基于高阶累积量的格型结构及算法

为了解决格型结构对噪声敏感的问题，从基于高阶累积量的均方误差（ＣＭＳＥ）准则出发，提出了一种基于高阶累积量的格型（ＣＬ）结构，并讨论了该结构具有的一些重要性质。在此基础上进一步推证了系统参数辨识的基于高阶累积量的Ｂｕｒｇ算法（ＣＢｕｒｇ），并给出了三阶ＣＢｕｒｇ算法的一种快速递归实现方案。仿真结果证明：就辨识结果的无偏性而言，这种ＣＢｕｒｇ算法的抗高斯噪声性能明显优于常规Ｂｕｒｇ算法而两种算法的运算量大体相当。

Cumulant-based lattice structure and algorithm

On the basis of Cumulant based Mean Square Error (CMSE) criteria, a new Cumulant based Lattice (CL) structure was proposed and some important properties associated with it were given. These properties are, in some ways, similar with those of conventional correlation based counterpart. Furthermore, this paper put forward a Cumulant based Burg (CBurg) algorithm and a fast recursive version in the case of the 3 rd order cumulant used for system identification. Simulation results show that, with trivial increasing in computational complexity, the proposed CBurg algorithm behaves rather better than the traditional Burg algorithm, in terms of unbiased results of identification, particularly in the lower SNR environment of Gaussian additive noise. Thus, the proposed CBurg algorithm can be used as an efficient tool to guarantee robust system identification for none Gaussian time series.