计算机工程与应用 ›› 2024, Vol. 60 ›› Issue (5): 299-306.DOI: 10.3778/j.issn.1002-8331.2210-0295

• 网络、通信与安全 • 上一篇    下一篇

融合DAE-LSTM的认知物联网智能频谱感知算法

段闫闫,徐凌伟   

  1. 1.青岛科技大学 信息科学技术学院,山东 青岛 266061
    2.数字化学习技术集成与应用教育部工程研究中心,北京 100039
  • 出版日期:2024-03-01 发布日期:2024-03-01

DAE-LSTM-Fused Intelligent Spectrum Sensing Algorithm for Cognitive Internet of Things

DUAN Yanyan, XU Lingwei   

  1. 1.School of Information Science, Qingdao University of Science and Technology, Qingdao, Shandong 266061, China
    2.Engineering Research Center of Integration and Application of Digital Learning Technology, Ministry of Education, Beijing 100039, China
  • Online:2024-03-01 Published:2024-03-01

摘要: 第五代(fifth-generation,5G)移动通信技术的兴起,推动了物联网(Internet of things,IoT)的发展。然而,随着物联网数据传输量的爆发式增长,频谱资源短缺问题越来越严重。频谱感知技术极大的提高了物联网频谱利用率。但是,物联网移动通信环境的复杂性高以及信号易畸变的特性,对现有的频谱感知算法提出了重大挑战。因此,提出了一种融合去噪自编码器(denoising autoencoder,DAE)和改进长短时记忆(long short term memory,LSTM)神经网络的智能频谱感知算法。DAE通过编码和解码过程挖掘移动信号的底层结构特征,改进的LSTM频谱感知分类器模型结合过去时刻信息特征对时序信号序列进行分类。与支持向量机(support vector machine,SVM)、循环神经网络(recurrent neural network,RNN)、LeNet5、学习矢量量化(learning vector quantization,LVQ)和Elman算法相比,该算法的感知性能提高了45%。

关键词: 认知物联网, 智能频谱感知, 去噪自编码器, 长短时记忆网络

Abstract: The rise of the fifth-generation (5G) mobile communication, the development of Internet of things (IoT) is promoted. However, with the explosive growth of IoT data transmission volume, the shortage of spectrum resources is becoming more and more severe. Spectrum sensing technology greatly improves the spectrum utilization of the Internet of things. However, the IoT mobile communication environment has the characteristics of high complexity and easy signal distortion, which poses a major challenge to the existing spectrum sensing. Thus, this paper proposes an intelligent spectrum sensing algorithm that fused with denoising autoencoder (DAE) and improved long short term memory (LSTM) neural network. DAE excavates the internal structural features of mobile signals through encoding and decoding. The improved LSTM spectrum sensing classifier model is designed to classify time series signal sequences combined with past moment information features. Finally, the proposed algorithm achieves 45% higher sensing accuracy than support vector machine (SVM), Elman, LeNet5, learning vector quantization (LVQ) and recurrent neural network (RNN) algorithms.

Key words: cognitive Internet of things (IoT) networks, intelligent spectrum sensing, denoising autoencoder (DAE), long short term memory (LSTM)