基于自适应粒子群优化的RBF毫米波信道建模研究

RBF neural network channel modeling of millimeter wave based on adaptive particle swarm optimization algorithm

基于毫米波室内无线信道测量数据,将机器学习(machine learning,ML)中的径向基函数(radial basis function,RBF)方法应用于毫米波信道建模中,建立了基于自适应粒子群优化(adaptive particle swarm optimization,APSO)的RBF神经网络信道参数预测模型,并与传统RBF算法的预测结果进行了比较.利用APSO-RBF模型对信道大尺度参数(large-scale channel parameter,LSCP)如路径损耗(path loss,PL)、时延扩展(delay spread,DS)等数据的特征进行了学习和预测.结果表明,APSO-RBF模型信道参数的预测值与实际测量值非常吻合,均方根误差(root-mean-square error,RMSE)较小,且预测曲线与原始测量值曲线的拟合度较好,该算法的学习性能和预测效果均优于传统RBF算法.另外,APSO-RBF模型在数据量波动较大的情况下对信道参数的变化有着良好的适应性,对5G毫米波信道参数可以取得较好的预测效果.

In this paper,the method of radial basis function(RBF)in machine learning is applied to the modeling of millimeter-wave channel based on millimeter wave indoor wireless channel measurement data.A RBF neural network channel parameter prediction model based on adaptive particle swarm optimization(APSO)is established,and the prediction results of the traditional RBF algorithm are compared.Specifically,the RBF model optimized based on APSO is used to learn and predict the characteristics of large-scale channel parameters(LSCP),such as path loss and delay spread.The results show that the predicted channel parameters of the APSO-RBF model is consistent well with the actual measured value.The learning performance and prediction effect of this algorithm are better than the traditional RBF algorithm,that is,the RBF algorithm has a smaller root-mean-square error(RMSE),and the predicted curve has a larger fitting degree with the original measured curve.In addition,the APSO-RBF model has good adaptability to the change of channel parameters in the case of large data fluctuation,and can achieve good prediction effect for 5G millimeter wave channel parameters.