Self-Representation and Generative Adversarial Learning-based Bluetooth Radio Map Reconstruction for Indoor Localization

Recently, there has been significant progress in indoor positioning research. While the wireless fingerprint-based positioning method offers excellent performance, its widespread use is hindered by the high cost of creating location-fingerprint databases. The rise of mobile intelligent devices has given way to a new sensing mode called “crowd sensing,” which enables the creation of low-cost location-fingerprint databases. This facilitates the widespread implementation of wireless signal fingerprint positioning methods. However, it's important to note that users participating in crowd sensing tend to gather in specific areas and are limited by event constraints, which means that user trajectories may not cover all detection areas. Consequently, certain areas may have limited or no wireless signal sensing information. To address this, a method for reconstructing radio map based on a small number of observation samples has been proposed in this paper to enable low-cost indoor positioning. To achieve high-quality radio map reconstruction, a combination of self-representation learning and generative adversarial learning architecture is used, with self-representation learning features guiding the generative adversarial learning process to achieve high-quality inference based on small samples. To validate the effectiveness of the proposed method, an indoor Bluetooth positioning system was built in our university's college building using ordinary mobile phone for signal acquisition and map construction. The results verified the superiority of the proposed method in map reconstruction and positioning accuracy compared to traditional methods.