1: \begin{abstract}

2: In this paper, we propose a fully distributed algorithm for joint clock skew and offset estimation in wireless sensor networks based on belief propagation.

3: {In the proposed algorithm, each node can estimate its clock skew and offset in a completely distributed and asynchronous way: some nodes may update their estimates more frequently than others using outdated message from neighboring nodes.

4: In addition, the proposed algorithm is robust to random packet loss.}

5: Such algorithm does not require any centralized information processing or coordination,

6: and is scalable with network size.

7: {The proposed algorithm represents a unified framework that encompasses both classes of synchronous and asynchronous algorithms for network-wide clock synchronization.}

8: {It is shown analytically that the proposed asynchronous algorithm

9: converges to the optimal estimates with

10: estimation mean-square-error at each node approaching the centralized  Cram\'{e}r-Rao bound

11:  under any network topology.}

12: Simulation results further show that {the convergence speed is faster than that corresponding to a synchronous algorithm}.

13: \end{abstract}