Denoising Refinement Diffusion Models for Simultaneous Generation of Multi-scale Mobile Network Traffic (2511.17532v1)

Abstract: Multi-layer mobile network traffic generation is a key approach to capturing multi-scale network dynamics, supporting network planning, and promoting generative management of mobile data. Existing methods focus on generating network traffic with a single spatiotemporal resolution, making it difficult to achieve joint generation of multi-scale traffic. In this paper, we propose ZoomDiff, a diffusion-based multi-scale mobile traffic generation model. ZoomDiff maps the urban environmental context into network traffic with multiple spatiotemporal resolutions through custom-designed Denoising Refinement Diffusion Models (DRDM). DRDM employs a multi-stage noise-adding and denoising process, enabling different stages to generate traffic with distinct spatial and temporal resolutions. It aligns the progressive denoising process of diffusion models with hierarchical network layers, including BSs, cells, and grids with different granularities. Evaluations on real-world mobile traffic datasets demonstrate that ZoomDiff achieves a performance improvement of at least 18.4% over state-of-the-art baselines on generation tasks at multi-scale traffic. The efficiency and generalization ability are also demonstrated, which indicates that ZoomDiff holds strong potential for generative mobile data management. The code of ZoomDiff is available at https://anonymous.4open.science/r/ZoomDiff-105E/.