Adaptive Modulation and Coding Based on Multi-Agent Reinforcement Learning for Power Emergency Communications

Abstract

Power emergency communications are critical to rescue work when some disasters happen. For the sake of alleviating the shortage of spectrum resources and maintaining system connections, an adaptive modulation and coding scheme is studied in this paper. For the target system, the principle of cognitive radio networks (CRNs) is involved and the users in power emergency communications are modelled as primary users (PUs) and secondary users (SUs) according to their communication requirements. A maximum entropy based multi-agent actor-critic (ME-MAAC) algorithm in deep reinforcement learning is proposed to train the system and achieve an optimal policy, in which different users can access the system with varying modulation and coding schemes. The simulation results show that the proposed ME-MAAC algorithm outperforms the Deep Q-Network (DQN) algorithm in accordance with efficiency and performance. The proposed adaptive modulation and coding (AMC) scheme can improve system connection rate and spectrum efficiency, that is, the users in power emergency communications can obtain more communications with limited power and spectrum resources. This paper provides an useful guidance for the design of practical power emergency communications.