HMT-Net: A Multi-Task Learning Based Framework for Enhanced Convolutional Code Recognition.

Area of Science:

• Electrical Engineering
• Computer Science
• Signal Processing

Background:

• Convolutional code recognition is crucial for non-cooperative communications like spectrum surveillance.
• Current deep learning methods often focus on single parameter recognition, neglecting inter-parameter correlations.
• There is a need for advanced techniques to improve the accuracy and efficiency of convolutional code identification.

Purpose of the Study:

• To propose a novel Hybrid Multi-Task Network (HMT-Net) for simultaneous recognition of convolutional code parameters.
• To leverage multi-task learning to capture inherent correlations between code rate and constraint length.
• To enhance convolutional code recognition accuracy in complex communication scenarios.

Main Methods:

• Developed HMT-Net, integrating dilated convolutions, attention mechanisms, and a Transformer backbone.
• Employed a Channel-Wise Transformer for efficient local and global feature extraction.
• Augmented the dataset with comprehensive sequences and extracted statistical features.

Main Results:

• HMT-Net achieved an average recognition accuracy of 2.89% higher than single-task models.
• Demonstrated significant performance gains: 4.57% in code rate and 4.31% in constraint length recognition compared to MAR-Net.
• Validated the effectiveness of multi-task learning in improving convolutional code parameter identification.

Conclusions:

• HMT-Net offers a robust and accurate solution for convolutional code recognition.
• The proposed framework shows significant practical value for intelligent signal analysis and spectrum management.
• Multi-task learning effectively addresses the limitations of single-parameter recognition in complex environments.