Comparative Study of Different BLE Fingerprint Reconstruction Techniques

Abstract

In radio-frequency Indoor Positioning Systems, the fingerprinting technique, which involves a laborious database construction stage to achieve good positioning results, has become a standard solution. Several researchers have tried to find methods to reduce the number of reference points – which need to be manually collected – while maintaining the positioning results, but have focused on other technologies, such as XBee or WLAN, and more specifically Wi-Fi. In this work, a comparison between different fingerprint generation techniques is made in a Bluetooth Low-Energy (BLE) Indoor Positioning System using Received Signal Strength Indicator measurements. Various databases are collected using three Android phones. These data are employed to train different methods that will generate fingerprints to reconstruct the missing points in the databases. The methods used are Inverse Distance Weighting, Support Vector Regression, Gaussian Process Regression and Generative Adversarial Networks. After this comparison, it has been found out that all these techniques could reconstruct part of the missing points and improve the positioning results. However, Generative Adversarial Networks achieved the best reconstruction throughout all of the databases.