Application Layer Agents for Mobile Network Congestion Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Mobile network operators face challenges in detecting network congestion and coverage issues across large numbers of cellular towers, as traditional methods are inefficient and limited, lacking effective means to monitor user experience and data usage at the application level, and failing to associate network performance degradation with specific data usage patterns.
Innovation Solution
A system that collects application data from users to monitor network performance and user experience at both front-end and back-end levels, using machine learning models to detect network congestions and coverage issues by analyzing metrics such as network speeds, latency, and traffic volume, and sends alerts to operators for optimization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional monitoring methods are used, then device complexity is reduced, but measurement precision of network performance degradation is insufficient
Solution Approach 1:
The patent introduces an intermediary system comprising application layer agents, data collection modules, and machine learning models that mediate between raw network data and congestion detection. This intermediary layer enables precise measurement of network performance degradation by collecting application-level metrics (video buffer levels, audio playout delays, HTTP transaction times) and processing them through trained models, thereby achieving high detection accuracy without requiring direct modification of network infrastructure.
Solution Approach 2:
The patent replaces traditional mechanical/network-layer monitoring methods with application-layer software-based monitoring. Instead of using network probes or hardware sensors at the infrastructure level, the system uses software agents embedded in applications to collect performance metrics, substituting physical monitoring mechanisms with software-based measurement and analysis.
2Loss of information
If application level monitoring is implemented, then loss of information about user experience is reduced, but ease of operation becomes more difficult
Solution Approach 1:
The patent implements self-service monitoring where applications automatically instrument themselves with data collection agents. The system collects user experience information autonomously through application-layer agents that gather metrics such as video buffer levels, audio playout delays, and HTTP transaction times without requiring manual configuration or user intervention. This automated self-monitoring preserves comprehensive user experience data while maintaining operational simplicity.
Solution Approach 2:
The patent establishes feedback loops where collected application-level metrics are continuously analyzed by machine learning models to detect congestion conditions. The system processes user experience data in real-time, compares it against learned patterns, and provides feedback for congestion detection and network optimization, thereby preserving information about actual user experience while automating the operational complexity.
3Productivity
If machine learning models are used for congestion detection, then productivity of network optimization is improved, but loss of time for model training and deployment is incurred
Solution Approach 1:
The patent applies preliminary action by pre-training machine learning models offline using historical network data and user experience metrics. The models are trained in advance to recognize patterns indicative of network congestion, and once trained, they can rapidly detect congestion conditions in real-time without requiring ongoing training. This preliminary model preparation enables high productivity in network optimization while minimizing the time loss associated with training and deployment.
Data Source
Figure 1
Figure 2
Figure 3A~3B
AI summary
In one embodiment, the system identifies (610) geographic areas covered by a communication network. The system determines (640), for each identified geographic area, a network performance metric for the identified geographic area based on a difference between: (1) a first average network speed (620) of the communication network in the identified geographic area during prior time periods in which the communication network is busy, and (2) a second average network speed (630) of the communication network in the identified geographic area during second prior time periods in which the communication network is not busy. The system compares the respective performance metrics of the geographic areas to a threshold network performance metric, which is determined (650) by a congestion-analysis machine learning (ML) model (220). The system identifies (670) traffic congestions in one or more of the identified geographic areas having a determined network performance metric below the threshold network performance metric.