Time synchronization is a critical piece of infrastructure for any distributed system. Distributed, wireless sensor networks make heavy use of synchronized time, but often have unique requirements in the scope, lifetime, and maximum error of the synchronization achieved, as well as the time and energy required to achieve it. Existing time synchronization methods need to be extended to meet these new needs.
We have presented an implementation of our own sensor network time
synchronization scheme, post-facto synchronization. This method
combines the oscillator frequency discipline provided by NTP with an
instantaneous phase correction provided by a simple synchronization
signal sent by a beacon. Our experiments have shown timing error for
a group of 10 nodes to be bounded near the limit of our clock
resolution of 
sec.
An important additional result is that the same error bound was achieved even when NTP no longer had an active external time or frequency standard, after an initialization period when it was allowed to estimate the local oscillator's frequency error. This is critical for sensor networks where limited energy reserves and the high energy cost of operating a wireless radio make standard NTP unsuitable for long-lived, low-power operation.
Although our current results are a preliminary laboratory study, we believe that post-facto synchronization over wireless radios will be able to support the same instantaneous creation of a short-lived but low-error synchronized timebase ever after a long period of radio silence. We propose ongoing research where we can
Portions of this work were supported by DARPA under grant No. DABT63-99-1-0011 as part of the SCADDS project, and was also made possible in part due to support from Cisco Systems.