Improving BLE Distance Estimation and Classification Using TX Power and Machine Learning: A Comparative Analysis

Distance estimation and proximity classification techniques are essential for numerous IoT applications and in providing efficient services in smart cities. Bluetooth Low Energy (BLE) is designed for IoT devices, and its received signal strength indicator (RSSI) has been used in distance and proximity estimation, though they are noisy and unreliable. In this study, we leverage the BLE TX power level in BLE models.We adopt a comparative analysis framework that utilizes our extensive data library of measurements. It considers commonly used state-of-the-art model, in addition to our data-driven proposed approach. The RSSI and TX power are integrated into several parametric models such as log shadowing and Android Beacon library models, and machine learning models such as linear regression, decision trees, random forests and neural networks. Specific mobile apps are developed for the study experiment. We have collected more than 1.8 millions of BLE records between encounters with various distances that range from 0.5 to 22 meters in an indoor environment. Interestingly, considering TX power when estimating the distance reduced the mean errors by up to 46% in parametric models and by up to 35% in machine learning models. Also, the proximity classification accuracy increased by up to 103% and 70% in parametric and machine learning models, respectively. This work is one of the first studies (if not the first) that analyze in depth the TX power variations in improving the distance estimation and classification.