Granted Patent B2

US 12,471,790 · App. 17/366,287 · Granted Nov 18, 2025

Inventors: Peter Richards (San Francisco, CA); Javier L. Prieto (Oakland, CA); Kevin Pu Weekly (Lafayette, CA); Christopher Hanrahan Sarantos (San Francisco, CA); Kyle Patrick Nadeau (Los Altos, CA); Aniket Sanjay Deshpande (Pleasanton, CA); Paul Francis Stetson (Piedmont, CA); Vahe Tshitoyan (Pleasanton, CA); Julian Maclaren (Menlo Park, CA)

Assignee: FITBIT, LLC

A61B5/02416A61B5/6824A61B5/7264A61B5/7475

Abstract

Claims (48)

1 . A method for generating a physiological metric, the method comprising:

obtaining a first PPG signal based on light received by a light detector from a first light source during a first temporal window during which a second light source is not emitting light;

obtaining a second PPG signal based on light received by the light detector from the second light source different from the first light source and during a second temporal window that is different from the first temporal window and during which the first light source is not emitting light;

removing a motion component from at least one of the first and second PPG signals based on information associated with a first distance between the first light source and the light detector and a second distance between the second light source and the light detector;

processing, via a trained machine learning model of a user device, the first and second PPG signals to produce an output corresponding to a heart rate probability distribution for the first and second PPG signals;

determining the physiological metric using the output corresponding to the heart rate probability distribution for the first and second PPG signals; and

causing the physiological metric to be displayed via a user interface on the user device.

2 . The method of claim 1 , wherein the heart rate probability distribution for the first and second PPG signals are combined to form a combined heart rate probability distribution.

3 . The method of claim 2 , further comprising:

determining a peak in the combined heart rate probability distribution; and

using the peak to select a heart rate estimation.

4 . The method of claim 1 , further comprising determining at least one confidence metric associated with at least one of first and second PPG signals, the confidence metric being indicative of a signal quality for a particular light path associated with each of the first and second PPG signals.

5 . The method of claim 4 , further comprising:

determining a sum of probabilities for all heart rates within a fixed number of beats-per-minute of the physiological metric; and

using the sum as the confidence metric.

6 . The method of claim 4 , further comprising:

determining a variance in the heart rate probability distributions for the first and second PPG signals; and

weighing the heart rate probability distributions based on the variance.

7 . The method of claim 4 , wherein the heart rate probability distribution for each of the first and second PPG signals are combined into a probability distribution combination over successive time points to allow temporal updates of heart rate that include the confidence metric.

8 . The method of claim 7 , wherein the probability distribution combination is weighted such that more recent measurements are given more weight than older measurements.

9 . A method for generating a heart rate estimation for a user device, the method comprising:

obtaining multiple photoplethysmography (PPG) signals based on light received by at least one light detector of the user device from multiple light sources during different temporal windows during which other light sources of the multiple light sources are not emitting light, wherein the multiple PPG signals each represent a different channel of light received by the at least one light detector from the multiple light sources;

processing, via a trained machine learning model of the user device, the multiple PPG signals to produce an output corresponding to a heart rate probability distribution for the multiple PPG signals;

determining the heart rate estimation in beats-per-minute using the output corresponding to the heart rate probability distribution for the multiple PPG signals; and

causing the heart rate estimation to be displayed via a user interface on the user device.

10 . The method of claim 9 , wherein the trained machine learning model comprises a neural network.

11 . The method of claim 9 , wherein the heart rate probability distribution for the multiple PPG signals are combined to form a combined heart rate probability distribution.

12 . The method of claim 11 , further comprising:

determining a peak in the combined probability distribution; and

using the peak to select the heart rate estimation.

13 . The method of claim 9 , further comprising determining at least one confidence metric associated with at least one of multiple PPG signals, the confidence metric being indicative of a signal quality for a particular light path associated with each of the multiple PPG signals.

14 . The method of claim 13 , further comprising:

determining a sum of probabilities for all heart rates within a fixed number of beats-per-minute of the heart rate estimation; and

using the sum as the confidence metric.

15 . The method of claim 13 , further comprising:

determining a variance in the heart rate probability distributions for the multiple PPG signals; and

weighing the heart rate probability distributions based on the variance.

16 . The method of claim 13 , wherein the heart rate probability distribution for each of the multiple PPG signals are combined into a probability distribution combination over successive time points to allow temporal updates of heart rate that include the confidence metric.

17 . The method of claim 16 , wherein the probability distribution combination is weighted such that more recent measurements are given more weight than older measurements.

18 . A wearable device for tracking a heart rate estimation of a user, comprising:

a plurality of light sources;

a light detector;

a memory; and

one or more processors, wherein the memory stores computer-executable instructions for causing the one or more processors to:

obtain multiple photoplethysmography (PPG) signals based on light received by the light detector from the plurality of light sources during different temporal windows during which other light sources of the plurality of light sources are not emitting light;

process, via a trained machine learning model of the one or more processors, the multiple PPG signals to produce an output corresponding to a heart rate probability distribution for the multiple PPG signals;

determine the heart rate estimation in beats-per-minute using the multiple PPG signals and the output corresponding to the heart rate probability distribution for the multiple PPG signals; and

cause the heart rate estimation to be displayed via a user interface on the wearable device.

Assignments (2)

Continuity (3)

References Cited (400)