Network Performance Probability Quantification via Grid-Based Confidence Intervals

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current Network Digital Twin solutions face challenges in accurately and reliably determining the confidence interval for Service Level Agreement (SLA) assurance probabilities, especially in scenarios with limited data samples and varying KPI densities across different locations.

Innovation Solution

The proposed method involves dividing a region of interest into dynamic grids based on KPI density, generating predicted service KPI values using Deep Neural Networks (DNNs) from radio KPI measurements, and computing confidence intervals using Bayes' theorem, incorporating weighting coefficients for measured and predicted values.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional methods are used to determine confidence intervals for SLA assurance probabilities, then the calculation can be performed with simple data, but the accuracy and reliability of the determination deteriorates especially in scenarios with limited data samples and varying KPI densities

Engineering Contradiction:
Improveaccuracy of confidence interval determinationVSAvoiddata samples required
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent divides the region of interest into multiple grids based on KPI density distribution. Each grid can be independently processed to determine confidence intervals, allowing the system to handle limited data samples in specific regions without compromising overall accuracy. This segmentation enables targeted analysis where data is available and appropriate confidence interval determination even with sparse data in particular areas.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent employs Deep Neural Networks to generate predicted service KPI values based on radio KPI measurements, transforming the type of data from direct service measurements to predicted values. This parameter transformation allows the system to work with limited measured data by inferring additional information through the DNN model, thereby improving confidence interval accuracy without requiring large quantities of direct service KPI samples.

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If the region is divided into dynamic grids based on KPI density, then the measurement precision for different locations is improved, but the device complexity increases

Engineering Contradiction:
Improvespatial precision of performance assessmentVSAvoidgrid division and processing complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The region is segmented into grids based on KPI density, which organizes the complex spatial data into manageable units. This segmentation simplifies the processing complexity by allowing independent analysis of each grid, making the overall system more tractable despite the increased structural complexity of dynamic grid division.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent uses Deep Neural Networks to create predicted versions of service KPI data based on radio KPI measurements. This copying approach generates synthetic service KPI values that mirror the distribution and characteristics of actual service measurements, reducing the need for complex direct measurements while maintaining spatial precision through the predicted values across different grids.

Inventive Principle:
Principle #26Copying

3Reliability

If predicted service KPI values are generated using Deep Neural Networks, then the reliability of SLA assurance probability is improved with limited data, but the computational resources and processing time increase

Engineering Contradiction:
Improvereliability of SLA assurance probabilityVSAvoidprocessing time for prediction and confidence interval calculation
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The Deep Neural Network model is pre-trained on historical data to generate predicted service KPI values. This preliminary training allows the model to quickly infer service KPIs from radio KPIs during actual network performance assessment without requiring time-consuming real-time training. The pre-trained model can rapidly process new data points and generate predictions, improving reliability while minimizing processing time during execution.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system transforms direct service KPI measurements into predicted values based on radio KPIs through the DNN model. This parameter transformation enables the system to work with readily available radio measurements rather than requiring extensive service KPI data, reducing the time needed for data collection and analysis while maintaining high reliability through the predictive model's accuracy.

Inventive Principle:
Principle #35Parameter changes

4Measurement precision

If weighting coefficients are applied to measured and predicted values in confidence interval calculation, then the accuracy of SLA compliance assessment is improved, but the complexity of the calculation method increases

Engineering Contradiction:
Improveaccuracy of SLA compliance assessmentVSAvoidcalculation method complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent introduces weighting coefficients as new parameters in the confidence interval calculation to balance the influence of measured and predicted values. By adjusting these weights, the system can optimize the accuracy of SLA compliance assessment. The weighting parameters allow flexible control over the contribution of different data sources, improving measurement precision while keeping the calculation framework relatively simple through parameter adjustment rather than complex structural changes.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP4568218A1Methods and apparatus for quantifying probability of expected network performance
Publication Date: 2025.06.11 NOKIA SOLUTIONS & NETWORKS OY
  • EP4568218A1 patent drawingFigure 1
  • EP4568218A1 patent drawingFigure 2
  • EP4568218A1 patent drawingFigure 3(a)~3(b)

AI summary

Described herein is a a method for evaluating performance of a network comprising a plurality of endpoint devices configured within a region of interest and at least one operations node, wherein the plurality of endpoint devices and the at least one operations node are configured to establish communication to each other for carrying out a service provided by an application installed on at least one of the plurality of endpoint devices, wherein the evaluation is based on determining a confidence interval for a probability of at least one performance indicator associated with the communication between the plurality of endpoint devices and the at least one operations node meeting a pre-determined target, wherein the at least one performance indicator comprises at least one signal-level indicator and/or at least one network-level indicator, wherein the method comprises: obtaining, based on communication-related data transmitted between the plurality of endpoint devices and the at least one operations node, measured values of the at least one performance indicator associated with one or more from the endpoint devices; dividing the region of interest into a plurality of grids based on a distribution of densities of said measured values of the at least one performance indicator within the region of interest, wherein the plurality of grids comprises one or more grids of interest each including at least one of said measured values; for a selected grid of interest, generating predicted values of the at least one performance indicator based on the at least one of said measured values; and for the selected grid of interest, determining said confidence interval based on said measured values and/or said predicted values and based on at least one target value of the at least one performance indicator set by the pre-determined target for the selected grid of interest.