Deep Reinforcement Learning-based Power Allocation for Downlink RSMA System

In this paper, we study the maximum sum rate for a downlink rate splitting multiple access (RSMA) system in long run. In the considered model, the RSMA technique is utilized at a powered base station (BS), where the data from the single BS is divided into two sub-data with different transmit power and receivers decode their received data using a successive decoding technique. To maximize the sum rate of the system, the optimization problem is first reformulated as a Markov decision process framework. Eventually, a deep reinforcement learning algorithm is applied to obtain the optimal solution and deal with the stochastic properties of the network environment. Simulation results show that the RSMA outperforms non-orthogonal multiple access (NOMA) and orthogonal frequency-division multiple access (OFDMA) in terms of system sum rate.