| Localization accuracy based on RSS (Received Signal Strength) is notoriously inaccurate in the application of wireless sensor networks. RSS is subject to shadowing effects, which is signal attenuation caused by stationary objects in the radio propagation. RSS are actually the result of decay over distances, and random attenuation over different directions. RSS measurement is also affected by antenna orientation. Starting from extracting the statistical orders in the function relationship between RSS and distance, this thesis first shows how non-metric MDS (Multi-Dimensional Scaling) is the suitable method for cooperative localization. Then, several issues are presented and discussed in the application of non-metric MDS, including determining full connections to avoid flip ambiguities, leveraging the proper initial estimation to avert from local minimum solutions, and imposing structural information to bend the localization result to a priori knowledge. To evaluate the solution, data were acquired from different scenarios including accurate radio propagation model, indoor empirical test, and outside empirical test. Experiment results shows that non-metric MDS can only combat the small scale randomness in the shadowing effects. To combat the large scale ones, macro-diversity approaches are further presented including rotating the receiver's antenna or collecting RSS from more than one mote in the same position. By averaging the measurements from these diversified receivers, simulation results and empirical tests show that shadowing effects can be greatly reduced. Also included in this thesis is how effective packet structures should be designed in the mote programming based on TinyOS to collect different sequences of RSS measurements and fuse them together.Keywords: Localization, MDS, Received Signal Strength, Wireless Sensor Network, Macro-Diversity... |
