Beamforming in Multi-User MISO Cellular Networks with Deep Reinforcement Learning

In multi-user multi-input single-output (MU-MISO) cellular networks, beamforming is an effective way to manage the inter-cell interference and intra-cell interference, and improve the achievable rate. However, finding the optional beamforming solution needs a centralized structure, which may be impractical in realistic scenario. In this paper, a distributed deep reinforcement learning (DRL) based beamforming algorithm is proposed in which each base station (BS) uses DRL to select the beamformers for its intended users in each cell. Besides, the channel orthogonality measure among intended users, on behalf of the intra-cell interference, is used as the state element of the DRL. Moreover, by applying the proposed method, the number of action elements can be reduced, thus the training complexity decreased. Compared with the benchmark algorithm, the simulation results demonstrate that this scheme could improve the system achievable rate. In a word, this paper provides another way for optimizing the beamforming problem in MU-MISO systems.