欠定盲分离时变混合矩阵的估计

摘要

针对欠定盲分离中时变混合矩阵的估计问题，在稀疏域二维最小偏差角算法的基础上，提出了一种改进的欠定盲分离时变混合矩阵估计算法。该算法通过判断原始阵各列上是否都有观测点聚集和聚集在原始阵上的观测点以外的点的聚集方向，来检测变化时刻；并利用基于点密度大区域检测算法估计混合矩阵。改进算法对于混合矩阵发生某些列增加、消失和变化时均能检测出变化，并且在大幅提高变化时刻检测概率和混合矩阵估计精度的同时，降低了复杂度。实验仿真结果表明，在20 dB信噪比时，混合矩阵估计精度提高了60%以上。%Aiming at the problem of time-varying mixing matrix estimation in underdetermined blind source separation, an improved time-varying mixing matrix estimation algorithm is presented based on the method of planar minimum offset angle of sparse domain. By determining whether the original columns has observed the signals and the direction of the observed signals that are not in the direction of the original matrix the change time instant can be detected. The mixing matrix can be estimated by the method called the large point density area detecting. The improved algorithm can detect the changes of the columns increasing, disappearing and changing. Moreover, the detection probability of the changed time instant and the estimation accuracy of the mixing matrix are substantially increased, while the complexity of the algorithm is decreased. Experimental results show that when SNR is 20 dB, the estimation accuracy of the mixing matrix is improved by more than 60%.