Skip to main content Skip to main navigation

Publication

AucRocket: High-precision Classification of Heterogeneous Time Series for Sustainable 6G Networks

Matthias Rüb; Michael Gundall; Oruc Kahriman; Niels Wessel; Hans Dieter Schotten
In: Sustainability and Deployability of AI Solutions toward AI-Native 6G Networks. IEEE Global Communications Conference Workshops (Globeccom Wkshps-2026), IEEE, 2026.

Abstract

Sustainability is a core objective for 6G, as telecom already accounts for a notable share of global energy use, with energy costs increasingly dominating network OPEX. 6G is also envisioned as AI-native, with autonomous agents perceiving, reasoning, and acting on network data to improve efficiency and performance. However, state-of-the-art time series classification methods rely on computationally expensive ensemble solutions which can limit their suitability for sustainable AI deployment in future wireless networks. We propose AucRocket, a lightweight time series classification approach that combines random con- volutional features with stretch-distribution-based pooling. On network telemetry collected from an Irish mobile network oper- ator, AucRocket outperforms state-of-the-art methods for service inference from downlink bitrate, demonstrating its applicability to network functionalities envisioned in 6G such as service- aware resource allocation and QoS management. Furthermore, an extensive evaluation on 112 heterogeneous datasets from the UCR archive shows that AucRocket significantly outperforms existing kernel-based time series classification approaches, including MultiRocket.

Projects