期刊:IEEE Transactions on Vehicular Technology [Institute of Electrical and Electronics Engineers] 日期:2026-01-01卷期号:: 1-16
标识
DOI:10.1109/tvt.2026.3743493
摘要
Automatic modulation recognition (AMR) is a key technique for spectrum monitoring, supporting radio-frequency security and efficient spectrum management. With the emergence of high-speed aerial and ground-based platforms, AMR in non cooperative scenarios faces increasing challenges from scarce labeled samples and propagation-related channel impairments, including Doppler shifts, multipath fading, and noise. To ad dress these challenges, this paper proposes a robust few-shot AMR method based on multi-view metric learning. Multiple complementary signal views are deterministically constructed from the same raw I/Q sequence and encoded by view-specific encoders to preserve view-dependent discriminative cues. An intra-class variance-aware adaptive metric is then introduced to down-weight unstable feature dimensions estimated from the support set, thereby improving the reliability of prototype-based matching under channel-induced perturbations. In addition, a query-aware multi-view fusion strategy adaptively integrates view-wise metric outputs according to query–prototype matching patterns, enabling query-dependent adjustment of the effective contributions of different views. Extensive experiments under various channel conditions and few-shot settings demonstrate that the proposed method outperforms representative state-of-the-art approaches and achieves improved robustness and generalization in complex channel environments.