Federated Learning-Enabled Smart Jammer Detection in Terrestrial and Non-Terrestrial Heterogeneous Joint Sensing and Communication Networks

In this paper, we propose a novel federated learning (FL) framework for detecting smart jamming in heterogeneous joint sensing and communication within terrestrial and non-terrestrial (HJSAC-TNT) networks. Our approach addresses the threat that signal-replicating smart jammers pose to unmanned aerial vehicle (UAV) operations by integrating a specially designed filtering technique, called a dynamic adaptive spectro-temporal resilience filter (DASTRF), into a local variational autoencoder (VAE) that has been enhanced with vision transformer (ViT) and long short-term memory (LSTM) units, called an FL-based ViT-LSTM-VAE. This setup effectively distinguishes between authentic signals and jamming interference by applying the DASTRF to the time-frequency distribution (TFD). It extracts discriminating features from unknown jamming without prior knowledge and refines waveform discrimination. Our FL framework significantly enhances the tradeoff between sensing and communication, thereby improving detection accuracy and jamming resistance with moderate time and resource complexity. This advancement ensures more reliable communications and secure target detection in complex network scenarios.