Energy-Delay Tradeoff for Multimodal Task Offloading in Semantic Communication-Enabled Low-Altitude MEC Networks Under Jamming

Abstract

Low-altitude uncrewed aerial vehicles (UAVs) equipped with mobile edge computing (MEC) capabilities can provide flexible computation services for latency-sensitive urban applications. However, the offloading of heterogeneous semantic tasks over jammed air-to-ground links requires the communication, computation, and mobility decisions to be coordinated. This paper studies a semantic communication (SemCom)-assisted MEC system in which ground devices generate multimodal tasks and a UAV acts as an aerial edge server in the presence of a malicious jammer. A unified model is established to capture multimodal computing demands, semantic transmission, and UAV movement. We then construct an energy-delay objective and optimize the binary offloading decisions, computing resource allocation, and UAV three-dimensional (3D) flight trajectory subject to semantic accuracy and mobility constraints. The resulting mixed-integer nonconvex formulation is addressed using an Exact Penalty-based Alternating and Soft Penalty-based Proximal Policy Optimization (EPA-SP3O) algorithm, where the EP component handles the discrete binary offloading indicator, whereas the learning component determines the continuous computing resources and 3D UAV trajectory. Numerical results show that EPA-SP3O achieves a more favorable balance between task completion time and UAV energy consumption than the considered benchmark schemes. The results also identify an altitude-dependent tradeoff between the elevation angle and the distance, providing insights into 3D trajectory design for semantic task offloading under jamming.