Localization using hidden Markov models in a multipath environment

Accurate and reliable positioning of passenger/user portable wireless devices can provide promising enhanced user function convenience and social benefits. This thesis aims to apply techniques at the radio frequency (RF) 2.4GHz Industrial, Scientific and Medical (ISM) communication band, for developing pedestrian localization algorithm in a harsh environment with, typically an indoor area where the global positioning system (GPS) technique is usually not available. The work involves algorithm design, experimental tests and application evaluation. Localization in an indoor environment remains a challenging problem to address. Firstly, the multipath propagation of RF signals due to reflection and diffraction, as well as signal occlusion due to non-line-of-sight propagation leads to significant uncertainties in modelling and calibration of the wireless channel, which in turn reduces the accuracy and reliability of many localization algorithms. Secondly, most of the existing localization systems have limited bandwidth, making time-of-arrival based methods less appealing as they rely on a higher system bandwidth for suitable time resolution for achievement of a desired localization accuracy. Thirdly, the measurement update rate in most existing infrastructure is too low to meet the practical real time localization requirement. The problem is exacerbated by missing measurements because of unreliable wireless links or short transmission range. These all pose significant challenges to existing localization techniques, such as those dedicated to the indoor localization systems. A cost effective solution with high accuracy and reliability is required. We propose a Bayesian localization scheme where prior information of the building environmental constraints (such as wall geometry) as well as the kinematic properties of the user (such as heading information, motion to different places or remaining in one place for a period of time), can be utilized for localization purpose. This can help to mitigate location ambiguity problems incurred by measurement uncertainties in a difficult environment and hence increase localization accuracy and reliability. Hidden Markov models (HMMs) turn out to be an effective framework to facilitate this Bayesian scheme for information fusion in the localization process. In addition, as an extension of HMMs, hidden semi-Markov models (HsMMs) are also employed to provide a more flexible Bayesian framework since the state sojourn time can be appropriately customized. The performances of HMMs and HsMMs are compared and evaluated using both simulated data and practical data. We further modify and extend the HsMM framework so that it can be utilized for localization in a varying environment, such as in the mixed line-of-sight and non-line-of-sight conditions. We demonstrate promising results that indicate that HsMMs provide a powerful framework for addressing various indoor localization problems. The algorithm developed here have wider applicability, and are not restricted to the type of the RF signal parameters where we have applied them. In this thesis, we mainly use the received signal strength (RSS) as our measurements for algorithm validation and performance evaluation.