Warped Linear Prediction for Sleep Disordered Breathing Characterization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for monitoring sleep disordered breathing (SDB) face challenges in accurately identifying and characterizing SDB events due to variations in sound patterns caused by changes in circumstances, such as medication or alcohol consumption, and struggle with differentiating SDB sounds from background noises, especially in low-frequency and high-frequency ranges.
Innovation Solution
A method involving a four-stage analysis approach that includes temporal segmentation, spectral characterization, identification of inhalation and exhalation phases, and generation of probability functions using Warped Linear Prediction (WLP) and Laguerre Linear Prediction (LLP) algorithms to differentiate SDB events from other sounds without requiring multiband analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional linear prediction methodologies are used for spectral analysis, then the analysis can be performed with standard methods, but the low-frequency modeling capabilities are poor and cannot accurately capture SDB sounds ranging from 100 Hz to several kHz
Solution Approach 1:
The patent transforms the audio signal from the time domain to the frequency domain using Fourier Transform, and then applies Warped Linear Prediction (WLP) with modified warping parameters to enhance low-frequency resolution. This parameter transformation allows accurate modeling across the full SDB frequency spectrum (100 Hz to several kHz) while maintaining computational efficiency.
2Measurement precision
If personalized signature data is collected for each patient to accurately identify SDB events, then the identification accuracy improves, but the complexity of the system increases and requires extensive personalized calibration
Solution Approach 1:
The patent creates a universal SDB detection system that uses population-based spectral templates and characteristic frequency patterns applicable to all patients. The system identifies SDB events based on universal acoustic features (spectral signatures, periodicity patterns, energy distribution) that are consistent across different individuals, eliminating the need for personalized calibration while maintaining high accuracy.
Solution Approach 2:
The system performs automatic adaptive learning during the monitoring period, where the algorithm automatically adjusts to individual patient characteristics without requiring manual personalized setup. The machine learning components continuously refine detection parameters based on the specific patient's breathing patterns, achieving personalized accuracy through self-service adaptation rather than pre-calibration.
3Measurement precision
If multiband analysis is used to differentiate SDB sounds across frequency ranges, then the spectral resolution improves, but the device complexity and filter design requirements increase significantly
Solution Approach 1:
The patent replaces complex mechanical filter banks with a computational approach using Fourier Transform and Warped Linear Prediction algorithms. This substitution achieves superior spectral resolution across all frequency bands without requiring physical filter design, simplifying the system while enhancing analytical capability.
4Loss of information
If the system monitors all breathing sounds continuously throughout the sleeping period, then complete SDB characterization is achieved, but the processing time and computational load increase
Solution Approach 1:
The patent divides the continuous sleep monitoring data into discrete breathing cycles and identifies candidate SDB events based on initial threshold criteria. The system then applies detailed spectral analysis only to these candidate segments rather than continuous processing, significantly reducing computational load while maintaining complete characterization of all SDB events.
Solution Approach 2:
The system applies a two-stage analysis where a coarse filter identifies potential SDB events using simple energy and periodicity thresholds, followed by refined spectral analysis only on these candidates. This partial application of full analysis to selected segments achieves complete SDB characterization with reduced processing time compared to analyzing all breathing sounds in detail.
Data Source
AI summary
A method of characterizing a patient's disordered breathing during a sleeping period includes performing a first partial characterization of a time axis of an audio signal in order to learn the most prominent and highly relevant events. Only at a later stage, i.e., after sufficient observation of the highly relevant events, is a full segmentation of the entire time axis actually carried out. Linear prediction is used to create an excitation signal that is employed to provide better segmentation than would be possible using the original audio signal alone. Warped linear prediction or Laguerre linear prediction is employed to create an accurate spectral representation with flexibility in the details provided in different frequency ranges. A resonance probability function is generated to further characterize the signals in order to identify disordered breathing. An output includes a characterization in any of a variety of forms of identified disordered breathing.