Description
The 3GPP AI/ML framework establishes standardized mechanisms for incorporating artificial intelligence and machine learning into mobile network operations. The architecture follows a distributed approach with AI/ML functions deployed at various network locations: near-real-time functions at the RAN Intelligent Controller (RIC) for radio optimization, non-real-time functions at the Service Management and Orchestration (SMO) for network-wide optimization, and core network functions for service intelligence. The framework defines standardized interfaces for data collection, model training, inference execution, and result distribution across network elements.
Key components include the AI/ML pipeline management system, which handles the complete lifecycle of ML models from training to deployment and monitoring. The NWDAF (Network Data Analytics Function) in the 5G core serves as a centralized analytics engine that can host ML models for network and service analytics. The RIC architecture supports xApps and rApps that implement ML algorithms for RAN optimization, with standardized interfaces (A1, E2) for data exchange and control. The framework also specifies data collection mechanisms, including standardized data sets, collection frequencies, and data formats to ensure interoperability between different vendors' AI/ML solutions.
The technical implementation involves several standardized procedures: data collection and preparation using defined data models, model training either centrally or distributed, model deployment to inference points, and continuous model monitoring and retraining. The framework supports various ML paradigms including supervised learning, reinforcement learning, and federated learning. For RAN optimization, ML models can predict traffic patterns, optimize beamforming, manage handovers, and allocate resources dynamically. In the core network, ML enables predictive QoS management, anomaly detection, and service experience optimization. The management system includes mechanisms for model versioning, performance monitoring, and fallback procedures to ensure network stability when ML models underperform.
Security aspects are integral to the design, with mechanisms for model integrity verification, data privacy protection, and secure model distribution. The framework addresses the computational requirements by defining capabilities for edge computing integration and distributed inference. Performance monitoring includes both traditional KPIs and ML-specific metrics like model accuracy, inference latency, and training convergence. The standardization ensures that AI/ML capabilities can be implemented consistently across multi-vendor networks while allowing innovation through open interfaces for custom ML applications.
Purpose & Motivation
AI/ML integration addresses the growing complexity of 5G and future 6G networks, which traditional rule-based optimization cannot manage effectively. As networks support diverse services with stringent requirements (ultra-low latency, ultra-high reliability, massive IoT), manual configuration and static optimization become impractical. The explosion of network data from connected devices, applications, and network elements creates opportunities for data-driven optimization that previous network generations couldn't fully exploit.
Historically, network optimization relied on expert knowledge, predefined rules, and periodic manual adjustments. This approach couldn't adapt quickly to changing conditions or discover complex patterns in network behavior. The limitations became particularly evident with 5G's introduction of network slicing, where each slice requires different optimization objectives that may conflict. Traditional methods also struggled with the scale of massive MIMO configurations, where beam management involves thousands of parameters that interact in complex ways.
The standardized AI/ML framework enables networks to become self-optimizing, reducing operational expenses while improving performance. It addresses specific challenges like energy efficiency optimization (reducing base station power consumption based on traffic predictions), mobility robustness (predicting and preventing handover failures), and load balancing (distributing traffic optimally across cells). By making AI/ML capabilities part of the standard, 3GPP ensures interoperability between different vendors' solutions and creates a foundation for network intelligence that will be essential for 6G's vision of truly autonomous networks.
Classification
Detected Changes Across Releases
from 3GPP Change RequestsSpecific changes extracted from the „Change history“ tables of 3GPP specifications (230 CRs across 5 releases). Complements the general historical overview above with the evidence-based evolution of this function.
In Release 15, the primary introduction for the AI/ML function was the definition of the Network Data Analytics Function (NWDAF) within the 5G system architecture. This was specified by adding the NWDAF to the core network description in the 23.501 specification. The release also included subsequent corrections to ensure the proper definition and resource modeling of this new analytics function.
In Release 16, the AI/ML function, embodied by the NWDAF (Network Data Analytics Function), saw significant enhancements in its operational scope and integration. The release introduced new analytics consumers like the UDM and OAM, expanded the types of analytics provided—such as UE mobility and abnormal behaviour analytics—and refined procedures for NWDAF discovery, selection, and service subscription. Furthermore, it clarified the inputs and outputs for analytics services and solidified the NWDAF's role in collecting data from sources like MDT/SON parameters to support network automation.
- Use of NWDAF analytics for decision of MICO mode parameters TS 23.501CR0837
- NEF service for NWDAF analytics TS 23.501CR0964
- CR for TS 23.501 Clarifications NWDAF Discovery and Selection TS 23.501CR0987
- CR for TS 23.501 Clarifications NWDAF Discovery and Selection TS 23.501CR1258
- Update the Analytics information provided by NWDAF TS 23.288CR0015
- Optionality of data to be collected by NWDAF TS 23.288CR0040
+ 21 more changes
In Release 17, the AI/ML function, centered on the NWDAF, introduced significant architectural enhancements for multi-instance deployments. Key new capabilities included procedures for the discovery, selection, reselection, and time coordination across multiple NWDAF instances, as well as mechanisms for ML model sharing and the aggregation of analytics from several NWDAFs. Furthermore, the release expanded NWDAF's data scope to support DN performance analytics and refined its integration by defining registration with the UDM and enhancements to event exposure procedures.
- NWDAF decomposition TS 23.288CR0205
- NWDAF - Data repository function TS 23.288CR0206
- Procedure for Multiple NWDAF Analytics aggregation TS 23.288CR0207
- Procedure for time coordination across multiple NWDAF instances TS 23.288CR0208
- Implementation of Enhancements on Event Exposure used by NWDAF in TS23.288 TS 23.288CR0216
- ML Model sharing between NWDAF instances TS 23.288CR0218
+ 66 more changes
In Release 18, 3GPP introduced several key enhancements for AI/ML, primarily focused on the NWDAF (Network Data Analytics Function). The release specifically added support for Federated Learning among multiple NWDAFs and introduced capabilities for AI/ML model information exchange and accuracy monitoring within the 5G Core. Furthermore, it expanded NWDAF's role in areas like application detection, resource monitoring for AI/ML-based services, and providing finer granularity of location information for service experience analytics.
- Alignment for UPF event exposure service to NWDAF via the SMF in TS 23.288 TS 23.288CR0551
- TS 23.288 Enhancement to Support AI/ML Data Transfer TS 23.288CR0554
- Enhancement on OSE for NWDAF assisting PCF in making URSP decisions TS 23.288CR0569
- NWDAF updates to assist resource monitoring of AI/ML-based services TS 23.288CR0575
- Federated Learning among Multiple NWDAFs in TS 23.288 TS 23.288CR0582
- NWDAF-assisted application detection in TS 23.288 TS 23.288CR0584
+ 92 more changes
In Release 19, the AI/ML function introduced comprehensive support for Vertical Federated Learning (VFL), including new general procedures for both training and inference between the NWDAF and AF, with the NWDAF acting as the VFL server. It also enhanced AI/ML-based positioning by enabling the LMF to retrieve ML models from the NWDAF and allowing the NWDAF to collect data from the LCS to train these models. Furthermore, the release added support for AI/ML model performance monitoring by the NWDAF and refined discovery procedures for VFL among network functions.
- General inference procedure for vertical federated learning TS 23.288CR1126
- Support for LMF to retrieve ML Model of AI/ML based positioning TS 23.288CR1164
- Registration and Discovery procedure for Vertical Federated Learning among NWDAF(s) and/or AF(s) with NWDAF as the VFL server TS 23.288CR1171
- Update the general inference procedure for vertical federated learning to resolve ENs TS 23.288CR1208
- General training procedure for Vertical Federated Learning between NWDAF(s) and AF(s) TS 23.288CR1134
- High-level description for Vertical Federated Learning when AF is as Server. TS 23.288CR1161
+ 25 more changes
Explore further
Broader topics and technologies where AI/ML plays a role.
Defining Specifications
3GPP specifications that define or reference AI/ML, with the latest known release. Sourced from the 3GPP document catalog — see methodology.
| Specification | Title | Release |
|---|---|---|
| TR 21.905 vj00 | 3GPP Technical Terms and Definitions | Rel-19 |
| TS 23.288 vk00 | 5GS Architecture Enhancements for Data Analytics | Rel-20 |
| TS 23.501 vk00 | 5G System Architecture Stage 2 | Rel-20 |
| TS 29.122 vj40 | T8 Reference Point for Northbound APIs | Rel-19 |
| TS 29.520 vj40 | 5G Network Data Analytics Services Stage 3 | Rel-19 |
| TS 29.530 vj00 | AF AI/ML Services Stage 3 Protocol | Rel-19 |
| TS 32.254 vj21 | Charging for Northbound APIs | Rel-19 |
| TS 33.501 vk00 | 5G Security Architecture and Procedures | Rel-20 |
| TS 38.300 vj00 | NG-RAN Overall Description | Rel-19 |
| TS 38.306 vj00 | NR UE Radio Access Capability Parameters | Rel-19 |