Learning for Visible Light Communications: Potential Scenarios and Applications

Visible light communications (VLC) are considered a key enabler technology for future wireless communications. The consumer light emitting diodes (LEDs) can be repurposed as optical access points (APs) providing a small and confined coverage footprint each. Taking into consideration the VLC characteristics and the interference/interaction with neighboring LEDs, the context-dependent propagation or the influence of effects such as blocking or shadowing, the use of specific channel models and network-centric resource management may fail because of the need for a large amount of parameters to properly model the VLC systems. In this context, learning methods may potentially solve these issues by exploiting previous measurements employed as training data. Although learning methods have been already proposed for radio-frequency (RF) systems, they may not be straightforwardly applied to VLC because of the particular characteristics of this technology. Available learning methods for VLC are today mostly based on synthetic datasets that may not capture the effects of these characteristics. In this article, the application of learning methods to VLC is analyzed presenting the differences with RF systems that make their application not a mere extension of the current RF approaches. We introduce the main issues to solve that may benefit from learning methods and the involved parameters. Moreover, the principles of a measurement platform are presented for generating realistic datasets. After that, the new thinking is presented for specific VLC applications as well as future challenges and research directions are provided.