A Distributed Kalman Filter Algorithm for Self-localization of Mobile Devices

Abstract: In many applications involving mobile devices it is essential to track their position. Moreover, it is usually desired, to perform this localization in a distributed fashion without using a central processing unit. In this case, only distance measurements to reference nodes which are in range can be utilized. It is proposed in this thesis to also use distance measurements to other moving objects, to improve the position estimation.The self-localization task is addressed in this work by introducing a distributed Kalman Filter. This state observer estimates the position of the moving objects based on distance measurements to neighboring devices and reference nodes. Additionally it was investigated if the performance of this filter could be improve by adding a data fusion step to the filter. In this case, every device additionally estimates the position of its neighbors. This generates multiple estimates for one object, which are afterwards fused using optimized weights. This allows the usage of more measurement information available in the network for the localization of one device. To compare the performance of the introduced algorithms, simulation results are given. A system with a static graph structure was investigated, as well as a system with a dynamic graph. It was found that the accuracy of the state estimation could be improved by introducing a data fusion step. Furthermore, it was seen that a higher average coupling among the nodes is necessary to ensure reliable performance when the graph is dynamic.