Federated learning-enabled jamming detection for stochastic terrestrial and non-terrestrial networks

In this paper, we present a novel federated learning (FL) algorithm, named Aggregated and Augmented Training Federated (AAT-Fed), tailored for stochastic, distributed, tactical terrestrial and non-terrestrial (SDT-TNT) network environments. Focusing on an SDT-TNT network with multiple clusters and potential unknown jammers, our approach addresses jammer detection through convolutional variational autoencoders (C-VAEs) within the FL framework. Leveraging the spectral correlation function (SCF) of the in-phase and quadrature (I/Q) representation of received signals, our method extracts discriminating features for jammer detection in the absence of prior knowledge about the jammers. AAT-Fed excels at managing the unique characteristics of the tactical TNT network, considering its stochastic nature and the heterogeneity in data distribution between network cells, leading to enhanced jamming detection accuracy. Comparative simulation results demonstrate AAT-Fed’s superior performance over FL and non-FL approaches, showcasing its effectiveness in providing accurate jamming detection at a low jamming-to-noise ratio.

Mots clés

jamming; autonomous aerial vehicles; servers; heuristic algorithms; scalability; knowledge engineering; feature extraction