A Dynamic Power Allocation Scheme in Power-Domain NOMA using Actor-Critic Reinforcement Learning

Non-orthogonal multiple access (NOMA) is one of the most promising technologies in the next-generation cellular communication. However, the effective power allocation strategy has always been a problem that needs to be solved in power-domain NOMA. In this paper, we propose a reinforcement learning (RL) method to solve the power allocation problem. In particular, in the power-domain NOMA, the base station (BS) simultaneously transmits data to the user under the constraint of the sum power. Considering that the power allocation assigned by the BS to each user can be used to optimize the energy efficient (EE) of the entire system, we propose the RL algorithm framework of the Actor-Critic to dynamically select the power allocation coefficient. A parameterized strategy is constructed in the Actor part, and then the Critic part evaluates it, and finally the Actor part adjust the strategy according to the feedback from the Critic part. Numerical results indicate that the proposed scheme can efficiently improve the EE of the entire system.