NOMA-Enhanced Intelligent Semantic Communication Networks using Deep Reinforcement Learning

This research explores an intelligent semantic communication network where the base station (BS) processes and transmits semantic data to users through a wireless link. To improve communication between the BS and users, we apply the non-orthogonal multiple access (NOMA) technique in this system. Within this system, we consider the delay model for transmitting original data from the BS to users using semantic communication. This model consists of three main phases: (i) extracting semantic data from the original data at the BS; (ii) transmitting this semantic data from the BS to the users; and (iii) reconstructing the original data from the received semantic data at the users. To optimize the overall system delay, we formulate a problem involving optimizing beamforming vectors and computation resource allocation at the BS. To address this problem, we propose a deep reinforcement learning (DRL) framework that utilizes the deep deterministic policy gradient algorithm. In simulations, we analyze the convergence of our proposed framework, which shows stable performance under varying algorithm parameters.