1: \begin{abstract}

2: Desynchronization approaches in wireless sensor networks converge

3: to time-division multiple access (TDMA) of the shared medium without

4: requiring clock synchronization amongst the wireless sensors, or indeed

5: the presence of a central (coordinator) node. All such methods are

6: based on the principle of \textit{reactive listening }of periodic

7: ``fire\textquotedblright{} or ``pulse\textquotedblright{} broadcasts:

8: each node updates the time of its fire message broadcasts based on

9: received fire messages from some of the remaining nodes sharing the

10: given spectrum. In this paper, we present a novel framework to estimate

11: the required iterations for convergence to fair TDMA scheduling. Our

12: estimates are fundamentally different from previous conjectures or

13: bounds found in the literature as, for the first time, convergence

14: to TDMA is defined in a \emph{stochastic sense}. Our analytic results

15: apply to the \textsc{Desync} algorithm and to pulse-coupled oscillator

16: algorithms with inhibitory coupling. The experimental evaluation via

17: iMote2 TinyOS nodes (based on the IEEE 802.15.4 standard) as well

18: as via computer simulations demonstrates that, for the vast majority

19: of settings, our stochastic model is within one standard deviation

20: from the experimentally-observed convergence iterations. The proposed

21: estimates are thus shown to characterize the desynchronization convergence

22: iterations significantly better than existing conjectures or bounds.

23: Therefore, they contribute towards the analytic understanding of how

24: a desynchronization-based system is expected to evolve from random

25: initial conditions to the desynchronized steady state. 

26: \end{abstract}