Fig 1.
Wireless communication network.
Fig 2.
Deep CNN architecture for AMC: (a) the overall architecture of DFENet, (b) the structure of DFE Block, (c) the structure of FEN block.
Table 1.
Comparison of DFENet with existing multi-branch and multiscale architectures.
Table 2.
Detailed architecture of the DFENet-4block model.
Fig 3.
Accuracy model with filter difference.
Fig 4.
Classification accuracy with different values of signal length.
Fig 5.
Accuracy performance of DFENet across 26 modulation types under varying SNR conditions.
Fig 6.
Confusion matrix of 26-modulation classification at +0 dB SNR.
Fig 7.
Confusion matrix of 26-modulation classification at +10 dB SNR.
Fig 8.
Output of the first convolutional layer at SNR = +0 dB.
Fig 9.
Output of the final fully connected layer at SNR = +0 dB.
Fig 10.
Output of the final fully connected layer at SNR = +10 dB.
Table 3.
Ablation study on the HisarMod2019 dataset.
Fig 11.
Classification accuracy with different numbers of DFE blocks.
Fig 12.
Macro averaged Precision, Recall, and F1-score of DFENet.
Fig 13.
Comparison of the classification accuracy of different single CNN models for HisarMod2019 dataset.
Table 4.
Complexity and computational time of the models.
Fig 14.
Comparison of the classification accuracy of different single CNN models for RadioMl2018 dataset.