Exploring Autoregressive Integrated Models for Time Synchronization in Wireless Sensor Networks

Time synchronization provides the basis for several applications in wireless sensor networks but the limited memory and computational power, and the use of low precision oscillators make the task of time synchronization non-trivial. In this demonstration, we present a novel time synchronization scheme that is based on time series analysis. To provide a general model for the practical behavior of low precision oscillators, autoregressive integrated moving average models are explored. Based on the analysis of experimental data, an autoregressive integrated model (ARI (1,1)) is derived. Unlike the resource hungry Kalman filter based formulations, the proposed scheme is resource efficient as it results in simple linear regression processing. Experiments are performed on real sensor devices including Zolertia and TelosB, where an accuracy below 1 clock tick1 is achieved.