| 引用本文: | 罗文宇， 钟云开， 邵 霞， 段臣续. 基于毫米波雷达感知的CNN⁃ConvLSTM多时刻阻塞预测方法[J]. 雷达科学与技术, 2023, 21(5): 531-538.[点击复制] |
| LUO Wenyu, ZHONG Yunkai, SHAO Xia, DUAN Chenxu. Multi⁃Time Blocking Prediction Method of CNN⁃ConvLSTM Based on Millimeter Wave Radar Perception[J]. Radar Science and Technology, 2023, 21(5): 531-538.[点击复制] |
|
| 摘要: |
| 针对当前毫米波雷达辅助无线通信单时刻阻塞预测无法适应高移动复杂场景的问题，提出一种结合卷积神经网络（CNN）和时空序列预测模型（ConvLSTM）的多时刻持续阻塞预测方法。该方法通过构造时间分布层存储CNN提取的热图特征，解决了卷积神经网络单次处理多张雷达热图可能存在的特征缺失问题，实现了连续热图特征的多帧分组。进而，利用ConvLSTM对输入多帧时空序列进行处理实现多时刻阻塞预测。利用DeepSense 6G真实场景数据的实验结果表明，该方法的多时刻预测结果均能达到90%的准确率和80%以上的F1?score，具备多时刻精确阻塞预测能力。本文方法在复杂动态环境下高频段通信的高可靠、低时延QoS保障方面具有重要的理论和应用价值。 |
| 关键词: 毫米波雷达 无线通信 卷积神经网络 时空序列预测模型 多时刻阻塞预测 |
| DOI：DOI:10.3969/j.issn.1672-2337.2023.05.009 |
| 分类号:TN957 |
| 基金项目:国家自然科学基金联合基金（No.U1804148）；河南省科技攻关项目（No.232102210141） |
|
| Multi⁃Time Blocking Prediction Method of CNN⁃ConvLSTM Based on Millimeter Wave Radar Perception |
|
LUO Wenyu, ZHONG Yunkai, SHAO Xia, DUAN Chenxu
|
|
School of Electronic Engineering, North China University of Water Resources and Electric Power, Zhengzhou 450046, China
|
| Abstract: |
| Aiming at the problem that the current millimeter?wave radar assisted wireless communication single time blocking prediction cannot adapt to the high movement complex scene, a multi?time continuous blocking prediction method combining convolutional neural network （CNN） and spatio?temporal series prediction model （ConvLSTM） is proposed. By constructing a time distribution layer to store the heat map features extracted by CNN, this method solves the problem of feature missing that may exist in the single processing of multiple radar heat maps by CNN, and realizes the multi?frame grouping of continuous heat map features. Then, ConvLSTM is used to process the input multi?frame spatio?temporal sequence to achieve multi?time blocking prediction. The experimental results using DeepSense 6G real scene data show that the multi?time prediction results of this method can reach 90% accuracy and more than 80% F1?score, and the proposed method has the ability of multi?time accurate blocking prediction. The method in this paper has important theoretical and application values in high reliability and low delay QoS guarantee of high frequency communication in complex dynamic environment. |
| Key words: millimeter⁃wave radar wireless communication convolutional neural network（CNN） spatio⁃temporal series prediction model （ConvLSTM） multi⁃time blocking prediction |
