Resource Allocation for Multi-cell Semantic Communication based on Deep Reinforcement Learning

Semantic communication is considered to be able to break through the limitations of Shannon's law and truly realizes the 6th generation mobile communication technology (6G). Most of the current semantic communication researches focus on signal detection, semantic extraction and recovery. The resource allocation in semantic communication system has not been completely considered. Meanwhile, with the development of communication system, the communication environment is increasingly complex, so that the high dynamic resource allocation problem is difficult to be solved by classical methods. Therefore, this paper formulates to solve multi-object, non-convex, dynamic resource allocation problems in multi-cell semantic communication systems through deep reinforcement learning. Based on Transformer model, we realize the transformation from bit information transmission rate to semantic information transmission rate. At the same time, we introduce semantic throughput (STP) as a metric of system semantic transmission rate. Moreover, we use deep reinforcement learning to solve sub-channel allocation and power allocation problems in dynamic systems. Compared with the traditional resource allocation method, that of semantic communication in this paper has larger STP under the same semantic confidence.