A Machine Learning-based Approach to NLOS Detection for Wi-Fi and BLE RSS Measurements

Wireless networks such as Wi-Fi and Bluetooth Low Energy are currently used to provide positioning solutions in GNSS-denied environments. These systems are based on the only physical observable that can be obtained from off-the-shelf smartphones, the received signal strength (RSS), which is correlated with the transmitter-receiver distance. In addition, other effects, such as the line of vision, makes RSS measurements unstable and noisy. Thus, distance information is extracted from several consecutive measurements. However, it is challenging to extract other information from its distribution. In this work, we study RSS measurements for Wi-Fi and BLE signals to predict the visibility measurement conditions. Results show that a 90 % prediction accuracy can be achieved using machine learning tools when comparing data from the same technology and conditions.