基于机器学习的无线信道簇的提取与轨迹追踪

The Extraction and Tracking Trajectory of Wireless Channel Tap Clusters Based on Machine Learning

针对时变无线信道抽头簇的提取和轨迹追踪提出了一种新方法:首先在时延-幅度维上采用反向传播(BP)神经网络对无线信道冲激响应(CIR)进行去噪,然后利用k-means聚类算法对有效抽头信号进行分簇,再用基于密度的空间聚类(DBSCAN)算法去除各个簇峰值抽头中的异常值,最后采用多项式拟合对去除异常值后的簇峰值抽头进行拟合,得到其时间变化轨迹.经过仿真和实测数据验证,该方法得到的簇峰值时间变化轨迹与根据几何关系得到的结果一致.

A new method for extraction and tracking trajectory of dynamic wireless channel tap clusters is proposed.First,the channel impulse response(CIR)denoising is achieved by back propagation(BP)neural network in time delay-amplitude dimension.Then effective taps are clustered by k-means clustering algorithm.Next,density-based spatial clustering of applications with noise(DBSCAN)algorithm is applied to remove the abnormal peak taps for every cluster.Finally,the trajectory of cluster peak taps is obtained by polynomial fitting.The simulation result shows that trajectory obtained by proposed method is approximate to geometric calculation result.Moreover,the analysis result of high speed railway measured data is consistent with the actual observations.