Tilted-angle insensitive received signal strength in visible light positioning systems using a deep neural network trained by synthetic data

Despite extensive research on received signal strength (RSS)-based visible light positioning (VLP), the receiver (RX) is assumed to stand vertically during the positioning process in most reported system designs. In this work, we propose a positioning strategy using a deep neural network (DNN) trained by synthetic data to address this problem. We further explicitly state the deficiencies in the current RSS-VLP algorithms when handling positioning problems involving RX orientation. Compared with existing RSS-VLP algorithms, our method can achieve high positioning accuracy even when the RX orientation is unknown. The results can further verify the feasibility of the system. In addition to the orientation predictability, the trained DNN can also regulate the algorithm time for each position.