一种改进的势函数欠定盲源分离算法

Improved laplace mixed model potential function algorithm for UBSS

针对原有的拉普拉斯混合模型势函数法复杂度高、随机选取部分观测数据点作为初始聚类中心的算法聚类结果不稳定、准确率低的问题,提出了一种改进的势函数欠定盲源分离算法.该算法在基于密度概念的基础上,以簇内距离小、簇间距离大为原则,选取部分高密度点作为势函数的初始聚类中心.理论分析与仿真实验表明,改进算法的复杂度大大降低,而估计准确度降低很少.在信噪比为10dB时,该算法仿真时间降为原始势函数法的5%;相对随机选取算法,在计算复杂度基本一致的前提下,该算法的估计准确度大大提高,源信号个数估计准确率由61%提高到85%,混合矩阵估计误差由0.47下降为0.27.

Aiming at the problem that the original Laplace Mixed Model Potential Function（LMMPF） algorithm has high complexity and the random initial cluster center algorithm has a low accuracy and stability,we propose an improved LMMPF algorithm.Based on the concept of density,we can choose some high-density data as the initial cluster centers.These data obey the principle that the distance between the data in the same group is small and the distance between groups is great.Theoretical analysis and experimental results show that compared to the original LMMPF algorithm the complexity of the new algorithm becomes much lower while the estimated accuracy is reduced only a little bit.When the Signal to Noise Ration（SNR）is 10 dB,the running time of the improved algorithm is reduced to 5%.Compared to the randomly-chosen algorithm,the new algorithm has a much higher accuracy：the accuracy rate of estimating the number of sources is raised from 61% to 85% and the mixing matrix estimated error is reduced from 0.47 to 0.27.