Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-12514509-B2
Publication Date
2026-01-06
Expiration Date
Abstract
The disclosed system improves the physiological estimates (e.g., heart rate, pulse oxygenation, etc.) extracted from photoplethysmography (PPG) signals captured by wearable health monitors (smart watches, fitness trackers, etc.) by employing three sign-data least mean squares (SDLMS) filters in a cascaded parallel combination (CPC) that each successively remove motion artifacts in the x-, y-, and z-dimensions. In some embodiments, a window function eliminates spectral content that is unlikely in view of recent frequency estimates and/or smoothing function smooths the physiological estimates using historical physiological data. The disclosed motion artifact removal system can achieve a level of accuracy using signals from a reflective-type PPG sensor that is typically only achieved via a transmissive-type (e.g., finger-worn) PPG sensor at rest. Specifically, in initial testing, the system was able to estimate the heart rates with less than 2 beats per minute (BPM) of root-mean-squared (RMS) error during periods of both rest and exercise.
Core Innovation
The invention describes a cascaded parallel combination (CPC) architecture for removing motion artifacts from photoplethysmography (PPG) data. The CPC architecture includes a first cascaded noise cancellation block and a second cascaded noise cancellation block, which generate corresponding outputs used for estimating physiological data. The first block includes sign-data least mean squares (SDLMS) filters configured to minimize error between SDLMS suboutputs and acceleration data across first, second, and third dimensions.
The second cascaded noise cancellation block includes recursive least squares (RLS) filters configured to minimize error between RLS suboutputs and acceleration data across the first, second, and third dimensions. A combination layer estimates physiological data by combining the first cascaded noise cancellation block output and the second cascaded noise cancellation block output according to a combination or mixing parameter. Each SDLMS filter is an adaptive filter that calculates a vector w(n) of optimized coefficients according to the disclosed update using a learning rate and sgn(x(n)) for acceleration sign.
In some embodiments, the CPC architecture further includes a window function that receives estimated physiological data and generates windowed physiological data. The window function is described as a mathematical function that is zero outside a predetermined threshold around an estimated frequency of the estimated physiological data. In further embodiments, a smoothing function receives the windowed physiological data and generates smoothed physiological data, including a moving average filter that uses the N most recent estimates within the windowed physiological data.
Claims Coverage
The document portion provides three independent claims, each centered on a CPC architecture or a corresponding method for motion-artifact removal from wearable PPG using two cascaded noise-cancellation paths and a combination layer. Across the independent claims, the main inventive features are the three-dimension cascaded SDLMS and cascaded RLS blocks, the use of acceleration data indicative of wearable motion, and the combination/mixing parameter to estimate physiological data; dependent claim refinements further add windowing and smoothing for processing estimated physiological data.
Cascaded parallel combination architecture with SDLMS and RLS noise-cancellation blocks
A cascaded parallel combination (CPC) architecture for removing motion artifacts, the CPC architecture comprising a first cascaded noise cancellation block with first, second, and third sign-data least mean squares (SDLMS) filters operating by minimizing error between SDLMS suboutputs and acceleration data in first, second, and third dimensions; a second cascaded noise cancellation block with first, second, and third recursive least squares (RLS) filters operating by minimizing error between RLS suboutputs and acceleration data in the first, second, and third dimensions; and a combination layer configured to estimate physiological data by combining the first cascaded noise cancellation block output and the second cascaded noise cancellation block output according to a combination or mixing parameter.
Adaptive SDLMS coefficient update using sign of acceleration
Wherein each SDLMS filter is an adaptive filter configured to calculate a vector w(n) of optimized coefficients according to the disclosed coefficient update with a learning rate and sgn(x(n)) being the sign of the acceleration data.
Dimension-wise successively cascaded motion-artefact removal using acceleration across three dimensions
A CPC architecture comprising: a first cascaded noise cancellation block generating a first suboutput, the first cascaded noise cancellation block including three SDLMS filters configured to receive acceleration data in one of three dimensions and successively remove motion artifacts in each of the three dimensions from PPG data; a second cascaded noise cancellation block generating a second suboutput, the second cascaded noise cancellation block including three RLS filters configured to receive acceleration data in one of three dimensions and successively remove motion artifacts in each of the three dimensions from the PPG data; and a combination layer configured to combine the first suboutput and the second suboutput according to a combination or mixing parameter.
Method for removing motion artifacts from wearable PPG using cascaded SDLMS and RLS blocks
A method for removing motion artifacts from photoplethysmography (PPG) data captured by a wearable PPG sensor using acceleration data indicative of the acceleration of the wearable PPG sensor in each of three dimensions, the method comprising receiving PPG data by a first SDLMS filter of a first cascaded noise cancellation block, generating SDLMS suboutputs by sequentially minimizing error between suboutputs and acceleration data in first, second, and third dimensions; receiving PPG data by a first RLS filter of a second cascaded noise cancellation block, generating RLS suboutputs by sequentially minimizing error between suboutputs and acceleration data in first, second, and third dimensions; and estimating physiological data by combining the first cascaded noise cancellation block output and the second cascaded noise cancellation block output according to a combination or mixing parameter.
Windowed estimated physiological data using a thresholded frequency window function
A CPC architecture further comprising a window function that receives estimated physiological data and generates windowed physiological data, wherein the window function is a mathematical function that is zero outside a predetermined threshold around an estimated frequency of the estimated physiological data.
Smoothed physiological data using a smoothing function and moving average of N most recent estimates
A CPC architecture further comprising a smoothing function that receives the windowed physiological data and generates smoothed physiological data, wherein the smoothing function comprises a moving average filter of the N most recent estimates in the windowed physiological data.
Across the independent claims and the provided dependent refinements, the core claim coverage is directed to CPC-based motion-artifact removal from wearable PPG using two cascaded adaptive noise-cancellation blocks (SDLMS across three acceleration dimensions and RLS across three acceleration dimensions) and a combination/mixing parameter to estimate physiological data. The disclosed adaptive SDLMS coefficient update uses a learning rate with sgn(x(n)), and further claim refinements process the estimated physiological data via a thresholded frequency window function and a smoothing function implemented as a moving average over the N most recent estimates.
Stated Advantages
Documented Applications
No documented applications found
Interested in licensing this patent?