GNSS Interference Signal Classification Based on Federated Learning

The global navigation satellite systems have provided space-time information services for both civilian and military application fields. However, they are susceptible to interference signals. Machine learning based identification methods have attracted much attention lately to identification interference signals. However, most of them require large amounts of data and conflict with data privacy protection. To solve these problems, this paper proposes a navigation interference identification method based on FedAvg framework with performance-weighted differential aggregation in a distributed network to protect the privacy of sub-network samples. The agents combine small amounts of new information with their prior knowledge to obtain enhanced generalization performance and improve recognition accuracy while avoiding overfitting to new data. Besides, to improve the suitability in complex scenario, spoofing signals as well as various jamming signals are considered simutaneously in this paper. The performance are compared with traditional CNN and FedAvg-based federated learning through the experiments conducted on the public interference dataset, the TEXBAT dataset of the University of Texas, and generated small sample dataset. The simulation results show that the proposed method can identify interference signal more effectively with an overall recognition rate of 99% without increasing algorithm complexity.