Cross-layer joint optimization for semantic communication-driven MEC systems via deep reinforcement learning

Abstract

The integration of semantic communication (SemCom) with mobile edge computing (MEC) has opened new avenues to improve task execution efficiency in intelligent networks. This paper proposes a cross-layer joint optimization framework for SemCom-driven MEC systems, aiming to minimize the weighted sum of task completion time and user energy consumption. Specifically, the framework jointly optimizes the semantic extraction factor at the application layer, task offloading decisions at the control layer, and communication and computational resource allocation at the network and physical layers. To address the non-convex and mixed-integer nature of the problem, a Deep Deterministic Policy Gradient (DDPG)-based algorithm was employed to efficiently search for solutions. The simulation results validate the effectiveness of the proposed approach and demonstrate that the integration of SemCom into MEC significantly improves the system performance. The findings offer practical insights for system engineers to design efficient MEC systems, reducing transmission overhead and energy consumption, especially in latency-sensitive applications such as autonomous driving and industrial Internet of Things.