Distributed Unknown Specific Emitter Identification Based on Federated Learning

The utilization of Specific Emitter Identification (SEI) in war defense has significantly improved the capability to identify and analyze enemy targets through the use of radio frequency fingerprint (RFF) extracted from received signals to identify specific emitters. The application of deep learning (DL) technology, specifically in wireless security authentication, has the potential to further enhance SEI. However, traditional machine learning approaches are centralized-based, which is not optimal for SEI due to the private nature of the emitter dataset, especially when the dataset are distributed among different organizations. Federated learning (FL) offers a solution by allowing multiple clients to cooperate in model training without dataset exchange. This paper introduces FL into SEI and proposes an open-set recognition framework. Unlike closed-set recognition, which only recognizes the emitters in the training set, open-set recognition is capable of identifying emitters that are not in the training set. Therefore the proposed framework is extremely suitable for the unknown emitter identification. The experimental results show that the proposed scheme achieves a high level of accuracy in identifying unknown emitters, even if the model is trained distributedly.