Research On Application Oriented Localization Problems In Wireless Sensor Networks

Due to owning the characteristics of low cost, robustness and ad hoc, WSN (wire-less sensor networks) has a wide range of potential applications in the fields of mili-tary defense, industrial control, environment monitoring and medical health-care. InWSN, location information is critical for sensor nodes'monitoring activities becausesensor data without complete coordinates is next to useless. Therefore, localization (thetechnology of acquiring sensor nodes'geographic information) is critical for deployingWSN system and is one of WSN's basic functions.In the past ten years, researchers have proposed lots of localization algorithms forWSN. However, most of these approaches usually meet different kinds of obstacleswhen being applied in practical circumstance such as imprecise localization informa-tion, poor scalability, high message complexity and low localization accuracy et al. Inorder to tackle these challenging problems, we begin our research from the angle of realpositioning system based on broad investigation and careful analysis. By studying thefour steps including collecting localization information, positioning initially, refininglocalization results and updating location information of a real positioning system care-fully, we propose novel and effective solutions to solve the problems existing in thesefour steps. The contributions of this thesis is as following:In the step of collecting localization information, we focus on the challengingproblem of revising raw collected information and propose a reputation-based revis-ing scheme (RRS) to achieve this goal. In RRS, we reevaluate the reliability of rawinformation by utilizing the neighboring relationship. After being processed throughRRS, the raw information is close to the information collected in ideal environment.According to our detailed simulations, we demonstrate DV-Hop utilizing revised local-ization information is capable to achieve about 34% higher accuracy than using the rawone. By building a simple testbed, we prove DV-Hop based on the revised informationcan improve its accuracy by about 16%.In the step of positioning initially, we proposed three effective localization algo-rithms including AFLA, ORM and DSBL to solve the problems of poor scalability andlow accuracy arising from applying traditional localization algorithms in real circum-stance. In AFLA, we first refine the hop-count information to obtain fine-grained hop-counts, then conduct initial localization to achieve initial position estimations, and fi-nally further improve the localization accuracy through refining scheme. We conduct comprehensive simulations to demonstrate that AFLA can achieve 30% higher averageaccuracy than the existing hop-count based algorithm. In the testbed we set up, AFLAcan improve the localization accuracy by about 24% on average. In ORM, we sam-ple limited RSSI values in several different directions and distances in advance. Thenbased on these information, we can endue the global radio strength distribution map ofthe positioning area. According to this map, the unknown nodes can conduct localiza-tion to acquire their coordinates. ORM only involves in collecting RSSI in small rangewhich makes it own good scalability. In order to demonstrate the performance of ourapproach, we build up a testbed with 14 MICAz motes to run ORM. The results showthat our method can outperform W-Centroid algorithm by about 26% in indoor environ-mentandasmuchas42%inoutdoorcircumstance. InDSBL,wediscoverthatinnearbyrange, direction affects RSSI values more than the physical distance does. Thus, DSBLjudges the area of unknown nodes according to the direction between nodes while notEuclidean distance. The system experiments present that DSBL can locate about 50%unknown nodes in the correct area while the traditional RSSI sequence based localiza-tion algorithm can only achieve around 17%.In the step of refining initial localization results, we propose CSM (confidencespring model) to further refine the initial positioning results by utilizing neighboringrelationship. In CSM, we assign unknown nodes different confidence level. Based oncorrelated neighboring information and their confidence level, we adjust the movingstep of spring model dynamically. As a result, our approach can both reduce the esti-mation error greatly and converge quickly. The system experiment results show thatCSM can decrease the initial estimation error by about 27% and achieve about 14%higher precision than traditional spring model.In the step of updating location information, we find out the mobiles in traditionaltracking systems report their location to server periodically, which will result in highpacketlossrateandrapidenergydepletionasthenumberofmobilesincrease. Inordertosolve these problems, we propose LUM (location updating mechanism). In LUM, onlydelegates report location periodically instead of each mobile node. Therefore, LUMcan save energy greatly through reducing the message complexity. To demonstrate theperformance of LUM, we deployed a prototype system with 38 Micaz Motes. Theresults show that LUM outperforms traditional approaches by at least45% less messagetransmission and 48% fewer energy depletion on average.