空间非平稳信道条件下XL-MIMO系统联合天线选择和波束成形  

Joint Antenna Selection and Beamforming in XL-MIMO Systems under Spatially Non-stationary Channel Condition

作　　者：褚宏云 贾帅 王鑫 闫桐菲 CHU Hongyun;JIA Shuai;WANG Xin;YAN Tongfei(School of Telecommunications and Information Engineering,Xi’an University of Posts and Telecommunications,Xi’an 710121,China)

机构地区：[1]西安邮电大学通信与信息工程学院,西安710121

出　　处：《电讯技术》2024年第12期1946-1954,共9页Telecommunication Engineering

基　　金：国家自然科学基金资助项目(62401467);陕西省自然科学基金项目(2202JQ-635)。

摘　　要：由于超大规模MIMO(Extremely Large-scale MIMO,XL-MIMO)系统中空间非平稳性的存在,使得部分天线对系统性能贡献较小,从而增加了系统能耗。通过天线选择并结合波束成形从而优化系统性能。以最小化基站发射功率为目标建模,在满足信干噪比和基站激活天线数的约束下,优化基站处的波束成形矩阵。由于该优化问题是典型的混合整数非线性规划问题,传统方法使用连续近似来求解,然而获得的解都是次优解。鉴于此,首先提出采用分支定界算法(Branch and Bound,BAB)求解上述优化问题,从而保证解的最优性。然而,BAB算法在处理大规模问题,特别是基站天线数大于128时,计算复杂度往往过高。为了解决这一问题,提出了一种基于图神经网络和多层感知机(Graph Neural Network and Multilayer Perceptron,GNN+MLP)的方法,通过利用GNN在BAB树的根节点提取一次全局特征,并在每个子节点利用MLP提取局部特征,通过结合全局特征和局部特征来训练一个二进制的节点分类器,以判断当前节点是否需要进一步分支,从而加速计算的过程。仿真结果表明,在天线数等于512时,GNN+MLP比BAB减少了54.2%的计算时间。The spatial non-stationarity in extremely large-scale MIMO(XL-MIMO)systems causes certain antennas to contribute less to system performance,thereby increasing energy consumption.To optimize system performance,antenna selection(AS)is combined with beamforming(BF).The problem is modeled with the objective of minimizing base station transmission power,while satisfying the constraint of interference-plus-noise signal-to-ratio(SINR)and the number of active antennas,by optimizing the beamforming matrix at the base station.Since this optimization problem is a typical mixed-integer nonlinear(MINP)programming problem,traditional methods rely on continuous approximation(CA)to solve it,yielding suboptimal solutions.In view of this,the authors propose using a Branch and Bound(BAB)algorithm to solve the optimization problem and ensure solution optimality.However,the BAB algorithm faces high computational complexity,especially when the number of base station antennas exceeds 128.To overcome this,a method based on the graph neural network(GNN)and multilayer perceptron(MLP)is This introduced.approach uses GNN to extract global features at the root node of the BAB tree,while MLP extracts local features at each child node.By combining global and local features,a binary node classifier is trained to decide whether further branching is necessary,thereby accelerating the computation process.Simulation results demonstrate that when the number of antennas is 512,GNN+MLP takes 54.2%less computation time than BAB.

关 键 词：超大规模MIMO 空间非平稳性 分支定界算法 图神经网络 多层感知机 

分 类 号：TN929.5[电子电信—通信与信息系统]