基于全局深度神经网络学习的通信系统最优重构方法  

Optimal reconfiguration method for communication systems based on global deep neural network learning

作　　者：丁丹 张美娟 杨柳 张伟 韩儒磊 DING Dan;ZHANG Meijuan;YANG Liu;ZHANG Wei;HAN Rulei(Space Engineering University,Beijing 101416,China)

机构地区：[1]航天工程大学,北京101416

出　　处：《航天工程大学学报》2025年第2期76-81,共6页

基　　金：智能化测运控教育部重点实验室基金资助项目(CYK2024-01-03)。

摘　　要：针对传统通信系统分模块设计与优化,难以精确拟合复杂信道环境的问题,提出一种基于全局深度神经网络(Deep Neural Network,DNN)学习的系统最优重构方法,将深度学习前沿技术和传统通信基础有机融合。通过构建全局DNN学习智能自优化架构模型,采用条件生成对抗网络信道生成器DNN完成对复杂动态信道的最优拟合,设计收发联动迭代重构训练过程并利用计算机进行仿真验证。研究了系统最优重构方法和高效在线学习技术以及递进式优化策略,设计了高效的智能自优化策略,完成了对时变信道的实时最优跟踪,实现了传输性能对动态电磁对抗环境的连续、最佳匹配。研究结果表明:相比传统方法,本方法保持误码率性能更优,有更强的抗干扰性能,且运算复杂度相当。To address the challenge of traditional communication systems'modular design and optimization,which struggle to accurately model complex channel environments,an optimal system reconstruction method based on global deep neural network(DNN)learning is proposed,which organically integrates cutting-edge deep learning technologies with foundational communication principles.By constructing an intelligent self-optimizing architecture model via global DNN learning,the conditional generative adversarial network channel generator DNN is used to complete the optimal fitting of complex dynamic channels.A joint transceiver iterative reconstruction training process is designed and verified by computer simulation.The optimal reconstruction method of the system,efficient online learning technology and progressive optimization strategy are studied.This framework enables real-time optimal tracking of time-varying channels and achieves continuous,optimal alignment of transmission performance with dynamic electromagnetic countermeasure environments.Experimental results demonstrate that,compared to traditional methods,the proposed approach maintains superior bit error rate performance,exhibits stronger anti-jamming capability,and retains equivalent computational complexity.

关 键 词：深度学习 全局深度神经网络学习 智能自优化策略 时变信道 误码率 

分 类 号：TN914.4[电子电信—通信与信息系统]