神经网络自适应噪声抵消系统的性能比较与仿真  

PERFORMANCE COMPARISON AND SIMULATION OF ADAPTIVE NOISE CANCELLATION SYSTEM BASED ON NEURAL NETWORK

作　　者：张磊[1] 李方鑫 王建新[1] 肖超恩[1] Zhang Lei;Li Fangxin;Wang Jianxin;Xiao Chaoen(Beijing Electronic Science and Technology Institute,Beijing 100070,China)

机构地区：[1]北京电子科技学院,北京100070

出　　处：《计算机应用与软件》2018年第12期263-268,共6页Computer Applications and Software

基　　金：中央高校基本科研业务费专项(2014GCYY04);北京市自然科学基金项目(4163076)

摘　　要：线性自适应滤波算法(LMS、NLMS、RLS)对非线性噪声抵消效果较差。针对这一问题,研究神经网络自适应噪声抵消系统中不同隐含层神经元节点数、不同隐含层传输函数、不同神经网络学习算法以及不同信噪比原始输入下系统的噪声抵消性能。建立结构为单层隐含层,且输入层、隐含层和输出层节点数为1-N-1结构的神经网络模型。通过仿真分析,优化神经网络自适应噪声抵消系统中,隐含层节点数经验公式的参数取值。结果表明该系统中噪声抵消效果受到神经网络结构的影响;对于原始输入信噪比在2～10 dB的信号,参考输入与原始输入中噪声非线性相关;选择传输函数为tansig,神经网络隐含层节点数使用优化后的参数取值,输出信号信噪比提高了1. 0～1. 5 dB。The linear adaptive filtering algorithm such as LMS, NLMS and RLS has a poor performance in handling nonlinear noise cancellation. To solve this problem, we studied the noise cancellation performance in the neutral network adaptive noise cancellation system with number of nodes in different hidden layers, transfer functions in different hidden layer, different neural network learning algorithms and originated input with different signal-to-noise ratio （SNR）. We established a neural network model with single hidden layer, in which the number of nodes in input layer, hidden layer and output layer was 1-N-1. Through simulation analysis, the range of parameter in empirical formula for the number of nodes in hidden layer was optimized in the adaptive noise cancellation system based on neural network. The results show that the structure of neural network has influence on the noise cancellation performance. For the original input signal with SNR of 2~10 dB, the reference input is nonlinearly related to the noise in original input, tansig is selected as the transfer function, and the number of hidden layer nodes in the neural network uses the optimized parameter value, and the SNR of the output signal improves by 1.0~1.5 dB.

关 键 词：神经网络 自适应噪声抵消系统 隐含层节点数 信噪比 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]