Spline nonlinear adaptive filters considering cross terms
Spline nonlinear adaptive filters considering cross terms
复制标题
DOI：
10.1016/j.sigpro.2021.108054
复制
发表时间：
2021
期刊：
影响因子：
--
通讯作者：
Shuji Saitou;Y. Sugita
中科院分区：
文献类型：
--
作者：
Shuji Saitou;Y. Sugita
最近提出的广义样条非线性自适应滤波器（GSNAF）可以建模系统的硬非线性低计算成本。在本文中，样条激活函数（SAF）的第一次扩展到一个二维函数的两组输入，以及工作在建模的非线性系统的交叉项是产品的时移输入信号。然后，为了防止更新的控制点的意义不大，更新算法推导出通过结合L1范数惩罚的控制点的平方误差成本函数。一些仿真结果表明，该滤波器可以提供相当或更好的性能相比，三阶沃尔泰拉滤波器，GSNAF。
The recently proposed generalized spline nonlinear adaptive filter (GSNAF) can model systems with hard nonlinearity at low computational cost. In this paper, a spline activation function (SAF) is first extended to a two-dimensional function with two sets of inputs to work well in modeling nonlinear systems with cross terms that are products of the time shifted input signals. Then, to prevent the updating of the control points of little significance, the updating algorithm is derived by combining a L1-norm penalty on the control points with the squared error cost function. Some simulations show that the proposed filter can provide comparable or better performance compared to third-order Volterra filters, and GSNAF.