- The paper introduces a hybrid GM-ANN mobility predictor that dynamically allocates power to mitigate misalignment in mobile VCSEL-based OWC systems.
- It demonstrates energy efficiency improvements of 30–35% over conventional schemes and 45–65% over reactive methods under varying mobility speeds.
- Simulations in realistic indoor settings confirm that the MAPC framework reliably maintains QoS by preemptively adapting to channel gain fluctuations.
Mobility-Aware Power Control Framework for VCSEL-Based Indoor Optical Wireless Communication
Context and Motivation
The proliferation of data-intensive indoor wireless applications necessitates robust, high-capacity connectivity solutions. Optical wireless communication (OWC) leveraging vertical-cavity surface-emitting lasers (VCSELs) offers multi-gigabit throughput, high spatial reuse, and enhanced security, but introduces severe sensitivity to user mobility due to its highly directional beams. Existing power control schemes, typically agnostic to mobility-induced channel dynamics, result in suboptimal energy utilization as they overprovision power to compensate for unpredictable misalignment. This paper introduces a mobility-aware power control (MAPC) framework that dynamically allocates power by predicting mobility-driven channel variations using a hybrid stochastic and learning-based model.
System and Mobility Modelling
The paper models an indoor OWC scenario with VCSEL-array access points (APs) mounted on the ceiling. Each AP forms highly directional Gaussian-profile beams targeted at mobile users equipped with angle diversity receivers (ADRs). The spatial and orientation dynamics of users (position, velocity, direction, elevation, and azimuth) critically impact channel gain, and hence, reliable power allocation.
The proposed mobility model fuses a Gauss-Markov (GM) process—capturing continuous and temporally correlated motion—with an LSTM-based artificial neural network (ANN) correction that models non-linear behavioral transitions (sudden turns, pauses, orientation changes). This hybrid GM-ANN methodology is empirically superior to conventional RWP or standalone GM models. The LSTM is trained on synthetic data reflecting both stochastic trajectories and realistic behavioral disruptions, producing short-term mobility predictions with low root mean squared error (RMSE), critical for channel gain estimations in highly directional OWC.
Channel Gain and Power Control
Channel modelling leverages the VCSEL beam profile, incorporating propagation distance, beam misalignment, and device orientation. The channel gain calculation includes the beam transformation through microlens arrays and user position-orientation dynamics. Users are associated to APs based on predicted maximal channel gain, discarding links violating field-of-view (FoV) constraints.
The power control optimization seeks to maximize system energy efficiency (EE), defined as the total achievable data rate per unit transmit power, subject to QoS, AP capacity, and hardware constraints. The non-convexity of the formulation precludes real-time MILP solving; instead, a CNN is trained offline on MILP solutions and during operation directly maps predicted channel gain matrices to near-optimal power allocations, enabling real-time implementation.
Numerical Results and Analysis
The paper demonstrates, via simulations in realistic indoor settings (5 × 5 × 3 m³ room, 12 APs, mobile users with walking speeds up to 1.5 m/s), that the MAPC approach materially outperforms three baselines:
- Conventional Power Control (CPC): Ignores mobility, reacts only to instantaneous channel info.
- Conservative Power Control (ConsPC): Allocates extra power as margin without mobility forecasting.
- Reactive Power Control (RPC): Responds post hoc to link degradation.
Strong numerical results include:
- MAPC achieves average EE gains of 30–35% over CPC/ConsPC and 45–65% over RPC across varying user loads and mobility speeds.
- MAPC exhibits considerably smaller EE degradation as mobility speed increases, due to proactive anticipation and correction of misalignment-induced channel fluctuations.
These results provide empirical evidence that mobility prediction tightly coupled to resource allocation yields significant gains in energy efficiency and robustness compared to legacy quasi-static and reactive schemes.
Implications and Future Directions
The practical implications of this MAPC paradigm include:
- Substantially improved energy efficiency, supporting sustainable high-capacity OWC deployments for dense, mobile indoor environments.
- Enhanced robustness to mobility-induced channel variation, reducing the risk of outage and maintaining QoS even under complex user trajectories and device orientation dynamics.
- The combination of mobility prediction and deep learning for optimization creates a viable real-time control architecture suitable for edge or cloud-based OWC management.
Theoretically, the joint modelling of mobility and channel dynamics sets a foundation for more granular resource allocation protocols, especially as VCSEL-based systems scale to dense IoT and AR/VR applications. Potential future developments include integration of blockage and shadowing effects, multipath channel modelling, extension to multipoint traffic demand prediction, and adaptive cross-layer scheduling responsive to both mobility and traffic dynamics.
Conclusion
This paper provides a comprehensive framework addressing the critical bottleneck in VCSEL-based indoor OWC: mobility-aware power allocation. By integrating a hybrid GM-ANN mobility predictor with learning-based power control, MAPC proactively adapts resource allocation in response to predicted channel conditions, maximizing energy efficiency and reliability. This work represents a step toward fully dynamic, context-aware indoor optical wireless networks. Further research can extend the model’s applicability to environments with obstructions and variable traffic demand, deepening its impact on practical OWC network design and operation (2604.22682).