RL-Assisted Power Allocation for Covert Communication in Distributed NOMA Networks

This article focuses on covert communication in a distributed network with multiple nonorthogonal multiple access (NOMA) systems, where each NOMA system is consisted of a transmitter, a legitimate public user, a covert user, and a warden. Power allocation for multiple transmitters in such network is a highly tricky problem, since it needs to addresses the issues of complex inter-NOMA system interference, constraints from both public users and covert users, and the optimization of overall network performance. We first conduct a theoretical analysis to depict the inherent relationship between the inter-NOMA system interference and transmit power of transmitters. With the help of the interference analysis, we then develop a theoretical framework for the modeling of detection error probability, covert rate, and public rate in each NOMA system. Based on these results and the constraints from both public users and covert users, we formulate the concerned power allocation problem as a Markov decision process, and further develop multiagent reinforcement learning (RL) algorithms to identify the optimal power allocation among transmitters to maximize the sum-rate of the overall network. Finally, numerical results are provided to illustrate the efficiency of our RL algorithms for power allocation in multi-NOMA networks.