Mobility-aware link adaptation in visible light communication systems using diverse reinforcement learning architectures

Visible Light Communication (VLC) has emerged as a promising solution for high-capacity indoor wireless networks; however, its practical deployment is severely challenged by user mobility, strong line-of-sight dependency, and frequent link blockages. Conventional mobility management and handover strategies in VLC systems predominantly rely on static thresholds or heuristic rules, which are inherently unable to adapt to rapidly time-varying channel conditions induced by user movement. To address these limitations, this paper adopts an intelligent learning-based framework for mobility-aware joint power and link adaptation in VLC networks. A mathematically grounded multi-objective reward function is formulated to jointly capture signal quality, handover efficiency, transmit power adaptation, and interruption penalties, and its robustness is validated through sensitivity analysis under varying weight configurations. This enables proactive and smooth handover decisions that adapt to instantaneous variations in received signal quality caused by user mobility and dynamic channel conditions. A realistic indoor VLC environment with mobility-aware channel dynamics is considered to evaluate the learning behavior under diverse movement scenarios. Simulation results demonstrate that the proposed lightweight learning scheme attains a steady-state total reward of approximately 14,200 while exhibiting the lowest reward standard deviation (50–100) and the tightest clustering across 50 independent runs (mean 14,150). In direct comparison, more complex deep-learning models reach a marginally higher reward of 14,500 but suffer from substantially larger variance and a significant increase in training time (1034.48 s versus 55.96 s), whereas the rule-based and greedy baselines remain limited to 7300 and 8000, respectively. Under minor LED-layout perturbations, the proposed solution shows negligible performance degradation (less than 1%), confirming exceptional generalization with minimal memory footprint (0.02 MB), thereby positioning it as the most balanced, stable, and deployment-ready solution for mobility-aware link adaptation in realistic multi-user VLC environments.