基于凸组合的同步长最大均方权值偏差自适应滤波算法

Maximum mean square deviation adaptive filtering algorithm with the same step-size via convex combination

针对NLMS和PNLMS滤波器对时变信道跟踪能力差的缺点,提出了一种同步长凸组合最大均方权值偏差(MSD,mean square deviation)算法。该算法将同步长的NLMS和PNLMS 2种不同类型的自适应滤波器进行凸组合,以最大均方权值偏差为准则,使新的滤波器能够在外界信道特性(稀疏、非稀疏和模糊态)时变的情况下,保持良好的随动性能,并在收敛的各个阶段均保持快速且稳定的均方特性。理论推导和仿真实验表明:该算法与NLMS、PNLMS和IPNLMS算法相比,在稀疏和非稀疏状态时能够保持四者中最快的收敛速度,并且在模糊状态时算法性能优于其余三者。另外,该算法仍保持较好的稳态均方性能。

Aimed at poor tracking performance of NLMS filter and PNLMS filter under time-varying channel,a same step-size convex combination of the maximum mean square deviation algorithm was presented.The algorithm convexly combined two different adaptive filters with the same step-size based on a criterion of maximum mean square deviation.So the proposed filter could keep good dynamic performance in the time-varying channel and stability of mean square characteristics in convergence stage.Theoretical analysis and simulation results show that in the sparse and non-sparse state the proposed algorithm indicates the fastest convergence rate compared with NLMS,PNLMS and IPNLMS algo-rithm.In the fuzzy state,the performance of proposed algorithm is superior to the above three.Additionally,the steady-state performance of mean square also keeps well.