Deep Learning for Selection Between RF and VLC Bands in Device-to-Device Communication

This letter focuses on the selection between radio frequency (RF) and visible light communications (VLC) bands for users exchanging data directly with each other via device-to-device (D2D) communication. We target to maximize the energy efficiency of D2D communication while the outage is minimized. Since the VLC channel can vary quickly due to the possible changes in irradiance and incidence angles, we aim to reach a quick band selection decision in a multi-user scenario based only on the knowledge of the received power and sum interference from all D2D transmitters at the individual D2D receivers. The proposed solution is based on a deep neural network making an initial band selection decision. Then, based on the DNN's output, a fast heuristic algorithm is proposed to further improve the band selection decision. The results show that the proposal reaches a close-to-optimal performance and outperforms the existing solutions in complexity, outage ratio, and energy efficiency.