Signal Event Detection Using Causal State Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting events in signals, particularly biomedical signals, are inadequate due to high memory and computational requirements, lack of robustness, and inability to function online on devices with limited resources, often relying on non-causal filters and transformations.
Innovation Solution
A method that detects distinct repetitive events in signals by selecting candidates, generating features from amplitude, time, and validity, and classifying them using internal states, ensuring online functionality and reduced memory usage through causal processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If complex feature generation or signal transformations are used, then detection performance is improved, but memory and computational requirements increase
Solution Approach 1:
The patent segments the signal analysis process into distinct stages: candidate detection, feature generation, state estimation, and candidate classification. Each stage processes only necessary information, avoiding unnecessary computations. The state estimation module maintains compact internal states rather than storing complete signal histories, significantly reducing memory requirements while preserving detection accuracy.
Solution Approach 2:
The patent transforms the signal processing approach by changing from storing raw signal data to maintaining compressed internal states that capture essential signal characteristics. This parameter transformation enables complex detection logic to operate with reduced memory footprints and computational loads while maintaining high detection performance.
2Measurement precision
If non-causal filters or transformations are used, then detection accuracy is improved, but online capability is lost
Solution Approach 1:
The patent performs preliminary candidate detection and feature extraction using only past and present signal values, preparing data structures that are ready for classification without requiring future signal values. This preliminary action enables real-time processing while maintaining detection accuracy, as the state estimation and classification can proceed causally.
3Device complexity
If fixed threshold decisions are used, then computational resources are reduced, but detection robustness deteriorates
Solution Approach 1:
The patent implements dynamic thresholding through state estimation, where detection thresholds adapt based on the estimated internal states of the system. This dynamic approach maintains robustness by adjusting to varying signal conditions while keeping computational requirements low, as the state estimation uses simple recursive updates rather than complex computations.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present disclosure provides a method, device and system for detecting at least two distinct repetitive events (300), especially peaks and troughs, from a signal (100), by comprising a candidate detection (201) that selects respective candidates (210) for the distinct repetitive events (300) from the signal (100); a feature generation (202) in which candidate features (240) are captured; a state estimation (203) that adjusts at least one internal state (230) based on at least one candidate feature (220); and a candidate classification (204) that determines the distinct repetitive events (300) from the respective candidates (210) based on at least one candidate feature (220) and at least one internal state (230).