Joint AMC and Resource Allocation for Mobile Wireless Networks Based on Distributed MARL

With the rapid development of intelligent devices, the fifth-generation (5G) mobile wireless networks are envisioned to support massive connections and higher capacity. To confront challenges on link inefficiency in traditional mobile wireless networks, the link adaptation technology is crucial for system capacity improvements and requires coordination with resource allocation strategy. In this paper, we consider a joint adaptive modulation and coding (AMC) and resource allocation (RA) in a wireless network, where multiple users share limited subcarriers and adaptively change modulation levels and transmit power with the target to maximize the long-term system throughput. Instead of using optimization theory-based methods with higher complexity, we propose an intelligent double deep Q-network (DDQN)-based AMC and RA algorithm, which regards users as agents that learn cooperatively from their past experiences and implement their policies distributively. Furthermore, to guarantee fairness among users, we re-design the multi-agent reinforcement learning (MARL) reward function to incorporate the attained proportional fairness of each user at the current cycle into our objective. Simulation results demonstrate that users successfully learn to collaborate in a distributed manner, which leads to improved throughput both of the single link level and the whole system level.