On the Impact of Time Synchronization on Quality of Information and Network Performance KARTIK ARIYUR†, THOMAS SCHMID‡, YUNJUNG YI†, ZAINUL CHARBIWALA‡, MANI B. SRIVASTAVA‡ †Honeywell Inc. 1985 Douglas Drive Golden Valley, MN {kartik.ariyur,yunjung.yi }@honeywell.com ‡Department of Electrical Engineering, University of California, Los Angeles 420 Westwood Plaza Los Angeles, CA {thomas.schmid,zainul,mbs }@ucla.edu Abstract—New hardware and technologies enable low-power low-cost distributed sensing systems. To realize certain appli- cations such as real-time event detection, target tracking, and system monitoring, time synchronization is essential. However, time synchronization both consumes limited battery power on the network nodes and can clog the bandwidth of the network. The choice of a time synchronization mechanism will depend on the application’s requirement of timing accuracy as well as its energy budget. We derive the effect of time synchronization accuracy upon real time estimation and detection problems. We also quantify the costs and benefits of time synchronization algorithms. The intuitive assumption that using higher stability clocks will automatically improve duty cycling performance, and thus decrease power consumption, does not always hold true. In this article, we present the link between clock stability, impact on duty cycling, and the possible bandwidth savings that can be achieved by using temperature compensated clocks or clock drift estimation techniques. This paper formalizes this relationship based on an analytical framework using representative appli- cations, namely, event detection and estimation. The analysis shows the impact of timing errors for different event durations, target speeds, number of sensors, and sampling frequencies. The analysis framework can also be used to estimate the maximum synchronization error each application can sustain while still achieving the desired Quality of Information (QoI). I. I NTRODUCTION New hardware and technology enable low-power inexpen- sive distributed sensor networks. To realize certain applications such as real time event detection, target tracking and system monitoring, time synchronization is essential. There exists substantial literature on synchronization techniques including Time-Sync Protocol for Sensor Network (TPSN) [4], Refer- ence Broadcast Synchronization (RBS) [3], elapsed time [7], and Flooding Time Synchronization Protocol (FTSP) [8]. In general, the goal of a time synchronization mechanism is to devise a scheme that improves accuracy with minimal energy consumption. However, there is clearly a trade off between This material is supported in part by the U.S. ARL and the U.K. MOD under Agreement Number W911NF-06-3-0001, by the NSF under award CNS-0614853, and by the Center for Embedded Networked Sensing at UCLA. Any opinions, findings and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the listed funding agencies. The U.S. and U.K. Governments are authorized to reproduce and distribute reprints for Government purposes notwithstanding any copyright notation herein. accuracy and energy consumption. It is not always be desirable to deploy the most accurate time synchronization as it can prematurely drain the batteries. The choice of a time synchro- nization mechanism will depend on the application’s demand on timing accuracy as well as the available energy budget. To select a proper time synchronization mechanism based on the requirements of the application, the relationship between application performance or QoI (Quality of Information) and time synchronization services are quantified in this article. We formalize the relationship between the performance of applications and time synchronization services based on an analytic framework using representative applications, namely, event detection and estimation. The paper is organized as follows. Section II quantifies the effect of the time-error between nodes upon the performance of detection algorithms, both for the case of no synchroniza- tion algorithm and for the case where synchronization algo- rithms are used. Section III introduces a novel nonlinear hybrid filter which is globally convergent and robust to synchroniza- tion errors. It also quantifies the effect of synchronization error upon the solution of least squares estimation problems across the network, which would automatically include batch least squares, recursive least squares and Kalman filtering. Section IV quantifies the trade-off between clock stabilization, power consumption, and synchronization across the network. In particular, it shows how much energy a duty-cycled system can gain by employing a more stable clock, and conversely, it shows the maximum energy such a clock system can use so that overall energy consumption is minimized. Upper and lower bounds on the number of resynchronization requests are determined using a 3 year temperature data set and modeling crystal drifts. II. A SIMPLE PROBABILISTIC DETECTOR Consider a system of N sensors that are identical and detect an event of time period τ such as a gunshot or an explosion. Let the probability that an event E has occurred, given that the sensor measurement s i exceeds a threshold T be given as: P(E |s i > T )= p ∀i, (1)