Opportunistic CR-NOMA Transmissions for Zero-Energy Devices: A DRL-Driven Optimization Strategy

To efficiently accommodate RF energy-harvesting (EH) capable device in a wireless network with prescheduled devices, this letter designs the deep reinforcement learning (DRL)-driven energy-efficient transmission strategy. The transmission strategy handles the EH-device uplink transmissions opportunistically using cognitive radio-inspired non-orthogonal multiple access (CR-NOMA) scheme while maximizing energy efficiency (EE). In this respect, firstly, we formulate the EE optimization problem of the EH-device while considering its RF circuit power consumption. Secondly, we divide the original, non-convex problem into a two-layer optimization problem, and solve it sequentially as i) we theoretically derive the optimal transmit power and time-sharing coefficient parameters from the first layer, and ii) using the derived parameters in the second layer, we solve the one-dimensional continuous space optimization problem through a DRL technique, recognized as a combined experience replay deep deterministic policy gradient (CER-DDPG). Finally, the numerical results show that, under different operational scenarios, the proposed approach outperforms benchmark DDPG and Stochastic algorithms in terms of EE.