Deep Learning-Based Resource Allocation for Device-to-Device Communication

In this paper, a deep learning (DL) framework for the optimization of resource allocation in multi-channel cellular systems with device-to-device (D2D) communication is proposed. Thereby, channel assignment and discrete transmit power levels D2D users, which are both integer variables, optimized maximization overall spectral efficiency whilst maintaining quality-of-service (QoS) users. Depending on availability state information (CSI), two different configurations considered, namely 1) centralized operation full CSI 2) distributed partial CSI, where latter case, encoded according to capacity feedback channel. Instead solving resulting problem each realization, DL proposed, optimal strategy arbitrary conditions approximated by neural network (DNN) models. Furthermore, we propose new training that combines supervised unsupervised methods local sharing achieve near-optimal performance while enforcing QoS constraints users efficiently handling variables based few ground-truth labels. Our simulation results confirm can be attained low computation time, underlines real-time capability proposed scheme. Moreover, our show not only but also encoding determined using DNN. easily extended design objectives.