Cell Performance Change-Point Analysis for Network Fault Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for analyzing cellular communications networks fail to distinguish between intentional network changes and network faults, leading to inefficiencies in corrective actions and root cause analysis.
Innovation Solution
A method involving time series analysis of performance data using linear regression models to identify change points, determine offset and angle differences, and classify anomalies as step or trend changes, enabling automated corrective actions and root cause analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If change point detection or anomaly detection methods are used to identify anomalously behaving cells, then problems can be detected sooner, but the methods cannot distinguish between intentional network changes and network faults
Solution Approach 1:
The patent segments the time series data into multiple segments around detected change points. By dividing the data into pre-change and post-change segments, the system can analyze the characteristics of each segment separately using linear regression, thereby identifying whether the change represents an intentional network modification or a fault condition based on the nature of the segmentation results.
Solution Approach 2:
The patent transforms the anomaly detection approach by changing the parameters analyzed - instead of merely detecting that a change occurred, the system calculates offset differences and angle differences between linear regression models of adjacent segments. These parameter changes enable classification of anomaly types (sudden vs. gradual) while maintaining rapid detection capability.
2Productivity
If automated anomaly detection is implemented, then corrective actions can be taken faster, but the lack of anomaly type classification reduces the efficiency of corrective actions
Solution Approach 1:
The patent implements a feedback mechanism where the results of linear regression analysis (offset difference and angle difference) provide information about the anomaly type. This feedback loop enables the system to not only detect anomalies quickly but also to classify them, thereby improving the effectiveness of subsequent corrective actions while maintaining high productivity.
Solution Approach 2:
The patent performs preliminary classification of anomalies by analyzing the characteristics of change points before corrective actions are initiated. By pre-classifying anomalies as sudden or gradual based on regression model comparisons, the system prepares appropriate response strategies in advance, making corrective actions more effective without delaying their execution.
3Measurement precision
If linear regression models are fitted before and after change points to determine offset and angle differences, then anomaly types can be classified, but the complexity of the analysis increases
Solution Approach 1:
The patent extracts only the essential features needed for anomaly classification by fitting linear regression models specifically to identify offset differences and angle differences at change points. Rather than analyzing all possible characteristics of the time series data, the method extracts and focuses on these two key parameters, achieving precise anomaly classification while managing analytical complexity.
Solution Approach 2:
The patent applies local quality analysis by fitting linear regression models only in the immediate vicinity of detected change points rather than analyzing the entire time series uniformly. This localized approach concentrates computational resources on the critical transition regions, improving measurement precision for anomaly classification while reducing overall device complexity by avoiding global analysis of all data points.
Data Source
AI summary
Analysis of operation of a communications network. The analysis is performed by obtaining (301) time series of performance data of a cell of the communications network; selecting (302) a change point in the time series; determining (303) a first linear regression model before the selected change point and a second linear regression model after the selected change point; determining (304) offset difference between the first linear regression model and the second linear regression model; and determining (306) anomaly type based on the offset difference.