Problem

Existing literature does not specifically examine AI/ML capabilities introduced in 3GPP Release 19 from the Service and System Aspects (SA) group's perspective. Integrating Artificial Intelligence and Machine Learning (AI/ML) into mobile networks faces significant challenges, including managing hierarchical computing for tasks split between devices and edge/cloud servers, efficiently distributing and sharing complex ML models, effectively implementing distributed and federated learning without compromising latency or data rates, and ensuring the availability of sufficient high-quality data and processing power for AI-driven network optimizations.

Method

The authors address the identified knowledge gap by providing a comprehensive and detailed account of the AI/ML-related enhancements introduced in 3GPP Release 19, specifically focusing on the contributions from the SA group. They systematically categorize these advancements into two overarching paradigms: "AI for Network," which involves leveraging AI to optimize traditional network functions and operations, and "Network for AI," which focuses on enhancing network capabilities to efficiently support various AI/ML applications. Furthermore, the paper outlines the approved study items for Release 20, offering insights into the future direction of AI/ML integration within 5G Advanced and 6G systems.

Results

Key Release 19 advancements include improved support for ML model transfer, distribution, and Federated Learning (FL), with defined system requirements for data rates in various scenarios and the introduction of direct Device-to-Device (D2D) connectivity to facilitate distributed inference and learning. Core Network functionalities were enhanced with AI/ML for more accurate location services, Vertical Federated Learning (VFL) support via the Network Data Analytics Function (NWDAF) for privacy-preserving model training, AI/ML-assisted policy control and Quality of Service (QoS) management to simplify and optimize parameter adjustments, and mechanisms for predicting and mitigating abnormal signaling storms. Furthermore, AI/ML was integrated to optimize media processing applications through model split scenarios, compression techniques, and new logical functions. The application layer saw enhancements with an AI/ML enablement layer and ML repository, while the Operations, Administration, and Maintenance (OAM) system extended its Management Data Analytics (MDA) framework to support advanced learning techniques such as transfer learning and reinforcement learning for efficient network management.

Implications

The integration of both "AI for Network" and "Network for AI" paradigms, while resource-intensive and requiring fundamental changes in network architecture, is deemed crucial for the evolution of future mobile networks given the widespread utility and acceptance of AI/ML. Ongoing standardization efforts in 3GPP Release 20 and future 6G systems indicate a strong commitment to further integrating AI/ML. This future work aims to enhance user-plane performance, improve network sustainability by evaluating AI/ML energy consumption, streamline ML model registration and discovery, and ensure robust application layer performance monitoring and service continuity across multi-operator domains.