基于端到端系统的导频插入和信道估计设计  

Pilot Insertion and Channel Estimation Design Based on End-to-End System

作　　者：卓娜娜 向路平 胡杰 杨鲲 ZHUO Nana;XIANG Luping;HU Jie;YANG Kun(Yangtze Delta Region Institute of University of Electronic Science and Technology of China(Huzhou),Huzhou 313000,China;School of Information and Communication Engineering,University of Electronic Science and Technology of China,Chengdu 611731,China;School of Intelligent Software and Engineering,Nanjing University(Suzhou Campus),Suzhou 215000,China)

机构地区：[1]电子科技大学长三角研究院(湖州),浙江湖州313000 [2]电子科技大学信息与通信工程学院,四川成都611731 [3]南京大学(苏州校区)智能软件与工程学院,江苏苏州215000

出　　处：《移动通信》2024年第6期34-39,51,共7页Mobile Communications

基　　金：国家自然科学基金重点项目“时延敏感的端到端传输理论和算法研究”(62132004)。

摘　　要：自深度学习技术驱动的端到端通信系统解决方案被引入后,通信系统的架构模式已经发生了巨大的变革。虽然该技术在AWGN信道上表现良好,但在无线信道环境变化情况下,其表现并不尽如人意。因此,当前普遍采用的策略是引入导频信号,将导频按照头部位置插入或固定间隔插入的方式插入数据流中,同时在端到端系统之外增设信道估计模块以补充信道信息。然而,这种做法实际上与端到端一体化的理念相矛盾。提出了一种综合的端到端系统,利用深度神经网络进行导频插入位置选择并设计接收机,使系统能够有效收集信道信息,进而顺利恢复传输数据。与传统的导频插入方案相比,该系统具备明显的性能优势,误码率降低1~2 dB。另外,针对各类信道环境,系统内部的导频插入算法都可以提供具有针对性且高效的导频插入方案,因此对通信环境具有极高的适应性。The architecture pattern of communication systems has undergone significant transformation with the introduction of end-to-end solutions driven by deep learning technology.While this technology performs well in AWGN channels,its performance is not satisfactory under varying wireless channel environments,Therefore,the prevailing strategy is to introduce pilot signals,which are inserted into the data stream either at header positions or at fixed intervals,while additional channel estimation modules are added outside the end-to-end system to supplement channel information.However,this approach is fundamentally contradictory to the concept of end-to-end integration.To this end,a comprehensive end-to-end system is proposed to utilize deep neural networks for selecting pilot insertion positions and designing receiver,enabling the system to effectively collect channel information and smoothly recover transmitted data.Compared with traditional pilot insertion schemes,this system demonstrates significant performance advantages,reducing the bit error rate by 1-2 dB.Furthermore,the pilot insertion algorithms within the system are tailored to various channel environments,providing targeted and eficient pilot insertion strategies,thus exhibiting a high degree of adaptability to communication environments.

关 键 词：端到端系统 信道估计 深度神经网络 导频插入 动态通信环境 

分 类 号：TN929.5[电子电信—通信与信息系统]