A wearable PPG motion artifact suppression method, system, device and storage medium fusing acceleration motion reference and blind source separation

Abstract

The application discloses a wearable PPG motion artifact suppression method and system fusing acceleration motion reference and blind source separation, equipment and a storage medium. The method synchronously collects red light PPG, infrared PPG and three-axis ACC signals and performs time alignment, identifies artifact risks based on motion intensity, jerk, posture mutation and spectral overlap; when the risk condition is met, adaptive filtering is performed with ACC as the reference, and blind source separation is performed on the first suppression signal, red light PPG, infrared PPG and ACC derived signals. The red light / infrared coherence and template beat residual are used to evaluate candidate components, and the segments are divided into three categories: recoverable, only rhythm positioning and non-recoverable, and artifact masks, quality reason codes and downstream available uses are output.

Description

Technical Field

[0001] This invention relates to the field of wearable physiological signal processing technology, and in particular to a wearable PPG motion artifact suppression method, system, device and storage medium that integrates acceleration motion reference and blind source separation. It can be used for preprocessing of heart rate, rhythm, blood oxygen and related physiological parameters in wrist-worn, finger-worn, ear-worn, chest patch or other patch devices. Background Technology

[0002] Photoplethysmography (PPG) signals reflect changes in blood volume throughout the heart cycle by illuminating tissue with a light source and detecting changes in reflected or transmitted light intensity. Wearable devices typically use PPG signals to estimate heart rate, blood oxygen saturation, respiratory parameters, or rhythm status. However, in everyday wearable scenarios, activities such as walking, running, arm swinging, fist clenching, device loosening, changes in skin contact pressure, and rapid posture changes can all introduce motion artifacts. These artifacts may closely approximate or overlap with the actual heart rate, causing PPG waveform distortion, peak drift, baseline perturbation, and multi-channel inconsistencies, leading to errors in downstream physiological parameters.

[0003] Existing solutions often employ single bandpass filtering, fixed threshold quality scoring, or simply use acceleration regression to eliminate motion noise. These solutions are effective for low-intensity motion or when the motion frequency is separated from the heart rate frequency. However, when the motion frequency enters the candidate heart rate band, there are sudden changes in wearing posture, increased differences in response between red and infrared channels, or the effective pulse morphology of the PPG is partially masked, problems such as over-recovery, retention of erroneous peaks, or misidentification of the motion cycle as the heart rate cycle can easily occur. Simply increasing the filtering intensity sacrifices the true pulse morphology, while simply removing high-motion segments reduces the availability of continuous monitoring.

[0004] Therefore, a processing scheme is needed that takes into account data sources, artifact causes, recovery boundaries, and downstream usage risks from the perspective of first principles: its input should simultaneously cover physiological observation signals and motion reference signals; its processing flow should identify the intensity and frequency overlap of motion's influence on PPG; its recovery algorithm should be able to both adaptively cancel using external motion references and separate potential physiological sources in multi-channel observations; its output should not only provide a single corrected waveform, but should clearly indicate whether each segment is recoverable, whether it can only be used for rhythm localization, or whether it must be removed or downweighted, and output interpretable quality reason codes and usable uses to downstream applications. Summary of the Invention

[0005] I. Technical problems to be solved

[0006] The technical problem to be solved by this invention is: to address the issue that wearable PPGs are affected by acceleration-related perturbations, abrupt changes in posture, changes in device contact, and overlap of motion frequency and heart rate frequency bands in motion scenarios, and to provide a PPG motion artifact suppression method that can identify artifact risks, perform adaptive suppression, separate physiological pulsation components, and clearly define recovery boundaries.

[0007] Specifically, the present invention addresses at least the following problems:

[0008] 1. In scenarios where PPG and ACC are acquired simultaneously, how to comprehensively assess the risk of artifacts based on motion intensity, jerk, abrupt changes in attitude, and spectral overlap;

[0009] 2. When motion reference is available but the motion and heartbeat frequency bands overlap, how can we avoid the weakening of the true beat or the incorrect preservation of the motion cycle caused by a single filter?

[0010] 3. In multi-channel PPG, how to use red / infrared coherence and template beat residual to determine whether the recovered signal still has physiological reliability;

[0011] 4. In cases where fragments cannot be reliably recovered, how to explicitly output elimination, deweighting, or use-only criteria to downstream algorithms, instead of outputting unconstrained pseudo-recovery results.

[0012] II. Technical Solution

[0013] To address the aforementioned technical problems, this invention provides a wearable PPG motion artifact suppression method that integrates acceleration motion reference and blind source separation. The method includes the following steps.

[0014] S101, synchronously acquires Red / IR PPG and ACC signals. The wearable device acquires the red channel PPG signal and the infrared channel PPG signal through the red light optical path and the infrared light path respectively, while simultaneously acquiring the ACC signal through a three-axis accelerometer. The system records a unified timestamp or estimates the time of the red channel PPG signal, the infrared channel PPG signal, and the ACC signal through clock drift estimation.

[0015] S102, Segment Segmentation and Preprocessing. The time-aligned PPG and ACC are divided into multiple analysis segments according to a preset window length, such as 4-second, 8-second, 10-second, or 15-second windows, with adjacent windows overlapping. For the PPG, detrending, bandpass filtering, outlier processing, and amplitude normalization are performed; for the ACC, gravity component estimation, modulus calculation, band energy calculation, and jerk calculation are performed.

[0016] S103, Motion Artifact Risk Assessment. Calculate at least the following indicators for each analysis segment:

[0017] Motion intensity: including ACC three-axis mode length root mean square, variance, peak-to-peak value, or target frequency band energy;

[0018] jerk: includes the root mean square of the first-order difference of acceleration, the peak value, or the proportion exceeding the threshold;

[0019] Attitude abrupt changes: including changes in the angle of gravity direction, changes in the rate of change of attitude angle, or abrupt changes in low-frequency acceleration components;

[0020] Spectral overlap: including the frequency difference between the PPG candidate frequency and the ACC frequency, the proportion of ACC energy within the PPG candidate heart rate band, and the coherence function or cross-spectral peak of PPG and ACC.

[0021] The system generates a motion artifact risk assessment result based on the above indicators. This result can be a risk score, a risk level, or a multi-label set of risk causes. When the motion frequency falls within the candidate heart rate frequency band, or when PPG and ACC are highly coherent within the candidate heart rate frequency band, the system increases the risk level and records the spectral overlap cause code.

[0022] S104, Adaptive Filtering. For segments that meet the risk conditions, the system uses the ACC three-axis signal, ACC magnitude, jerk, attitude change, and its delay term as reference inputs, and employs LMS, RLS, or Kalman adaptive filtering to estimate motion artifact components. Let the PPG observation be y(n), the ACC reference vector be x(n), and the filter-estimated artifact be m(n), then the first suppression signal can be expressed as:

[0023] s(n) = y(n) - m(n)

[0024] Where m(n) is dynamically updated based on the adaptive filter parameters and x(n). LMS is suitable for real-time scenarios with limited computational resources, RLS is suitable for rapidly changing motion scenarios, and Kalman filtering is suitable for scenarios where physiological pulsations and motion artifacts are modeled as state variables.

[0025] S105, Blind Source Separation. The system constructs an observation matrix from the first suppression signal, multi-channel PPG signal, and ACC derived signal, and performs blind source separation using ICA, PCA, or CCA. The goal of blind source separation is to obtain candidate physiological pulsation components and candidate motion artifact components from the mixed observations. The system selects candidate physiological pulsation components according to the following criteria:

[0026] The correlation with the ACC signal and ACC derived signals is lower than the motion-related threshold;

[0027] It has a stable main peak within the physiologically feasible heart rate frequency band;

[0028] The candidate heart rate changes between adjacent segments do not exceed the physiologically feasible range;

[0029] The red and infrared PPG channels have sufficient coherence near the candidate heart rate frequency;

[0030] The single-beat pattern has a low residual compared to the historical template or local dynamic template.

[0031] S106, Red / Infrared Coherence and Template Beat Residual Auxiliary Evaluation. For devices containing red and infrared channels, the system calculates the amplitude ratio, phase difference, coherence function, and peak consistency of the two channels within the candidate heart rate frequency band. When both the red and infrared channels support the same candidate beat rhythm, the recovery reliability of that segment is improved; when the main peaks of both channels are separated, coherence is low, or the phase relationship is abnormal, the recovery reliability is reduced.

[0032] The system also constructs a dynamic pulsation template. After peak localization, single-beat truncation, phase alignment, and amplitude normalization of candidate physiological pulsation components, the differences between the template pulsation and the template pulsation are calculated. The template pulsation residuals may include mean squared error, dynamic time warping distance, peak-valley position deviation, or inverse indices of morphological correlation coefficient. The template is updated based on high-quality fragments; low-quality fragments are not involved or are included in the template update with low weight.

[0033] S107, Fragment Classification. The system classifies fragments based on motion artifact risk, filter convergence, blind source separation stability, red / infrared coherence, and template beat residuals:

[0034] Recoverable fragments: motion risks are interpretable and stable physiological pulsations are preserved after adaptive filtering and blind source separation, meeting the quality requirements for downstream parameter calculations;

[0035] Rhythm-localizable segments: The timing or location of the pulsation can still be located, but the waveform amplitude, morphology, or multi-channel consistency is insufficient to support the calculation of blood oxygen, HRV, or morphological parameters;

[0036] Unrecoverable fragments: Excessive motion intensity, drastic jerk or posture changes, spectral overlap that cannot be separated, red / infrared inconsistency, excessive template beat residual, signal saturation, or unstable blind source separation, resulting in the inability to reliably identify physiological beats.

[0037] S108, Output Quality Results. The system performs removal or downweighting on irrecoverable segments, restricts downstream applications for segments that can only be rhythmically located, and outputs corrected PPG waveforms or candidate beat sequences for recoverable segments. The system also outputs an artifact mask, a quality reason code, and downstream usability information. The artifact mask identifies the artifact status of each sampling point or segment; the quality reason code explains the reason for the quality degradation; and the downstream usability information controls whether the segment can be used by heart rate, HRV, blood oxygen, rhythm analysis, or other algorithms.

[0038] III. Beneficial Effects

[0039] Compared with the prior art, the present invention has at least the following technical effects:

[0040] 1. By synchronizing PPG and ACC acquisition and aligning the time, motion reference and physiological observation are placed on a unified time axis, improving the accuracy of motion artifact localization;

[0041] 2. By combining the determination of exercise intensity, jerk, abrupt changes in posture, and spectral overlap, it can not only identify whether the exercise is intense, but also whether the exercise frequency may be mistaken for the heart rate.

[0042] 3. By using LMS, RLS or Kalman adaptive filtering and ACC reference to dynamically estimate motion artifact components, the mismatch problem of fixed filter parameters under different wearing conditions can be reduced;

[0043] 4. Enhance the separability of physiological pulsation and motor components by blind source separation using ICA, PCA, or CCA in the case of mixed multi-channel PPG and motion reference;

[0044] 5. By using red / infrared coherence and template beat residual constraints to recover the results, we avoid outputting spurious recovered waveforms based solely on frequency peaks;

[0045] 6. By classifying fragments into three categories—recoverable, rhythm-localizable only, and non-recoverable—the recovery boundaries are clearly defined, reducing the contamination of non-recoverable fragments on heart rate, blood oxygen, HRV, or rhythm analysis;

[0046] 7. By using artifact mask, quality reason code, and downstream usability output, subsequent algorithms can select, eliminate, or downweight signal segments according to their intended use, thereby improving system-level reliability and interpretability. Attached Figure Description

[0047] Figure 1 is a flowchart of the PPG motion artifact suppression method provided in the embodiment of the present invention (overall flowchart of PPG and ACC synchronous motion artifact suppression).

[0048] Figure 2 is a schematic diagram of PPG and ACC synchronous acquisition and time alignment provided in an embodiment of the present invention;

[0049] Figure 3 is a schematic diagram of the motion artifact risk assessment module provided in an embodiment of the present invention;

[0050] Figure 4 is a schematic diagram of the joint processing flow of adaptive filtering and blind source separation provided in an embodiment of the present invention;

[0051] Figure 5 is a schematic diagram of the evaluation of red light / infrared coherence and template beat residual quality provided in an embodiment of the present invention;

[0052] Figure 6 is a schematic diagram of fragment classification, artifact mask, quality reason code and downstream usable use output provided in the embodiment of the present invention (fragment classification and downstream use control chart).

[0053] Figure 7 is a block diagram of the system module structure provided in an embodiment of the present invention;

[0054] Figure 8 is a schematic diagram of the electronic device structure provided in an embodiment of the present invention. Detailed Implementation

[0055] The present invention will be further described below with reference to the accompanying drawings and embodiments. It should be understood that the following embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the present invention. Features in the various embodiments can be combined with each other unless otherwise specified.

[0056] Example 1: Real-time PPG motion artifact suppression method in wrist-worn devices

[0057] In this embodiment, the wearable device is a wrist-worn device, which includes at least one PPG sensor, a red LED, an infrared LED, an optional green LED, a photodetector, a triaxial accelerometer, a processor, and a memory. The PPG sampling rate can be from 25 Hz to 256 Hz, and the ACC sampling rate can be the same as the PPG or higher and resampled for alignment.

[0058] When acquiring PPG and ACC signals, the device records a timestamp for each sampling block. If the PPG and ACC signals come from different sampling clocks, the processor estimates the clock drift based on the timestamp difference and cross-correlation within the window, and maps them to a unified time axis through linear interpolation, spline interpolation, or multiphase resampling. The time-aligned signal is divided into a sliding window with a length of 8 seconds and a step size of 2 seconds. The window length and step size can be adjusted according to application requirements.

[0059] For each window, the processor first removes DC drift and anomalous spikes from the PPG, then performs bandpass processing within the physiologically feasible frequency band. For ACC, the processor calculates the triaxial module lengths:

[0060] a_mag(n) = sqrt(ax(n)^2 + ay(n)^2 + az(n)^2)

[0061] And calculate jerk:

[0062] j(n) = a_mag(n) - a_mag(n-1)

[0063] The processor also estimates the gravity direction g(n) based on the triaxial acceleration after low-pass processing and calculates the change in the gravity direction angle within the window as an indicator of attitude change. When the jerk peak exceeds the threshold or the attitude angle change exceeds the threshold, it indicates that the device or limb has undergone a rapid state change, and the PPG contact pressure and optical path may change abruptly.

[0064] Subsequently, the processor calculates the frequency domain characteristics of PPG and ACC. The processor performs Fast Fourier Transform or Autoregressive Spectral Estimation on the PPG and ACC moduli to obtain the main peaks of PPG and ACC within the candidate heart rate band. If the frequency difference between the two main peaks is less than a preset frequency difference threshold, or if the coherence between PPG and ACC within the candidate heart rate band is greater than a preset coherence threshold, then a risk of spectral overlap is identified. This risk indicates that the exercise cycle may be confused with the actual heartbeat cycle.

[0065] When the window risk is low, the processor can retain the original PPG preprocessing results and proceed to quality evaluation. When the window risk meets the medium-high risk criteria, the processor performs adaptive filtering. Taking LMS as an example, the reference input vector x(n) may include ax(n), ay(n), az(n), a_mag(n), j(n) and several delay terms, and the filter output is:

[0066] m(n) = w(n)^T x(n)

[0067] s(n) = y(n) - m(n)

[0068] w(n+1) = w(n) + mu * s(n) * x(n)

[0069] Where y(n) is the PPG observation signal, m(n) is the estimated motion artifact, s(n) is the first suppression signal, w(n) is the filter weight, and mu is the step size. If the motion state changes rapidly, the processor can use RLS to converge faster; if the system has established a state-space model, Kalman filtering can be used to jointly estimate the physiological pulsation amplitude, baseline, and motion artifact as state variables.

[0070] After adaptive filtering, the processor constructs the observation matrix X. For devices with red and infrared channels, X can include the red PPG, infrared PPG, first suppression signal, ACC principal components, and jerk sequence. The processor performs blind source separation using ICA, PCA, or CCA to obtain multiple candidate components. If ICA is used, the processor searches for statistically independent components; if PCA is used, the processor decomposes along the variance direction and removes motion-dominant components; if CCA is used, the processor can use the delayed embedded signal to find components with periodic stability.

[0071] The processor calculates the correlation with ACC, spectral concentration, candidate heart rate stability, and pulsatility stability for each candidate component. Components highly correlated with ACC are labeled as candidate motion artifact components. Components with physiologically feasible dominant frequencies, low correlation with ACC, continuity in adjacent windows, and low morphological residuals are selected as candidate physiological pulsatility components.

[0072] For the red and infrared channels, the processor calculates their coherence functions and phase differences near the candidate heart rate frequencies. Physiological beats should generally exhibit correlated rhythms in both the red and infrared channels, although the amplitudes may differ. If a candidate component appears only in a single channel and is not supported in the other, the processor reduces its reliability.

[0073] The processor also maintains a dynamic template beat. The template beat consists of single-beat waveforms from several recent high-quality windows. For the current candidate physiological beat component, the processor performs peak localization, extracts beat segments around each peak, and performs phase alignment and amplitude normalization on the beat segments. The mean square error, morphological correlation, or dynamic time warping distance between the current beat and the template beat constitutes the template beat residual. A smaller residual indicates that the morphology is close to historically reliable beats, while a larger residual indicates that artifacts or peak mislocalization may still exist.

[0074] Finally, the processor determines the window category based on rules or a machine learning classifier. Example rules are as follows:

[0075] If the motion risk is low or moderate, the adaptive filtering converges, the blind source separation candidate physiological components are stable, the red light / infrared coherence is high and the template beat residual is low, then it is marked as a recoverable fragment.

[0076] If the candidate rhythm position is stable, but the template beat residual is high, the red / infrared amplitude relationship is unstable, or the waveform shape is insufficient to support amplitude correlation calculation, it is marked as a rhythm-localizable segment.

[0077] If the motion intensity is extremely high, the jerk is violent, the attitude change is significant, the spectrum overlap cannot be separated, the blind source separation result is unstable, the red light / infrared light is inconsistent, or the template beat residual is too large, it is marked as an unrecoverable fragment.

[0078] For recoverable fragments, the system outputs corrected PPG, pulse peak position, fragment quality score, and available uses. For fragments that can only be rhythmically localized, the system outputs the pulse peak position or rhythmic event, but disables or downweights the uses of blood oxygen, HRV, and morphological parameters. For unrecoverable fragments, the system marks the fragment as an unrecoverable artifact in the artifact mask and sends a rejection or low-weight instruction to the downstream parameter estimation module.

[0079] Example 2: System Structure

[0080] As shown in Figure 7, this embodiment provides a wearable PPG motion artifact suppression system that integrates acceleration motion reference and blind source separation, including a synchronous acquisition module, a motion risk determination module, an adaptive filtering module, a blind source separation module, a quality evaluation module, and a segment classification and output module.

[0081] The synchronous acquisition module receives signals from the PPG and ACC sensors and performs timestamp correction, drift estimation, and resampling alignment. The module outputs PPG and ACC sequences on a unified timeline.

[0082] The motion risk assessment module is used to calculate motion intensity, jerk, posture change, and spectral overlap indices. This module outputs not only the total risk score but also the cause code, such as high motion intensity, sudden increase in jerk, posture change, and spectral overlap.

[0083] The adaptive filtering module is used to estimate and subtract motion artifact components based on the ACC reference. This module can select one or more implementations among LMS, RLS, and Kalman filtering, depending on the device's computational resources, the rate of change of motion state, and real-time requirements.

[0084] The blind source separation module is used to separate candidate source signals from multi-channel observations. This module can select ICA, PCA, or CCA based on the number of input channels and computational resources. When the number of channels is insufficient, the observation dimensionality can be expanded through delayed embedding, ACC derived features, or a combination of PPG before and after filtering.

[0085] The quality assessment module is used to calculate red / infrared coherence and template beat residuals. This module can also calculate signal saturation, contact failure index, peak confidence, and local signal-to-noise ratio to aid in classification.

[0086] The fragment classification and output module generates three categories of results: recoverable, rhythm-localization-only, and non-recoverable. It outputs an artifact mask, a quality cause code, and downstream applications. The artifact mask can be output at the sampling point, beat, window, or event granularity. Downstream applications can include heart rate estimation, HRV analysis, blood oxygenation estimation, rhythm localization, sleep analysis, or exercise recovery analysis.

[0087] Example 3: Device and Storage Medium

[0088] This embodiment provides an electronic device, including a processor, a memory, a PPG sensor interface, and an ACC sensor interface. The memory stores a computer program, and when the processor executes the computer program, it implements the method described in Embodiment 1. This electronic device can be a smartwatch, smart bracelet, finger clip device, ear-worn device, chest patch, medical monitoring device, mobile terminal, or edge computing gateway.

[0089] This embodiment also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, it implements the method described in Embodiment 1. The computer-readable storage medium may be a read-only memory, random access memory, flash memory, solid-state memory, memory card, or other non-transitory storage medium.

[0090] Example 4: Output Data Structure

[0091] In one implementation, the system outputs the following quality results:

[0092] {

[0093] "segmentId": "seg_0001",

[0094] "timeRangeMs": [120000, 128000],

[0095] "segmentClass": "recoverable",

[0096] "artifactMask": [0, 0, 1, 1, 0],

[0097] "qualityScore": 0.86,

[0098] "reasonCodes": ["SPECTRAL_OVERLAP", "MOTION_REFERENCE_REMOVED"],

[0099] "usableFor": ["HEART_RATE", "RHYTHM_LOCALIZATION"],

[0100] "restrictedFor": ["SPO2", "HRV"],

[0101] "restoredPpgRef": "ppg_restored_seg_0001"

[0102] }

[0103] Here, `segmentClass` can be set to `recoverable`, `rhythm_only`, or `unrecoverable`, corresponding to recoverable segments, rhythm-only segments, and unrecoverable segments, respectively. `reasonCodes` is used to record quality reason codes. `usableFor` and `restrictedFor` are used to constrain downstream algorithm calls.

[0104] The data structure described above is merely an example. In actual implementations, binary structures, protocol buffers, database records, message queue events, or memory objects can be used.

[0105] Attached diagram

[0106] Figure 1: Overall Flowchart

[0107] Figure 1 illustrates the complete workflow from PPG / ACC synchronous acquisition, time alignment, windowing, motion risk assessment, adaptive filtering, blind source separation, quality evaluation, fragment classification to quality result output. The figure should highlight the cascading relationship between adaptive filtering and blind source separation, as well as the three outputs: artifact mask, quality cause code, and downstream usable applications.

[0108] Figure 2: Schematic diagram of synchronous acquisition and time alignment

[0109] Figure 2 includes the PPG multi-channel sampling axis, the ACC three-axis sampling axis, timestamp correction, drift compensation, and a unified time axis after resampling. It can be shown in the figure that PPG and ACC sampling rates are different but mapped to the same analysis window.

[0110] Figure 3: Motion Artifact Risk Assessment Module Diagram

[0111] Figure 3 illustrates four types of risk inputs: motion intensity, jerk, attitude change, and spectral overlap. Each type of input enters the feature calculation unit, and then the risk fusion unit, outputting the risk level and quality reason code. The spectral overlap branch should display the PPG main peak, ACC main peak, and coherence determination.

[0112] Figure 4: Joint processing diagram of adaptive filtering and blind source separation

[0113] Figure 4 shows that the ACC reference input enters the LMS / RLS / Kalman adaptive filter, the filter estimates motion artifacts and subtracts them from the PPG observation to obtain the first suppressed signal; the first suppressed signal, the red channel PPG signal, the infrared channel PPG signal and the ACC derived signal enter the ICA / PCA / CCA blind source separation module, and output candidate physiological pulsation components and candidate motion artifact components.

[0114] Figure 5: Evaluation of Red / Infrared Coherence and Template Impedance

[0115] Figure 5 illustrates the coherence calculation of red PPG and infrared PPG within the candidate heart rate band, as well as the residual calculation between candidate single beats and dynamic template beats. The figure should indicate how insufficient coherence and excessive template residuals affect the quality assessment.

[0116] Figure 6: Control Chart of Fragment Classification and Downstream Use

[0117] Figure 6 shows that each analysis segment is divided into three categories: recoverable, rhythm-localization-only, and non-recoverable. Recoverable segments flow to downstream modules such as heart rate, blood oxygen, and HRV; rhythm-localization-only segments flow only to the rhythm localization or heart rate coarse estimation modules; non-recoverable segments are discarded or downweighted. The figure also shows the artifact mask and reason code outputs.

[0118] Figure 7: System Module Structure Diagram

[0119] Figure 7 shows the data connection relationships between the synchronous acquisition module, motion risk determination module, adaptive filtering module, blind source separation module, quality evaluation module, and fragment classification and output module.

[0120] Figure 8: Schematic diagram of electronic device structure

[0121] Figure 8 illustrates the processor, memory, PPG sensor, ACC sensor, communication module, power module, and display or output interface. The processor executes a program in the memory to implement the method of the present invention.

Claims

1. A wearable PPG motion artifact suppression method integrating acceleration motion reference and blind source separation, characterized in that, include: Simultaneously, red channel PPG signals, infrared channel PPG signals, and triaxial acceleration ACC signals generated by wearable devices worn on the human body are acquired, and time alignment is performed on the red channel PPG signals, infrared channel PPG signals, and triaxial acceleration ACC signals. The time-aligned red channel PPG signals, infrared channel PPG signals, and triaxial acceleration ACC signals are divided into multiple analysis segments. For each analysis segment, motion intensity, acceleration jerk, attitude change index, and spectral overlap index between PPG signals and ACC signals are calculated to obtain motion artifact risk assessment results. When the motion artifact risk assessment result meets the preset risk conditions, the PPG signal is adaptively filtered using the ACC signal as the motion reference to obtain a first suppression signal; blind source separation is performed on the observation signal, including the first suppression signal, the red channel PPG signal, the infrared channel PPG signal, and the ACC derived signal, to obtain candidate physiological pulsation components and candidate motion artifact components; the candidate physiological pulsation components are quality evaluated based on the coherence between the red PPG channel and the infrared PPG channel and the template beat residual; based on the motion artifact risk assessment result and the quality evaluation, the analysis segment is divided into recoverable segments, rhythm-localizable segments only, and unrecoverable segments; unrecoverable segments are eliminated or downweighted, and a quality result including at least an artifact mask, a quality cause code, and downstream usable applications is output.

2. The method according to claim 1, characterized in that, The time alignment includes mapping the red channel PPG signal, the infrared channel PPG signal, and the ACC signal to a unified time axis based on the sampling timestamp, device clock drift estimation, cross-correlation peak, and / or interpolation resampling; wherein the analysis segment includes a sliding window segment, and adjacent analysis segments have a preset overlap rate.

3. The method according to claim 1, characterized in that, The motion intensity includes the root mean square, variance, and frequency band energy of the ACC triaxial module, or a combination thereof; the acceleration jerk is the first-order difference or derivative of acceleration with respect to time; the attitude change index is determined based on the gravity direction estimation, the low-frequency component of the triaxial acceleration, or the attitude angle change rate.

4. The method according to claim 1, characterized in that, The spectral overlap indicators include the distance between the main peak frequency of PPG and the main peak frequency of ACC, the proportion of ACC energy within the candidate heart rate band of PPG, the frequency domain coherence function of PPG and ACC, cross-spectral peak values ​​or combinations thereof; When the spectral overlap index indicates that the motion frequency falls within the candidate heart rate frequency band, the risk level of the motion artifact risk assessment result is increased.

5. The method according to claim 1, characterized in that, The adaptive filtering includes at least one of LMS filtering, RLS filtering, or Kalman filtering; wherein, the motion-related artifact components are estimated by using the ACC triaxial signal, ACC magnitude, jerk, attitude change amount, and / or its delay term as reference inputs, and the artifact components are subtracted from the PPG signal.

6. The method according to claim 1, characterized in that, The blind source separation includes at least one of Independent Component Analysis (ICA), Principal Component Analysis (PCA), or Canonical Correlation Analysis (CCA); wherein, candidate physiological pulsation components are selected from the blind source separation results based on the correlation between candidate components and ACC signals, the periodic stability of candidate components, the spectral concentration of candidate components, and the consistency of red / infrared PPG channels.

7. The method according to claim 1, characterized in that, The template beat residual is calculated by normalizing the amplitude and aligning the phase of the single beat waveform in the candidate physiological beat components with the dynamically updated beat template; When the template beat residual exceeds the first threshold and the red light / infrared coherence is lower than the second threshold, the corresponding analysis segment is marked as an unrecoverable segment or a segment that can only be rhythmically located.

8. The method according to claim 1, characterized in that, The downstream applications include at least one of heart rate estimation, heart rate variability analysis, rhythm localization, blood oxygenation estimation, and those not applicable to physiological parameter calculation; the quality cause codes include at least one of high exercise intensity, jerk spike, abrupt posture change, spectral overlap, red / infrared inconsistency, excessive template beat residual, unstable blind source separation, and signal saturation.

9. A wearable PPG motion artifact suppression system integrating acceleration motion reference and blind source separation, characterized in that, include: The synchronous acquisition module is used to acquire and time-align the PPG signal and the ACC signal; The motion risk assessment module is used to calculate motion intensity, jerk, attitude change index, and spectral overlap index, and generate motion artifact risk assessment results; the adaptive filtering module is used to perform LMS, RLS, or Kalman adaptive filtering with the ACC signal as the motion reference. The blind source separation module is used to perform ICA, PCA, or CCA blind source separation on the first suppression signal, the red light channel PPG signal, the infrared channel PPG signal, and the ACC derived signal; The quality assessment module is used to evaluate candidate physiological pulsation components based on red / infrared coherence and template beat residuals; the fragment classification and output module is used to classify the analyzed fragments into recoverable fragments, rhythm-localizable fragments, and non-recoverable fragments, and output artifact mask, quality reason code, and downstream usable uses.

10. An electronic device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and executable by the processor, wherein the processor, when executing the computer program, implements the method according to any one of claims 1 to 8.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method according to any one of claims 1 to 8.

Images