QoE-Aware Resource Allocation for AIGC Services in Low-Altitude Economy Networks

Abstract

In this paper, we investigate an unmanned aerial vehicle (UAV)-enabled low-altitude wireless network, where multiple UAVs provide diffusion-based image generation services to a large number of ground users. We first model the image generation performance, characterized by contrastive languageimage pre-training (CLIP) and Fréchet inception distance (FID), as a nonlinear function of the number of inference steps. Then, we formulate a fixed-pricing Stackelberg resource allocation problem, subject to constraints on UAV computation resources, transmit power, system bandwidth, and service energy consumption. To solve this problem, we exploit the discrete concavity of user utility to characterize inducible inference steps and the associated minimum GPU resource requirements, and develop a closed-form enabled dual decomposition algorithm that yields a feasible resource allocation with a computable optimality gap. Simulation results demonstrate that the proposed algorithm can successfully fulfill the image generation demands, while achieving high service profit.