Research On Deep Beam Tracking Technology In Internet Of Vehicles

The Internet of Vehicles is one of the hottest applications in the scenario of low latency and high reliability for the fifth generation(5G)mobile communication.While beam tracking technology is the key technology for millimeter wave in massive multiple input multiple output(MIMO)systems.In the scenario of high-speed mobile vehicular network,it is difficult for the beam tracking technology to balance the beam coverage and beam interference;therefore,based on deep reinforcement learning,this thesis studies the beam tracking technology of vehicular networks in two scenarios of a single-cell and multi-cell junction.Research on intelligent beam tracking technology for multiple vehicles in single-cell scenario.Considering the single-cell scenario that millimeter wave massive MIMO system periodically provides services for multiple mobile vehicles with multiple beams.To maximize the success probability of information transmission in each beam tracking cycle,an intelligent beam tracking scheme that considering beam coverage and inter-vehicle interference is proposed based on deep reinforcement learning.The communication model of single-cell with multiple vehicles is firstly established,the corresponding reinforcement learning model is formulated according to the communication process,and the optimal objective is also proposed.Then,a three-stage single-cell intelligent beam tracking scheme is designed to optimize the communication success probability:(1)Long Short-Term Memory(LSTM)network is adopted to predict the vehicle trajectory,and the predicted position replaces the historical trajectory to reduce the state space dimension;(2)All vehicles are divided into vehicle clusters according to inter-vehicle distance,and the joint target is decomposed to reduce the complexity of joint decision making;(3)The optimal joint beam is solved based on deep reinforcement learning,and the beam contribution is decomposed to reduce the action space dimension.The simulation results show that the proposed algorithm has better ability to balance interference and coverage under different vehicle speeds than other comparison schemes,and can achieve higher successful transmission probability;the effects of environment parameters,beam width and discrete beam set size on the system performance are given in order to provide a reference for practical scenarios.Research on intelligent cooperative beam tracking technology for multi-cell.The scenario of millimeter wave beams periodically providing services for multiple vehicles on multiple lanes at the edge of multiple cells is further considered in this thesis.The main objective is still to maximize the average successful transmission probability in each beam tracking cycle.Cell handover of vehicles at the cell edges should be considered,therefore the dual choice of cell cooperation mode and beam direction are taken into account for beam tracking scheme.In order to balance the contradiction between beam coverage,inter-vehicle interference and intercell interference,an intelligent cooperative beam tracking scheme based on deep reinforcement learning is proposed.First,the communication model is formulated and the successful transmission probability is taken as the optimal objective.Secondly,a thee-stage multi-cell intelligent beam tracking scheme is designed.The LSTM network is employed for trajectory prediction,the frequency resources are adopted for vehicle cluster division,and the joint beam action is solved based on deep learning and reinforcement learning.In the third stage,to reduce the complexity of the scheme,a two-layer model of inner and outer loops is proposed to select the cell cooperation mode and the joint beam direction,respectively.In each layer of the network,the deep network is used to simulate the relationship between the network input and the desired action,and the network training label is constructed according to reinforcement learning;the network converges through partial training.The simulation results show that the proposed scheme has better communication performance than other schemes,and also provides a reference for actual systems to appropriately divide vehicle clusters under different antenna numbers.