摘要
针对使用CAD软件设计射频微波电路繁琐且耗时长等缺点,提出一种新颖的带外部输入的非线性自回归(NARX)神经网络逆向建模方法。此方法采用具有激励函数的NARX神经网络(DAFNN)为模型以提高网络的泛化能力,利用支持向量机(SVM)替代模型的前馈部分完成数据分类,解决设计中的多解问题。然后应用于可以覆盖多个频段的可重构功率放大器中,实验表明,该方法在精度方面分别优于直接逆向建模方法和自适应η逆向建模方法99.86%和81.32%,计算速度方面优于直接逆向建模方法31.72%,可以降低射频微波可重构功率放大器的设计复杂度、缩短其设计时间。
Using simulation softwares to design RF microwave circuits is cumbersome and time-consuming.Aiming at these defects,a novel nonlinear auto-regressive with exogenous inputs(NARX)neural network inverse modeling is proposed.In this method,the neural network with dynamic activation functions(DAFNN)is used as the model to improve the generalization ability of network.And the inverse modeling method in this paper uses the support vector machines(SVM)to replace the feedforward part of the DAFNN neuron model to complete the data classification,which can solve the problem of multiple solutions in design.Then it is applied to the reconfigurable power amplifier which can cover multiple frequency bands.The experiment shows that the method’s accuracy is superior to the direct reverse modeling method and the adaptiveηinverse modeling method 99.86%and 81.32%respectively,the calculation speed is better than the direct reverse modeling method 31.72%.It can reduce the design complexity of RF microwave reconfigurable power amplifier and shorten its design time.
作者
南敬昌
臧净
高明明
胡婷婷
NAN Jing-chang;ZANG Jing;GAO Ming-ming;HU Ting-ting(School of Electronic and Information Engineering,Liaoning Technical University,Huludao 125105,China)
出处
《微波学报》
CSCD
北大核心
2019年第5期51-56,共6页
Journal of Microwaves
基金
国家自然科学基金(61971210,61372058)
国家自然科学基金青年科学基金(61701211)
辽宁省特聘教授项目(551806006)
辽宁省高校重点实验室项目(LJZS007)
关键词
带外部输入的非线性自回归(NARX)神经网络
逆向建模
DAFNN神经元模型
支持向量机
可重构功率放大器
nonlinear auto-regressive with exogenous inputs(NARX)neural network
inverse modeling
neural network with dynamic activation functions(DAFNN)neuron model
support vector machine
reconfigurable power amplifier
