Visible Light-Based User Position, Orientation and Channel Estimation Using Self-Adaptive Location-Domain Grid Sampling

Abstract

In this paper, visible light-based positioning (VLP) is studied. VLP is greatly challenging because (i) it is essentially a non-convex optimization problem since the visible-light received signal strength (RSS) is nonlinear with the user equipment (UE) position; and (ii) in addition to the UE location, the visible light RSS also depends on the UE orientation and small-scale channel gains, which are unknown in practice. This complicates the VLP problem due to the enlarged searching space. To address these challenges, we propose a location-domain grid sampling scheme. Specifically, the location-domain grid sampling can potentially partition the location space into small cells, and hence the non-convexity challenge of RSS-based VLP is mitigated. In addition, using the location-domain grid sampling, we transform VLP into a sparse recovery problem. A novel group sparse learning (GSL) algorithm with self-adaptive location-domain grids is proposed to achieve an efficient RSS-based VLP solution, via exploring the inherent sparse structure. The convergence of our GSL algorithm is established. Thanks to the adaptivity of dynamic location-domain grids, the required number of location-domain grids can be significantly reduced, compared with conventional fixed-grid-based GSL solutions. Moreover, the proposed GSL-based VLP method jointly learns the UE location, orientation and channel gain, thus achieving a robust RSS-based VLP solution against parameter uncertainties. Finally, our simulation result verifies the large performance gain of the proposed RSS-based VLP solution over state-of-the-art VLP baselines, thanks to our self-adaptive grid sampling and problem-specific group sparse learning.