基于神经网络的射频功放行为模型研究*

摘要：为了设计和优化高线性功率放大器，在系统级仿真中，构建功率放大器精确的行为模型是极为重要的。本文在ADS中基于飞思卡尔半导体芯片MRF5P21180进行功放电路设计，运用BP和RBF以及在这两种网络的基础上所提出的级联模型BP-RBF对从设计的功放中提取出的电压数据进行MATLAB拟合仿真，通过电压均方根误差（RMSE）大小的比较进行模型精确性的验证，训练次数多少及收敛时间长短进行模型训练速度的对比，最终得出BP-RBF较其他两种模型有更好的拟合功能，对于射频功放行为模型的研究具有重要的参考价值。

Research on behavioral models for RF power amplifier based on neural network

Abstract: In order to design and optimize high-linearity power amplifier (PA), it is very important to build correct PA behavioral model in system-level simulation. This article mainly used Freescale semiconductor chip MRF5p21180 to carry on amplifier circuit design in the ADS .Carried on the MATLAB fitting simulation to the extracted data by BP and RBF as well as Cascading model BP-RBF. BP-RBF was presented on the basis of the two networks.Compared with voltage RMS error (RMSE) to validate the model accuracy, Compared with epochs and and convergence time to validate the model training speed, eventually the result derived that BP-RBF has better fitting function than other two models,having the important reference value to study Behavior model of RF power amplifier.