引用本文: |
-
黄佳,李晓娜,张仕响,高宗宁,孟晨,魏冬.基于跨层数据融合分析的移动通信网络空口干扰识别技术研究[J].信息安全学报,已采用 [点击复制]
- Huang Jia,Li Xiaona,Zhang Shixiang,Gao Zongning,Meng Chen,Wei Dong.Research on Air Interface Interference Recognition Technology in Mobile Communication Networks Based on Cross-Layer Data Fusion Analysis[J].Journal of Cyber Security,Accept [点击复制]
|
|
摘要: |
由于移动通信频段干扰信号多种多样,特征提取困难,违规用频、恶意干扰等行为难以快速自动化识别,干扰信号对通信网络的影响难以精确评估。本文针对这一问题,基于空口采集数据和基站测量上报数据分别提出了上下行通信链路的空口干扰识别技术,构建干扰识别模型,实现了干扰自动化识别,并评估了干扰识别准确度。本文在终端侧,结合信号特征提取以及深度学习的优势,通过提取信号的时频资源块(Resource Block,RB)占用特征,构造时频资源块特征图,利用半监督学习的生成对抗网络(Generative Adversarial Network,GAN)学习正常通信业务情况下的移动通信时频资源占用行为,并利用训练的模型对干扰信号进行准确识别,解决干扰信号特征提取困难,难以自动化识别问题;在基站侧,利用基站导出的信号强度、信号质量和信噪比等信道测量数据,使用长短时记忆网络(Long Short-Term Memory,LSTM)自编码器模型学习正常信号模式以判别干扰,并评估干扰对通信网络的影响。该方法结合了空口采集数据和基站测量数据,实现了终端侧在语音以及视频正常业务行为情况下六种干扰的检测,以及基站侧对定频干扰实现精准识别,并评估了干扰对通信网络的影响。实验结果表明,终端侧方法对 6 种干扰总体识别 F1 分数达 0.95 以上;基站侧方法对定频干扰识别 F1 分数达 0.99,优于同等条件下的单类支持向量机(One-Class Support Vector Machine,OCSVM)、主成分分析(Principal Component Analysis,PCA)与孤立森林(Isolation Forest,IForest)方法。 |
关键词: 跨层数据融合 干扰识别 信号处理 图像识别 GAN 网络 异常检测 |
DOI:10.19363/J.cnki.cn10-1380/tn.2025.04.24 |
投稿时间:2024-01-28修订日期:2024-03-28 |
基金项目:国家重点研发计划 |
|
Research on Air Interface Interference Recognition Technology in Mobile Communication Networks Based on Cross-Layer Data Fusion Analysis |
Huang Jia, Li Xiaona, Zhang Shixiang, Gao Zongning, Meng Chen, Wei Dong
|
(Institute of Information Engineering,CAS) |
Abstract: |
Due to the diverse interference signals in mobile communication frequency bands and the difficulty of feature extraction, the rapid and automated identification of activities such as unauthorized frequency use and malicious interference
is challenging. The impact of interference signals on communication networks is difficult to assess accurately. This paper suggests methods to find air interface interference for uplink and downlink communication. It collects air interface data and base station measurement data. It constructs an interference recognition model, automatically identifies interference, and evaluates accuracy. This paper, at the terminal side, combines the advantages of signal feature extraction and deep learning. By extracting the time-frequency resource block (RB) occupancy features of signals, time-frequency resource block feature maps are constructed. Utilizing a semi-supervised learning model, Generative Adversarial Network (GAN), the study focuses on learning the normal time-frequency resource block occupancy behavior of mobile communication under regular communication scenarios. The trained model is then employed for accurate identification of interference signals, addressing challenges related to the difficulty of interference signal feature extraction and automated recognition. On the base station side, signal metrics such as Reference Signal Received Power, Reference Signal Received Quality, and Signal-to-Interference-plus-Noise Ratio, exported by the base station, are utilized. A Long Short-Term Memory (LSTM) autoencoder model is employed to learn normal signal patterns for interference detection. The paper also evaluates the impact of interference on communication networks. This method combines air interface data collection and base station measurement data, achieving the detection of six types of interference at the terminal side under normal voice and video business scenarios. Additionally, on the base station side, it achieves precise identification of fixed-frequency interference and assesses the impact of interference on communication networks. Experimental results demonstrate that the terminal-side method achieves an overall F1 score above 0.95 for the identification of the six types of interference. The base station-side method achieves an F1 score of 0.99 for the identification of fixed-frequency interference, surpassing single-class Support Vector Machine (OCSVM), Principal Component Analysis (PCA), and Isolation Forest (IForest) methods under equivalent conditions. |
Key words: Cross-Layer Data Fusion Interference Recognition Signal Processing Image Recognition Generative Adversarial Networks Anomaly Detection. |