CellSyn: Large-Scale Multi-Dimensional Cellular Traffic Data Synthesis via Deep Generative Models

Abstract

Cellular networks empower smartphone users to access online services, and associated traffic usage data hold immense potential for applications such as network planning, traffic prediction, mobility modeling, and personalized recommendations. However, limited data accessibility has been a bottleneck for data-driven research in the network communication community. Previous research has explored data synthesis, but the generation of cellular traffic data with diverse characteristics, including user attributes, application usage traffic, and base station (BS) associations, still remains a challenge. In this paper, we propose CellSyn, a novel framework designed to generate large-scale multi-dimensional synthetic cellular traffic data. CellSyn consists of three modules: i) a user attribute generation module to model various attribute information; ii) a temporal traffic usage generation module to capture dynamic traffic variations; iii) a spatial BS sequence generation module to learn heterogeneous BS association patterns. Experiments conducted on two private large-scale cellular traffic datasets and a public dataset illustrate CellSyn’s superior fidelity and utility over several state-of-the-art baselines, and show a favorable empirical privacy-utility trade-off under distance-based privacy metrics. We have released our source code and synthetic dataset for community research https://github.com/LPines/synthetic-cellular-traffic-dataset