A Federated Adversarial Learning Approach for Robust Spectrum Sensing

This paper proposes a robust federated learning approach against adversarial attacks for the spectrum sensing application. It is the process of detecting and analyzing radio frequency spectrum to identify available frequencies for transmission spectrum sensing in communication systems. The proposed system utilizes federated learning to train a semantic segmentation model for spectrum sensing in environments with radar and wireless communication systems. The adversarial training at the client level is used to enhance model resilience and the methodology for aggregating local model updates into a robust global model. According to the results, adversarial training strengthens federated learning models against these attacks. Empirical validation through case studies demonstrates the robustness of federated learning against adversarial attacks.