依据迭代系数状态因子分段的变步长NLMS算法

针对固定步长的归一化LMS算法（NLMS）存在不能同时兼顾收敛速度与稳态误差的问题，本文提出一种依据迭代系数状态因子进行分段的变步长NLMS算法。该变步长NLMS算法采用迭代系数状态因子作为表征迭代系数与实际系数的逼近状态的指标。当迭代系数状态因子值大于1，则说明迭代系数有偏离真实系数的趋势，此时采用步长因子较大的变步长方案；反之，说明迭代系数有逼近真实系数的趋势，应该采样步长因子较小的变步长方案。这样的自适应选择措施使得算法具有较强的收敛能力。理论分析和实验表明:在同样实验条件下，本文算法能够获得比其他文献更快的收敛速度和更小的稳态误差。 

Segmental Variable Step-Size Normalized LMS Algorithm Dependent on the State Parameter of Interation Coefficient

Fixed step size Normalized LMS algorithm can’t step out the dilemma of fast convergence rate and low excess mean-square error. To solve this problem, this paper proposed a Segmental Variable Step-Size Normalized LMS algorithm Dependent on the state parameter of iteration coefficient. This Normalized LMS algorithm employs the state parameter of iteration coefficient to express the approximation degree between iteration coefficient and real coefficient. When the value of state parameter of iteration coefficient is larger than 1, which indicates that iteration coefficient tends to deviate from the real coefficient, at this time, variable step-size scheme that can provide larger step-size parameter is needed. However when the value of state parameter of iteration coefficient is smaller than 1, which indicates that iteration coefficient tends to approach the real coefficient, at this time, variable step-size scheme that can provide smaller step-size parameter is needed. This adaptive selection of variable step-size scheme enables the proposed NLMS algorithm to have better convergence performance. Analysis and experiment results show that: under the same experimental condition, the proposed algorithm can obtain faster convergence rate and lower excess mean-square error than other literatures.