用二次判据神经元实现的神经滤波器

Neural Filter Implemented by an Artificial Neuron With Quadratic Criterion Function

基于神经网络理论及滤波理论提出了一种新型神经滤波器———广义自适应神经滤波器 (GANF)。GANF是用神经运算器代替级叠滤波器中的布尔运算器后得到的一类非线性自适应神经滤波器。讨论了其结构与特性 ,证明了最优GANF的平均绝对误差 (MAE)上限即为最优级叠滤波器的MAE。同时 ,通过计算机仿真实验 。

On the basis of neural network theory and filter theory this paper presents a new neural filter-generalized adaptive neural filter(GANF).GANF is a class of nonlinear adaptive filter achieved by replacing the Boolean function operator with a neural operator. The paper discusses its structure and properties and proves that the upper bound of the mean absolute error(MAE)of the optimal GANF is the MAE of optimal stack filter. Moreover it verifies the superioriy of GANF in Gaussian noise suppression.