Localization For Wireless Sensor Networks Via Matrix Completion

In recent years, wireless sensor networks gained rapid development, and has brought a revolution of information perception by its low-cost, large-scale and self-organization features. In WSNs, the location information of sensor nodes is essential for a variety of applications. Therefore, how to achieve precise localization is one of the key technologies of WSNs. Existing WSN node localization algorithms still have defects on localization accuracy, robustness and localization costs. A deep research and discussion is made in this paper based on the matrix completion theory in Compressed Sensing filed.Firstly, this thesis improves the existing matrix completion model by introducing regularization technique to sift outliers and Gaussian noises, and further present a novel matrix completion algorithm based on norm regularization. On this basis, a noise-tolerant localization algorithm which only need small amounts of range measurements is proposed for traditional WSNs whose sensor nodes are deployed randomly in a fixed area. By utilizing the natural low rank property of Euclidean distance matrix, the reconstruction of partially sampled and noisy distance matrix is formulated as a norm-regularized matrix completion problem and the node positions can be further determined by MDS method. Simulation results demonstrate that this algorithm can achieve a high accuracy from only a small fraction of distance measurements and resist various types ranging noises. Moreover, a robust localization algorithm based on fingerprint matrix completion is designed for indoor scenario. By utilizing the low rank property of fingerprint matrix, the complete fingerprint database can be recovered from a small collection of fingerprint data, which reduces the cost of the construction of fingerprint database. What’s more, the norm-regularized matrix completion can sift noises in fingerprint matrix efficiently, which improves the robustness of this localization algorithm.