Enhancing 3D Indoor Visible Light Positioning With Machine Learning Combined Nystrom Kernel Approximation

Abstract

Optical wireless communication (OWC) is emerging as a pivotal technology for next-generation broadcast networks, with visible light communication (VLC) poised to meet the escalating demands of advanced radio frequency systems. This study focuses on enhancing visible light positioning (VLP), recognized for its precision, simplicity, and cost-effectiveness, which are essential for accurate indoor localization and responsive location-based services. Central to our approach is the integration of advanced machine learning (ML) techniques, which fundamentally transform the accuracy and efficiency of 3D indoor positioning systems. We introduce an advanced VLP framework where ML is leveraged not merely as an adjunct but as the primary driver of innovation, significantly refining the processing of received signal strength (RSS) indicators. The methodology centers around a system comprising four light-emitting diodes (LEDs) arranged in a star geometry, optimized for precise spatial localization. We evaluate three distinct methodologies: a foundational star-shaped configuration for baseline position estimation, a repeated unit cell strategy to extend the four-LED configuration to a larger positioning area, and a sophisticated implementation employing Nystrom kernel approximation. This integration of Nystrom approximation into our ML framework drastically enhances the system's predictive accuracy, achieving an exceptional average relative root mean square error (aRRMSE) of 2.1 cm in a simulated setup. The results demonstrate that ML, especially combined with the application of the Nystrom kernel approximation, significantly elevates the precision and operational efficiency of traditional VLP systems, setting new benchmarks for accuracy in indoor 3D positioning technologies and fostering advancements towards more sophisticated and adaptable communication networks.