Dynamic Power Allocation Scheme for NOMA Uplink in Cognitive Radio Networks Using Deep Q Learning

Non-orthogonal multiple access (NOMA) is a promising technique to satisfy a host of access demand and provide higher throughput. It enables that multiple users are multiplexed with different power levels by using superposition coding at the transmitter side and successive interference cancellation at the receiver side. In this paper, we deal with the long-term throughput maximization of an uplink NOMA in the cognitive radio network (CRN). The secondary users (SUs) have limited capacity battery, hence, SUs equipped with an energy harvester can harvest energy from solar sources to prolong their operations. Particularly, a combination of NOMA and time division multiple access (TDMA) is proposed in order to reduce the complexity of massive wireless communication systems. By taking into account the practical applications, a deep Q learning algorithm is employed to maximize the long-term throughput of the system, where the agent (i.e secondary base station (SBS)) can interact with an environment to learn about system dynamics. As a result, the SBS learns how to allocate optimal transmission energy to SUs in each time slot. Simulation results demonstrate that the proposed scheme can achieve better performance than conventional schemes.