Distributed Linear Combination Estimators for Localization Based on Received Signal Strength
Abstract
Locating the position of a radio frequency device is indispensable in many wireless applications. The most famous method is the Global Positioning System (GPS), which uses trilateration with satellites, is generally unavailable for indoor devices and expensive for large networks. Therefore, this dissertation aims to develop and discuss accurate, fast, low-cost, energy-efficient, and robust localization algorithms especially based on the received signal strength (RSS).
This dissertation proposes a distributed and iterative estimator by linearly combining location estimates from maximum likelihood based range estimates. In non- cooperative cases where unknown-location (blindfolded) devices only utilize the in- formation from known-location devices (anchors), each combining weight is proportional to the reciprocal of the estimated distance squared between the blindfolded node and an anchor. The numerical simulations demonstrate that the proposed LC estimator has similar error behaviors to the maximum likelihood estimator (MLE) and fewer computations under various topologies and noisy wireless environments. If the parameters for the RSS model are unknown, they are estimated by the least square and/or maximum likelihood methods. The accuracy difference of the linear combination estimators by estimated and perfect parameters is acceptable and decreasing as more anchors are deployed.
In cooperative localization, a blindfolded node uses information from not only anchors but also other blindfolded nodes. The combining weight is now proportional to the reciprocal of the estimated distance squared and the transmitter?s positioning error. After being mainly compared with the distributed maximum likelihood estimator by coordinate descent method and the distributed weighted-multidimensional scaling (dwMDS) method, the LC estimator performs well in accuracy, computation time, and the use of wireless transmissions under various topologies, connectivities, and noisy environments. Moreover, the estimation error is clipped by upper and lower bounds. The drawback is that the convergence is not guaranteed, although non-convergent cases rarely happen. For the connectivity issue, placing more nodes with smaller transmitting ranges results in fewer connected nodes and less power consumption. However, to improve localization of an existing system, the relative costs of node and consumed power must be considered to determine the lowest cost system. Finally, the density of blindfolded nodes is two to three times to the density of anchors to achieve the same accuracy.