Deep Reinforcement Learning for Channel Selection and Power Control in D2D Networks

As a promising candidate to alleviate the mobile traffic explosion, device-to-device (D2D) technology enables the direct communications between proximal devices. To mitigate the mutual interference among D2D pairs and improve the spectrum efficiency, this paper investigates a weighted-sum-rate (WSR) maximization problem in multi-channel D2D networks. Particularly, we propose a deep reinforcement learning based scheme for each D2D pair to make decisions independently on the channel selection and power control. In contrast to the conventional methods that require instantaneous global network information, the proposed scheme only needs local information and outdated feedbacks. Simulation results demonstrate that, in terms of the WSR, the proposed scheme outperforms the conventional suboptimal fractional programming algorithm that requires the instantaneous global network information.