Multi-Task Based Deep Learning Approach for Open-Set Wireless Signal Identification in ISM Band

• DOI: 10.1109/tccn.2021.3118456
• Publication Date: 2021-10-10T15:49:45Z
• Authors: Jialiang Gong; Xiaowei Qin; Xiaodong Xu
• Keywords: Robustness; Identification

Wireless signal identification plays an important role in effectively implementing spectrum monitoring and management. However, ISM (Industrial, Science Medical) band, it becomes a challenging task due to the heterogeneity of variously emerging wireless techniques, part potential unknown spectral occupants may even hinder feasibility identification. To overcome such difficulties, we focus on open-set recognition (OSR) this paper present multi-task learning architecture based deep neural network for identifying known occupants. A novel structured extension counterfactual GAN (CountGAN) is proposed introduce multi-tasking take advantage modulation-domain information from captured signal, thus improving representation individual signals further enhancing model robustness adaptability scenarios. In particular, Circle-loss metric extreme value theory are also applied make tighter clearer decision boundaries identification, enhance optimization between classes. Numerical results indicate that framework consistently outperforms state-of-the-art OSR algorithms several baselines task, both terms convergence performance classification accuracy.