Deep Reinforcement Learning Based Power Allocation for D2D Network
Zhengran Bi, Wenan Zhou
Abstract
In device-to-device (D2D) networks, when D2D communication shares cellular network spectrum resources, it will cause serious co-channel interference. The problem of solving interference is complicated because of the large number of D2D users and the fading of the wireless channel. Aiming at this problem, this paper proposes a centralized Deep Reinforcement Learning (DRL) algorithm to solve the power allocation problem of D2D communication in time-varying environment. The algorithm regards D2D network as a multi-agent system, and represents the channel as a Finite State Markov Channel (FSMC). In particular, the method obtains power control by maximizing system capacity and user experience quality, respectively, while considering actual time-varying channels and D2D interference. Our results indicate that the proposed method outperforms the conventional reinforcement learning methods in terms of system throughput and user experience quality.