Leveraging Time-domain Fingerprinting for Joint LiFi Position and Orientation Estimation

Abstract

To support performance requirements for smart services in 6G, user positioning is a crucial component. Indoor user position and orientation estimation based on Light Fidelity (LiFi) system is considered as a promising technology, due to its high precision, along with its ease of installation. The main bottleneck of user position and orientation estimation in LiFi is a non-linearity between the metrics, such as the received signal strength (RSS), position and orientation. A deep learning-based estimation methodology holds promise for addressing this issue, because it can learn complex propagation features dependent on user position and orientation. To fully capitalize on this advantage in the time-domain, we propose utilizing both time-series RSS and its received time, i.e. time-of-arrival (ToA) fingerprints, along with a novel neural network architecture named Deep RSS-ToA Fusion Network (DRTFNet). Simulation results demonstrate that the proposed DRTFNet achieves positioning accuracy of less than 3 cm and orientation accuracy of less than 3 degrees, outperforming both the basic Convolutional Neural Network (CNN) architecture using only RSS data and other baseline systems with more light sources.