Abstract
Although the existing time synchronization protocols in wireless sensor networks (WSNs) are
efficient for short periods, many applications require long-term synchronization among the nodes, for
example, coordinated sleep and wakeup modes, and synchronized sampling. In such applications, experiments
have shown that even clock skew correction cannot maintain long-term clock synchronization and
a quadratic model of clock variations can better capture the dynamics of the actual clock model involved,
hence increasing the resynchronization period and conserving significant energy. This paper derives the
maximum likelihood (ML) estimator for all the clock parameters in a two-way timing exchange model
with exponential delays. The same estimation procedure can be applied to one-way timing exchange
models with little modification.