Transformer based HF communication demodulation.

Area of Science:

• Electrical Engineering
• Computer Science
• Signal Processing

Background:

• Shortwave channels face significant impairments like multipath and fading in low signal-to-noise ratio (SNR) environments, increasing bit error rate (BER).
• Traditional demodulation techniques, such as Least Squares (LS) and Minimum Mean Square Error (MMSE), struggle to meet communication requirements under these conditions.

Purpose of the Study:

• To propose a novel deep learning-based demodulation algorithm for shortwave channels to overcome limitations of traditional methods.
• To enhance demodulation accuracy and communication quality in challenging low-SNR environments.

Main Methods:

• A multi-channel deep learning approach using a Transformer-based network, feeding real and imaginary signal parts as separate inputs.
• Integration of a convolutional neural network (CNN) module for effective local feature extraction within the network architecture.

Main Results:

• The proposed deep learning algorithm significantly outperforms traditional LS and MMSE methods in demodulation performance within -10 dB to 10 dB SNR range.
• The Transformer-based network achieved consistent SNR gains of 1-5 dB compared to benchmark methods including CNN, GRU, and CNN-RNN models.
• Demonstrated superior performance of the Transformer-based demodulation receiver under low-SNR, high-frequency channel conditions.

Conclusions:

• The developed Transformer-based deep learning demodulation algorithm offers a significant advancement for shortwave communication systems.
• The findings highlight the potential for enhanced reliability and stability in shortwave communications through advanced deep learning techniques.
• This research provides valuable insights for future development of robust communication systems operating in adverse channel conditions.