Decentralized Algorithms for Sequential Network Time Synchronization

Accurate clock synchronization is important in many distributed applications. Standard algorithms, such as the Network Time Protocol (NTP), essentially rely on pairwise offset estimation between adjacent nodes. Some recent work introduced more elaborate algorithms for clock offset estimation, which take into account the algebraic constraints imposed on the sum of offsets over network cycles, using a least-squares framework. These algorithms are iterative and decentralized in nature, requiring several cycles of local communication among neighbors for convergence. In this paper, we extend this approach towards a sequential estimation framework, which allows to incorporate initial time estimates along with their uncertainty, as well as multiple rounds of pairwise measurements. We propose a decentralized implementation of the estimation algorithm that employs only local broadcasts and establish its convergence to the optimal centralized solution. We also present some simulation results to illustrate the performance benefits of the suggested algorithms.