离滤波器过长时原算法效率低下的问题。仿真结果表明:新算法在保持了DoubleLMS算法良好分离性能的基础上,降低了原算法的复杂度,提高了计算效率。

To improve the performance of the blind source separation double least mean squeres（ Double-LMS） algorithm which is computationally inefficient if the separation filters are too long,the Double-LMS is converted to a frequency-integration algorithm with the help of Parseval＇s theorem.Fast Fourier transform is exploited to implement the proposed algorithm to overcome the low computational efficiency problem due to the too long separation filters. Simulations show that the new algorithm works as well as the Double-LMS algorithm and has higher computation efficiency.