Sensor Fused Indoor Positioning Using Dual Band WiFi Signal Measurements

Abstract: A ubiquitous and accurate positioning system for mobile devices is of great importance both to business and research due to the large number of applications and services it enables. In most outdoor environments this problem was solved by the introduction of the Global Positioning System (GPS). In indoor or suburban areas however, the GPS signals are often too weak to enable a reliable position estimate. Instead, other techniques must be utilized to provide accurate positioning. One of these is trilateration based on WiFi signal strengths. This is an auspicious technology to use partly because of the large number of access points (APs) in our everyday environment, and partly due to the possibility of measuring signal strength with a normal smartphone. The technique is further enabled by the move to include transmitters at 2.4 as well as 5 GHz in modern APs, providing a better basis for accurate position estimations. Furthermore, the motion sensors present in today’s smartphones are accurate enough to provide a short-time estimate of the user’s movement with high accuracy. In this thesis, both of these technologies are used to develop an accurate method for indoor positioning, and the contributions can be summed up into two points. The first contribution is an investigation of the behavior of two WiFi frequencies, 2.4 and 5 GHz, where their time dependent noise is proven to be almost uncorrelated with each other. This is then exploited to develop aWiFi-only trilateration algorithm by the use of a particle filter (PF), where the only restriction is that the locations of the APs need to be known. The second contribution is adding an accelerometer and a gyroscope to the algorithm, to provide a more accurate estimation. A step counter is developed using the accelerometer, and the gyroscope detects changes in heading while the WiFi signal strengths give information about the position. This makes it possible to alongside the position also estimate both heading and step length, while still keeping the only restriction of knowing the AP locations. The resulting algorithm produces position estimates with a mean error less than two meters for a specific use case, and around three meters when a more lenient user behavior is allowed.