【作者】 张凌；

【导师】 归琳；

【作者基本信息】 上海交通大学 ， 信息与通信工程， 2021， 博士

【摘要】 5G/B5G无线通信系统以sub-6 GHz与毫米波为通信频段,通过配置大规模天线和密集基站应对无线信号的路径损耗。毫米波使能的大规模多输入多输出(Massive Multiple-input Multiple-output,Massive MIMO)与超密集组网技术为无线信号传输带来丰富的空域资源,允许用户空分多址接入以实现能效的“绿色通信”。高度的时频资源复用导致严重的多用户干扰问题,5G/B5G背景下的多天线预编码设计依然存在如下挑战:低频段系统天线阵维度低,预编码优化自由度受限;毫米波Massive MIMO系统混合预编码非凸联合优化问题难求解;多小区Massive MIMO系统干扰对齐实现复杂。在上述背景下,本文结合典型场景研究了面向5G/B5G空分多址系统的多天线预编码技术,主要成果如下:针对低频段多用户MIMO空分多址系统,本文提出了基于符号预检测的相长干扰预编码方案,通过发端波束成形设计直接优化收端检测误符号率(Symbol Error Rate,SER)。该方案根据收端符号检测SER函数扩展发端预编码优化空间,通过控制干扰下的接收信号星座落点降低SER。基于凸性证明与Karush-Kuhn-Tucker条件分析,本文证明了最优的预编码设计为信道向量的线性组合,提出了基于可行方向投影修正的相长干扰预编码算法(Projection-Revised Feasible Direction Scheme for Constructive Interference Precoding,PRFD-CIP)寻找全局最优解。PRFD-CIP算法以低SER用户的性能损伤换取高SER用户的性能提升,在用户分簇间的SER协调平衡中优化预编码设计。算法基于可离线获取的梯度值寻找优化变量的更新方向,避免SER导函数的复杂表达,通过融合投影修正过程压缩待优化参量维度,以高效的可行方向设计缩短算法收敛时间。仿真结果表明:PRFD-CIP相比基于最小均方误差的相长干扰预编码算法,能获得1.8 d B以上的信噪比增益。针对毫米波Massive MIMO空分多址系统,本文引入非监督学习的思想,提出了基于聚类的混合预编码方案处理非凸约束,通过射频结构设计提升预编码效率。针对全连接结构,本文以充足射频链配置下的理想设计为基础,将射频链配置不足时的预编码问题建模为聚类问题,提出基于层次聚类的全连接混合预编码算法(Hierarchical Agglomerative Clustering based Fully-connected Hybrid Precoding,HAC-FHP),挖掘理想射频链设计中的相关性,以单射频链取代具有高相关性的理想射频链簇,减少射频链受限导致的性能损失。针对自适应连接结构,本文将射频链与大规模天线选择性分簇连接的问题建模为聚类问题,提出基于K-均值聚类的自适应连接混合预编码算法(K-means based Adaptively-connected Hybrid Precoding,KM-AHP),挖掘全数字结构中天线间的相关性,充分利用模拟预编码结构中天线与射频链连接的灵活性提升混合预编码性能。仿真结果表明:在发端信噪比为-5 d B时,相比数模解耦设计方案,HAC-FHP的频谱效率与全数字结构的差距降低了40%,KM-AHP的频谱效率提升了12 bps/Hz。针对多小区Massive MIMO空分多址系统,本文提出了混合结构下的多小区干扰子空间对齐方案,通过收端干扰信号空间优化提升用户频谱效率。本文从全数字预编码切入,证明了对于用户单数据流传输,当发射天线数量不少于系统用户总数时,收端干扰子空间具有最高的优化自由度,并给出实现干扰子空间对齐的预编码解空间的矩阵表达,直观体现多天线信道的空间自由度与预编码优化自由度间的关联。进一步考虑混合结构导致的干扰泄漏问题,本文提出基于干扰子空间近似对齐的混合预编码算法(Approximated Interference Subspace Alignment scheme with Hybrid Precoding,AISA-HP),通过射频链的正交匹配设计与干扰子空间的交替优化控制干扰泄漏的能量,提升边缘用户频谱效率。仿真结果表明:相比干扰对齐方案,AISA-HP以更低的算法复杂度提升了23 bps/Hz频谱效率平层。综上所述,本文围绕sub-6 GHz/毫米波传输以及多小区交叠部署三类典型空分多址通信场景,提出基于符号预检测、聚类以及干扰子空间近似对齐的多天线预编码方案。所提方案从发端波束成形、射频结构设计以及收端干扰信号空间的角度展开优化,有效降低了信号检测SER、提高了多用户传输的频谱效率。本文为面向5G/B5G空分多址系统的多天线预编码技术研究提供了创新性的设计理念与关键技术支撑。更多还原

【Abstract】 Millimeter wave(mm Wave)and sub-6 GHz have been determined as alternative bands in 5G/Beyond 5G(B5G),which enables large scale antennas and dense base stations to combat the path loss.Massive multiple-input multiple-output(Massive MIMO)and ultra dense network bring abundant spatial resources for wireless communication,which contributes to high energy efficiency with space division multiple access.The deep multiplexing of time-frequency resources poses new challenges for multi-antenna precoding in multi-user systems,such as the limited feasible precoding region in low frequency transmission caused by the low dimensional antenna array,the non-convex optimization for hybrid precoding in mm Wave Massive MIMO systems,and the difficulties for interference alignment in multi-cell Massive MIMO systems.Focusing on the challenges above,this dissertation studies multi-antenna precoding for 5G/B5 G space division multiple access systems.The main research results are as follows.For low-frequency multi-user MIMO SDMA systems,this dissertation proposes a symbol pre-detection based constructive interference precoding(CIP)scheme,which directly optimizes the detection SER with beamforming.The proposed scheme reduces the SER by controlling the received constellation under the impact of the multi-user interference with an expanded feasible region of CIP,which is defined by the symbol detection SER function.Based on the proof of convexity and Karush-Kuhn-Tucker conditions,we prove that the optimal precoded signal for SER minimization is a linear combination of channel vectors,and propose the projection-revised feasible direction scheme for CIP(PRFD-CIP)to seek for the global optimal solution.In PRFD-CIP,the performance of the high SER users is improved with certain performance loss of the low SER users.The precoding is optimized with the balancing of SER among users.To avoid the complex expression of the derivative of the SER function,the updating direction in PRFD-CIP is developed with several gradient values which can be obtained via off-line calculation.The dimension of the optimization parameters is compressed with the concept of projection based revision,which enhances the efficiency of the feasible direction and contributes to less convergence time.Simulation results show that compared with the existing minimum mean square error based CIP scheme,PRFD-CIP obtains a signal-to-noise ratio(SNR)gain of more than 1.8d B.For mm Wave Massive MIMO SDMA systems,the thought of clustering in unsupervised learning is adopted to handle the non-convex constrains,which develops a clustering based hybrid precoding scheme with improved RF design.For fully-connected hybrid precoding(FHP),based on the ideal design with sufficient RF chains,we formulate the design with limited RF chains as a clustering problem,and propose a hierarchical agglomerative clustering based FHP scheme(HAC-FHP)to exploit the correlation among RF chains in the design of the ideal sufficient case.The RF chains with high correlation in the ideal design are replaced by an individual efficient RF chain,which reduces the performance loss caused by the limited RF chains.For adaptively-connected hybrid precoding(AHP),we formulate the design of the connection between RF chains and large scale antennas as a clustering problem,and propose a K-means based AHP scheme(KM-AHP)to exploit the correlation among antennas in full-digital structure.The flexibility in AHP is utilized to improve the connection between the antennas and RF chains.Simulation results show that compared with the classic decoupling scheme with SNR =-5 d B,the spectrum efficiency gap between HAC-FHP and full digital precoding is reduced by 40%,while KM-AHP can obtain the spectrum efficiency gain of 12 bps/Hz.For multi-cell massive MIMO SDMA systems,this dissertation proposes a interference subspace alignment(ISA)scheme with hybrid precoding for multi-cell systems,which improves the spectrum efficiency by optimizing the interference space.Considering the full digital structure first,we prove that when the number of transmit antennas is not less than that of overall users,each of which requires a single stream,ISA can be achieved with the maximal degree of freedom for interference subspace design.The ISA solution space is further provided in a matrix form which reflects the spatial resources brought by large scale antennas.Further,the approximated interference subspace alignment scheme with hybrid precoding(AISA-HP)is proposed to control the power of interference leakage,with the iterative optimization between the orthogonal matching designed RF chains and interference subspace.Simulation results show that compared with the interference alignment scheme,AISA-HP can improve the ceiling of spectrum efficiency by 23 bps/Hz with lower computational complexity.In summary,this dissertation proposes novel CIP,hybrid precoding and cooperative precoding schemes for low-frequency/mm Wave and multi-cell SDMA systems.The precoding schemes are improved in terms of beamforming,RF design and interference space,which contributes to low SER in symbol detection and high spectrum efficiency of the overall system.The methods in this dissertation provide innovative design concept and key technical support for the research of multi-antenna precoding in 5G/B5 G space division multiple access systems.更多还原