Answers are grounded in this paper and cite their sources.
Compares classification accuracy of deep learning models for automatic modulation classification, indicating input type significance.
In this paper, we use various deep learning models for automatic modulation classification (AMC) of received signals and compare the classification performance according to different input types. To enable a fair comparison, a dataset consisting of 26 analog and digital modulation schemes is generated. Five input types are considered: time-domain inphase and quadrature components, time-domain magnitude and phase, frequency-domain real and imaginary components, frequency-domain magnitude and phase, and time-frequency spectrogram. The classification performance is evaluated using five representative deep learning models, including convolutional neural network (CNN), long short-term memory (LSTM), residual network (ResNet), Transformer, and a hybrid CNN-LSTM model, under various signal-to-noise ratio (SNR) conditions. Simulation results show that the LSTM model with time-domain inphase and quadrature inputs achieves the best classification accuracy with a significant margin compared to the other models and input formats. For the remaining models, the time-domain inputs generally outperform the frequency-domain inputs, while no significant performance difference is observed between the two time-domain types. These results demonstrate that both the choice of deep learning model and input type have a substantial impact on the performance of deep learning-based modulation classifiers.
No takes yet. Share an insight, caveat, or question.
Kang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: