Deep Reinforcement Learning-based Resource Allocation and Mode Selection for Semantic Communication

In this paper, we aim to solve the joint resource allocation and mode selection problem, in which an agent adaptively allocates communication users to appropriate resource units and toggles between bit and semantic transmission modes while determining the count of transmitted semantic symbols in semantic communication mode. Specifically, in contrast to the common yet unrealistic assumptions of prior research, which posits the possibility of limitless data transmission over infinite periods, our focus shifts towards the realities of unsaturated traffic conditions, where users transmit a finite amount of data within restricted time frames. In order to evaluate the efficiency of data transmission within the semantic domain under unsaturated traffic conditions, we propose a short-term semantic transmission rate (SR), as an evaluation metric of the joint problem. Under these unsaturated traffic scenarios, the challenge emerges from the need to address a combinatorial issue, optimizing resource allocation, transmission mode selection, and symbol lengths simultaneously across the time-frequency axis. This task is compounded by the high degree of complexity and a significant number of unknown variables, making it a formidable challenge for conventional optimization techniques to solve effectively. In response, we propose a deep reinforcement learning-based method that in each time step allocates users to each resource units, determines the communication transmission mode, and selects data size according to communication environment and users' packet states. Extensive experiments demonstrate superior performance over conventional schemes in terms of semantic transmission performance.