Time synchronization remains as a challenging task in wireless sensor networks that face severe resource constraints. Unlike previous work's aiming at pure clock accuracy, this paper proposes On-Demand Synchronization (ODS), a design to achieve efficient clock synchronization with customized performance. By carefully modeling the error uncertainty of skew detection and its propagation over time, ODS develops a novel uncertainty-driven mechanism to adaptively adjust each clock calibration interval individually rather than traditional periodic synchronization, for minimum communication overhead while satisfying the desired accuracy. Besides, ODS provides a nice feature of predictable accuracy, allowing nodes to acquire the useful information about real-time qualities of their synchronization. We implemented ODS on the MICAz mote platform, and evaluated it through testbed experiments with 33 nodes as well as simulations obeying real world conditions. Results show that ODS is practical, flexible, and quickly adapts to varying accuracy requirements and different traffic load in the network for improved system efficiency.
I. INTRODUCTIONTime synchronization is one of the most fundamental and widely employed middle-ware services in wireless sensor networks (WSN) [10]. It allows nodes in the network to have a common notion of time, either respect to a global reference or among themselves [9]. Accurate time synchronization is critical for saving communication energy [26][28], promoting localization accuracy [1][24], optimizing surveillance coverage [18][34], extending network lifetime [14][15] and improving system security [4]. However, due to extremely limited resources at each low-cost sensor node (e.g., poor clock quality, limited computation and communication capabilities, ultra tight energy budgets, etc), time synchronization remains as a challenging task in the WSN community. Previous research in this field mostly focused on how to do time synchronization for better clock accuracy (e.g., AD [33], TSS [19], RBS [20], TPSN [21], FTSP [22], VHT [5], etc). In spite of microsecond (µs) level accuracy achieved [5][7][22], they are not designed to provide a precise and convenient trade-off between service performance and energy efficiency, which complicates their deployment for energy sensitive applications with different timing requirements. In practice, systems depend on diverse synchronization qualities. For example, the bridge surveillance project [18] demands tens-of-µs clock accuracy for effective data acquisition; to detect jamming attacks in the network [4], the deviation of a couple of mil-