Characterizing Wi-Fi and cellular based mobile localization

• Main Author: Chua, Hong Chiat
• Other Authors: Li Mo
• Format: Final Year Project
• Language: English
• Published: 2016
• Subjects: DRNTU::Engineering
• Online Access: http://hdl.handle.net/10356/69113
• Institution: Nanyang Technological University

Localisation is the process of determining an object’s location in an environment. There are a wide range of uses for localisation. Some examples include targeted advertising and navigation. In indoor environment, there are many access points so localisation using Wi-Fi fingerprinting is more accurate. However, in outdoor environment, signal may not always be available for localisation using Wi-Fi fingerprinting. On the other hand, localisation using Cellular fingerprinting is less accurate in indoor environment but signal is always available even in outdoor environment. By making a localisation scheme that combine the advantage of Wi-Fi fingerprinting and the advantages of Cellular fingerprinting, the new localisation scheme will be both accurate and the signal will always be available. The experiment begins with data collection regarding received signal strength indicator (RSSI) at different indoor and outdoor locations. An algorithm is then developed to predict the ground truth of a location by comparing the RSSI value of the ground truth with the RSSI value of location that is in database. A linear regression model is also built to estimate the distance error between predicted location and the ground truth. There will be 4 linear equations, one for Wi-Fi at indoor, one for Wi-Fi at outdoor, one for Cellular at indoor and one for Cellular at outdoor. The attribute used for linear equation are resolution and difference. To improve accuracy of prediction model, at each location, the localisation scheme that have a lower predicted distance error will be used for predicting the ground truth. In order to evaluate if the linear regression model built is good, tests will be carried out along a path which is a combination of locations of different environment. The linear regression model is considered good if the predicted location is close to the ground truth. For all 4 linear equations, the p value obtained is very small. This means that the attribute resolution and difference are significant attribute for the linear equation. The cumulative distribution function that is obtained also shows that Selection localisation scheme, which is the localisation scheme that choose Wi-Fi or Cellular based on lower predicted distance error, increases the accuracy of localisation as compared to Wi-Fi or Cellular only. By knowing when to use Wi-Fi or Cellular for localisation, accuracy of localisation will be improved.