On structural identification of 2D regression functions for indoor Bluetooth localization

In-door localization of mobile devices is a common problem for many current and future applications, for example to control infrastructure services or for personalized in-building navigation systems. Sufficiently capable Bluetooth support is often available in off-the-shelf mobile devices such as mobile phones, which makes Bluetooth an attractive technology for cheap and widely available in-door localization systems. However, Bluetooth has been optimized to deal with effects of radio frequency transmission such as reflection and multi-path propagation. It therefore produces highly non-linear relationships between the distance of devices and their perceived signal strength. In this paper, we aim to identify these relationships for a specific dataset of 2D device positions using structural identification methods. Driven by an extended genetic algorithm, we aim to find optimal mappings in form of non-linear equations for x and y coordinates, thus producing formal regression functions.