A Novel User Selection Massive MIMO Scheduling Algorithm via Real Time DDPG

Massive MIMO tremendously improves system capacity in the fifth generation (5G) wireless network with spacial diversity. Multi-user Multi-Input Multi-Output (MU-MIMO) user selection and resource allocation becomes more complex and challenging with a significantly increasing number of antenna in 5G network. In this paper, we aim to identify the user channel correlation and select users in the MU-MIMO system via the deep deterministic policy gradient (DDPG). Considering a short timing budget for real-time scheduling, we innovatively tailor DDPG into a short latency decision making process by asynchronously training update. By modeling MU-MIMO scheduling into a markov decision process (MDP), the proposed real-time DDPG (RT-DDPG) algorithm can be well-trained and outputs the selected users with their frequency domain resource simultaneously on all subbands, satisfying the minimum correlation among selected users. Numerical results validate the proposed RTDDPG user scheduling algorithm scheme drops 56% of latency comparing with traditional DDPG and significantly outperforms the traditional MU-MIMO user scheduling algorithms in system throughput.