Feature-Imitation Federated Learning: An Efficient Approach for Specific Emitter Identification in Low-Resource Environments

As communication technology evolves, specific emitter identification (SEI) gains significance in areas like wireless network security and IoT device identification. While big data and deep learning have spurred centralized identification methods, they often falter in low-resource, real-world settings due to data dispersion and heterogeneity, as well as limited computational power. Addressing these challenges, this paper presents feature-imitation federated learning (FIFL), a novel SEI approach for resource-constrained environments. FIFL utilizes a global classifier, refined through Kullback-Leibler divergence, to manage feature prediction alignment. Simulation results on actual data demonstrate FIFLs effectiveness in overcoming global model drift, ensuring accuracy and reliability even amidst diverse, distributed data in resource-limited settings.