The team tackles a scalability problem in network-based Global Navigation Satellite Systems (GNSS), where centralized processing can become a bottleneck. Their architecture estimates receiver states and creates network-wide satellite-correction products using decentralized coordination across a network of receivers.

They model the receiver network as a changing graph and use a graph-aware learning method to set how information should be mixed over time. Receivers then run an online “gradient-tracking” diffusion process, combining local observations with compact exchanges to reach consensus on satellite corrections and self-localization, with reference stations able to broadcast consensus products for precise positioning services.