ML-Enhanced Indoor Positioning Using Photovoltaic Cells for Light-Based Internet of Things

The development of 6G technology is driving the development of sustainable Internet of Things (IoT)-based sensor networks, emphasising low-cost and environmentally friendly solutions. Light-Based IoT (LIoT) emerges as a promising approach, leveraging the unlicensed optical spectrum for energy autonomy through photovoltaic energy harvesting and Visible Light Communication (VLC) via indoor luminaires acting as Optical Access Points (OAPs). This research explores integrating Machine Learning (ML) techniques for indoor positioning within LIoT sensor networks, supporting the intermittent, energy-autonomous operation. We propose an ML-based indoor positioning method that utilises OAPs for data preprocessing and model training, capitalising on existing energy harvesters (EH) for both power generation and position detection. A proof-of-concept LIoT node prototype was developed and tested against conventional supervised ML algorithms. Results demonstrate that these ML models can achieve distance accuracy with error margins within 7 cm and positioning accuracy within 20 cm. This study highlights the design approaches and potential of integrating EH systems with ML-based positioning. By enabling indoor location detection, this integration allows LIoT-based devices to communicate, harvest energy, and acknowledge their position using the same illumination from OAPs.