Invisible Traffic Estimation via Singular Value Decomposition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods fail to effectively infer invisible traffic information flowing through a target network, as they rely on temporal estimation and assume uniform sampling and low variability in traffic matrices, which do not hold for networks with high variability and irregular scattering.
Innovation Solution
A system and method that uses singular value decomposition to determine the low effective rank of traffic matrices, allowing for the estimation of invisible traffic by constructing linear estimators and employing techniques like principal component regression and ridge regression to predict traffic flowing through other networks, even when only a partial subset of data is available.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If temporal estimation methods are used to infer invisible traffic, then estimation can be performed, but the estimation accuracy deteriorates due to strong correlation between temporally sequenced traffic matrices
Solution Approach 1:
The patent transitions from temporal estimation (time-based) to spatial estimation (network-topology-based) by utilizing the low effective rank property of traffic matrices. Instead of estimating traffic at different time points, the system estimates invisible traffic by analyzing the spatial structure and correlations within the traffic matrix itself, leveraging the fact that traffic matrices have low effective rank and can be accurately represented by a small number of principal components.
Solution Approach 2:
The patent creates a simplified representation (copy) of the complex traffic matrix by retaining only the top k principal components that capture the essential structure. This low-rank approximation serves as a copy that preserves the critical traffic flow patterns while discarding redundant information, enabling accurate estimation of invisible traffic elements without requiring complete observation of all traffic data.
2Loss of information
If existing traffic matrix completion methods are applied, then some traffic information can be inferred, but the methods fail to handle high variability and irregular scattering of traffic data
Solution Approach 1:
The patent changes the approach from assuming uniform sampling and low variability to explicitly modeling high variability and irregular scattering. By using robust statistical methods and low-rank matrix factorization, the system can handle cases where traffic data is sparsely and irregularly sampled across different network elements, still achieving reliable estimation by capturing the underlying low-rank structure that persists despite the variability.
3Measurement precision
If network operators collect complete traffic data, then accurate traffic engineering decisions can be made, but the complexity of data collection and processing increases
Solution Approach 1:
The patent extracts only the essential information needed for traffic engineering by identifying and utilizing the top k principal components that capture the majority of traffic flow patterns. Instead of processing complete traffic matrices with all their elements, the system extracts and works with a reduced set of dominant patterns, significantly lowering computational complexity while maintaining measurement precision for traffic flow analysis.
Solution Approach 2:
The patent applies partial action by estimating only the invisible traffic elements that are most critical for network optimization, rather than attempting to reconstruct the entire traffic matrix. By focusing computational resources on inferring only the missing portions of traffic data that affect network performance, the system reduces processing complexity while achieving sufficient accuracy for engineering decisions.
Data Source
AI summary
This disclosure is directed to techniques for inferring traffic information or estimating total volume of traffic/data flowing through a target network/entity, wherein only a partial subset of inferred traffic information or volume of data is available to a predictor entity/network that infers such traffic information. In an embodiment, such partial subset of total traffic can either be made available to the entity/network for inferring and estimating total traffic or such partial data can actually flow through the entity/network.