By Yali Yan (CT3 Chair), Zhenning Huang (China Mobile, WID Rapporteur), Xuefei Zhang (Huawei, WID Rapporteur)

First published December 2024, in Highlights Issue 09

As 5G network supports more and more telecommunication scenarios and considerably higher data exchange and processing, it is becoming increasingly urgent to optimize the service experience, as well as improve the network efficiency in an automated, real-time and flexible manner.

3GPP has specified a network automation architecture to leverage 5G information exposure and network data analytics. It enhances network performance, operational assurance and resource utilization and enables self-optimization and automated management, as shown in Figure 1.

The Network Data Analytics Function (NWDAF) specified from the 5G's initial Release (Rel-15), is designed to control the management of sophisticated network data, as specified in 3GPP TS 29.520 [1].

Upon reception of the analytics request from the consumer, the NWDAF collects network data, service data, management data and/or UE performance data from dedicated data sources, then processes and generates the reliable analytics information to assist the consumer to improve the user experience and optimize the network performance.

A consumer (e.g., the PCF or NSSF) may subscribe to the network slice load level analytics information to assist in deriving the policy/Quality of Service (QoS) decisions and/or performing the network slice selection process. The analytics information offered by the NWDAF can be either statistical events that happened in the past, or predictive information of the future.

In Release 16, 3GPP extended the data sources so that the NWDAF can collect different kinds of data from more sources, such as the 5GC Network Functions (NF), Application Functions (AF) and/or Operation Administration and Maintenance (OAM), then process the collected data to perform analytics to support more scenarios, for example, the Abnormal UE behavior analytics scenario which is used to detect ping-pong handovers and prohibit the communication of the hijacked UE in a timely manner.

In Release 17, the NWDAF was decomposed into two functionalities, i.e., the Model Training Logical Function (MTLF) and the Analytics Logical Function (AnLF). The MTLF trains Machine Learning models and exposes new training services, and the AnLF performs inference, derives analytics information and provides analytics services. The Data Collection Coordination Function (DCCF) as defined in 3GPP TS 29.574 [2], the Messaging Framework Adaptor Function (MFAF) as defined in 3GPP TS 29.576 [4], and the Analytics Data Repository Function (ADRF) as defined in 3GPP TS 29.575 [3] were specified as new functions into the network automation architecture to improve the efficiency of data collection, analytics exposure and data storage.

Release 18 covers additional functionality, e.g., the Packet Flow Description (PFD) Determination performs data analytics on existing PFD information and User-Plane traffic, and provides analytics in the form of new or updated PFDs to the analytics consumer.

Meanwhile, in order to ensure more accurate calculation, prediction, and decision, the NWDAF supports computing the accuracy of the ML models and analytics, Federated Learning, i.e., training a ML Model across multiple decentralized NWDAF instances without exchanging and sharing the local data set in each NWDAF, and the analytics and data exchange in the roaming scenario, etc.

The standardization timeline of the network automation architecture evolution and typical use cases definition from Release 15 all the way to Release 19 are illustrated in figure 2 with some use cases described in further detail below.

Use Case 1: Service Experience

The Service Experience case takes the Mean Opinion Score (MOS) for services, like the audiovisual streaming or the non-audiovisual service (e.g. the V2X and Web Browsing service). The NWDAF collects the the Quality of Experience (QoE) metrics and performance data, e.g., the average packet delay, average packet loss rate and throughput from the AF. Other input data includes the QoS flow level network data from the 5GC NFs and the radio-related data, e.g., RAN throughput for uplink and downlink from the OAM.

With the collected data, the NWDAF can offer diverse, multi-level analytical capabilities that support fine granularity experience prediction and analysis. For instance, the NWDAF can provide analytics for an individual UE or a group of UEs, different access types, multiple applications, as well as different network slices. The operator takes the analytics into account to measure the actual user’s service experience and to identify the opportunities for network optimization.

Use Case 2: NF load analytics

The NF Load analytics refers to analyzing the load status of 5G NFs to assist the consumer in optimizing the NF selection process. Once the target NFs are indicated by the consumer, the NWDAF will interact with the NRF to collect the targeted NF’s load status and the resource usage related data, e.g., the usage of virtual CPU, memory, and disk from the OAM. Further, the NWDAF can also retrieve the traffic usage reports from the UPF and the UE movement information from the AF. With this data, the NWDAF produces the NF load analytics result including the NF type, NF instance ID, NF status, NF resource usage, NF load, and NF peak load. The NF load analytics is very helpful for an operator for, e.g., capacity planning. Some NFs can use it to select less-loaded NFs, e.g., SMF selection by the AMF or UPF selection by the SMF to improve the network efficiency.

In summary, since the introduction of 5G, 3GPP has already specified various intelligent network and Network Automation use cases, mainly focusing on optimizing network resources utilization and improving user experience. Meanwhile, the service-oriented architecture design ensures highly efficient and fully secure interactions between the Network Functions within the mobile core network.

3GPP is actively and continuously enhancing 5G network intelligence in Release 19. In particular, TSG CT (mainly driven by the CT3 working group) continues to build elaborated protocol descriptions and efficient interface design to support more functionalities and use cases, such as AI-assisted positioning enhancement, vertical federated learning, QoS policy recommendation and network abnormal behavior mitigation and prevention.

For more about CT groups go to www.3gpp.org/3gpp-groups

References

[1]        3GPP TS 29.520: "5G System; Network Data Analytics Services; Stage 3".

[2]        3GPP TS 29.574: "5G System; Data Collection Coordination Services; Stage 3".

[3]        3GPP TS 29.575: "5G System; Analytics Data Repository Services; Stage 3".

[4]        3GPP TS 29.576: "5G System; Messaging Framework Adaptor Services; Stage 3".