Parameter Estimation for Mobile Positioning Applications

The availability and reliability of mobile positioning algorithms depend on both the quality of measurements and the environmental characteristics. The positioning systems based on global navigation satellite systems (GNSS), for example, have typically a few meters accuracy but are unavailable in signal denied conditions and unreliable in multipath environments. Other radio network based positioning algorithms have the same drawbacks. This thesis considers a couple of cases where these drawbacks can be mitigated by model-based sensor fusion techniques. The received signal strength (RSS) is commonly used in cellular radio networks for positioning due to its high availability, but its reliability depends heavily on the environment. We have studied how the directional dependence in the antenna gain in the base stations can be compensated for. We propose a semiempirical model for RSS measurements, composed of an empirical log-distance model of the RSS decay rate, and a deterministic antenna gain model that accounts for non-uniform base station antenna radiation. Evaluations and comparisons presented in this study demonstrate an improvement in estimation performance of the joint model compared to the propagation model alone. Inertial navigation systems (INS ) rely on integrating inertial sensor measurements. INS as a standalone system is known to have a cubic drift in the position error, and it needs supporting sensor information, for instance, position fixes from GNSS whenever available. For pedestrians, special tricks such as parametric gait models and step detections can be used to limit the drift. In general, the more accurate gait parameters, the better position estimation accuracy. An improved pedestrian dead reckoning (PDR) algorithm is developed that learns gait parameters in time intervals when direct position measurements (such as GNSS positions) are available. We present a multi-rate filtering solution that leads to improved estimates of both gait parameters and position. To further extend the algorithm to more realistic scenarios, a joint classifier of the user’s motion and the device’s carrying mode is developed. Classification of motion mode (walking, running, standing still) and device mode (hand-held, in pocket, in backpack) provides information that can assist in the gait learning process and hence improve the position estimation. The algorithms are applied to collected data and promising results are reported. Furthermore, one of the most extensive datasets for personal navigation systems using both rigid body motion trackers and smartphones is presented, and this dataset has also been made publicly available