一种深度残差网络构建的5G无线智能传播模型

为建立更为准确的全覆盖、全应用、全频谱的5G无线信道模型,提出通过认知无线电与深度神经网络相结合的方法研究无线电波传播特性。根据传统无线传播模型并考虑到不同传播环境,根据信道大尺度衰落特性(包括路径损耗、阴影衰落和小尺度衰落特性)的统计结果,通过BP算法提取特征,并应用FeatureTools进行深度特征综合建立特征方程,计算特征变量与传播损耗的相关系数,进行相关系数的置信区间及变量独立性检验,最终筛选出22个特征并排序。基于深度残差网络建立传播路径损耗的回归模型,结合批正则化过拟合测算平均接收功率,为建立更精确的无线信道模型提供了量化依据,并最终在测试数据集上取得均方根误差8.36(本地)和10.03(云端)的成绩,对工程实践具有较强的参考价值。

5G Wireless Intelligent Propagation Channel Modeling Based On Deep Residual Network

In order to establish a more accurate 5G wireless channel model with full coverage,full application and full spectrum,the propagation characteristics of radio waves are studied by combining cognitive radio with deep neural network.According to the traditional wireless propagation model and simultaneous interpreting of different propagation environments,the scheme extracts features based on the BP algorithm,according to the statistical results of the large-scale fading characteristics of the channel,including the path loss,shadow fading and small scale fading characteristics,and applies FeatureTools to the depth feature synthesis to establish the characteristic equation,and calculates the designed characteristic variables and propagation loss.Finally,22 features are selected and sorted.Based on the residual network(ResNet),the regression model of propagation path loss is established.Through estimation of the reference signal receiving power with batch regularization over fitting,the specific parameters of the neural network are set reasonably,and a more accurate wireless signal model is established.Finally,the root mean square errors of 8.36(local) and 10.03(cloud) are achieved in the test data set,thus greatly reducing the cost of base station construction and improving the network construction efficiency.