Deep reinforcement learning based dynamic power allocation for uplink device-to-device enabled cell-free network

In this paper, we consider device-to-device enabled uplink cell-free communication between external users with the base station. By exploiting the channel gain differences, external and cellular users are multiplexed into the transmission power domain and then non-orthogonally scheduled for transmission with the same spectrum resources. Successive interference can-cellation is then applied at the base station to decode the message signals. We introduce an effective deep reinforcement learning (DRL) scheme to optimise the worst-case user rate through the dynamic power allocation of both external and cellular users. We also compare the performance of the DRL scheme under zero-forcing beamforming and conjugate beamforming methods. Simulation results verify the effectiveness of the DRL method for guaranteeing the user fairness through the worst-case rate maximisation.