Time Synchronization for Wireless Sensor Networks Using Adaptive Linear Prediction

Time synchronization is a crucial component in wireless sensor networks (WSN), especially for location-aware applications. The precision of time-based localization algorithms is closely related to the accuracy of synchronization. The estimation of synchronization errors in most of the existing time synchronization algorithms is based on some statistical distribution models. However, these models may not be able to accurately describe the synchronization errors due to the uncertainties in clock drift and message delivery delay in synchronization. Considering that the synchronization errors are highly temporally correlated (short-term correlation), in this paper, we present an adaptive linear prediction synchronization (ALPS) scheme for WSN. By applying linear prediction on synchronization errors and adaptively adjusting prediction coefficients based on the difference between the estimated values and the real values, ALPS enhances synchronization accuracy at a relatively low cost. ALPS has been implemented on the Tmote-sky platform. As experiment results demonstrate, compared with RBS and TPSN, ALPS cuts synchronization cost by almost 50% while achieving the same accuracy; compared with DMTS and PulseSync, ALPS reduces the MSE (mean square error) of synchronization errors by 41% and 24%, respectively, with the same cost.