摘要
针对FIR滤波器的神经网络设计法,提出一种泛函连接人工神经网络的改进算法。通过设置不同的加权误差函数值来控制各个样本的学习率,改善了网络的学习效果;制定了神经网络训练集的选取规则,使用该规则选取样本对网络进行训练,可设计通带阻带截止频率指标精确可控的滤波器,克服了现有算法只能设计具有通带截止频率的滤波器和不能精确控制任意截止频率的不足。仿真结果表明所提出的方法能很好地满足设计要求。
To the neural network design method of Finite Impulse Response(FIR) filter,an improved algorithm of Functional Link Artificial Neural Network(FLANN) is advanced.It ameliorates learning effect by setting different weighted function values to control the learning rate of every sample,besides a training set selection rule is set down.Through training samples selected by the rule,a FIR filter with precisely controlled passband and stopband cut-off frequencies specifications can be designed.The algorithm overcomes the shortage of existing others which only can design a filter with passband cut-off frequency and can not precisely control arbitrary cut-off frequency.The result of simulation shows that the method proposed in this paper can satisfy the design requirement.
出处
《计算机工程与应用》
CSCD
北大核心
2009年第23期82-84,共3页
Computer Engineering and Applications
关键词
泛函连接人工神经网络
加权误差函数
训练集选取规则
FIR滤波器
Functional Link Artificial Neural Networks (FLANN)
weighted error function
training set selection rules
FIR (FiniteImpulse Response) filters