자동변조인식을 위한 conformer기반 신경망 성능 분석

다국어 초록 (Multilingual Abstract)

Automatic modulation recognition (AMR) is essential in modern electronic warfare support for identifying modulation schemes in communication signals. This study adopts the Conformer architecture, originally developed for speech recognition, to address the AMR task. The proposed model employs convolutional preprocessing and Conformer blocks to extract both local and global features from raw in-phase/quadrature data and classifies 11 modulation types. Evaluated on the RadioML 2016.10A dataset, the model outperforms CNN and Transformer baselines, achieving 90.8% accuracy at an SNR of 18 dB—an improvement of 5.5 percentage points over the Transformer model. These results demonstrate the effectiveness of the Conformer architecture for wireless signal classification.