Time synchronization has been an active issue for decades. Stricter timing, energy considerations, external influences, and so on have spurred new studies. Two variants can be found: synchronize the clocks of distributed devices or provide a formula that deals with unsynchronized clocks. In both types, specific control transmissions are sent and received to tune local clocks or to adjust the formula.

Bennett and colleagues provide a new synchronization method, which is data based (more precisely, event based). If an event occurs in a certain instant and several devices detect it, they will be able to take that event as a synchronization point. It is a smart approximation. Furthermore, the time adjustment can be done offline at a central server. Obviously, this approach is liable to errors: events detected in some devices but not others, time misalignment, and so on. The authors have included methods to enhance the detection of events and the pruning of those not relevant. Detection relies on peaks in the mutual information of either the template-based flows or entropy of the signals. After that, several analyses are applied: delay model mapping with logical outlier rejection, quality of matching between signals, and timing outlier detection. These procedures reject those not sufficiently strong or out of a given timing window.

These authors provide experimental results that show a reduction in the synchronization error without including network transmissions. Communications to transmit data from sensors are used to obtain synchronized events. They obtain a reduction between 66 and 98 percent of the median synchronization error. That means a reduction in drift error of 90 percent. Thus, this shows how well it works. It would be interesting to compare this method with standard network-based synchronization mechanisms. It is still a very interesting and worthwhile paper.