Overview
This tutorial provides an overview of AI/ML advanced technologies in 5G Core (5GC) based on recent 3GPP specifications. The integration of AI/ML capabilities into 5G systems aims to enhance network automation, improve performance, and enable new AI/ML-based services.
AI/ML Model Management
The management of AI/ML models in 5G systems is facilitated through a set of key functional components and services. At the core of this framework is the Model Training Logical Function (MTLF), which is responsible for training ML models using network data and predefined algorithms. Complementing the MTLF is the Model Inference Logical Function (MILF), which utilizes the trained models to perform real-time inference on incoming data, enabling rapid decision-making and predictive analytics. These functions are supported by comprehensive ML model provisioning and information services, which manage the lifecycle of ML models from deployment to updates and retirement. This integrated approach ensures that 5G networks can dynamically adapt to changing conditions and requirements, leveraging the power of AI/ML to optimize performance and enhance service delivery across the network infrastructure.
Network Data Analytics Function (NWDAF)
The Network Data Analytics Function (NWDAF) serves as a cornerstone for AI/ML integration in 5G Core, offering a comprehensive suite of capabilities. It efficiently gathers data from diverse network functions (NFs) and Operations, Administration and Maintenance (OAM) systems, leveraging this information to conduct sophisticated data analytics using AI/ML techniques. The NWDAF then disseminates these valuable analytics and predictions to other NFs and OAM, enabling data-driven decision-making across the network. With the advent of Release 18, the NWDAF's functionality has been significantly enhanced. It now supports ML model training and inference directly within the function, allowing for more dynamic and adaptive analytics. Additionally, the ability to aggregate analytics from multiple NWDAFs has been introduced, providing a more holistic view of network performance. Furthermore, the new release facilitates the seamless transfer of analytics context and subscriptions between NWDAFs, ensuring continuity and efficiency in analytics services across the network.
AI/ML Management and Orchestration
The management and orchestration of AI/ML in 5G systems encompass several critical aspects that ensure the effective integration and operation of intelligent technologies within the network infrastructure. At the heart of this framework is the comprehensive AI/ML model lifecycle management, which oversees the entire process from initial training through deployment to ongoing inference operations. This lifecycle approach ensures that models remain current and effective in addressing evolving network challenges. Equally important is the coordination of AI/ML functions across various 5G system domains, facilitating seamless interaction and data exchange between different network components to maximize the benefits of AI/ML integration. Furthermore, the framework places a strong emphasis on trustworthiness management for AI/ML models, implementing rigorous validation, verification, and monitoring processes to ensure the reliability, security, and ethical operation of AI-driven decision-making within the 5G ecosystem. This holistic approach to AI/ML management and orchestration enables 5G networks to harness the full potential of artificial intelligence while maintaining the highest standards of performance and integrity.
Support for AI/ML-based Services
The integration of AI/ML within the 5G system is characterized by several key features that enhance its functionality and effectiveness. One of the primary features is the capability for AI/ML model and data distribution and sharing across the 5G network, which facilitates collaboration and resource optimization among various network functions. This is complemented by support for distributed and federated learning, allowing models to be trained collaboratively across multiple locations without the need to centralize sensitive data, thereby improving privacy and efficiency. Additionally, the system incorporates robust traffic characterization and establishes performance requirements specifically tailored for the transfer of AI/ML models. This ensures that model transfers are executed efficiently, meeting the necessary latency and reliability standards essential for real-time applications. Together, these features create a dynamic environment that maximizes the potential of AI/ML technologies within the 5G framework, enabling more intelligent and responsive network operations.
Conclusion
The integration of AI/ML technologies in 5GC represents a significant advancement in network intelligence and automation. These capabilities enable more efficient network operations, improved service quality, and support for innovative AI/ML-based applications. As 5G continues to evolve, we can expect further enhancements and refinements in the use of AI/ML within the core network.
Reference