DiWi: A Transformer-Based Digital Twin for Wireless Mobility

Understanding device mobility in wireless networks is essential for multiple purposes: optimizing space usage, managing intelligent buildings, or improving network efficiency. However, the use of real mobility data raises significant privacy concerns. In this work, we propose DiWi (a Digital twin for Wireless mobility), a Transformer-based model to generate spatiotemporal mobility traces that mimic real-life behavior while preserving user privacy. We validate the utility of DiWi by comparing real and synthetic datasets, and by demonstrating its usefulness in a series of use cases related to mobility management, resource consumption, and privacy enhancements. We also confirm that DiWi is secure by evaluating empirical privacy metrics, such as, direct leakage, similarity searches, or membership inference. Our results illustrate that DiWi serves to generate realistic and useful mobility patterns without exposing identifiable user traces, making it a valuable tool for privacy-preserving mobility analysis. Furthermore, we investigate how enforcing differential privacy affects the generative performance of the model.

DiWi: A Transformer-Based Digital Twin for Wireless Mobility Articles