Blind Recognition of Channel Coding Types Based on the CTN Model

This paper addresses the issues of high data demands and coarse recognition granularity in blind channel coding recognition for non-cooperative communications. A hybrid model, CTN, is proposed, which integrates CNN with the self-attention mechanism of Transformers.The model uses CNNs to extract local features and suppress noise, while the Transformer component captures global semantic dependencies to enhance sensitivity to differences in coding structures. Experimental results demonstrate that CTN performs excellently across six different coding types. The average recognition accuracy across the full SNR range reaches 81.6%. In low SNR conditions (-10 to 4dB), CTN outperforms the best comparison model, ResNet, by approximately 3.5%. In high SNR conditions (4 to 20dB), CTN's performance approaches 100%.Additionally, CTN offers advantages such as low training cost, fewer parameters, and reduced recognition delay, combining high recognition accuracy with low complexity.