Two-clocks synchronization for networked sensors

In this paper we deal with the problem of estimating the most likely values of both the relative offset and rate of two distributed clocks, one in a device that acquires sensor data and the other in a remote computer that receives them. This two-clock synchronization problem is a common issue, for example, when time-stamped sensor data is requested remotely for executing real-time control actions. There are a number of approaches for solving that problem, being particularly intuitive those based on geometry (convex closures- or polygon-based). Surprisingly, they define geometrical regions in the space of local times, different from the space of the estimated parameters, thus additional transformations are required to deduce the final estimates. Here we propose a more natural method that works directly on the space of parameters and does not depend on the probability distribution of the time delays. It obtains relevant reductions in computational cost while keeping soundness and good estimation properties. Our experiments confirm the sound behavior of the algorithm and its low expected computational complexity with respect to others.