动量项盲源分离算法及其性能优化策略

动量项技术是用来改善自适应盲源分离算法分离性能的有效手段，但算法在融入动量项后，其收敛特性对于动量因子的选取数值较为敏感，且算法的稳态性能仍要受到步长参数的限定.本文首先给出了动量项盲源分离算法的设计原理，分析了现有算法存在的两个缺陷性问题；然后利用梯度下降法构造了具有在线调整特性的动量因子自适应迭代规则，通过对动量因子的实时更新以消除固定动量因子算法的性能缺陷；在此基础上，基于凸组合理论设计了不同步长参数下两个变动量因子算法的自适应优化组合方案，从而在一定程度上缓解了步长参数对于算法性能的限定.在不同环境下进行的仿真实验表明，本文针对动量项盲源分离算法所设计的优化策略能够有效消除其所存在的缺陷问题，在确保快速收敛的同时，又能获取较小的稳态误差.

Momentum Term Based Blind Source Separation Algorithm and Its Performance Modified Strategies

Momentum term technology is an effective solution to improve the performance of the adaptive blind source separation (BSS) algorithm,but the convergence property of the momentum term based BSS algorithm is very sensitive to the fixed momentum factor,and its performance in steady state is also restricted by the step size. Firstly,the principle of the momentum term based BSS algorithm as well as its two disadvantages were presented and analyzed in this paper. Then,in order to eliminate the first disadvantage of the momentum term based algorithm using the fixed momentum factor,we structured a variable momentum factor algorithm with the adaptive adjustment property based on the gradient descent method. On this basis,by virtue of the convex combination theory,an adaptive combination of tow variable momentum factor algorithms with different step size was proposed to alleviate the performance restriction caused by the step size. The simulation results in different conditions demonstrate that the proposed modified strategies got the optimization balance between the fast convergence speed and small steady-state error,and effectively avoid the two drawbacks of the momentum term based BSS algorithm.