Optical Instruments, Volume. 45, Issue 6, 60(2023)

Research on classification of high order QAM modulation with improved Transformer network

Yi AN and Lan XIANG
Author Affiliations
  • School of Opitcal-Electrical and Computer Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China
  • show less
    Figures & Tables(14)
    QAM constellation diagram without carrier frequency offset and phase offset (SNR is 20 dB)
    Time-domain scatter plots of and for 16 QAM and 32 QAM (SNR is 20 dB)16 QAM和32 QAM的 与时域散点图(SNR 为 20 dB)
    Constellation diagrams of 16 QAM and 32 QAM with carrier frequency offset of 500 Hz or carrier phase offset of π/6 (SNR is 20 dB)
    Time-domain scatter plots of and for 16 QAM and 32 QAM with a carrier frequency offset of 500 Hz (SNR is 20 dB)载波频偏为500 Hz时16 QAM和32 QAM的与时域散点图(SNR 为 20 dB)
    Time-domain scatter plots of and for 16 QAM and 32 QAM with a carrier phase offset of π/6 (SNR is 20 dB)载波相偏为π/6时16 QAM和32 QAM的与时域散点图(SNR 为 20 dB)
    Network framework
    Computational process of multi-head attention mechanism
    Recognition rate without frequency offset and phase offset
    Recognition rate with frequency offset 500 Hz
    Recognition rate with phase shift π/6
    Confusion matrix on three datasets (SNR is 20 dB)
    Training and testing modulation classification accuracy and loss curves of three datasets (SNR is 20 dB)
    • Table 1. Parameter output of each layer network

      View table
      View in Article

      Table 1. Parameter output of each layer network

      网络层名称网络输出备注
      data preprocessing$ (Batch,N) $添加CLK
      embedding$ (Batch,N,D) $$ D $为词向量维度
      mutil-head attention input$ (Batch, h, $$ N,D/h) $$ h $为注意力头数
      mutil-head attention output$ (Batch,N,D) $多头注意力计算
      Add & Norm$ (Batch,N,D) $归一化
      FFN in$ (Batch,N,4D) $非线性连接
      FFN out$ (Batch,N,D) $
      Add & Norm$ (Batch,N,D) $归一化
      Linear$ (Batch,D) $取出CLK
      Linear & Softmax$ (Batch,M) $调制分类
    • Table 2. Comparison of the results in this article and the results obtained by the other methods

      View table
      View in Article

      Table 2. Comparison of the results in this article and the results obtained by the other methods

      调制格式文献[5]文献[8]本文
      16 QAM10 dB20 dB6 dB
      32 QAM14 dB30 dB8 dB
      64 QAM10 dB20 dB10 dB
      128 QAM14 dB25 dB12 dB
      256 QAM25 dB14 dB
    Tools

    Get Citation

    Copy Citation Text

    Yi AN, Lan XIANG. Research on classification of high order QAM modulation with improved Transformer network[J]. Optical Instruments, 2023, 45(6): 60

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category:

    Received: Feb. 26, 2023

    Accepted: --

    Published Online: Feb. 29, 2024

    The Author Email:

    DOI:10.3969/j.issn.1005-5630.202302260027

    Topics