Automatic Modulation Recognition (AMR) is a critical technology for intelligent wireless communication systems, but the deployment of high-performance deep learning models is often hindered by their substantial computational and memory requirements. To address this challenge, this paper proposes a multi-level knowledge distillation network, namely MLD-Net, for creating a lightweight and powerful AMR model. Our approach employs a large Transformer-based network as a teacher to guide the training of a compact and efficient Reformer-based student model. The knowledge contained in the large model is transferred across three distinct granularities: at the output level, to convey high-level predictive distributions; at the feature level, to align intermediate representations; and at the attention level, to propagate relational information about signal characteristics. This comprehensive distillation strategy empowers the student model to effectively emulate the teacher’s complex reasoning processes. Experimental results on the RML2016.10A benchmark dataset demonstrate that MLD-Net achieves state-of-the-art performance, outperforming other baseline models across a wide range of signal-to-noise ratios while requiring only a fraction of the parameters. Extensive ablation study further confirms the collaborative contribution of each distillation level, validating that the proposed MLD-Net is an effective solution for developing lightweight and efficient AMR networks for edge deployment.
Modulation recognition, as one of the key technologies in the field of wireless communications, holds significant importance in applications such as spectrum resource management, interference suppression, and cognitive radio. While deep learning has substantially improved the performance of Automatic Modulation Recognition (AMR), it heavily relies on large amounts of labeled data. Given the high annotation costs and privacy concerns, researching semi-supervised AMR methods that leverage readily available unlabeled data for training is of great significance. This study constructs a semi-supervised AMR method based on dual-student. Specifically, we first adopt a dual-branch co-training architecture to fully exploit unlabeled data and effectively learn deep feature representations. Then, we develop a dynamic stability evaluation module using strong and weak augmentation strategies to improve the accuracy of generated pseudo-labels. Finally, based on the dual-student semi-supervised framework and pseudo-label stability evaluation, we propose a stability-guided consistency regularization constraint method and conduct semi-supervised AMR model training. The experimental results demonstrate that the proposed DualBranch-AMR method significantly outperforms traditional supervised baseline approaches on benchmark datasets. With only 5% labeled data, it achieves a recognition accuracy of 55.84%, reaching over 90% of the performance of fully supervised training. This validates the superiority of the proposed method under semi-supervised conditions.