摘要
本文针对信号处理中一类常用的非线性模型，提出了切比雪夫正交多项式神经网络（TOPNN）。与多层感知器（MLP）相比，其计算量可显著减少。本文还参考自适应IIR滤波器理论，提出了TOPNN的递归结构和类递归结构，讨论了其特点和应用范围。此外，本文还给出了（1）类递归结构TOPNN的在线辩识算法及其在非线性系统辩识中的应用，（2）递归结构TOPNN的滤波算法及其在非线性相关噪声抵消中的应用。最后在计算机仿真中将TOPNN与MLP进行了对比。
For a common kind of nonlinear in signal processing, Tchebycheff Orthogonal Polynomial Neural Network (TOPNN) is put forward in the paper. Compared with muiti-layered perceptron (MLP), its computational cost is much less. In reference of the theory of adaptive IIR filters, the recursive structure and quasi-recursive structure of TOPNN are also put forward,and their characteristics and areas of application are discused.In addition, also given in the paper are (1) the on-line identification algorithm of quasi-recursive TOPNN with its application in nonlinear system identification, (2) the filtering algorithm of recursive TOPNN with its applicatin in nonlinear correlated noise cancellation.Finally,TOPNN is compared with MLP in computer simulation.
出处
《信号处理》
CSCD
北大核心
1996年第4期362-368,349,共8页
Journal of Signal Processing
关键词
正交多项式
神经网络
递归结构
信号处理
orthognal polynomials, neural networks, system identification, noise elimination/recursive structure
