Dual Loop Meta-Learning for Limited-Sample Modulation Recognition: A Collaborative Model-Agnostic Strategy

Abstract

Modulation recognition in electronic support measures systems is challenging due to the scarcity of labeled data. To address this, we propose a Dual-Loop Meta-Learning network for automatic modulation recognition. The network framework consists of an inner loop, where a novel loss function inspired by metric learning optimizes the feature space for task adaptation with limited data, and an outer loop, where a co-learner regularization module mitigates overfitting and optimizes meta-initialization parameters. Importantly, the co-learner operates only during meta-training, ensuring no overhead during meta-testing. Experimental results indicate that our proposed algorithm outperforms state-of-the-art few-shot learning methods.