Federated IoT Traffic Classification for Low-Delay Edge Management

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing network traffic classification systems, particularly in IoT devices, face challenges in accurately distinguishing between IoT and non-IoT traffic, leading to incorrect prioritization and management of network traffic, especially when using rule-based systems and deep packet inspection.

Innovation Solution

Implementing federated machine learning techniques to train global machine learning models using parameters from multiple edge devices, allowing local classification of network traffic as IoT or non-IoT traffic, and managing traffic based on these classifications, including prioritization and selective compression.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If rule-based systems with deep packet inspection are used to classify network traffic, then classification can be performed at the gateway, but accuracy in distinguishing IoT traffic from non-IoT traffic deteriorates due to similar traffic patterns and non-standard port numbers

Engineering Contradiction:
Improvetraffic classification accuracyVSAvoidclassification system complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent replaces rule-based classification systems with machine learning models that automatically learn traffic patterns. Instead of manually configuring rules for port numbers and protocols, the ML models analyze traffic characteristics and automatically distinguish IoT from non-IoT traffic, improving accuracy while reducing the complexity of system configuration and maintenance.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system employs federated learning where each terminal device independently trains local ML models using its own traffic data. This self-service approach allows devices to adapt to their specific traffic patterns without requiring centralized configuration or complex gateway-based deep packet inspection, improving both accuracy and reducing system complexity.

Inventive Principle:
Principle #25Self-service

2Ease of operation

If centralized remote systems perform traffic classification, then system management is simplified, but network delays increase due to the need for constant communication with remote systems

Engineering Contradiction:
Improvesystem management easeVSAvoidnetwork delay
Core Design Contradiction:
Ease of operationVSLoss of time

Solution Approach 1:

The patent divides the centralized classification system into distributed local ML models at each terminal device. Each device independently performs classification locally, eliminating the need for constant communication with remote systems and reducing network delays. The segmentation maintains ease of operation through automated local decision-making while dramatically improving response time.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system pre-trains ML models at each terminal device using historical traffic data before actual traffic classification is needed. This preliminary action allows devices to perform accurate real-time classification without waiting for remote system instructions, reducing delays while maintaining simplified management through centralized model distribution.

Inventive Principle:
Principle #10Preliminary action

3Measurement precision

If local machine learning models are trained at each terminal device using federated learning, then traffic classification accuracy improves by leveraging local traffic characteristics, but device computational requirements and energy consumption increase

Engineering Contradiction:
Improvelocal traffic classification accuracyVSAvoiddevice energy consumption
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent implements partial training where devices perform only local model training using their own traffic data, rather than processing all possible features or running exhaustive training algorithms. This partial action achieves sufficient local accuracy for traffic classification while significantly reducing computational load and energy consumption compared to full centralized processing.

Inventive Principle:
Principle #16Partial or excessive action

Solution Approach 2:

The system uses a centralized server as an intermediary to coordinate the federated learning process, aggregate gradients from multiple devices, and distribute updated global models. This intermediary approach allows devices to perform lightweight local training with reduced energy consumption while still benefiting from improved accuracy through periodic model updates from the aggregated global knowledge.

Inventive Principle:
Principle #24Intermediary (Mediator)

4Measurement precision

If deep packet inspection is used to identify IoT traffic, then traffic classification can be performed, but processing time increases and network performance deteriorates

Engineering Contradiction:
ImproveIoT traffic identification accuracyVSAvoidnetwork processing throughput
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent extracts essential traffic classification features at the application layer without performing deep packet inspection of the actual data payloads. By taking out only the necessary metadata and traffic pattern information needed for classification, the system maintains accurate IoT traffic identification while avoiding the computational overhead and processing delays associated with inspecting packet contents.

Inventive Principle:
Principle #2Taking out (Extraction)

Data Source

PatentEP3846085B1Managing internet of things network traffic using federated machine leaning
Publication Date: 2025.07.02 HUGHES NETWORK SYST
  • EP3846085B1 patent drawingFigure 1
  • EP3846085B1 patent drawingFigure 2
  • EP3846085B1 patent drawingFigure 3

AI summary

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for training and using machine learning models to classify network traffic as IoT traffic or non-loT traffic and managing the traffic based on the classification. In some implementations, machine learning parameters of a local machine learning model trained by the edge device is received each of at least a subset of a set of edge devices. The machine learning parameters received from an edge device are parameters of the local machine learning model trained by the edge device based on local network traffic processed by the edge device and to classify the network traffic as Internet of Things (IoT) traffic or non-loT traffic. A global machine learning model is generated, using the machine learning parameters, to classify network traffic processed by edge devices as IoT traffic or non-loT traffic.