NWDAF Model Accuracy Measurement via Cross-Dataset Validation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

In communications networks, varying accuracy of Network Data Analytics Functions (NWDAFs) due to different training datasets and algorithms leads to inaccurate model selection by consumers, resulting in suboptimal performance.

Innovation Solution

A method for data analytics entities to test the accuracy of their machine-learning models on different datasets through an accuracy measurement service, facilitated by a network function repository entity, allowing for model updates and improvements based on higher accuracy models.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If multiple NWDAF instances with different training datasets and algorithms are deployed to provide diverse analytics, then the versatility and coverage of network analytics are improved, but the accuracy and reliability of model selection deteriorate due to varying model performance

Engineering Contradiction:
Improveanalytics coverageVSAvoidmodel selection accuracy
Core Design Contradiction:
Adaptability or versatilityVSReliability

Solution Approach 1:

The patent implements a feedback mechanism where NWDAF instances provide accuracy information about their machine learning models to the NRF. Consumers can query this accuracy information before selecting a model, creating a feedback loop that enables informed decision-making. This resolves the contradiction by allowing diverse analytics coverage while maintaining reliable model selection through transparency of model performance metrics.

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The NRF acts as an intermediary between NWDAF instances and consumers. It collects, stores, and manages accuracy information from multiple NWDAF instances, then provides this information to consumers upon request. This intermediary role resolves the contradiction by centralizing accuracy information management, enabling consumers to make reliable model selections while preserving the diversity of analytics provided by multiple NWDAF instances.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Ease of operation

If consumers select NWDAF models based on reported accuracy without verification, then the ease of operation is improved, but the measurement precision of model accuracy deteriorates due to potential inaccuracies in self-reported metrics

Engineering Contradiction:
Improvemodel selectionVSAvoidaccuracy measurement
Core Design Contradiction:
Ease of operationVSMeasurement precision

Solution Approach 1:

The patent merges the accuracy measurement function into the NWDAF instance itself. Each NWDAF instance applies its own machine learning model to a dataset to measure accuracy, then reports this measured accuracy to the NRF. This combination of self-measurement and central registration resolves the contradiction by maintaining ease of operation through automated processes while improving measurement precision through actual performance testing rather than unverified claims.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

NWDAF instances perform self-service by automatically measuring their own model accuracy using available datasets and reporting this information to the NRF without external intervention. This self-service approach resolves the contradiction by maintaining ease of operation through automation while improving measurement precision through actual performance evaluation. Consumers benefit from readily available, verified accuracy information without complex verification processes.

Inventive Principle:
Principle #25Self-service

Data Source

PatentEP4140156B1Methods, apparatus and machine-readable media relating to data analytics in a communications network
Publication Date: 2025.06.25 TELEFONAKTIEBOLAGET LM ERICSSON (PUBL)
  • EP4140156B1 patent drawingFigure 1
  • EP4140156B1 patent drawingFigure 2
  • EP4140156B1 patent drawingFigure 3

AI summary

There is provided a method performed by a first data analytics entity for a communications network. The first data analytics entity has access to a first dataset of network data. The method comprises: receiving a request message from a second data analytics entity for the communications network, the request message comprising a model generated by the second data analytics entity using a machine-learning algorithm based on a second dataset to which the second data analytics entity has access, and an indication of an analytic to be calculated by the model; applying the model to the first dataset to measure the accuracy of the model; and transmitting a response message to the second data analytics entity comprising an indication of the accuracy of the model when applied to the first dataset.