LAUSR

Visible light positioning (VLP) systems have shown significant promise for indoor localization, providing advantages, such as environmental friendliness, immunity to electromagnetic interference, and enhanced security. Nevertheless, existing VLP solutions predominantly focus on small-scale scenarios and generally necessitate complex receiver architectures to manage co-channel interference (CCI) resulting from identical carrier frequencies employed by multiple light emitting diode transmitters. Such interference severely compromises localization accuracy in large-scale deployments. In this article, we introduce a novel compressed sensing (CS)-based VLP framework, reframing the large-scale indoor localization problem as a sparse signal Recovery task. This framework simplifies receiver designs and effectively mitigates CCI arising from frequency reuse. Additionally, we propose a graph-neural-network-based localization algorithm that leverages the sparse coefficient fingerprints obtained through CS reconstruction to achieve high-precision positioning. To further optimize system performance, we design a neural-network-generated measurement matrix characterized by adjustable illumination properties and low mutual coherence, enhancing robustness against noise and further improving localization accuracy. Simulation results demonstrate that the proposed system achieves an average localization accuracy of 1.9 cm at a signal-to-noise ratio of 25 dB within a repeatable test unit measuring 200 cm $\times $ 200 cm $\times $ 300 cm. Such units can be flexibly and arbitrarily deployed in large-scale environments, consistently maintaining the reported localization accuracy. Compared to conventional methods employing the fast Fourier transform, our approach achieves a 53.98% improvement in positioning performance. Consequently, the proposed system represents a scalable, accurate, and practical solution ideally suited for complex, large-scale indoor positioning scenarios.