基于改进环境编码的无线电环境地图预测方法  

Radio Environment Map Prediction Method Based on Improved Environment Coding

作　　者：田茂源 冯菊[1] 唐彪 薛文杰 TIAN Maoyuan;FENG Ju;TANG Biao;XUE Wenjie(Institute of Electromagnetics,Southwest Jiaotong University,Chengdu 610031,China)

机构地区：[1]西南交通大学电磁场与微波技术研究所,四川成都610031

出　　处：《无线电工程》2024年第11期2610-2617,共8页Radio Engineering

基　　金：四川省自然科学基金(2022NSFSC0494);国家自然科学基金(61801405)。

摘　　要：深度学习相比传统预测无线电环境地图(Radio Environment Map,REM)方法有着预测时间快、精度高的优点。然而模型为了保证预测的精度往往需要设计复杂的网络进行训练,导致耗费大量的训练时间。为降低模型的训练时长,实现REM快速构建,提出了一种改进深度学习模型结合环境编码的REM构建方法。该方法在深度学习网络结构中,通过利用轻量级视觉转换器(Mobile Vision Transformer,MobileViT)模块替换传统模型的卷积模块,增加了模型的全局视野;在输入数据预处理中,引入电波传播机理,提高了模型的可解释性与图像的一维熵,通过结合经验公式计算的路径损耗与天线位置图进行环境编码,进而与城市环境地图作为共同输入。仿真结果表明,改进模型在训练阶段具有更快的收敛速度;所提出的数据预处理方法能够对模型训练起到加速的作用。Deep learning has the advantages of faster prediction speed and higher accuracy than the traditional method of predicting Radio Environment Map(REM).However,in order to ensure the accuracy of prediction,it's required to design complex model network for training,which leads to a lot of training time.In order to reduce the training time of the model and realize the rapid construction of REM,a method with an improved deep learning model combined with environment coding is proposed for REM construction.In the deep learning network structure,the Mobile Vision Transformer(MobileViT)module is used to replace the convolution module of the traditional model,which increases the global view of the model.In the input data pre-processing,the radio wave propagation mechanism is introduced to improve the interpretability of the model and the one-dimensional entropy of the image.The environmental coding is carried out by combining the path loss calculated by the empirical formula and the antenna position map,and then the resulting encoding map is used together with the urban environment map as the common input.The simulation results show that the improved model has a faster convergence rate in the training stage.The proposed data preprocessing method can accelerate the model training.

关 键 词：电波传播 无线电环境地图 路径损耗预测 深度学习 图像生成 

分 类 号：TN011[电子电信—物理电子学]