Multiscale Digital Signal Feature Extraction Method
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for feature extraction from digital signals face a trade-off between collecting comprehensive information and processing speed, often requiring significant memory resources, especially when dealing with large volumes of data like Internet traffic analysis for cyberattack detection.
Innovation Solution
A multiscale feature extraction method that segments digital signal samples into groupings of varying sizes, applying specific operators to derive representative values that capture all information without omitting individual samples, and iteratively refining these values using multiple operators to achieve compacted representations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional feature extraction methods are used to collect comprehensive information from digital signals, then measurement precision is improved, but productivity deteriorates due to slow processing speed and excessive memory usage
Solution Approach 1:
The patent divides the digital signal into multiple overlapping groupings or windows of different sizes. Each grouping is processed independently to extract features, allowing parallel computation and reducing the computational burden on the entire signal. This segmentation enables comprehensive information collection from all regions while maintaining fast processing through distributed computation.
Solution Approach 2:
The patent introduces a new dimension by processing the signal at multiple scales or resolutions simultaneously. Instead of processing the entire signal at one resolution, it creates groupings of varying sizes (different dimensions) and extracts features from each scale. This multi-scale approach captures both fine-grained and coarse-grained information, improving measurement precision while the selective processing at different scales maintains productivity.
2Measurement precision
If traditional feature extraction methods process all digital signal samples, then measurement precision is improved, but loss of substance increases due to excessive memory consumption
Solution Approach 1:
The signal is divided into overlapping groupings that can be processed in smaller chunks rather than loading the entire signal into memory. Each grouping is processed independently, and results are aggregated. This segmentation allows comprehensive feature extraction from all signal regions while using only a fraction of the memory that would be required to store and process the complete signal at once.
Solution Approach 2:
The patent applies different processing strategies to different regions of the signal based on their characteristics. By creating groupings of different sizes and applying appropriate operators to each, it extracts the most relevant information from each local region. This localized processing ensures that all important information is captured (maintaining measurement precision) while avoiding the memory overhead of uniform processing of the entire signal.
Data Source
AI summary
A method of digital signal feature extraction comprises steps of: (a) segmenting samples of the digital signal to form a set of groupings each comprising a subset of the samples, with each grouping having endpoints spaced apart by a current grouping size; (b) applying an operator, which is associated with the desired feature to be extracted, to the subset of the samples of each grouping to derive a representative value therefor corresponding to the grouping size; and (c) repeating step a), but based on a different grouping size, and repeating step b) on the set of groupings formed based on the different grouping size, with the operator being adapted to correspond to the different grouping size. The set of groupings formed in step a) collectively includes all of the samples of the signal. One endpoint of at least one grouping is intermediate the endpoints of another one of the groupings.