Intelligent Channel Prediction And Beam Tracking Schemes In 5G Mobile Scenarios

The fifth-generation communication systems use the millimeter wave band to increase communication capacity,but face the problem of high propagation loss.The loss can be effectively compensated by deploying large-scale antenna arrays with beamforming methods.However,in mobile scenarios,frequent channel estimation and beam alignment are required to ensure the quality of the communication link,which brings a large frequency guide overhead and feedback overhead,as well as high hardware implementation cost and signal processing complexity.Therefore,this thesis focuses on low overhead channel estimation and beam alignment schemes in multiple-input multiple-output systems,which are divided into the following two points:(1)To address the problem of high channel estimation overhead,a convolutional neural network(CNN)-based channel prediction method is proposed,by jointly using trajectory prediction and channel reconstruction.First,a CNN is adopted to learn the mapping from the planned route and the location of the mobile terminal to the moving direction,which in turn predicts multiple target locations on the trajectory;second,a CNN is used to learn the mapping from the channel of K location terms near the target location to the channel of the target location,which is used to achieve channel prediction for the predicted trajectory.Simulation results show that,the CNN-based method significantly outperforms the FCN-based channel estimation method,and outperforms the K-nearest neighbor interpolation method when K<6.(2)To address the problem of frequent beam alignment,a low-overhead beam tracking scheme is investigated in this thesis,by exploiting the spatial consistency property of adjacent position channels.The scheme is optimized with the objective of balancing system performance and beam adjustment frequency.To address the difficulty of solving the optimization problem,three deep learning models are used in this thesis to design the proposed scheme and train the models in an unsupervised learning manner.Simulations are performed using the channel dataset collected by ray tracing.Results show that,the proposed scheme can effectively balance the system performance and beam tuning frequency,in which the twodimensional convolution-based beam tracking scheme has the best performance.