Recognition And Multipath Clustering Methods For Channel LOS/NLOS Scenarios Based On Machine Learning

Radio channel modeling has been an important research topic of wireless communications.The performance of communication system is affected by channel characteristics.A large number of channel measurement data has shown that wireless multipath components(MPCs)tend to exhibit cluster structure distribution in realistic environment.In order to balance the complexity and accuracy of channel model,the model based on cluster structure has been widely used in current channel modeling.In the line-of-sight(LOS)and non-line-of-sight(NLOS)scenarios,the distributions of MPCs are different,thus the channels exhibit different statistical characteristics.Therefore,the distinction between LOS and NLOS is of great significance for wireless channel characterization and modeling.In addition,the distinction between the two channel scenarios is also important for positioning related technologies,such as the wireless positioning technologies based on time of arrival or time difference of arrival.Because of the different characteristics of MPCs in the two scenarios,different clustering algorithms need to be designed to improve the accuracy of MPC clustering.In this thesis,the identification methods of LOS and NLOS scenarios in wireless channels and the clustering methods of MPCs are studied and the main contributions are summarized as follows:1.In the study of the identification of channel LOS and NLOS scenarios,the existing identification schemes are reviewed firstly,in which the identification method based on the changes of propagation characteristics is widely used,such as the identification scheme based on the Ricean K factor of channel small scale fading.However,the Ricean K factor is relatively sensitive to the changes of propagation environment,hence it is often difficult to find an accurate identification threshold for NLOS scenario.In this regard,an identification method based on convolutional neural network is proposed in this thesis,which uses the measurement data of the vehicular channels as the training and testing samples of the network.The specific structure and core parameters of the network are designed based on the characteristics of channels and the feedback results of the network training,and the optimal network structure and parameter configuration are obtained.The performance of the algorithm is verified by different datasets,and the practicality and feasibility of the designed neural network are discussed.In the comparisons of the existing identification algorithms,such as the Ricean K factor based method and some other machine learning algorithms,the performances of different algorithms are simulated and tested.The experimental results show that the convolutional neural network designed in this thesis has the highest identification accuracy.2.The influence of time varying channels is considered in the clustering of MPCs for the LOS scenario.It is necessary to use dynamic clustering algorithms to conduct parameter analysis and channel modeling.A tracking based clustering algorithm of MPCs is designed,in which the method of maximizing the total probability is used to identify the motion trajectories of the multipath components of the continuous snapshots in the time-varying channels,and MPCs are clustered based on the motion trajectories.A dynamic channel simulator is established to simulate the dynamic changes of MPCs in MIMO channels,and the accuracy of the designed MPC clustering algorithm is verified by the simulation.3.In the study of MPCs clustering in the NLOS scenario,the dispersed distribution of cluster structure,the large variations of the number of MPCs in different clusters and the uncertain cluster recognition are considered to improve the clustering algorithm based on density.The MPCs are clustered by the MPC distance based on power weighting,which improves the accuracy of clustering.The performance of the designed MPC clustering algorithm is validated by the simulated data in the NLOS scenario,which are simulated by using the general channel simulator.