Published in IET Wireless Sensor Systems Received on 26th November 2010 Revised on 1st March 2011 doi: 10.1049/iet-wss.2010.0091 ISSN 2043-6386 Spatio-temporal modelling-based drift-aware wireless sensor networks M. Takruri 1 S. Rajasegarar 2 S. Challa 3 C. Leckie 4 M. Palaniswami 2 1 Centre for Real-Time Information Networks (CRIN), University of Technology, Sydney, Australia 2 Department of Electrical and Electronic Engineering, The University of Melbourne, Melbourne, Australia 3 NICTA Victoria Research Laboratory, The University of Melbourne, Melbourne, Australia 4 NICTA Victoria Research Laboratory, Department of Computer Science and Software Engineering, The University of Melbourne, Melbourne, Australia E-mail: maen.takruri@uts.edu.au Abstract: Wireless sensor networks are deployed for the purpose of monitoring an area of interest. Even when the sensors are properly calibrated at the time of deployment, they develop drift in their readings leading to erroneous network inferences. Based on the assumption that neighbouring sensors have correlated measurements and that the instantiations of drifts in sensors are uncorrelated, the authors present a novel algorithm for detecting and correcting sensor measurement errors. The authors use statistical modelling rather than physical relations to model the spatio-temporal cross-correlations among sensors. This in principle makes the framework presented applicable to most sensing problems. Each sensor in the network trains a support vector regression algorithm on its neighbours’ corrected readings to obtain a predicted value for its future measurements. This phase is referred to here as the training phase. In the running phase, the predicted measurements are used by each node, in a recursive decentralised fashion, to self-assess its measurement and to detect and correct its drift and random error using an unscented Kalman filter. No assumptions regarding the linearity of drift or the density (closeness) of sensor deployment are made. The authors also demonstrate using real data obtained from the Intel Berkeley Research Laboratory that the proposed algorithm successfully suppresses drifts developed in sensors and thereby prolongs the effective lifetime of the network. 1 Introduction Wireless sensor networks (WSNs) are an important and promising field of research that has a lot of prospective applications in all aspects of our lives [1]. A WSN usually consists of cheap sensors with limited processing capabilities and energy resources. However, as a group they can accomplish more complex sensing tasks than the expensive individual traditional sensors. Moreover, they can be deployed in inaccessible areas without the need for carefully engineering their position of deployment and their communication topology [2]. Typically, a WSN consists of large number of sensor nodes left unattended for long periods of time. This makes them prone to failures due to either lack of energy or due to the harsh environmental conditions surrounding them [3]. This emphasises the importance of implementing fault-tolerant algorithms for the successful deployment of WSNs [4]. Sensor nodes also tend to develop drift in their measurements as they age. The drift we consider in this context is a slow, unidirectional long-term change in the sensor measurement. In addition to drift, sensor nodes suffer from bias in their measurements [5]. This poses a major problem for the end application, as the data from the network becomes progressively useless. Traditionally such errors are corrected by site visits where an accurately calibrated sensor is used to calibrate other sensors. In a large-scale sensor network, constituted of cheap sensors, there is a need for frequent recalibration. Owing to the size of such networks, it is impractical and cost prohibitive to manually calibrate them. Hence, there is a significant need for auto-calibration [3] in sensor networks. We address the sensor drift/bias problem assuming that neighbouring sensors in a network observe correlated data, that is, the measurements of one sensor are related to the measurements of its neighbours. Furthermore, the physical phenomenon that these sensors observe also follows some spatial correlation. Hence, in principle, it is possible to predict the data of one sensor using the data from other closely situated sensors [3, 4]. This predicted data provides a suitable basis to correct anomalies in a sensor’s reported measurements. The sensor bias estimation and correction problem has been well studied in the context of the multi-radar tracking problem. In the target tracking literature, the problem is usually referred to as the registration problem [6, 7]. When the same target is observed by two sensors (radars) from two different angles, the data from those two sensors can be fused to estimate the bias in both sensors. In the context of image processing of moving objects, the problem is referred to as image registration [8, 9]. 110 IET Wirel. Sens. Syst., 2011, Vol. 1, Iss. 2, pp. 110–122 & The Institution of Engineering and Technology 2011 doi: 10.1049/iet-wss.2010.0091 www.ietdl.org