<p>With the application of deep learning in spectrum sensing, the accuracy of detection has been significantly improved. This paper proposes a novel spectrum sensing method using continuous wavelet transform (CWT) and deep learning. The method first utilizes CWT to process the input and obtain the matrix of wavelet coefficients. Subsequently, the convolutional layers and LSTM units are employed to extract the time-frequency features from the coefficient matrix. Additionally, a residual structure is incorporated to further enhance network performance. The optimal structure and hyperparameters of the model are determined through parametric experiments. Simulation results demonstrate that the proposed method performs better than the widely used deep learning-based spectrum sensing methods. In particular, the sensing accuracy of this method surpasses that of other methods by more than <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11760_2025_4069_Article_IEq1.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="31" /> </InlineMediaObject> <EquationSource Format="TEX">\(25\%\)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>25</mn> <mo>%</mo> </mrow> </math></EquationSource> </InlineEquation> when the signal-to-noise ratio is equal to −&#xa0;16 dB. Finally, this paper develops a lightweight spectrum sensing hardware platform and deploys the method onto it.</p>

错误:搜索内容不能为空,请输入英文关键词
错误:关键词超出字数限制,请精简
高级检索

Deep CWT-ResLSTM for spectrum sensing in cognitive radio

  • Shengping Zhao,
  • Wei Xue,
  • Yidong Xu

摘要

With the application of deep learning in spectrum sensing, the accuracy of detection has been significantly improved. This paper proposes a novel spectrum sensing method using continuous wavelet transform (CWT) and deep learning. The method first utilizes CWT to process the input and obtain the matrix of wavelet coefficients. Subsequently, the convolutional layers and LSTM units are employed to extract the time-frequency features from the coefficient matrix. Additionally, a residual structure is incorporated to further enhance network performance. The optimal structure and hyperparameters of the model are determined through parametric experiments. Simulation results demonstrate that the proposed method performs better than the widely used deep learning-based spectrum sensing methods. In particular, the sensing accuracy of this method surpasses that of other methods by more than \(25\%\) 25 % when the signal-to-noise ratio is equal to − 16 dB. Finally, this paper develops a lightweight spectrum sensing hardware platform and deploys the method onto it.