Research On Communication Optimization Of Cellular-connected UAV Based On Machine Learning

With the rapid deployment and development of the fifth generation mobile communication,the traditional unmanned aerial vehicle industry is rapidly developing towards networking,digitization and intelligence.Replacing the traditional peer-to-peer communication link with the global ubiquitous cellular network,the Cellular-Connected UAV can operate and transmit data remotely without terrain and distance constraints.It provides 24/7 online services and has comprehensive monitoring capabilities,which greatly improves the performance and applications of the UAV.Nowadays,the Cellular-Connected UAVs are widely used in power patrols,public security patrols,traffic patrols,as well as in agriculture,logistics and exploration and mapping.The 5G+UAV mode is fully enabling the development and construction of various industries in China,and has become the core force and important means of low-altitude economic development in the new era.The cellular network supporting low-altitude vehicle communications has become an important research direction for UAV communications.However,the three-dimensional air-ground channel of the UAV is quite different from the ground channel.Increased line-of-sight link probability and higher flight height also cause serious interference problems,and the high-speed stereo movement of the UAV will also affect the communication link.At the same time,the existing cellular network environment basically only supports service quality control and communication for ground users.Air users in low altitude airspace are higher than antennas and are covered by side lobes of downward tilting antennas,and network-connected UAVs require more stringent and complex communication link rates and latencies.Support levels for network-connected UAVs in multi-mission scenarios are still being explored under existing network conditions.Based on UAV communication theory and practical problems in application,aiming at the interference and communication level of network-connected UAV,the channel quality of network-connected UAV is simulated and measured,and a multi-point Collaboration Scheme Based on machine learning and a route planning scheme are proposed on the base station side and the user side respectively.The results show that the communication quality and work efficiency of network-connected UAV can be significantly improved.Firstly,UAV channel is studied.Based on the existing research on UAV channel propagation characteristics and representative low-altitude channel literature,typical air channel models are analyzed,including several large-scale and small-scale fading channel models.The unmanned aerial vehicle channel model is compared with the traditional terrestrial channel model to describe its characteristics and influencing factors.The UAV ground-air communication scenarios are classified according to 3GPP and ITU protocol reports and standards,and channel model parameters under different scenarios are given.The communication quality of the UAV is simulated under the system level simulation environment and the building environment.At the same time,a real network UAV system is set up for field test,and the communication requirements of the network UAV task scene are analyzed.The support of existing communication environment for low altitude network UAV is evaluated,and data reference is provided for subsequent optimization.Secondly,multi-point collaboration technology is applied on the base station side to solve the network-connected UAV interference coordination problem.Because of its flight height and high line-of-sight link probability,unmanned aerial vehicle（UAV）has more severe common frequency interference and complex base station coverage problems.To solve the above problems,a network-connected UAV interference suppression scheme for dynamic point selection in multi-CSI process with multi-mode switching is presented.User channel state information is shared between base stations and transmission points,and physical shared channel is provided for users to transmit data through air user CSI information to select the best transmission point.A multi-mode switching scheme based on SVM support vector machine is applied to the network-connected UAV channel.Considering the UAV collaboration delay constraint,the transfer point can be adjusted reasonably by controlling the switching policy,and the corresponding switching policy can be selected independently according to the UAV speed and delay requirements.The simulation results show that the throughput of air users can be improved and the bit error rate can be reduced effectively with limited collaboration delay.Finally,a route planning scheme for single and multiple UAV systems based on reinforcement learning is proposed on the user side.When the network conditions are certain,the communication service level and energy consumption obtained by the UAV are optimized by reasonable route planning.In a single unmanned aerial vehicle environment,space is discretized into planar grids,and reward functions based on communication parameters and path wind bias parameters are constructed.Achieve the most effective overall UAV route through intensive learning.In multi-UAV systems,the environment is further extended to complex urban buildings and communication environments.Based on the area patrol of multiple unmanned aerial vehicle systems,each unmanned aerial vehicle generates video data from a cloud camera and transmits it to a base station using the unmanned aerial vehicle or server as input after map processing.When using a trained network model to input untrained area maps and parameters,high data throughput and safe and efficient routes are still achieved.