摘要
为了提高建模精度,提出了广义改进型径向基函数神经网络模型。首先,在径向基神经网络的输入层中使用延时抽头以仿真功放的线性记忆效应;然后,对每个抽头进行非线性级数展开,用于模拟功放的非线性记忆效应;最后,在非线性级数展开模块中引入超前包络项和滞后包络项,用于模仿功放的超前包络效应和滞后包络效应。文中使用三载波WCDMA信号驱动Doherty射频功率放大器进行测试,实验结果表明,与传统功放模型相比,广义改进型径向基神经网络模型能够更准确地拟合射频功放的特性,其归一化均方误差可以达到-41 d B。
In order to improve modeling accuracy, a generalized improved radial basis function (RBF) neural network model is proposed. Firstly, the delay tap is used in the input layer of RBF neural network to simulate the linear memory effect of power amplifier. Then, the nonlinear series expansion of each tap is used to simulate the nonlinear memory effect of power amplifier. Finally, the leading envelope term and lagging envelope term are introduced into the linear expansion module to simulate the leading envelope effect and lagging envelope effect of power amplifier, respectively. The three-carrier WCDMA signal is used to drive Doherty RF power amplifier for testing. Experimental results show that, compared with the traditional power amplifier model, the modified RBF neural network model can more accurately fit the characteristics of RF power amplifier, and the normalized mean square error can reach -41 dB.
作者
江明玉
刘太君
叶焱
许高明
JIANG Mingyu, LIU Taijun, YE Yan,XU Gaoming(Faculty of Electrical Engineering and Computer Science, Ningbo University, Ningbo 315211, Chin)
出处
《移动通信》
2018年第3期64-69,共6页
Mobile Communications