一种结构噪声抑制的非线性滤波器算法及模型

Algorithm Deduction and Model Design of Non-linear Filters for Structure Noise Reduction

针对线性滤波器难以滤除超声回波信号中结构噪声的问题,从分析超声回波结构噪声和缺陷信号的产生、传播衰减及频率响应模型出发,推导了一种基于神经网络NARX(Nonlinear Auto-Regressive Exogenous Input)结构的非线性滤波器模型及其改进RTRL(Real Time Recurrent Learning)算法,该算法模型动态地建立了超声回波与缺陷信号之间的数学映射关系,利用该映射关系实现对缺陷信号中结构噪声抑制。仿真实验证实了该算法模型建立的正确性和有效性。

In the ultrasonic signal processing, the structure noise was difficult to be removed by linear filters. A neural network non-linear filters NARX （Nonlinear Auto-Regressive Exogenous Input） model with its improved RTRL （Real Time Recurrent Learning） algorithm was deduced from the analysis of the generation, attenuation of propagation, and frequency response model of the flaw signals with structure noise. The mathematic map between the clutter and flaw signals was built by the neural network through a dynamic process, and based on the map, the structure noise reduction from the flaw signals was realized. The correctness and the validity of the models with its algorithms were provided by the simulations.