Drift estimation using pairwise slope with minimum variance in wireless sensor networks

Time synchronization is mandatory for applications and services in wireless sensor networks which demand common notion of time. If synchronization to stable time sources such as Coordinated Universal Time (UTC) is required, employing the method of flooding in order to provide time synchronization becomes crucial. In flooding based time synchronization protocols, current time information of a reference node is periodically flooded into the network. Sensor nodes collect the time information of the reference node and perform least-squares regression in order to estimate the reference time. However, least-squares regression exhibits a poor performance since sensor nodes far away from the reference node collect the time information with large deviations. Due to this fact, the slopes of their least-squares line exhibit large errors and instabilities. As a consequence, the reference time estimates of these nodes also exhibit large errors. This paper proposes a new slope estimation strategy for linear regression to be used by flooding based time synchronization protocols. The proposed method, namely Pairwise Slope With Minimum Variance (PSMV), calculates the slope of the estimated regression line by considering the pairwise slope between the earliest and the most recently collected data points. The PSMV slope is less affected by the large errors on the received data, i.e. it is more stable, and it is more computationally efficient when compared to the slope of the least-squares line. We incorporated PSMV into two flooding based time synchronization protocols, namely Flooding Time Synchronization Protocol (FTSP) and PulseSync. Experimental results collected from a testbed setup including 20 sensor nodes show that PSMV strategy improves the performance of FTSP by a factor of 4 and preserves the performance of PulseSync in terms of synchronization error with 40% less CPU overhead for linear regression. Our simulations show that these results also hold for networks with larger diameters and densities. © 2012 Elsevier B.V. All rights reserved.