Search

The rapid advancement of the Internet of Things (IoT) and intelligent vehicular networks has led to the widespread deployment of massive long-range (LoRa) nodes. These nodes play a crucial role in facilitating the sharing of critical information towards various vehicles. However, a significant challenge that persists is the secure authentication of LoRa. Recently, radio frequency fingerprinting (RFF) emerges as a robust potential solution, offering both rapid and scalable identification capabilities. By harnessing machine learning (ML) techniques, the inherent hardware-level, unique, and resilient features of RFF can be thoroughly explored. Given the inherent non-identical (heterogeneity) of LoRa networks and the non-stationary characteristics of chirp signals, conventional approaches prove inadequate in effectively capturing the RFF features. To surmount these challenges, a novel scheme termed as “deep federated learning network (DFLNet)” is proposed. DFLNet facilitates distributed RFF learning while ensuring privacy preservation within diverse LoRa networks. Addressing the heterogeneity of datasets, we incorporate an adaptive optimizer that contributes to the efficient convergence improvement. The DFLNet adaptively manages computational demands and privacy concerns through the utilization of mobile edge computing (MEC) servers, thus enabling distributed learning. Empirical evaluations substantiate the efficacy of DFLNet, attaining an impressive identification accuracy of up to approximately 96.7% utilizing merely around 5000 samples per client, while concurrently ensuring a robust privacy preservation.