基于卷积神经网络的 AFDM 信号检测模型
Convolutional Neural Network-Based Signal Detection model for AFDM
黄伟清1
作者信息
- 1. 北京邮电大学人工智能学院，北京，100876
- 折叠
摘要
仿射频分复用（AFDM）因其对双选择衰落信道的高度适应性，已成为下一代高移动性通信中颇具潜力的波形方案。然而，AFDM系统中的最优信号检测计算复杂度极高，而传统线性及迭代检测器在恶劣信道条件下常面临性能损失。为应对这些挑战，本文提出了一种新颖的基于深度学习的信号检测框架，将传统检测算法--特别是最小均方误差（MMSE）和消息传递（MP）--与一维卷积神经网络（1D-CNN）相结合。该方案利用传统检测器提供的检测结果作为先验信息，使CNN能够有效学习并抑制残余干扰与非线性信道失真。仿真结果表明，所提出的检测器在显著超越独立MMSE与MP算法，实现了更优的误码性能，在高移动性环境中展现出卓越的鲁棒性，且保持了可观的计算复杂度。
Abstract
Affine Frequency Division Multiplexing (AFDM) has emerged as a promising waveform candidate for next-generation high-mobility communications due to its resilience against doubly dispersive channels. However, optimal signal detection in AFDM systems remains computationally prohibitive, while traditional linear and iterative detectors often suffer from performance degradation under severe channel conditions. To address these challenges, this paper proposes a novel deep learning-based signal detection framework. We design an architecture that integrates traditional detection algorithms, specifically Minimum Mean Square Error (MMSE) and Message Passing (MP), with a One-Dimensional Convolutional Neural Network (1D-CNN). The proposed scheme utilizes coarse estimates from these traditional detectors as prior information, enabling the CNN to effectively learn and suppress residual interference and non-linear channel distortions. Simulation results demonstrate that the proposed detector significantly outperforms standalone MMSE and MP algorithms, achieving superior error performance and robustness in high-mobility environments while maintaining manageable computational complexity.关键词
通信与信息系统，AFDM，信号检测，深度学习，CNNKey words
Communication and Information Systems/Affine Frequency Division Multiplexing (AFDM)/Signal Detection/Deep Learning/Convolutional Neural Network (CNN)/引用本文复制引用
黄伟清.基于卷积神经网络的 AFDM 信号检测模型[EB/OL].(2026-01-22)[2026-01-25].http://www.paper.edu.cn/releasepaper/content/202601-50.学科分类
通信/无线通信
评论