Recursive time-synchronization in sensor networks

Clock synchronization is a critical challenge in wireless sensor networks. Noisy measurements of relative offset and skew between pairs of nodes can be obtained by an exchange of time-stamped packets. The problem then can be cast as a linear estimation problem, and many distributed algorithms have been proposed recently. These algorithms are iterative, involving exchange of estimates with 1-hop neighbors. These message exchanges could be time-stamped to yield additional measurements of clock offsets and skews. We propose a distributed algorithm to fuse the new measurements with the current estimates. We show that the estimates produced by the proposed protocol converge in mean square to the true parameter values. More importantly, the error variance of the estimates even after a small number of iterations are far lower than what the earlier protocols achieved upon convergence. Simulations also indicate that the algorithm is robust to intermittent link and node failures.