Multi-Agent Deep Reinforcement Learning for Resource Allocation in the Multi-Objective HetNet

Resource allocation in a heterogeneous network is an NP-hard problem, especially in 5G network scenarios. Multi-objective optimization in resource allocation is a challenging task that cannot be solved by the conventional optimization algorithm. In this paper, we propose a distributed multi-agent deep reinforcement learning (MADRL) for joint resource allocation to maximize spectrum efficiency (SE) and energy efficiency (EE). We employ the distributed learning in a stochastic geometry-based realistic heterogeneous network, where multiple femto base stations, a fixed number of macro and pico base stations in addition to randomly distributed mobile users are deployed. We propose a distributed MADRL multi-objective optimization problem (MADRL-MOP) framework to validate the performance. The simulation results demonstrate that the DDPG-based MADRL-MOP framework can not only handle the joint resource allocation problem effectively but also achieve better spectrum efficiency as well as convergence performance.