摘要
本文提出了一种用神经网络算法来设计任意幅频响应二维FIR线性相位数字滤波器的新方法,其主要思想是使频率响应平方误差函数最小化.根据给定的任意幅频响应指标,按该算法可直接获得滤波器系数.为保证该算法的稳定性,提出并证明了该算法的收敛定理.文中给出了滤波器优化设计实例,计算机仿真结果表明由该方法设计的任意幅频响应二维数字滤波器波动小,算法收敛速度快,稳定性强.
This paper provides a novel neural networks algorithm (NNA) for the design of 2-D FIR linear-phase digital filters with arbitrarily shaped amplitude-frequency responses, the main idea is to minimize the squared-error function in the frequency-domain. By using the NNA, the coefficients of the designed filter can be obtained directly from the specified amplitude-frequency response. The convergence theorem is presented and proved to illustrate the stability of the NNA. The result of the optimal design example show that the ripple of the filter is tiny, and the proposed approach is of fast convergence and powerful stability.
出处
《电子学报》
EI
CAS
CSCD
北大核心
2005年第5期950-953,共4页
Acta Electronica Sinica
基金
国家自然科学基金(No.50277010)
高校博士点基金(No.20020532016)
湖南省教育厅科研项目(No.04C073)
湖南省科技计划项目(No.03GKY3115
No.04FJ2003)
湖南省杰出青年基金(No.03JJY1010)
教育部新世纪优秀人才支持计划和湖南大学撷英计划
关键词
二维数字滤波器
线性相位
神经网络
收敛定理
优化设计
Algorithms
Convergence of numerical methods
FIR filters
Frequency domain analysis
Frequency response
Functions
Mathematical models
Optimization
Stability
Theorem proving
Two dimensional