Federated graph equalization for orientation-robust cell-free visible light communication with angle-diversity receivers

Abstract

Visible light communication (VLC) is attractive for dense indoor wireless access because the light-emitting diode infrastructure can provide illumination and data transmission simultaneously. However, practical VLC access points are highly sensitive to user orientation, non-uniform illumination, co-channel interference and privacy limitations in collecting user-side waveforms for machine learning. This paper proposes a new cell-free VLC architecture in which distributed ceiling light-emitting diode access points jointly serve mobile users equipped with angle-diversity photodiode receivers. A federated graph neural equalizer is introduced to learn the orientation-dependent optical channel and residual non-linear distortion without transferring raw received samples to the controller. The graph representation treats light access points and user terminals as nodes, while edge attributes include received signal strength, geometric distance, receiver tilt and blockage indicators. A complete intensity-modulation/direct-detection signal model, orientation-aware channel gain formulation and resource-control optimization are developed. Monte Carlo simulations in a 6 m × 6 m × 3 m indoor room show that the proposed scheme improves network sum rate by 26.9 % compared with cell-free weighted minimum mean square error processing without federated learning and reduces orientation-induced outage by 74.4 % compared with strongest-light association. The results indicate that federated graph learning is a promising privacy-preserving tool for robust next-generation LiFi access.