Cellular Network Root Cause Analysis via Machine Learning Graphs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for root cause analysis in cellular networks are inadequate due to reliance on manual rules and thresholds, which are not scalable or adaptable to the increasing complexity of modern networks.
Innovation Solution
An automated and data-driven approach using machine learning (ML) to analyze issues in cellular networks by constructing a graph-shaped dataset considering Key Performance Indicators (KPIs), handover statistics, and physical factors, allowing for graph-based methods to identify root causes and adapt to network changes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If manual rules and thresholds are used for root cause analysis, then domain knowledge can be applied, but the solution is not scalable and cannot adapt to increasing network complexity
Solution Approach 1:
The patent replaces manual rule-based systems with machine learning models that automatically learn from network data. The ML models process complex network metrics and relationships without requiring explicit manual rule definitions, enabling the system to adapt to evolving network conditions and complexities dynamically.
Solution Approach 2:
The system transitions from static manual rules to dynamic machine learning models that continuously learn and adapt to changing network conditions. The ML models can automatically update their understanding of network behavior patterns, making the solution adaptable to new network configurations and issues without manual intervention.
2Device complexity
If aggregated cell level information is used, then the solution architecture is simplified, but diagnostic information for understanding root causes is insufficient
Solution Approach 1:
The patent segments the analysis into multiple levels: individual cell metrics, neighbor cell metrics, and relational metrics between cells. This segmentation allows the system to maintain detailed diagnostic information for root cause analysis while organizing the complex data structure in a manageable way through the graph-based approach.
Solution Approach 2:
The patent introduces a new dimension by representing network cells and their relationships as a graph structure. This graph representation adds spatial and relational dimensions to the data, enabling the system to capture both individual cell characteristics and their interrelationships without requiring complex multi-dimensional data structures.
3Device complexity
If only source cell and outgoing relational metrics are considered, then the analysis is simpler, but neighbor cells' own metrics and incoming relational metrics are not considered
Solution Approach 1:
The patent extracts and incorporates multiple types of metrics from different sources: source cell metrics, neighbor cell metrics, outgoing relational metrics, and incoming relational metrics. By extracting these separate components and integrating them into a unified graph-based analysis framework, the system achieves comprehensive root cause identification while managing analysis complexity systematically.
4Productivity
If rule-based solutions are built on domain knowledge, then initial analysis capability is provided, but the solutions are static and hard to adapt as networks evolve
Solution Approach 1:
The patent implements self-service through machine learning models that automatically learn from network data without requiring continuous manual rule updates. The models autonomously adapt to evolving network conditions by learning from historical and current network metrics, eliminating the need for manual rule maintenance while preserving strong initial analysis capabilities.
Data Source
AI summary
A computer-implemented method for analyzing issues in a cellular network is provided. The cellular network includes a plurality of cells, including source cells and neighbor cells. The method includes building a network graph representing features of the cells. The method includes identifying, using the network graph, sub-graphs for each source cell indicating a network issue. The method includes, for each network issue of the source cell for each sub-graph, ranking each feature. The method includes, for each source cell, training, using a feature set identified, a first machine learning (ML) model. The method includes training a second ML model, using patterns identified by each cluster, to classify an unidentified pattern of one or more neighbor cells contribution to the source cell network issue and identify root cause information for the issue.