Downlink User Selection And Beamforming Design For Multi-user MIMO Systems

By making full use of spatial resources,Massive multi-input multi-output(MIMO)technology obtains extremely high multiplexing gain and multi-user diversity gain,which is regarded as one of the key technologies to improve the performance of communication systems.In multi-user system,it is necessary to eliminate the inter-user interference through beamforming technology.The difficulty of obtaining users' channel information limits the number of users that the base station can serve simultaneously.Therefore,user selection is also required.The resource allocation in downlink transmission of multi-user MIMO systems was studied in this paper,including beamforming,user selection,and the joint optimization.Channel modeling for the multi-user MIMO system was firstly studied and the system capacity was also analyzed.The key sessions of resource allocation in multi-user MIMO system were studied.The optimization problem of the beamforming vector was modeled as a power minimization problem with a given signal-to-interference-noise ratio target.The best beamforming vector under different SNR conditions was obtained.The joint design of beamforming at the transmitter and receiver in the multi-user MIMO system was studied.A multi-user beam selection scheme based on graph learning was proposed: the beam pair selection was modeled as a multi-classification problem.By using the graph neural networks,the hidden information contained in users' channel information and topological relationships was discovered.The channel parameters(including angles of departure,angles of arrival,phases,time of arrival,and received powers,etc.)of multi-user communication scenarios in urban streets were generated in the ray tracing simulator.Pre-processing and feature extraction were performed,and the model training and calibration were completed in Tensorflow.The simulation results showed that this method can obtain the global optimization of the system with high efficiency,and avoid exhaustive search,which effectively reduces the training overhead.The same performance as the traditional beam sweeping method was achieved with a smaller beam training overhead.When the number of multi-users in the system was 8,the beam training overhead was reduced to 80% compared with the traditional beam sweeping method.This paper proposed a greedy user selection algorithm by considering user fairness,which can improve the algorithm complexity of the direct user selection method as well as the user fairness.This algorithm selects users according to the criteria of weighted sumrate maximization,updates user weights in a proportionally fair manner,and reduces the computational complexity by optimizing the session of power allocation.Simulation results show that the algorithm can obtain a high system sum rate while ensuring user fairness and computational efficiency.This algorithm was superior to other direct user selection algorithms in terms of fairness,and its performance is stable.The fairness factor was always remained above 0.95.The calculation time was significantly smaller than other direct user selection algorithms,and it increased linearly with the increase of the number of users to be selected.