Reasearch On Out-of-band Information Aided Millimeter Wave Beam Selection Technology

In order to meet the needs of mobile services,millimeter wave with enormous unallocated bandwidth has been one of the communication bands of 5G systems.Compared with traditional communication systems with low frequency signal,millimeter wave signal will suffer from more severe propagation loss,which braings a huge challenge to establish a reliable communication link.In order to overcome the unacceptable loss of millimeter wave,beamforming has become a key technology in millimeter wave communications.In practical systems,the implementation of beamforming is established on the knowledge of channel state information.Thanks to the short wavelength of millimeter wave,larger antenna arrays can be deployed in millimeter wave systems,but this also means that the channel matrix is huger,which leads to a lot of training overhead for traditional beamforming methods.With the diversification and intellectualization of the future network systems,there is a large amount of out-of-band information available in the network systems,such as the location information of user terminals.Therefore,we consider utilizing the out-of-band information reasonably to assist the beam selection process in millimeter wave systems,which will reduce the beam selection overhead as much as possible on the premise of ensuring the communication performance.The existing researches on out-of-band information assisted millimeter-wave communication do not make full use of out-of-band information and the application scenarios are relatively limited,without considering the relationship between channel changes and terminal state.Based on the spatial consistency of millimeter wave channel,this thesis studies the out-of-band information assisted beam selection algorithm in terminal static scenarios and mobile scenarios.The main research work of this thesis is as follows:1.The spatial consistency of millimeter wave channel is studied,and the beam selection process of millimeter wave system is assisted by location and attitude information for reducing the overhead.Based on the characteristics of millimeter wave channel,this thesis focuses on the spatial consistency characteristics of millimeter wave channel,derives the relationship between the change of channel parameters and the change of terminal location and attitude,and then analyzes the distribution of codewords in space.According to the correspondence between the optimal beam and terminal location-attitude,the traditional beam selection problem is transformed into a classification problem with the terminal location and attitude information.The beam selection classification problem is a nonlinear multi-classification problem,and we use a support vector machine model with the kernel function to solve this beam selection classification problem.In order to improve the classification accuracy,we design a custom kernel function suitable for the beam selection classification problem according to the linear space definition and beam selection strategy.Simulation results show that the proposed location-attitude aided beam selection algorithm can greatly reduce the beam selection overhead without loss of performance.2.The influence of terminal movement on channel parameters is analyzed,and the out-of-band information aided beam prediction technology is studied in terminal mobile scenario.Terminal movement will cause frequent beam switching,by analyzing the relationship between the terminal motion state and the channel parameters,we adopt the neural network model to output the candidate beam set at the next moment by taking the motion state information,such as the location,attitude,speed,and angular velocity of the terminal,and the current access beam index as input.When the current access beam can not meet the communication requirements due to terminal movement,the terminal will quickly select an appropriate beam from the candidate beam set output by the neural network model for beam switching.Simulation results show that the candidate beam set can effectively narrow the search range of beam selection and reduce the beam switching overhead.3.Considering the problem of noise labels in practical systems,a noisy label cleaning algorithm based on random label propagation is proposed.In machine learning based out-of-band information assisted millimeter wave communication systems,the problem of noise label may exist,which will greatly degrade the system performance.By analyzing the intrinsic relationship between the training samples,we generate the homogeneous regions of training samples based on the spatial consistency of millimeter wave channel,and construct the spatial consistent probability transition matrix.Based on the predefined probability transfer matrix,we apply the random label propagation algorithm to propagate true label information between training samples,so as to eliminate the noise labels in the training dataset.Simulation results show that the proposed noise label cleaning algorithm can effectively remove the noise labels in the training dataset and improve the performance of the out-of-band information assisted communication system.