A Power Allocation Scheme Based on Deep Reinforcement Learning in HetNets

Femtocells are being increasingly deployed to meet the capacity requirements of the next-generation heterogeneous mobile networks (HetNets). However, co-channel interference among cells has become a significant factor which limits the performance of dense HetNets. Reinforcement learning (RL) is regarded as a promising tool for resource management in femtocell networks. In this work, power allocation is modeled as a Markov Decision Process and the dense HetNet is modeled as a multi-agent network in which each base station is treated as an agent. Deep Q network (DQN) is adopted to solve the problem of huge states space caused by the combination of all agents’ states and manage the resources of HetNets in an efficient way. Compared with previous work that based on simulations, the proposed approach can improve the system capacity effectively.