基于BP神经网络的射频信号包络线预测研究  

Research on RF signal envelope prediction based on BP neural network

作　　者：王雪然 周岩 陈建 杨龙祥 WANG Xueran;ZHOU Yan;CHEN Jian;YANG Longxiang(School of Automation and Artificial Intelligence,Nanjing University of Posts and Telecommunications,Nanjing 210023,China)

机构地区：[1]南京邮电大学自动化学院、人工智能学院,江苏南京210023

出　　处：《微电子学与计算机》2024年第6期1-10,共10页Microelectronics & Computer

摘　　要：包络线跟踪电源相较于传统恒压供电大大提高了功放效率,但对硬件要求更高且会产生较高的额外时延。针对以上问题本文提出了一种基于BP神经网络的射频信号包络线的数据预测方案,用于提前生成控制开关变换器的基准信号。首先,基带数据经正交频分复用(Orthogonal Frequency Division Multiplexing,OFDM)技术生成射频信号的包络线数据。其次,将处理好的数据用于训练包络线跟踪电源的预测模型。最后,改变OFDM调制的子载波数以及选取不同正交振幅调制方式(Quadrature Amplitude Modulation,QAM)分别训练网络。在12子载波下,分别选取16QAM~512QAM之间的6种映射方式进行仿真实验:512QAM在最佳隐含层节点16下训练得到的均方根误差值(Root Mean Squared Error,RMSE)为0.3143;在映射方式为16QAM,64QAM映射方式下,分别选取12、28、52子载波进行仿真,52子载波64QAM映射下的RMSE值最大为0.1752。预测结果中RMSE值均较小,满足预测误差的要求,且预测结果图中包络线与实际包络线拟合效果很好。通过对BP网络预测模型与传统调制模型浮点运算次数的计算,求取52子载波映射方式为64QAM的信号包络,计算次数节省率可以达到49.40%。The envelope tracking power supply greatly improves the efficiency of power amplifier compared with the traditional constant voltage power supply.However,Envelope Tracking(ET)technology requires high hardware conditions and produces high extra delay.In this paper,a data prediction scheme of RF signal envelope based on BP neural network is proposed,which is conducive to generating the reference signal of control switch converter in advance.Base band data generates envelope data of RF signal by Orthogonal Frequency Division Multiplexing(OFDM)modulation technology,which is used to train the prediction model.Finally,the network is trained under different subcarrier numbers and mapping modes to obtain and the feasibility of the model is verified after a better number of nodes is selected.The results show that the method proposed in this paper can achieve accurate envelope data prediction with the maximum Root Mean Squared Error(RMSE)value of 0.1752.Moreover,the predicted envelope fits well with the actual envelope,which verifies the feasibility of the proposed model.By calculating the number of floating point operations on the BP neural network prediction model,and the saving rate of the calculation times can reach 49.40%.

关 键 词：包络线跟踪电源 OFDM调制 BP神经网络 均方根误差 正交振幅调制 

分 类 号：TN911.3[电子电信—通信与信息系统]