基于自适应神经网络的二维线性相位FIR滤波器优化设计  

Optimum design of two-dimensional linear-phase FIR filters based on adaptive neural network

作　　者：李目[1,2] 何怡刚[1] 刘祖润[2] 周少武[2] 

机构地区：[1]湖南大学电气与信息工程学院,湖南长沙410082 [2]湖南科技大学信息与电气工程学院,湖南湘潭411201

出　　处：《电路与系统学报》2011年第2期94-98,共5页Journal of Circuits and Systems

基　　金：国家杰出青年科学基金(50925727);国家自然科学基金项目(50677014;60876022);高校博士点基金项目(20060532016);湖南省科技计划项目(2010J4);广东省教育部产学研项目(2009B090300196)

摘　　要：提出一种基于自适应三角函数基神经网络的二维线性相位FIR滤波器优化设计方法。该方法根据二维线性相位FIR滤波器幅频响应特性,采用三角函数基神经网络优化算法计算滤波器系数,同时在神经网络训练过程引入自适应学习率算法,提高神经网络的学习效率和收敛速度。通过训练神经网络的权值,使二维线性相位FIR滤波器幅频响应与理想幅频响应之间的误差平方和最小。文中给出了二维线性相位圆形低通滤波器优化设计实例,仿真结果表明,该方法设计二维线性相位FIR滤波器具有运算量小、速度快和稳定性好等优点。A novel design method of two-dimensional（2-D） linear-phase FIR filters based on adaptive triangle function basis neural network is presented.According to the amplitude-frequency response characteristics of 2-D linear-phase FIR filters,coefficients of the filters are calculated using triangle function basis neural network algorithm.In the training of neural network,adaptive learning algorithm is applied in order to enhance learning efficiency and convergence rate.By training neural network weights,the algorithm makes the squared sum of amplitude-frequency response error between the designed FIR filter and the ideal filter least in the whole pass band and stop band.The examples of 2-D linear-phase circular low-pass filters are also given by using the algorithm in the paper.The simulation results have shown the method has little calculation amount,fast rate and strong stability in the design field of 2-D linear-phase FIR filters.

关 键 词：三角函数 神经网络 自适应学习率 二维圆形低通滤波器 

分 类 号：TN713[电子电信—电路与系统]