Deep-Reinforcement-Learning-Based Resource Allocation in ultra-dense network
Rui Huangi, Jiangbo Si, Jia Shi, Zan Li
Abstract
In this paper, a joint spectrum and power allocation problem is formulated to maximize the total achievable transmit rate for the ultra-dense network (UDN). To solve this optimization problem in UDN, we propose a cooperative multi-agent reinforcement learning (CMARL) framework, where double deep Q-network (DDQN) is deployed to allocate the discrete spectrum and power resources. In addition, considering that the multiple agents cooperatively make resource allocation for total transmit rate maximization, we propose a new Q function to guarantees a better performance and stability. Finally, simulation is performed to evaluate the performance of CMARL scheme. It is shown that our proposed CMARL scheme can achieve a larger total transmission rate than conventional resource allocation algorithms and distributed reinforcement learning approach.