Geometry-Aware Contrastive Learning for Few-Shot Automatic Modulation Recognition

Standard Self-Supervised Learning (SSL) for Automatic Modulation Recognition (AMR) struggles with ineffective isotropic augmentations, spectral instability, and semantic drift. To address these challenges, we propose Dynamic-Consistency Contrastive Learning (DyCo-CL), a geometry-aware framework that couples Virtual Adversarial Augmentation (VAA) with a semantic consistency loss. We provide a theoretical analysis indicating that this strategy acts as an implicit spectral regularizer for the encoder, enabling stable manifold exploration. Complementing this, our Signal-Adaptive Swin Backbone with fixed-window attention improves structural stability by constraining attention locality, while a Hybrid Knowledge Fusion module anchors representations with physical priors. Experiments on RML benchmarks show that DyCo-CL achieves a 6.27% accuracy gain in 1-shot settings over prior methods.

Reliable recognition of wireless signal types is important for modern communication systems, but current AI methods often struggle when only a small amount of labeled data is available. In this work, we propose a new learning framework that helps AI models learn signal patterns more stably and effectively by generating meaningful signal variations and preserving important signal structures during training. We also design the model to focus on stable local signal patterns and incorporate prior communication knowledge to improve robustness. Experiments on standard wireless communication benchmarks show that our method significantly outperforms previous approaches, especially in extremely low-data settings where only one labeled example per signal type is available. This work could help enable more reliable and efficient wireless communication systems in real-world environments.