基于条件神经过程的星上多用户检测算法  

On-board Multi-User Detection Algorithm Based on Conditional Neural Process

作　　者：刘轶伦 金亮[2] 李佳立[2] 朱立东[1] LIU Yilun;JIN Liang;LI Jiali;ZHU Lidong(National Key Laboratory of Science and Technology of Communications,University of Electronic Science and Technology of China,Chengdu 611731,China;Space Star Technology Co.,Ltd.,Beijing 100095,China)

机构地区：[1]电子科技大学通信抗干扰技术国家级重点实验室,四川成都611731 [2]航天恒星科技有限公司,北京100095

出　　处：《天地一体化信息网络》2021年第4期60-65,共6页Space-Integrated-Ground Information Networks

基　　金：国家重点研发计划项目(No.2019YFB1803102);国家自然科学基金面上项目(No.61871422)。

摘　　要：全球全天候无缝覆盖的特性使得卫星通信成为6G潜在的重要组成部分,而实现卫星智能化的一个重要前提是卫星具有星上处理能力。多用户检测是无线通信中抑制多址干扰的经典方法,如最小均方误差、高斯过程回归等算法,由于检测过程中需要求逆矩阵,算法复杂度通常为立方级,难以直接应用到处理能力受限的卫星平台。条件神经过程结合神经网络低复杂度和高斯过程小样本的特性,利用神经网络将高斯过程参数化,避免求逆矩阵,从而降低计算复杂度。研究条件神经过程在多用户检测中的应用,仿真结果表明,在降低复杂度的同时,条件神经过程还极大地提升了多用户检测的误码率性能。With the characteristics of all-terrain,all-weather and seamless coverage,satellite communications have become a potentially important part of 6G.An important prerequisite for achieving satellite intelligence is that the satellite have on-board processing capabilities.Multi-user detection(MUD)is a classic method of suppressing multiple access interference(MAI)in wireless communication,such as MMSE,Gaussian processregression(GPR)and other algorithms.Due to the inverse matrix required in the detection process,the algorithm complexity is usually cubic,and it is diffi cult to directly apply to satellite platforms because of its limited processing capabilities.The conditional neural process combined the characteristics of the low complexity of the neural network and the data-effi cient of the Gaussian process.The neural network was used to parameterized the Gaussian process to avoided the inversion of the matrix,thereby reduced the computational complexity.The application of conditional neural process in MUD was studied.The simulation results showed that,while reduced complexity,conditional neural process also greatly improved the performance of bit error rate(BER).

关 键 词：卫星通信 多用户检测 高斯过程 条件神经过程 低复杂度 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]