Statistical Location Detection

Introduction

Localization – determining the approximate physical position of a user/device on a site – can be seen as an important enabling service in Wireless Sensor Networks (WSNs). The Global Positioning System (GPS) (Hofmann-Wellenhof et al. 1997) provides an effective localization technology outdoors and its popularity and the host of location-based services it has spawned is a testament to the importance of location information. The GPS technology though is unreliable in downtown urban areas and not functional indoors. Moreover, GPS receivers are expensive and power-hungry making them inappropriate for many WSN applications that emphasize very low-cost low-power sensor nodes.

A reliable indoor localization service would be extremely useful and would give rise to a plethora of innovative applications including: asset and personnel tracking in hospitals, warehouses, and other large complexes; locating faulty sensors in building automation applications; intelligent audio players in self-guided museum tours; intelligent maps for large malls and offices, smart homes (Hodes et al. 1997, Priyantha et al. 2000); as well as surveillance, military and homeland security related applications. Moreover, a location detection service is an invaluable tool for counter-action and rescue (Meissner et al. 2002) in disaster situations.

For these reasons, localization has received widespread attention in the literature and many approaches have been developed. A large class of localization systems uses special hardware (e.g., infrared sensors, ultrasound) which necessitates the deployment of a special-purpose WSN just for this purpose. Several related works are described in the following section. We are instead interested in a localization approach that can use WSN features found in virtually all existing platforms. Specifically, all WSN nodes carry a radio to communicate with each other. That radio is often rather rudimentary and the only information on the received RF signals one can obtain is signal strength. Received signal strength depends on the location of the transmitting sensor and the objective is to exploit this information to reveal the transmitter’s location. At the same time, we are interested in an approach that is general enough to exploit additional RF or other information that could be obtained with more sophisticated hardware, for instance signal angle-of-arrival and signal time-of-flight. As we will see, we are able to deal with any vector of available observations about the transmitting sensor.

The approach we develop in this work starts with a “discretization” of the localization problem by splitting the coverage area into a set of regions. The problem is to determine the region where a sensor node we seek resides. Quantities like signal strength are highly variable indoors due to the dynamic character of the environment leading to multipath and fading in the propagation of RF signals. For example, the propagation environment inside a building is highly complex and dynamic as there are multiple reflections, doors that may be open or shut, and people (acting as RF energy absorbers and reflectors) that are constantly moving.