基于机器学习的太赫兹信道预测建模研究  

Research on Terahertz Channel Prediction Modeling Based on Machine Learning

作　　者：王世豪 李双德 刘芫健[1] 梁静宜 蒋晨晨 WANG Shihao;LI Shuangde;LIU Yuanjian;LIANG Jingyi;JIANG Chenchen(College of Electronic and Optical Engineering&College of Flexible Electronics(Future Technology),Nanjing University of Posts and Telecommunications,Nanjing 210023,China)

机构地区：[1]南京邮电大学电子与光学工程学院、柔性电子(未来技术)学院,江苏南京210023

出　　处：《无线电通信技术》2024年第5期914-920,共7页Radio Communications Technology

基　　金：国家自然科学基金(62371248);南京邮电大学引进人才自然科学研究启动基金(NY222059)。

摘　　要：针对6G移动通信的通信场景复杂化、数据海量化,以及传统信道建模方法带来的测量成本昂贵、建模复杂度高等挑战,将机器学习中的反向传播神经网络(Back Propagation Neural Network,BPNN)应用到室内太赫兹信道建模中,有效降低了建模复杂度,提高了建模效率。建立了基于遗传算法(Genetic Algorithm,GA)和蚁群算法(Ant Colony Optimization,ACO)混合优化的BPNN信道参数预测模型,对太赫兹无线信道的大小尺度特性进行了学习与预测,并与传统的BPNN模型、GA-BP和ACO-BP的预测结果进行了比较,验证了所建立模型的准确性和有效性。结果表明,遗传蚁群反向传播(Genetic Algorithm-Ant Colony Optimization-Back Propagation,GA-ACO-BP)模型的预测值和实际值间的误差更小、拟合度更高,该模型的预测性能相较于其他3种模型更优。基于GA-ACO混合优化的BPNN能够在小数据量的情况下对信道参数进行学习和预测,可用于未来基于测量的无线信道建模分析中。The intricate communication scenarios in 6G mobile communication pose significant challenges,including high modeling complexity,prohibitive measurement costs,and overwhelming data volumes.Back Propagation Neural Network(BPNN)from machine learning is applied to indoor terahertz channel modeling to overcome these challenges.This approach effectively reduces modeling complexity and improves modeling efficiency.A BPNN channel parameter prediction model based on a hybrid optimization of Genetic Algorithm(GA)and Ant Colony Optimization(ACO)is established to study and predict large-and small-scale characteristics of terahertz wireless channels.Prediction results are compared with traditional BPNN model,GA-BP,and ACO-BP,and the accuracy and effectiveness of the established model are verified.Results indicate that the error between the predicted and actual values of Genetic Algorithm-Ant Colony Optimization-Back Propagation(GA-ACO-BP)model is smaller and a better fit.The model demonstrated superior prediction performance compared to other three models.BPNN based on GA-ACO hybrid optimization can learn and predict channel parameters with a small amount of data,making it applicable for future measurement-based wireless channel modeling analysis.

关 键 词：太赫兹 信道建模 射线跟踪 机器学习 

分 类 号：TN929.5[电子电信—通信与信息系统]