AI/ML for Beam Management in 5G-Advanced: A Standardization Perspective
Qing Xue, Jiajia Guo, Binggui Zhou, Yongjun Xu, Zhidu Li, Shaodan Ma
TL;DR
Beam management (BM) in mmWave 5G NR faces high signaling overhead and latency due to exhaustive beam sweeping. The paper contrasts legacy non-AI BM with AI-enabled BM frameworks, focusing on SBP (spatial beam prediction) and TBP (temporal beam prediction) and detailing deployment options on either gNB or UE, with beam sets $\mathcal{A}$ (predict) and $\mathcal{B}$ (measure). It discusses standardization perspectives for 5G-Advanced (Release 18), highlighting scope items such as accuracy, overhead, latency, and issues around complexity, generalization, collaboration, and lifecycle management of AI-enabled BM. The findings aim to guide industrial adoption and the development of interoperable AI-native BM protocols, addressing data privacy, robustness, and real-time inference challenges to realize proactive and efficient beam management in future networks.
Abstract
In beamformed wireless cellular systems such as 5G New Radio (NR) networks, beam management (BM) is a crucial operation. In the second phase of 5G NR standardization, known as 5G-Advanced, which is being vigorously promoted, the key component is the use of artificial intelligence (AI) based on machine learning (ML) techniques. AI/ML for BM is selected as a representative use case. This article provides an overview of the AI/ML for BM in 5G-Advanced. The legacy non-AI and prime AI-enabled BM frameworks are first introduced and compared. Then, the main scope of AI/ML for BM is presented, including improving accuracy, reducing overhead and latency. Finally, the key challenges and open issues in the standardization of AI/ML for BM are discussed, especially the design of new protocols for AI-enabled BM. This article provides a guideline for the study of AI/ML-based BM standardization.