Outdoor Visible Light Positioning using Convolutional Neural Networks

In this study, 2-D positioning is examined in a 60 $\text{m} \times 5 \text{m}$ area using a Visible Light Communication (VLC) link employing urban lighting infrastructure, the so-called Infrastructure to Vehicle (I2V) scenario, where a deep learning-based method is applied for this purpose. In this work, 4 photodiodes are used in the receiver. Also, 3 street lights in a row are considered as three transmitters. As a result, 12 received power signals from three transmitters are used for four receivers to estimate the position. The results obtained using the proposed architecture show an improvement in average positioning error of up to $37 \%$ compared to similar works, under various scenarios including different transmitter distances, road widths, and transmitter heights. This performance improvement varies depending on the mentioned parameters, with the positioning error increasing as the distance between transmitters increases. Additionally, in these study, the channel conditions, signal transmission, and reception are considered much closer to reality compared to similar works, although the signal transmission method used is simpler. The cost of this improvement is an increase in computational complexity and doubling of the positioning computation time compared to the similar work where the deep network architecture used is of the Multilayer Perceptron type.