基于深度学习的仿射频分复用接收机设计  

Deep Learning-based Receiver Design for Affine Frequency Division Multiplexing

作　　者：黄鹏飞 李强 HUANG Pengfei;LI Qiang(College of Information Science and Technology,Jinan University,Guangzhou 510632,China)

机构地区：[1]暨南大学信息科学技术学院,广东广州510632

出　　处：《无线电工程》2025年第4期739-748,共10页Radio Engineering

基　　金：国家自然科学基金青年项目(62201228);广州市科技计划项目(2024A04J0191);中央高校基本科研业务费专项资金(21624405)。

摘　　要：针对双选择性衰落信道下的仿射频分复用(Affine Frequency Division Multiplexing,AFDM)设计了基于深度学习的接收机。构建深度神经网络(Deep Neural Network,DNN),使用训练数据对DNN进行离线训练,在线部署在接收机上以输出传输的比特。仿真结果表明,当导频和数据之间存在保护间隔时,基于深度学习的接收机能够达到传统信道估计的误比特率(Bit Error Rate,BER)性能。当导频和数据间的保护间隔缩减时,基于深度学习的接收机的BER性能优于传统的信道估计方案。在存在导频-数据干扰的情况下,该接收机比现有方案更具鲁棒性。In this research,a deep learning-based receiver is designed for Affine Frequency Division Multiplexing(AFDM) over doubly selective fading channels.A Deep Neural Network(DNN) is constructed,trained offline using training data,and then deployed online at the receiver to output transmitted bits.Simulation results demonstrate that the proposed deep learning-based receiver achieves comparable Bit Error Rate(BER) performance with respect to the conventional channel estimation scheme when a guard interval exists between the pilot and data.Notably,as the guard interval between the pilot and data decreases,the BER performance of the deep learning-based receiver surpasses that of the traditional channel estimation scheme.Moreover,the proposed receiver exhibits greater robustness than existing schemes in the presence of pilot-data interference.

关 键 词：仿射频分复用 深度神经网络 双选择性衰落信道 信道估计 符号检测 

分 类 号：TN929.5[电子电信—通信与信息系统]