融合多层注意力机制与CNN-LSTM的反向散射信道预测  

Backscatter Channel Prediction by Incorporating CNN-LSTM and Multi-layer Attention Mechanism

作　　者：徐双 文永新 刘文斌 李佳龙 李灯熬 赵菊敏[2,3,4] XU Shuang;WEN Yongxin;LIU Wenbin;LI Jialong;LI Dengao;ZHAO Jumin(College of Computer Science and Technology(College of Data Science),Taiyuan University of Technology,Taiyuan 030024,China;College of Electronic Information and Optoelectronic Engineering,Taiyuan University of Technology,Taiyuan 030024,China;Key Laboratory of Big Data Fusion Analysis and Application of Shanxi Province,Taiyuan 030024,China;Intelligent Perception Engineering Technology Center of Shanxi,Taiyuan 030024,China)

机构地区：[1]太原理工大学计算机科学与技术学院(大数据学院),太原030024 [2]太原理工大学电子信息与光电工程学院,太原030024 [3]大数据融合分析与应用山西省重点实验室,太原030024 [4]山西省智能感知工程研究中心,太原030024

出　　处：《小型微型计算机系统》2024年第9期2278-2284,共7页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(62102280)资助;山西省基础研究计划项目(20210302124167)资助;山西省重点研发计划项目(202102020101001)资助;国家自然科学基金面上项目(61972273)资助;山西省关键核心技术和共性技术研发攻关专项项目(2020XXX007)资助.

摘　　要：反向散射通信系统频谱资源十分有限且易受链路突变性影响,信道预测是提高其频谱资源利用率和通信质量的一种有效方法.但大多数现有预测方法的预测精度偏低、依赖完全已知的信道状态信息、适用性受限.为此,本文提出了一种融合多层注意力机制与卷积神经网络-长短期记忆网络(Convolutional Neural Networks-Long Short-Term Memory,CNN-LSTM)的信道预测方法.利用CNN模型与注意力机制提取接收信号强度序列的特征,并进一步使用LSTM模型与注意力机制提取其跨时间步长的特征,从而实现对信道指标的预测.最后,借助商用阅读器与标签采集3种不同场景下的信道状态数据,并基于Tensorflow与Keras验证了所提预测方法性能.结果表明,融合多层注意力机制与CNN-LSTM的信道预测方法具有较强的场景适用性,且其预测准确性较高.Backscatter communication system has very limited spectrum resources and is susceptible to link bursts.Channel prediction is an effective method to improve its spectrum resource utilization and communication quality.However,most existing prediction methods have low prediction accuracy,rely on completely known channel state information,and have limited applicability.To this end,a channel prediction method that incorporates multi-layer attention mechanisms and Convolutional Neural Networks-Long Short-Term Memory(CNN-LSTM)is proposed.The features of the received signal strength sequence are extracted by using a CNN model with the attention mechanism,and the features across time steps are further extracted by using an LSTM model with the attention mechanism to realize the prediction of the channel metrics.Finally,the channel state data of three different scenes are collected with the help of commercial readers and tags,and the performance of the proposed prediction method is verified based on Tensorflow and Keras.The results show that the channel prediction method incorporating multi-layer attention mechanisms and CNN-LSTM is more applicable,and its prediction accuracy is higher.

关 键 词：反向散射通信 信道预测 卷积神经网络 长短期记忆网络 多层注意力机制 

分 类 号：TP927[自动化与计算机技术]