基于深度展开的大规模MIMO低轨卫星预编码方法  

Deep Unfolding-assisted Precoding for Massive MIMO LEO Satellite Communications

作　　者：王茂富 尤力[1,2] 周慧斌 孔庆付 高西奇 WANG Maofu;YOU Li;ZHOU Huibin;KONG Qingfu;GAO Xiqi(National Mobile Communications Research Laboratory,Southeast University,Nanjing 210096,China;Purple Mountain Laboratories,Nanjing 211111,China)

机构地区：[1]东南大学移动通信全国重点实验室,江苏南京210096 [2]紫金山实验室,江苏南京211111

出　　处：《移动通信》2024年第10期98-105,共8页Mobile Communications

基　　金：中央高校基本科研业务费专项资金资助(2242022k60007);中央高校基本科研业务费专项资金资助(2242023k5003)。

摘　　要：低轨卫星通信系统能够提供低成本、低功耗、高灵活性的发射需要,为实现下一代移动通信系统的空天地海一体化发挥重要作用。将大规模MIMO技术应用于低轨卫星通信系统中,能够显著提升低轨卫星通信系统的能量效率与频谱效率,其中预编码作为大规模MIMO系统中的关键技术而被广泛研究。WMMSE方法是优化预编码问题的一种方法,尽管可以有效地找到局部最优解,但它需要矩阵求逆、二分搜索等操作,导致算法复杂度较高。鉴于此,提出了一种基于深度展开的大规模MIMO低轨卫星预编码方法。首先,采用Dinkelbach算法和WMMSE方法交替优化求解能效最大化问题。进而将迭代算法的每一次迭代展开为类似于神经网络每一层的结构,通过将矩阵求逆用一阶泰勒展开替代并引入矩阵形式的可训练参数,以降低算法复杂度。仿真结果表明,所提方法能够在逼近采用交替优化的低轨卫星预编码方法性能同时有效降低复杂度。The low earth orbit satellite communication system can provide low-cost,low-power consumption and high-flexibility for transmission,which plays an important role in realizing the space-air-ground-sea integration of the next-generation mobile communication system.The application of massive multiple-input multiple-output(MIMO)technology to low earth orbit satellite communication systems can significantly improve the energy efficiency and spectral efficiency,and precoding has been widely studied as a key technology in massive MIMO systems.The weighted minimum mean-square error(WMMSE)method is an approach to optimize the precoding problem.Although it can effectively find the local optimal solution,it requires operations such as matrix inversion and binary search,resulting in high algorithm complexity.Therefore,this paper proposes a precoding method for massive MIMO low earth orbit satellites based on deep unfolding.Firstly,the Dinkelbach algorithm and the WMMSE method are used to alternately optimize the energy efficiency maximization problem.Then,each iteration of the iterative algorithm is expanded into a structure similar to that of each layer of the neural network,and the trainable parameters in the form of matrices are introduced by replacing the matrix inversion with the first-order Taylor expansion to reduce the complexity of the algorithm.Simulation results show that the proposed method can approach the performance of the low-orbit satellite precoding method with alternate optimization and effectively reduce the complexity.

关 键 词：低轨卫星 预编码 深度展开 大规模MIMO 

分 类 号：TN929.5[电子电信—通信与信息系统]