Federated Learning for Edge Network Performance Prediction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing approaches for predicting Key Performance Indicators (KPIs) for base stations require large amounts of data to be centralized, which can be impractical due to bandwidth limitations and data geo-fencing regulations. Additionally, current federated learning methods assume random distributions of data across sites, which is not always the case.

Innovation Solution

A disciplined approach to federated learning is implemented, where base stations train locally on their own datasets and share only model weights with a centralized node. This approach allows for accurate prediction of KPI degradations without transferring raw data, reducing latency, improving security, and minimizing data transfer costs.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If data is centralized in a cloud repository for training machine learning models, then model prediction accuracy is improved, but data transfer bandwidth requirements increase and data security is compromised

Engineering Contradiction:
Improvemodel prediction accuracyVSAvoiddata transfer volume
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent extracts only the essential model parameters (weights and biases) from the training process and transfers them to edge devices, rather than transferring entire datasets. This extraction principle reduces data transfer volume while maintaining model accuracy, as edge devices can locally train models using transferred parameters and their own local data.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent introduces a two-stage training process where a centralized server first trains a model and transfers it to edge devices, which then perform local fine-tuning. This intermediary approach allows the system to benefit from both centralized data aggregation and local data privacy, resolving the contradiction between accuracy and data transfer requirements.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Reliability

If raw data is transferred to the cloud for model training, then comprehensive model learning is achieved, but data privacy and security are compromised

Engineering Contradiction:
Improvemodel learning effectivenessVSAvoiddata privacy risk
Core Design Contradiction:
ReliabilityVSObject-affected harmful factors

Solution Approach 1:

The patent extracts only model parameters (weights and biases) from the training process and transfers them to edge devices, rather than transferring entire datasets. This extraction principle reduces data transfer volume while maintaining model accuracy, as edge devices can locally train models using transferred parameters and their own local data.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent enables edge devices to perform self-service model training by transferring trained models from a centralized server to local devices, where they can independently fine-tune models using local data. This self-service capability maintains data privacy while achieving effective model learning, as no raw data leaves the edge devices.

Inventive Principle:
Principle #25Self-service

3Quantity of substance

If federated learning is used to train models locally at edge devices, then data transfer is reduced, but model accuracy deteriorates due to non-random data distributions

Engineering Contradiction:
Improvedata transfer volumeVSAvoidmodel prediction accuracy
Core Design Contradiction:
Quantity of substanceVSMeasurement precision

Solution Approach 1:

The patent applies preliminary action by first training a model centrally with comprehensive data, then transferring this pre-trained model to edge devices before local fine-tuning. This preliminary training ensures that edge devices start with a robust baseline model, compensating for the limitations of non-random local data distributions and maintaining high accuracy.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent introduces a two-stage training process where a centralized server first trains a model and transfers it to edge devices, which then perform local fine-tuning. This intermediary approach allows the system to benefit from both centralized data aggregation and local data privacy, resolving the contradiction between accuracy and data transfer requirements.

Inventive Principle:
Principle #24Intermediary (Mediator)

4Stability of the object's composition

If models are trained centrally and deployed to edge devices, then model consistency is improved, but actuation latency increases

Engineering Contradiction:
Improvemodel consistencyVSAvoidactuation latency
Core Design Contradiction:
Stability of the object's compositionVSLoss of time

Solution Approach 1:

The patent segments the model training process into centralized pre-training and local fine-tuning stages. This segmentation allows the system to maintain model consistency through centralized training while reducing actuation latency through local deployment and inference, as edge devices can immediately apply trained models to local KPI predictions without cloud round-trips.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies preliminary action by first training a model centrally with comprehensive data, then transferring this pre-trained model to edge devices before local fine-tuning. This preliminary training ensures that edge devices start with a robust baseline model, compensating for the limitations of non-random local data distributions and maintaining high accuracy.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP3891933B1Predicting network communication performance using federated learning
Publication Date: 2025.05.28 TELEFONAKTIEBOLAGET LM ERICSSON (PUBL)
  • EP3891933B1 patent drawingFigure 1
  • EP3891933B1 patent drawingFigure 2
  • EP3891933B1 patent drawingFigure 3

AI summary

Federated learning is used to predict network communication performance at an edge node (10-1). The edge node (10-1) trains a local model (14-1) of network communication performance over one or more rounds of training, based on a local training dataset (22-1) and based on multi-node training information (20-1) received in each round of training. This information (20-1) comprises information about local models (14-2,...14-N) at other respective edge nodes (10-2,...10-N) as trained based on local training datasets at the other edge nodes (10-2,...10-N). After or as part of each round of training, the edge node (10-1) transmits control signaling that indicates an accuracy of the local model (14-1) as trained through that round of training, that indicates whether another round of training is needed or desired, and/or that indicates whether any further information (20-1) is needed or desired. The edge node (10-1) predicts network communication performance at the edge node (10-1) based on the trained local model (14-1).