Neural-Network-Based Blockage Prediction and Optimization in Lightwave Power-Transfer-Enabled Hybrid VLC/RF Systems

In this article, we consider a simultaneous lightwave information and power transfer-enabled indoor visible light/radio frequency (RF) hybrid communication system. In this system, several luminaries are mounted on the ceiling of a room and sensors receive lightwave power and information from luminaries. Each sensor harvests energy from lightwave signals and uses them for uplink communication using RF signals. In general, blockages due to the movement of humans are common in indoor systems which results in severe performance degradation. We consider blockages due to such human movements in our system. An optimization problem is formulated to maximize the uplink weighted sum rate considering uplink orthogonal multiple access (OMA) and non-OMA cases. To find the optimal beamforming matrix and time allocation parameters, a lightweight artificial neural network architecture is proposed. Our solution is capable of predicting the human blockages and accordingly optimizing the beamforming matrices and time allocation parameters leading to a near-optimal uplink sum rate. Specifically, up to 30% rate improvement is observed compared to zero-forcing beamforming when the number of blockages is more than five. Further, the use of NOMA for uplink results in up to 25% sum rate improvement.