.png){kind=icon}
.png){kind=icon}
Return
Deep Learning for Channel Code Type Recognition
.png){kind=icon}
PRE
AI
DOI：10.1109/TVT.2025.3605146.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Channel code type recognition is critical for enabling receivers to discern codes without prior knowledge. Despite the promise of deep learning approaches in this field, they often encounter difficulties in reconciling accuracy with computing complexity due to insufficient attention to task-specific attributes. The absence of publicly accessible datasets has further hindered research progress. This paper presents the inaugural publicly accessible dataset for channel code type recognition and assesses the performance of several benchmark models on it. We propose CCTRNet, a CNN-Transformer model for Channel Code Type Recognition, which considers task-specific attributes such as irregular dependency distributions and noisy interference. CCTRNet leverages the synergy between the adaptive local feature extractor and multi-head attention mechanism to enhance the model's capacity to handle intricate input patterns and guarantee robustness under noisy conditions. Experiments demonstrate that CCTRNet exceeds existing benchmarks in accuracy while decreasing computational complexity, with a 2.34% improvement in average accuracy and a 41.11% decrease in the volume of parameters relative to the optimal baseline model.
Keywords:
Channel code type recognition
deep learning
CNN
transformer
public dataset
Journal
IF:
7.1
Papers:
1.8W
Citations:
6.6W