NWDAF: The Central Analytics System for 5G Networks

As telecommunication networks evolve, the 5G Core (5GC) architecture has introduced a fundamental need for automation and intelligent management. Simply put, networks have become too complex to be managed reactively. The key element enabling this transformation is the NWDAF (Network Data Analytics Function) – a dedicated data analytics function.

NWDAF acts as the central analytical “brain” of a 5G network. Its task is to collect data from the entire infrastructure, process it and deliver insights in the form of statistics or predictions. It is precisely these analytics that enable the proactive optimization of network operations and services. How are companies using NWDAF to enhance data analytics capabilities? Read on to learn more.

Internal architecture: how NWDAF works

NWDAF is not a monolith. According to the 3GPP standard, its functionality is logically divided into several key components that work together:

• AnLF (Analytics Logical Function): The main analytics engine. It’s responsible for inference, which involves using pre-built models to analyze incoming real-time data and generate predictions for other network functions.
• MTLF (Model Training Logical Function): The “model factory.” This function handles the training of machine learning (ML) models based on collected historical data. The resulting trained models are then deployed to the AnLF.

Additionally, the following network functions are not strictly components of NWDAF. However, they collaborate with the above functions to ensure optimal resource utilization in the network.

• DCCF (Data Collection Coordination Function): An intelligent data collection coordinator. It acts as a broker to optimize the process of retrieving information from the network. If multiple AnLF instances need the same data (e.g., statistics from a single AMF), the DCCF collects it only once and distributes it to all interested parties, which prevents excessive network load.
• ADRF (Analytics Data Repository Function): The analytics archive. This is a dedicated database that stores two key things: historical data (aggregated and cleaned by NWDAF) and trained ML models (created by the MTLF). It serves as the data source for the MTLF during the training process and as a model library for the AnLF.

Data collection from the 5G ecosystem

To provide valuable analytics, NWDAF must integrate data from the vast network ecosystem. This data is collected from multiple sources:

• Network Functions (NFs): This is the primary source of information about network status and user sessions. NWDAF obtains data from functions like AMF (Access and Mobility Management Function), SMF (Session Management Function), PCF (Policy Control Function), UDM (Unified Data Management) and AF (Application Function).
• OAM (Operations, Administration and Management) Systems: Global operational data, performance statistics for individual network elements and device tracking data (MDT) are gathered from here.
• Data Repositories (ADRF/UDR): NWDAF can retrieve information from dedicated repositories, such as historical data or ML models (from ADRF) and subscription data (from UDR via UDM).

To optimize this process, especially when collecting large amounts of data (so-called “bulked data”), NWDAF collaborates with coordinating functions like the DCCF (Data Collection Coordination Function).

Exposing analytics (service exposure)

NWDAF does not collect data for its own sake. Its purpose is to deliver insights to other network functions (consumers) so they can make intelligent decisions. This is done within the Service Based Architecture (SBA) using defined (Nnwdaf) service interfaces.

The two main mechanisms are:

• Subscription (Nnwdaf_AnalyticsSubscription): Network functions (e.g., PCF, NSSF, AMF) “subscribe” to notifications. NWDAF proactively informs them when a specific analytical event occurs (e.g., “predicted load for slice X will exceed 80%”).
• Request (Nnwdaf_AnalyticsInfo): An NF can also “query” NWDAF for specific information one time (a request/response model).

Additionally, NWDAF offers specialized services for managing the ML model lifecycle, including their distribution (Nnwdaf_MLModelProvision), which is crucial for Federated Learning (FL), among other things.

Analytical capabilities: what exactly does NWDAF analyze?

The range of analytics generated by NWDAF is broad and serves various optimization purposes:

1. Network load analytics

NWDAF provides statistics and forecasts regarding load levels for:

• Network Slices (Slice/NSI Load): Analysis of the number of UE registrations or active PDU sessions for a specific network slice (S-NSSAI). This is crucial for the NSSF (for slice selection) and PCF (for policy) to manage resources.
• Network Functions (NF Load): Information on the current and predicted load of other NFs (e.g., AMF, SMF), which allows for intelligent load balancing.

2. UE Behavior Analytics

Understanding how end-user devices behave:

• UE Mobility: Predicting the location and movement trajectory of users. This is used to optimize mobility management (e.g., handovers).
• UE Communication: Analysis of communication patterns (e.g., traffic volume, frequency) helps optimize inactivity timers or routing.
• Abnormal Behavior: Identifying UEs or groups of UEs exhibiting anomalies, such as unusual mobility, generating excessive signaling traffic, or suspected of participating in a DDoS attack.

3. QoS and Performance Analytics

Monitoring the real-world quality of service:

• Observed Service Experience (OSE): NWDAF provides averaged user experience metrics (e.g., an estimated MoS score for a voice call), rather than relying solely on raw network parameters.
• Network Performance: Analysis of statistics and predictions concerning, for example, gNB (base station) resource consumption, the success rate of PDU session establishments, or successful handover rates.
• Dispersion Analytics: Analysis of the dispersion (spread) of data volume and transactions generated by users in a specific location or network slice.

Enhancing 5G network management with an experienced partner

NWDAF, thanks to its broad integration capabilities and advanced analytics, is transforming 5G network management from a reactive model (fixing problems) to a proactive and predictive one (preventing problems). It is becoming the central mechanism enabling intelligent, data-driven decisions to maintain and optimize the complex telecommunications infrastructure.

Stiff competition, dynamic market standards and ever-increasing customer expectations mean operators need to integrate evolving solutions, like 5G, that differentiate offers and attract users. That’s why, for over 25 years, global telecommunications providers have been turning to Software Mind. Want to know how our experts can support your 5G strategies? Contact us by filling out this form.

FAQ

What is NWDAF?

NWDAF stands for Network Data Analytics Function, which is a key part of 5G core networks as it gathers and analyzes network data. As such, it provides valuable insights and facilitates intelligent network management and data-driven strategies.

How does NWDAF work?

NWDAF has a modular architecture and four key functionalities. First, there is AnLF (Analytics Logical Function), which is the main analytics engine, responsible for inference. Next, there is MTLF (Model Training Logical Function), which handles the training of machine learning (ML) models based on collected historical data. The resulting trained models are then deployed to the AnLF. Beyond NWDAF architecture, there is also DCCF (Data Collection Coordination Function), which is an intelligent data collection coordinator that acts as a broker to optimize the process of retrieving information from the network. The final component is ADRF (Analytics Data Repository Function), which, though also separate from NWDAF architecture plays an important role, as its dedicated database stores historical data (aggregated and cleaned by NWDAF) and trains ML models (created by the MTLF). It serves as the data source for the MTLF during the training process and as a model library for the AnLF.

How does NWDAF provide analytics?

NWDAF integrates data from vast network ecosystems and multiple sources, including

Network Functions (NFs): AMF (Access and Mobility Management Function), SMF (Session Management Function), PCF (Policy Control Function), UDM (Unified Data Management) and AF (Application Function). OAM (Operations, Administration and Management) Systems: Global operational data, performance statistics for individual network elements and device tracking data (MDT) and Data Repositories (ADRF/UDR).

Does NWDAF operate within the Service Based Architecture (SBA)?

Yes – using defined service interfaces. The two main mechanisms are Subscription (Nnwdaf_AnalyticsSubscription): Network functions (e.g., PCF, NSSF, AMF) “subscribe” to notifications. NWDAF proactively informs them when a specific analytical event occurs (e.g., “predicted load for slice X will exceed 80%”) and Request (Nnwdaf_AnalyticsInfo): An NF can also “query” NWDAF for specific information one time (a request/response model). NWDAF also offers specialized services for managing the ML model lifecycle, including their distribution (Nnwdaf_MLModelProvision), which is crucial for Federated Learning (FL), among other things.

What does NWDAF analyze?

The range of analytics generated by NWDAF provides a wide range of analytics, including Network load analytics, UE Behavior Analytics and QoS and Performance Analytics.