Key Technologies And Characteristic Mode Analysis Methods For 5G Base Station Antennas

In order to meet the requirements of larger transmission bandwidth,lower delay and higher system spectral efficiency in future mobile communication systems,the combination of millimeter wave technology and large-scale antenna arrays provides strong support for the evolution of mobile communication technology.As the size of the array increases,not only the performance improvement,but also higher hardware cost,power consumption and backhaul link overhead.In order to reduce hardware complexity and base station power consumption,digital-analog hybrid precoding has become an effective solution to balance performance and power consumption,and the joint optimization of digital domain and analog domain undoubtedly increases the complexity of traditional algorithm design and optimization problem solving.In recent years,with the development of artificial intelligence(AI),how to use deep learning technology(DL)to solve complex optimization problems in traditional communication systems has received extensive attention.As an excellent tool for dealing with non-convex and high-computational problems,deep learning provides a new solution for solving complex mixed precoding problems.In this thesis,with the background of hybrid precoding technology in Time Division Duplexing(TDD)system,several common optimization problems are discussed and studied in detail,and datadriven solutions are proposed for different problems.At the same time,this thesis verifies that the proposed scheme has lower cost and better performance than traditional algorithms through analysis and simulation.Firstly,considering that the acquisition of downlink channel state information(CSI)is the premise of designing a precoding scheme,and the asymmetrical characteristics of the uplink and downlink radio frequency links lead to the destruction of the channel reciprocity of the TDD system.Based on this non-ideal factor,this thesis discusses and analyzes the reciprocity calibration problem under the hybrid architecture,and proposes a deep learning-assisted over-the-air calibration scheme.We takes the mean square error of the actual downlink channel and the channel after calibration as the evaluation indicator,and under the condition of low training overhead,it is verified that the calibration error of the proposed deep learning calibration scheme can reach a lower level than the traditional scheme.Secondly,since energy efficiency(EE)is a key indicator to evaluate system performance and power consumption,this thesis researches on the multi-carrier energy efficiency optimization problem in a single-cell multi-user scenario.In order to ensure the minimum quality of service of users,the modeling of the optimization problem introduces the minimum Signal to Interference plus Noise Ratio(SINR)constraint,which makes the traditional solution method more complicated.Based on this,this thesis deduces the conventional solutions to such problems in detail,and further proposes an unsupervised learning-assisted energy efficiency optimization algorithm.A convolutional neural network is designed for the hybrid precoding problem,and its performance is verified by multiple indicators such as system energy efficiency,spectrum efficiency,and base station power consumption.Compared with the traditional Dinkelbach iterative optimization algorithm,the performance is improved,and the computational complexity is lower.Finally,this thesis extends the study to multiple base station cooperation scenarios.Considering that in the traditional cooperation scheme,the base stations need to share a large amount of real-time CSI,which leads to huge backhaul link overhead and information exchange delay.Based on this,this thesis designs a data-driven solution to transfer a large amount of data exchange to the offline training stage,and utilizes the system spectral efficiency as an evaluation indicator to verify the performance.The simulation results show that the solution based on deep learning can achieve similar or even better system performance than the traditional schemes with the reduce of real-time data interaction.