基于神经网络的无线信道场景识别

Wireless channel scenario recognition based on neural network

无线信道场景识别对于无线资源调度和系统性能的优化等具有重要意义.文中基于QuaDriGa平台研究了反向传播神经网络(back propagation neural network,BPNN)和卷积神经网络(convolutional neural networks,CNN)在无线信道场景识别中的应用.首先,利用QuaDriGa生成不同场景下的信道冲激响应(channel impulse response,CIR),并提取时延扩展、角度扩展等信道参数.然后,对于BPNN,直接利用其对不同场景的参数进行训练;对于CNN,需要经过"抽头移动、数量级微调、自相关"等操作将一维的CIR转化为二维图像再进行训练.最后,计算识别准确率并利用K折交叉验证该两种模型的泛化能力.结果表明,CNN比BPNN识别精度高,但BPNN识别效率更高,二者均可用于未来信道场景的智能感知和识别.

Wireless channel scene identification is of great significance for wireless resource scheduling and system performance optimization.Based on the QuaDriGa platform,this paper studies the application of back propagation neural network(BPNN)and convolutional neural networks(CNN)in wireless channel scene recognition.Firstly,based on QuaDriGa,the channel impulse response(CIR)in different scenarios is generated,and channel parameters such as delay spread and angle spread are extracted.In this paper,BPNN is used to train the parameters of different scenes.For CNN,it is necessary to convert the one-dimensional CIR into a two-dimensional image and then train it through operations such as"tap movement,order of magnitude fine-tuning,and autocorrelation".Finally,the recognition accuracy is calculated and the K-fold cross-validation is used to verify the generalization ability of the two models.The results show that CNN has higher recognition accuracy than BPNN,but BPNN has higher recognition efficiency.Both can be used for intelligent sensing and recognition of future communication network scenarios.