Research On Localization Technology In Wireless Sensor Network: Theory And Experiment

Wireless Sensor Networks (WSNs), as a major development and transformation in thefield of information, has attracted widely attention in today’s emerging technologies, andindicates very broad application prospects both in military and civilian fields. Localization,as one of the key technologies in WSNs, faces many challenges due to the sensor node’slow-cost and low-power design, as well as the harsh and complexity working environment.Localization in large-scale WSNs can be generally divided into two categories: range-based and range-free approaches. The range-based approaches can achieve high localizationaccuracy, but their performance is much dependent on the accuracy of the estimated pairwisedistance or angle, which costs extra expensive hardware to measure, and, therefore, they arenot suitable for large-scale deployment of WSNs. Compared with the range-based one,range-free localization approaches are almost based on the connectivity, instead of pairwisedistance or angle measurements by complex and expensive additional hardware, and moresuitable for large-scale deployment of WSNs. However, range-free approaches can onlyprovide coarse localization.This thesis focuses on the range-free localization in WSNs, and conducts the researchfrom theoretical and experimental aspects respectively. In the theoretical studies, we find thereason that leads to the poor localization accuracy by range-free localization:hop-distanceambiguity problem where a node has a same distance estimation to all of its one-hop neigh-bors. To address this problem, we define a new measure, called Regulated NeighborhoodDistance (RND) by relating the proximity of two neighbors to their neighbor partitions. Inexperimental studies, we build the experimental platform based on the widely used Mica2prototype, and conduct our experimental studies with the platform. In particular, the inno-vations in the thesis are concluded as the following:(1) To address the hop-distance ambiguity problem, we define a new measure, calledRegulated Neighborhood Distance (RND) by relating the proximity of two neighbors totheir neighbor partitions. The basic idea of RND is motivated from the observation that two neighboring nodes are closer, if they have more common neighbors. That is, we can use thenumber of common neighbors between two neighboring nodes to measure their closeness.Then, we analysis the features of RND under finite and infinity node density conditions,and further propose the RND-based localization algorithm, called DV-RND, to improve theDV-Hop localization by using our new RND-based distance estimation technology. What’smore, the computation of RND is only based on the one hop neighbor information and intro-duces no additional hardware requirements. We compare our scheme with two typical algo-rithms, namely, the DV-Hop and the localization method called DV-CNED, via simulationsin different network scenarios, which include grid deployment, random uniform deploy-ment, non-uniform deployment and uniform deployment with a coverage hole. Simulationresults show that DV-RND performs better than the DV-Hop and DV-CNED algorithm interms of lower localization error in the above scenarios.(2) To address the optimal neighborhood definition problem, we propose an optimalPacket Reception Ratio (PRR) threshold selection algorithm to adaptively adjust the RND-based distance estimation. We consider a rather general radio propagation model, where theexpectation of received power is a non-increasing function of the distance between a pair oftransmitter and receiver. For wireless networks with such a general model, we first revisitthe definition’hop’ and’neighboring node’ based on the Packet Reception Rate (PRR), andthen propose an improved RND-based localization scheme. The performance of the RND-based localization scheme is dependent on how to choose an appropriate PRR threshold todefine’hop’ and’neighboring node’. In the improved RND-based localization, the anchorsadaptively adjust the PRR threshold to improve localization accuracy for different numbersof transmitted packets. Simulations are used to evaluate the proposed scheme for the shad-owing model and a polynomial fitting model that is obtained from our field experiments.Results show that the optimal selection of PRR threshold leads to significant improvementsof localization accuracy. Furthermore, our scheme achieves localization accuracy muchhigher than that of the peer DV-Hop algorithm with the same setting.(3) To address the environment-dependence of Received Signal Strength (RSS) basedapproaches, we propose Adaptive Weighted Centroid Localization algorithm AWCL-IRWbased on adaptive Inverse RSS indicator Weights (IRW), which modeled by the g-th power of the inverse RI (RSS Indicator). Anchors use self-learning algorithm to obtain theiroptimal weight exponent g. The AWCL-IRW successfully combine highly environment-dependence RSS with the low complexity Weighted Centroid Localization (WCL). What’smore, we construct the experimental platform, which is composed of20Mica2nodes.and conduct the RSS experiments under different conditions, which include differentdistances, different directions, different transmit power. We verify the AWCL-IRW with theplatform in different fading environments, which include the cement ground environmentand the wooded grass environment. The experimental results show that, in the cementground environment (20×30(m2)), the average localization error by the AWCL-IRWalgorithm is less than2(m). In the wooded grass environment (40×50(m2)), the averagelocalization error by the AWCL-IRW algorithms is less than4(m); Compared with theCentroid Localization (CL) algorithm and the WCL-Distance algorithm, the AWCL-IRWalgorithm achieves better localization accuracy.