Federated Learning-Based Jamming Detection for Tactical Terrestrial and Non-Terrestrial Networks

In this paper, we propose federated learning (FL)-based jamming detection algorithms for a stochastic, distributed, tactical terrestrial and non-terrestrial (SDT-TNT) network. Specifically, we consider an SDT-TNT network with multiple clusters, in which multiple unknown jammers might be present. Moreover, we employ the spectral correlation function (SCF) on local servers to estimate the cyclostationary properties of the received waveforms. We then use the SCF to train local convolutional autoencoders (CAEs). In the inference phase, we jointly use the latent representation of the trained CAE and the kernel density estimation (KDE) to detect the existence of jammers. Our proposed methods show very promising results in jamming detection and outperform non-FL approaches. We further demonstrate that using the SCF feature provides higher accuracy than using In-phase/Quadrature-phase (I/Q) features.