A Doppler-Resilient Covert Communication Scheme for Federated Learning in the IoUT

Abstract

In the Internet of Underwater Things (IoUT), autonomous underwater vehicle (AUV) motion introduces residual delay-Doppler spread, phase perturbations, and intra-packet coherence loss after conventional synchronization and nominal Doppler compensation, which degrade the reliability and covertness of existing federated learning (FL)-based covert communication schemes. To address this challenge, we propose FedStealth, a Doppler-resilient covert communication scheme for practical IoUT conditions. FedStealth constructs a private signal subspace via a Doppler-dependent hypergraph with a shared constraint to suppress motion-induced phase distortion, and performs correntropy-guided pairing optimization of embedding coordinates to minimize phase-flip errors and enhance message recovery. At the legitimate receiver, phase-invariant differential decoding enables reliable message recovery under post-compensation phase perturbations. For the warden, the embedding and pairing design constrain the perturbation energy so that the embedded update remains hard to distinguish from a normal FL update under hypothesis testing. Theoretical analyses and real-world experiments validate FedStealth’s reliability and covertness under practical IoUT conditions.