Dual Residual Denoising Autoencoder with Channel Attention Mechanism for Modulation of Signals.

Area of Science:

• Signal Processing
• Machine Learning
• Communications Engineering

Background:

• Low signal-to-noise ratio (SNR) degrades modulation signal demodulation accuracy and classification.
• Existing denoising methods struggle with complex modulation schemes and low SNR environments.

Purpose of the Study:

• To propose a novel denoising autoencoder method, DRdA-CA, for improving SNR of modulation signals.
• To enhance the accuracy of modulation signal demodulation and classification under low SNR conditions.
• To reduce the bit error rate (BER) in signal demodulation.

Main Methods:

• A double-residual denoising autoencoder with a channel attention mechanism (DRdA-CA) was developed.
• The autoencoder incorporates a modified Squeeze-and-Excitation (SE) ResNet module for channel attention.
• Dual residual connections were integrated into encoding and decoding modules to optimize network performance.

Main Results:

• The DRdA-CA method significantly improved SNR for eight modulation types (-12 dB to 8 dB).
• Average SNR gains of 8.38 dB for 16QAM and 8.27 dB for 64QAM were achieved.
• Classification accuracy increased by 67.59–74.94%, and BER was reduced by up to 63.5% for 16QAM and 46.7% for 64QAM compared to benchmarks.

Conclusions:

• The proposed DRdA-CA effectively enhances SNR and denoising performance for modulation signals.
• The integration of channel attention and dual residual connections is crucial for optimal results.
• DRdA-CA offers superior performance in improving modulation signal quality for communication systems.