Study Of Wireless Sensor Network Localization Algorithm

Wireless sensor network is one of the hotspots on which many countries spent a lot of efforts and money in recent years. Wireless sensor network has unique characteristic in its network design. It is different from the traditional wireless communication network owing to the physical limitation such as more sensor node, low energy, less memory and inefficiency in processing. As one of the foundation of the application of wireless sensor network, the conventional GPS locating is not suitable for the localization algorithm as its high cost. A new breakthrough is essential the for wireless sensor network localization algorithm.My thesis introduce the distance measurement and former localization algorithm of wireless sensor network at first, and gives a detailed introduction of the MDS localization algorithm. On the basis of MDS-MAP, I further introduce the interative majorization and the least squared estimator based on Taylor series expand, which optimizes the multidimensional scaling and brings up the RSSI based MDS-LSE localization algorithm, which can self-adapt to different channel attenuation index. The algorithm first divide the whole network into several local networks, then use the iterative majorized multidimensional scaling algorithm to localize in local network. MDS-MAP converge rapid but has local minimal problem, sometimes fail to get the optimal solution. Then merge all the local networks and solve the inaccuracy by means of the Taylor series expansion which brought up by the introduction of the least squared estimator. Taylor series expansion is a recursion algorithm needs initial estimate value, as a result, it has the advantage of accurate localization. Moreover, on the condition of the RSSI measurement, even if the channel attenuation index is unknown, which still can be achieved by the least squared estimator based on Taylor series expand. Yet, the least squared estimator's converging speed is rapid only when the initial estimate value is proper. On the other hand, if the initial estimate value differs greatly from the actual result, the value of interative will be big, even far from converge. What's more, every recursion will have matrix processing, great amount of calculation is needed. The core of this thesis's MDS-LSE is the combined algorithm which uses estimate value calculated by multidimensional Scaling algorithm as the initial value for the LSE based on Taylor series expansion. Since the location estimate value after the process of multidimensional scaling is similar to the accurate one, the least squared estimator can be converged rapid and can save the compute time and self-adapt the enviorment.Finally,my thesis simulate the interative majorized multidimensional scaling algorithm on the platform of MATLAB and Freesacle ZigBee (1321xEVK) in small-scaled local network, and compares the algorithm raised up in this thesis with MDS-MAP(P) by experiment simulation. The experiment shows that the MDS-LSE is marked by high accuracy, robust and self-adapt.