Embedding Covert Signals in Federated Learning: IEEE CNS 25 Poster

This paper presents a covert communication frame-work for federated learning (FL) systems, where the server embeds a covert message into the global model update. The embedding is designed to make the transmission statistically indistinguishable from standard model updates. We analyze the decoding error probability and achievable covert rate under noisy Rayleigh fading channels. Experimental results demonstrate that covert embedding has negligible impact on FL performance, with test accuracy preserved after sufficient training rounds.