Fault Isolation in OTT Broadband Networks Using Predictive Modeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing fault isolation techniques for broadband networks are inadequate in measuring video streaming quality and identifying issues in over-the-top (OTT) content delivery, as they focus on traditional methods like loop health and modulation parameters, which do not account for user experience and are not continuous or autonomous.
Innovation Solution
A machine learning-based system that uses supervised learning algorithms to infer fault groups and diagnose root causes of problems in OTT video streaming by analyzing quality of experience metrics across network components, including publishers, CDNs, devices, and fiber nodes, employing a counterfactual predictive approach to isolate faults and predict quality of experience.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional fault isolation techniques (loop health, modulation parameters, FEC) are used, then network infrastructure health can be monitored, but video streaming quality and user experience cannot be accurately measured
Solution Approach 1:
The patent introduces QoE metrics as an intermediary layer between traditional network infrastructure monitoring and user experience assessment. These metrics act as mediators that translate complex network conditions into measurable video streaming quality indicators, enabling accurate measurement without requiring direct complex infrastructure analysis.
Solution Approach 2:
The patent replaces traditional mechanical/physical monitoring methods (loop health checks, modulation parameter analysis) with data-driven machine learning models that analyze QoE metrics. This substitution enables more precise video streaming quality measurement by using statistical and algorithmic approaches instead of conventional network monitoring techniques.
2Productivity
If continuous autonomous fault isolation is implemented, then user experience can be continuously monitored, but system complexity and computational resources increase
Solution Approach 1:
The patent implements self-service through automated machine learning models that continuously analyze QoE metrics and autonomously identify fault conditions without requiring manual intervention. The system performs self-diagnosis by comparing observed metrics against learned patterns, enabling continuous fault isolation while reducing operational complexity through automation.
Solution Approach 2:
The patent applies preliminary action by pre-training machine learning models on historical QoE data and network conditions before deployment. This preliminary preparation enables the system to quickly and efficiently perform continuous fault isolation without requiring complex real-time computations, as the models have already learned patterns during the training phase.
3Loss of information
If traditional monitoring methods are used, then infrastructure health can be assessed, but root cause analysis for video streaming issues cannot be performed
Solution Approach 1:
The patent implements feedback loops where QoE metrics are continuously collected, analyzed by machine learning models, and used to generate fault diagnoses that can trigger automated remediation actions. This feedback mechanism ensures complete fault diagnosis information by continuously monitoring the system state and adjusting diagnoses based on observed changes in video streaming quality.
Solution Approach 2:
The patent creates a universal fault diagnosis system that handles multiple types of video streaming issues through a single machine learning-based platform. The system can diagnose various root causes (network congestion, encoder issues, CDN problems, device compatibility) using the same QoE metric analysis framework, enabling comprehensive automated fault diagnosis across diverse failure modes.
Data Source
AI summary
Fault isolation in over-the-top content (OTT) broadband networks is disclosed. Network topology information associated with a network service provider is received. Session information associated with one or more streaming sessions is received. A predictive model is generated for predicting session quality at least in part by using at least some of the network topology and session summary information as features. The predictive model is used to determine a first prediction of session quality using a first set of feature values. A second set of feature values is generated at least in part by replacing a first feature value in the first set of feature values with a replacement value. The predictive model is used to determine a replacement prediction of session quality using the second set of feature values including the replacement value with which the first feature value was replaced. Based at least in part on the first prediction and the replacement prediction, an impact of the first feature value on session quality is determined. A fault in a content delivery ecosystem is isolated based at least in part on the determined impact of the first feature value on session quality.