比例最小均方/四阶算法及其在系统辨识中的应用

Proportionate least mean square/fourth algorithm and its application for system identification

针对非高斯噪声环境下稀疏系统参数辨识问题,提出一种基于比例更新机制的最小均方/四阶(LMS/F)自适应滤波算法(PLMS/F)。该方法以混合均方/四阶准则(MS/FE)为代价函数,其包含了误差的高阶项,具有解决非高斯噪声问题的优势。引入比例更新机制,从而可根据算法当前时刻权重变化特征来调整各权重参数的步长,因此具有良好的跟踪性能。使用梯度下降法设计了阈值参数自适应更新机制以进一步改进算法稳态性能。此外,分析了所提算法的平均和均方收敛性。应用具有稀疏特征的FIR系统参数模型对所提算法实现了在非高斯噪声环境中的参数辨识。仿真实验结果表明,该算法可以有效辨识模型参数,且具有较低的稳态误差和强的鲁棒性。

For sparse system parameter identification problem under non-Gaussian noise environment,the least mean square/fourth(LMS/F)algorithm based on proportionate update scheme,namely PLMS/F,is proposed.The PLMS/F algorithm with the mixed square/fourth error criterion(including higher order term of error)has advantage of solving non-Gaussian problem.In addition,it has good tracking ability via the introduced proportionate update scheme which adjusts the step size of each weight parameter according to the weight change feature at instant time.A threshold adaptively update approach is developed to further improve the performance of the PLMS/F algorithm.Furthermore,we perform the mean and mean square convergence analysis of the proposed algorithm.Taking the FIR system parameter model with sparse feature as the object,the identification under the non-Gaussian environment is realized by using the proposed algorithm.Simulation results show that the proposed PLMS/F and BCPLMS/F algorithms can identify the model parameter efficiently,and have lower steady-state misalignment and stronger robustness in comparison to other algorithms.