ALLSTaR — Automated LLM-Driven Scheduler Generation and Testing for Intent-Based RAN
IEEE Transactions on Mobile Computing · 2026-11 · 25 (11) · 20461-20478
DOI: 10.1109/tmc.2026.3711510 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Mobile Computing, 2026-11. Verfasst von Maxime Elkael und 4 weiteren Autor:innen. Die Arbeit wurde bisher 4-mal zitiert (Crossref) und nennt 69 Literaturangaben. Gefördert von: National Telecommunications and Information Administration (NTIA)’s Public Wireless Supply Chain Innovation Fund, Army Research Laboratory Cooperative Agreement, U.S. NSF.
Abstract (Originalsprache)
The evolution toward open and programmable O-RAN and AI-RAN networks creates new opportunities for IBN to optimize RAN operations according to dynamic operator requirements. However, applying intents to the scheduler, which directly controls resource allocation and performance, remains challenging. Existing approaches largely rely on coarse-grained network slicing and cannot adapt policies to individual users, channel conditions, and traffic patterns. Although the literature provides many scheduling algorithms, their practical use is limited by heterogeneous implementations, insufficient evaluation in production environments, and the effort required to develop high-performance code. We present ALLSTaR (Automated LLM-driven Scheduler generation and Testing for intent-based RAN), a framework for automated intent-driven scheduler design, implementation, evaluation, and selection. ALLSTaR interprets natural-language intents, extracts scheduling algorithms from the research literature using Optical Character Regnition (OCR) and LLMs, generates functional scheduler code, and matches operator intents to suitable schedulers. The generated schedulers are deployed as O-RAN dApps, enabling on-the-fly execution and testing on a production-grade, multi-vendor, 5G-compliant testbed. Using this pipeline, we conduct the most extensive Over-the-Air (OTA) experimental comparison of schedulers to date, evaluating 18 algorithms synthesized from the academic literature. Their measured performance profiles drive our Intent Based Scheduling framework, which dynamically selects and deploys the scheduler that best satisfies the current intent. We demonstrate use cases beyond slicing-based optimization, including fine-grained adaptation to buffer status, physical-layer conditions, and heterogeneous traffic.
Förderung
- National Telecommunications and Information Administration (NTIA)’s Public Wireless Supply Chain Innovation Fund
- Army Research Laboratory Cooperative Agreement
- U.S. NSF
- Qualcomm, Inc.
Zitieren
APA
Elkael, M., Polese, M., Prasad, R., Maxenti, S., & Melodia, T. (2026). ALLSTaR — Automated LLM-Driven Scheduler Generation and Testing for Intent-Based RAN. IEEE Transactions on Mobile Computing, 25(11), 20461-20478. https://doi.org/10.1109/tmc.2026.3711510
Vancouver
Elkael M, Polese M, Prasad R, Maxenti S, Melodia T. ALLSTaR — Automated LLM-Driven Scheduler Generation and Testing for Intent-Based RAN. IEEE Transactions on Mobile Computing. 2026;25(11):20461-20478. doi:10.1109/tmc.2026.3711510
BibTeX
@article{Elkael2026,
author = {Elkael, Maxime and Polese, Michele and Prasad, Reshma and Maxenti, Stefano and Melodia, Tommaso},
title = {ALLSTaR — Automated LLM-Driven Scheduler Generation and Testing for Intent-Based RAN},
journal = {IEEE Transactions on Mobile Computing},
year = {2026},
volume = {25},
number = {11},
pages = {20461--20478},
doi = {10.1109/tmc.2026.3711510}
}
EUVIMED