Matching Game Based Resource Allocation Scheme for Adaptive Semantic and Bit Communication Networks

In this paper, we propose an adaptive semantic and bit communication framework for image transmission, where each user is allowed to select either semantic communication (SemCom) or bit communication (BitCom) mode. As no closed-form expression for the peak mean square error (PMSE) that is a key parameter to characterize the difference between original and corresponding recovered images exists, we propose a data regression approach to approximate the PMSE by the Gompertz function. Due to the different characteristics of the two types of communication modes, the cost functions are designed for both SemCom and BitCom modes, which take the delay, energy consumption, and PMSE into consideration. Specifically, for the SemCom mode, we analyze the required computation resources for semantic feature extraction and recovery and the corresponding computation delay and energy consumption are integrated into the SemCom cost function design. Then, an overall user cost minimization problem is formulated to jointly optimize the communication mode decision, channel selection, and computation resource allocation. To solve the formulated problem, we propose a particle swarm optimization based many-to-one matching game with one-side cooperation (PSO-MMOC) algorithm, where a many-to-one matching game with one-side cooperation is established to determine the communication mode and channel selection and the particle swarm optimization algorithm is adopted to allocate the computation resource in each step of the matching game. We further prove the convergence and stability of the proposed PSO-MMOC algorithm. Simulation results are provided to demonstrate the superiority of the proposed scheme.