We investigate periodic time synchronization of networks without centralized control, which can be modeled as a problem of aligning local variables taking values on a circle. Synchronization is investigated in a two-dimensional network with nearest neighbor interaction. We introduce a class of circular averaging algorithms, where nodes decide their timings based on neighbor information only, without considering their own time. Compared to firefly synchronization based on pulse coupled oscillators, these algorithms are more robust and stable when new nodes joins the system.