基于最小二乘支持向量机的时变信道建模

Time-Varying Channel Modeling Using Least Square Support Vector Machine

基于2.55GHz市区微蜂窝多输入多输出信道实测数据,将机器学习中的最小二乘支持向量机(LS-SVM)算法应用于时变信道参数的建模中,建立了基于遗传算法(GA)优化的LS-SVM信道参数预测模型,对信道参数如时延扩展、接收端的水平角度扩展和垂直角度扩展的数据特征进行了学习,并实现了准确预测;同时通过与反向传播神经网络模型以及传统的LS-SVM模型进行比较,验证了算法的有效性.基于GA优化的LS-SVM模型能够在有限数据量下对信道参数的变化有着良好的适应性,可实现非线性时变信道参数的准确预测.

Based on 2.55 GHz urban microcellular multiple-input multiple-output(MIMO)channel measurement data,the least squares support vector machine(LS-SVM)method was applied on time-varying channel model.Specifically,a genetic algorithm(GA)based LS-SVM(GA+LS-SVM)model was established for channel parameter prediction.Based on GA+LS-SVM model,the time-varying channel parameters,such as delay spread,horizontal angle spread and vertical angle spread of receiver,were investigated and predicted accurately.Moreover,the GA+LS-SVM model was compared with back propagation neural network and traditional LS-SVM algorithms to verify the effectiveness of the algorithm.In summary,with limited amount of data the GA based LS-SVM model can better adapt to non-linear timevarying channel to realize the accurate prediction of nonlinear time-varying channel parameters.