融合动量项的步长自适应盲源分离算法  

Step-size adaptive blind source separation algorithm with adding momentum term

作　　者：李春腾 蒋宇中[1] 刘芳君 张曙霞[1] LI Chun-teng;JIANG Yu-zhong;LIU Fang-jun;ZHANG Shu-xia(College of Electronic Engineering,Naval Univ. of Engineering,Wuhan 430033,China;Academy of Mathematics and Computer Science,Yunnan Nationalities Univ.,Kunming 650500,China)

机构地区：[1]海军工程大学电子工程学院,武汉430033 [2]云南民族大学数学与计算机科学学院,昆明650500

出　　处：《海军工程大学学报》2019年第3期107-112,共6页Journal of Naval University of Engineering

摘　　要：为进一步缓解盲源分离算法收敛速度与稳态误差之间的矛盾,首先在自然梯度算法的基础上,通过融合动量项改善算法的收敛速度,基于分离性能指标的步长自适应减小稳态误差;然后,给出了所提算法的模型图,同时考虑分离性能和计算复杂度,选择合适的融合动量项算法,并设计了算法的近似最优参数,有效避免了算法的分段收敛;最后,合理选择步长与动量项的权重系数,有效改善了分离性能与收敛速度。仿真结果表明:该算法在一定程度上缓解了上述矛盾,并具有较低的计算复杂度。In order to further alleviate the contradiction between the convergence speed and the steady-state error of blind source separation algorithm,on the basis of the natural gradient algorithm,the convergence speed of the algorithm is improved by adding momentum term,and the steady-state error is reduced by step-size adaptive based on the separation performance index.The model of the proposed algorithm is given,and the appropriate algorithm about momentum term is selected by considering the separation performance and computational complexity.The approximate optimal parameters are designed to avoid the segmented convergence of the algorithm.The reasonable selection weight coefficients between the step size and the momentum terms effectively improve the separation performance and the convergence rate.The simulation results show that the algorithm alleviates the above contradiction to a certain extent and has low computational complexity.

关 键 词：盲源分离 自然梯度算法 动量项 自适应步长 分离性能指标 计算复杂度 

分 类 号：TN911.7[电子电信—通信与信息系统]