Probabilistic Sensor Data Fusion for Biomedical Signal Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Mechanical and biomedical sensors face challenges in accurately filtering out noise and artifacts, particularly in mobile or ambulatory settings, due to overlapping frequency noise bands and nonlinear, non-stationary noise patterns, which limits the extraction of reliable biomedical signals and the measurement of a narrow spectrum of medical parameters.
Innovation Solution
A probabilistic model-based system that combines data from multiple sensors using a dynamic state-space model and probabilistic digital signal processor to filter, estimate, and extract additional information, effectively handling nonlinear and non-stationary noise by iteratively processing probability distribution functions and integrating sensor data with physical models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional filtering techniques are used to remove noise from sensor data, then some noise reduction is achieved, but the filtering is unreliable when noise frequencies overlap with signal frequencies or when noise patterns are nonlinear and non-stationary
Solution Approach 1:
The patent introduces probability distribution functions as an intermediary layer between the raw sensor data and the final signal extraction. Instead of directly filtering the signal, the system models the probabilistic characteristics of both signal and noise, allowing for more reliable discrimination even when frequencies overlap. This intermediary probabilistic modeling approach resolves the contradiction by providing a framework that maintains measurement precision while improving filtering reliability in non-stationary conditions.
Solution Approach 2:
The system dynamically changes parameters by iteratively updating probability distribution functions based on incoming sensor data. Rather than using fixed filtering parameters, the system adapts its probabilistic model parameters in real-time to match the non-stationary characteristics of the noise and signal, thereby maintaining both reliability and precision across varying operating conditions.
2Adaptability or versatility
If data from multiple sensors is integrated to extract more parameters, then the spectrum of measurable medical parameters expands, but the complexity of data processing and noise filtering increases
Solution Approach 1:
The patent merges data from multiple sensors by integrating their probability distribution functions into a unified probabilistic model. Instead of processing each sensor independently and then combining results, the system combines the probabilistic representations at the model level, which simplifies the overall processing complexity while enabling the extraction of multiple medical parameters from fused sensor data.
Solution Approach 2:
The probabilistic data processing framework serves multiple functions simultaneously: it filters noise, extracts signals, estimates parameters, and handles data fusion. This universal approach allows the system to expand the spectrum of measurable parameters without proportionally increasing processing complexity, as the same probabilistic machinery handles all these tasks integratedly.
3Ease of manufacture
If simple filters are used to remove artifacts, then processing is computationally simple, but artifacts that resemble real processes (such as ectopic beats) cannot be removed reliably
Solution Approach 1:
The system moves from fixed-parameter simple filters to adaptive probability distribution functions whose parameters change dynamically based on the data. This allows the system to maintain computational tractability while significantly improving artifact removal reliability, as the probabilistic model can distinguish between artifacts and real processes based on their statistical characteristics rather than relying on simple frequency-based filtering.
Data Source
AI summary
A probabilistic digital signal processor using data from multiple instruments is described. In one example, an analyzer is configured to: receive discrete first and second input data, related to a first and second sub-system of the system, from a first and second instrument, respectively. A system processor is used to fuse the first and second input data into fused data. The system processor optionally includes: (1) a probabilistic processor configured to convert the fused data into at least two probability distribution functions and (2) a dynamic state-space model, the dynamic state-space model including at least one probabilistic model configured to operate on the at least two probability distribution functions. The system processor iteratively circulates the at least two probability distribution functions in the dynamic state-space model in synchronization with receipt of updated input data, processes the probability distribution functions, and generates an output related to the state of the system.