Motion noise suppression system and method based on body surface multi-sensor

Abstract

The application provides a motion noise suppression system and method based on a body surface multi-sensor, and mainly relates to the technical field of sensors. The motion noise suppression system comprises a body surface wireless transmission structure for being attached to a human body surface, the body surface wireless transmission structure comprises a plurality of functional modules arranged on a flexible substrate, the plurality of functional modules at least comprise a first module and a second module, the first module and the second module have different equivalent electromagnetic characteristics in a target working frequency band, so as to form a boundary state at a boundary region formed by the contact of the first module and the second module, and the boundary state is used for guiding the propagation of a wireless signal along the boundary region; at least two acceleration sensors are respectively used for synchronously collecting a chest position acceleration signal and an abdominal position acceleration signal in a motion state; and a signal processing unit is wirelessly coupled with the at least two acceleration sensors through a wireless transmission channel formed in the boundary region.

Description

Technical Field

[0001] This application relates to the field of sensor technology, and more specifically, to a motion noise suppression system and method based on multiple sensors on the body surface. Background Technology

[0002] With the rapid development of wearable electronics, flexible sensing and wireless communication technologies, body surface sensor networks have shown broad application prospects in fields such as vital sign monitoring, motion behavior recognition and human-computer interaction.

[0003] However, existing wireless body surface sensor networks mostly adopt a monolithic, customized structure with a fixed network topology, making rapid reconfiguration difficult to adapt to changing needs. Furthermore, mechanical disturbances generated by the human body during dynamic movement can superimpose motion artifacts onto physiological signals. Traditional signal processing methods struggle to effectively suppress noise under high-intensity exercise, leading to decreased accuracy in monitoring parameters such as heart rate and respiration. Summary of the Invention

[0004] In view of this, this application provides a motion noise suppression system and method based on multiple sensors on the body surface.

[0005] One aspect of this application provides a motion noise suppression system based on multiple sensors on the body surface, wherein the system includes: a wireless transmission structure for adhering to the human body surface, the wireless transmission structure including multiple functional modules disposed on a flexible substrate, the multiple functional modules including at least a first module and a second module, the first module and the second module having different equivalent electromagnetic characteristics in a target operating frequency band to form a boundary state at the boundary region formed by the contact between the first module and the second module, the boundary state being used to guide the wireless signal to propagate along the boundary region; at least two accelerometers, respectively used to synchronously acquire chest position acceleration signals and abdominal position acceleration signals during motion; and a signal... The signal processing unit is wirelessly coupled to the at least two accelerometers via a wireless transmission channel formed in the boundary region. The signal processing unit is configured to: input the chest position acceleration signal and the abdominal position acceleration signal into a first adaptive filter; use the abdominal position acceleration signal as a reference signal to perform noise estimation on the motion noise component in the chest position acceleration signal to obtain a motion noise estimation signal; input the chest position acceleration signal and the motion noise estimation signal into a second adaptive filter; use the motion noise estimation signal as a reference signal to perform noise cancellation processing on the chest position acceleration signal to obtain a physiological signal after motion noise suppression, wherein the physiological signal includes a heartbeat signal.

[0006] According to an embodiment of this application, the system further includes: a third accelerometer for synchronously acquiring lower abdominal position acceleration signals during motion; the signal processing unit is wirelessly coupled to the third accelerometer via the wireless transmission channel, and the signal processing unit is further configured to: input the abdominal position acceleration signal and the lower abdominal position acceleration signal into a third adaptive filter, using the lower abdominal position acceleration signal as a reference signal, perform noise estimation on the motion noise component in the abdominal position acceleration signal to obtain a second motion noise estimation signal; input the abdominal position acceleration signal and the second motion noise estimation signal into a fourth adaptive filter, using the second motion noise estimation signal as a reference signal, perform noise cancellation processing on the abdominal position acceleration signal to obtain the physiological signal after suppressing motion noise, wherein the physiological signal includes a respiratory signal.

[0007] According to the embodiments of this application, the first adaptive filter, the second adaptive filter, the third adaptive filter and the fourth adaptive filter are all recursive least squares adaptive filters.

[0008] According to an embodiment of this application, the signal processing unit is further configured to: divide the physiological signal after suppressing motion noise into multiple consecutive time windows based on a preset window step size; perform frequency domain transformation on the signal within each time window, and calculate the signal power in a first signal frequency band and the noise power in a second signal frequency band based on the frequency domain transformation result; wherein, the first signal frequency band is the frequency band range where the heartbeat signal is located, and the second signal frequency band is the frequency band range that does not include the first signal frequency band; calculate the signal-to-noise ratio (SNR) within each time window based on the signal power and the noise power; obtain the target motion state based on the comparison result of the average SNR of the multiple consecutive time windows and multiple motion state thresholds, wherein the target motion state includes at least a stationary state, a walking state, a jogging state, and a running state; and dynamically adjust the parameters of the first adaptive filter and the second adaptive filter according to the target motion state so that the first adaptive filter and the second adaptive filter adapt to the noise characteristics of the motion state.

[0009] According to an embodiment of this application, the signal processing unit is further configured to: identify the target motion state as the stationary state when the average signal-to-noise ratio is greater than or equal to a first motion state threshold; identify the target motion state as the walking state when the average signal-to-noise ratio is less than the first motion state threshold and greater than or equal to a second motion state threshold; identify the target motion state as the jogging state when the average signal-to-noise ratio is less than the second motion state threshold and greater than or equal to a third motion state threshold; and identify the target motion state as the running state when the average signal-to-noise ratio is less than the third motion state threshold.

[0010] According to an embodiment of this application, the signal processing unit is further configured to: increase the filtering length of the first adaptive filter and the second adaptive filter when the target motion state is the sprinting state or the jogging state; decrease the filtering length of the first adaptive filter and the second adaptive filter when the target motion state is the walking state; and directly output the physiological signal after suppressing motion noise when the target motion state is the stationary state.

[0011] According to the embodiments of this application, both the first module and the second module include periodically arranged conductive units; the conductive units are one of square metal patches, circular metal patches, or ring structures; the conductive units of the first module and the conductive units of the second module have different geometric parameters, so that the first module and the second module form the boundary state at the boundary region.

[0012] According to an embodiment of this application, the first module and the second module are arranged in different ways in the plane, so as to form at least one boundary state at the junction of the first module and the second module.

[0013] According to embodiments of this application, the flexible substrate is a fabric substrate or a polymer film substrate, and the plurality of functional modules are fixed on the flexible substrate based on at least one of printing, laser cutting or hot pressing bonding methods.

[0014] Another aspect of this application provides a motion noise suppression method based on multiple body surface sensors, applied to the aforementioned system. The method includes: acquiring a chest position acceleration signal and an abdominal position acceleration signal; inputting the chest position acceleration signal and the abdominal position acceleration signal into a first adaptive filter, using the abdominal position acceleration signal as a reference signal, and performing noise estimation on the motion noise component in the chest position acceleration signal to obtain a motion noise estimation signal; inputting the chest position acceleration signal and the motion noise estimation signal into a second adaptive filter, using the motion noise estimation signal as a reference signal, and performing noise cancellation processing on the chest position acceleration signal to obtain a physiological signal after motion noise suppression, wherein the physiological signal includes a heartbeat signal.

[0015] The embodiments of this application utilize boundary states formed between multiple functional modules at the physical layer to construct a low-loss, directional wireless transmission channel on the body surface. This frees data transmission between sensor nodes and the signal processing unit from dependence on free-space wireless links, reduces the absorption and scattering of wireless signals by human tissue, and ensures communication stability during movement. Furthermore, at the signal processing layer, the embodiments of this application employ a collaborative configuration of chest and abdominal accelerometers and a cascaded architecture of two-step adaptive filtering. Using the abdominal signal as a noise reference, they accurately estimate and eliminate motion noise components in the chest signal, thereby accurately extracting the heartbeat signal from a strong noise background. This solves the problem of signal distortion caused by motion artifacts in traditional wearable devices under dynamic conditions, providing fundamental hardware and algorithmic support for physiological signal monitoring during movement. Attached Figure Description

[0016] The above and other objects, features and advantages of this application will become clearer from the following description of embodiments with reference to the accompanying drawings:

[0017] Figure 1 A schematic diagram of a motion noise suppression system based on multiple body surface sensors according to an embodiment of this application is shown.

[0018] Figure 2 This illustration schematically shows a signal processing flow diagram of a motion noise suppression system according to an embodiment of the present application;

[0019] Figure 3 A schematic diagram of a motion noise suppression system based on multiple body surface sensors according to another embodiment of this application is shown.

[0020] Figure 4 This illustration schematically shows a signal processing flow diagram of a motion noise suppression system according to another embodiment of this application;

[0021] Figure 5This schematically illustrates a flowchart of the synchronous data acquisition and processing of the heartbeat signal extraction channel and the respiratory signal extraction channel according to an embodiment of this application;

[0022] Figure 6 This schematic diagram illustrates the topological patterns of the first module and the second module according to a specific embodiment of this application;

[0023] Figure 7 This schematic diagram illustrates the boundary state formed between the first module and the second module according to a specific embodiment of this application.

[0024] Figure 8 The diagram illustrates the number of wireless propagation channels corresponding to different permutations and combinations according to specific embodiments of this application;

[0025] Figure 9 A flowchart illustrating a motion noise suppression method according to an embodiment of this application is shown schematically. Detailed Implementation

[0026] The embodiments of this application will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of this application. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of this application for ease of explanation. However, it will be apparent that one or more embodiments may be implemented without these specific details. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concepts of this application.

[0027] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0028] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0029] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0030] With the rapid development of wearable electronics, flexible sensing, and wireless communication technologies, surface sensor networks are showing broad application prospects in fields such as vital sign monitoring, motion behavior recognition, and human-computer interaction. By distributing multiple sensor nodes on the human body surface and achieving wireless interconnection, the collaborative acquisition and transmission of multi-source physiological information can be realized.

[0031] In the body surface environment, human tissue exhibits significant absorption and scattering effects on radio electromagnetic waves, leading to high link loss and poor stability in traditional free-space wireless communication methods. Currently, most studies in this field utilize artificial electromagnetic structures or metamaterial fabrics to construct wireless transmission paths on the body surface to verify the feasibility of body surface sensor networks. However, existing wireless body surface sensor networks often employ monolithic or highly customized structures, with their network topology essentially fixed after design, making rapid reconfiguration difficult based on changes in the number of sensors, their placement, or application requirements.

[0032] Furthermore, under dynamic usage conditions, human movement generates complex mechanical disturbances to the sensor, superimposing significant motion artifacts onto the collected physiological signals. Existing signal processing methods struggle to effectively suppress motion noise during high-intensity activities such as walking and running, leading to a significant decrease in the monitoring accuracy of key physiological parameters such as heart rate and respiration, failing to meet the needs of continuous and stable monitoring in everyday activities.

[0033] In view of this, the motion noise suppression system and method based on multiple surface sensors proposed in this application constructs a low-loss, directional wireless transmission channel on the body surface at the physical layer using boundary states formed between multiple functional modules. This frees data transmission between sensor nodes and signal processing units from dependence on free-space wireless links, reduces the absorption and scattering of wireless signals by human tissue, and ensures communication stability during motion. Furthermore, the embodiments of this application, at the signal processing layer, utilize the coordinated configuration of chest and abdominal accelerometers and a cascaded architecture of two-step adaptive filtering. By using abdominal signals as a noise reference, motion noise components in chest signals are accurately estimated and eliminated, thereby accurately extracting heartbeat signals from a strong noise background. This solves the problem of signal distortion caused by motion artifacts in traditional wearable devices under dynamic conditions, providing basic hardware and algorithmic support for physiological signal monitoring during motion.

[0034] Specifically, embodiments of this application provide a motion noise suppression system based on multiple surface sensors, comprising: a surface wireless transmission structure for adhering to the surface of a human body, the surface wireless transmission structure including multiple functional modules disposed on a flexible substrate, the multiple functional modules including at least a first module and a second module, the first module and the second module having different equivalent electromagnetic characteristics in a target operating frequency band to form a boundary state at the boundary region formed by the contact between the first module and the second module, the boundary state being used to guide the wireless signal to propagate along the boundary region; and at least two accelerometers, respectively used to synchronously acquire chest position acceleration signals and abdominal position acceleration signals during motion. The signal processing unit is wirelessly coupled to at least two accelerometers via a wireless transmission channel formed in the boundary region. The signal processing unit is configured to: input the chest position acceleration signal and the abdominal position acceleration signal into a first adaptive filter, using the abdominal position acceleration signal as a reference signal, perform noise estimation on the motion noise component in the chest position acceleration signal to obtain a motion noise estimation signal; input the chest position acceleration signal and the motion noise estimation signal into a second adaptive filter, using the motion noise estimation signal as a reference signal, perform noise cancellation processing on the chest position acceleration signal to obtain a physiological signal after motion noise suppression, the physiological signal including a heartbeat signal.

[0035] It should be noted that the motion noise suppression system and method based on multiple body surface sensors as defined in the embodiments of this application can be used in the field of sensor technology. The motion noise suppression system and method based on multiple body surface sensors as defined in the embodiments of this application can also be used in any field other than sensor technology, such as wearable sensing devices, physiological signal monitoring, and human-computer interaction technology. The application fields of the motion noise suppression system and method based on multiple body surface sensors as defined in the embodiments of this application are not limited.

[0036] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with specific embodiments and the accompanying drawings.

[0037] Figure 1 A schematic diagram of a motion noise suppression system based on multiple body surface sensors according to an embodiment of this application is shown.

[0038] like Figure 1 As shown, the motion noise suppression system includes a body surface wireless transmission structure 110, at least two acceleration sensors such as a first acceleration sensor 121 and a second acceleration sensor 122, and a signal processing unit 130.

[0039] The body surface wireless transmission structure 110 can represent a physical structure that can conform to the human body surface, such as the chest or abdomen, to provide a low-loss, directional propagation path for wireless signals in a body surface environment. The body surface wireless transmission structure 110 can serve as the physical carrier for wireless communication between sensor nodes, including a flexible substrate and multiple functional modules.

[0040] The flexible substrate refers to the underlying support material in the wireless transmission structure 110 on the body surface. It possesses mechanical flexibility and biocompatibility, enabling it to adapt to the curved shape of the human body surface and maintain stable contact with the human body surface, such as skin, during movement. The flexible substrate can be a fabric substrate, a polymer film substrate, or other flexible media materials.

[0041] Multiple functional modules refer to multiple structural units with specific electromagnetic properties disposed on a flexible substrate. These modules are not ordinary wires or circuits, but artificial electromagnetic structures composed of periodically arranged conductive units, used to construct the physical path for wireless signal transmission. In one example, multiple functional modules can be fixed to the flexible substrate based on at least one of the following methods: printing, laser cutting, or thermoforming.

[0042] Multiple functional modules include at least two functional modules with different electromagnetic properties, such as the first module 111 and the second module 112. This difference stems from the two modules having different topological phases or different topological invariants, resulting in different equivalent electromagnetic properties (such as equivalent impedance, dielectric constant, or permeability) within the target operating frequency band. By adjusting the geometric parameters or arrangement of the conductive units within the module, its topological phase can be controlled, thereby constructing module regions with different electromagnetic properties on the same flexible substrate. The target operating frequency band can represent a pre-configured wireless communication frequency range. In one example, the target operating frequency band can be configured as the Bluetooth communication band 2.4GHz-2.48GHz.

[0043] The boundary region can represent the interface formed by the first module and the second module in contact with each other or arranged adjacently on the flexible substrate. Due to the topological phase difference between the first module 111 and the second module 112, a topological interface state, such as a boundary state, will be formed in this interface region.

[0044] A boundary state can represent an electromagnetic wave propagation state. Physically, it manifests as the suppression of wireless signal propagation perpendicular to the boundary, while a low-loss directional propagation path is formed along the boundary. The key condition for efficient transmission in this boundary state is that the module units on both sides of the boundary satisfy a specific matching relationship, thereby ensuring the stability of mode coupling. This boundary state allows wireless signals to be localized and guided to propagate within the boundary region, avoiding ineffective radiation into free space or human tissue.

[0045] A wireless transmission channel can represent a physical path formed by a boundary region, used to guide the directional propagation of wireless signals. Specifically, such as... Figure 1 As shown, a wireless transmission channel can refer to a continuous, closed, or through boundary path formed by the alternating arrangement of the first module 111 and the second module 112, where each boundary segment can guide the continuous propagation of wireless signals.

[0046] like Figure 1 As shown, the first accelerometer 121 and the second accelerometer 122 are arranged near the boundary region and wirelessly coupled to the signal processing unit 130 through this wireless transmission channel. Specifically, the signals collected by the sensor nodes enter the boundary region through near-field electromagnetic coupling and are directionally transmitted to the signal processing unit 130 along the boundary state-guided path, rather than relying on a traditional free-space wireless link.

[0047] The first accelerometer 121 is used to collect acceleration signals at the chest position, and the second accelerometer 122 is used to collect acceleration signals at the abdominal position. The chest position corresponds to the thoracic region of the human body, where the acceleration signal contains a mixture of cardiac impulses, respiratory movements, and noise components caused by overall body movement. The abdominal position corresponds to the abdominal region of the human body, where the acceleration signal is primarily composed of respiratory movements and motion noise, with a relatively weak cardiac impulse component.

[0048] In one example, at least two accelerometers are configured to synchronously acquire acceleration signals, ensuring precise time alignment between the reference and target signals in adaptive filtering. Synchronous acquisition is achieved through a hardware triggering mechanism: the signal processing unit 130 broadcasts a synchronization beacon to each sensor node via a wireless transmission channel. Upon receiving the synchronization beacon, each sensor node begins sampling at the same time reference. The sampling clock is calibrated with the synchronization beacon using a local crystal oscillator, ensuring that the sampling time error of each channel is less than 1 millisecond.

[0049] The signal processing unit 130 can represent an electronic processing module with signal receiving, storage, and computing capabilities. The signal processing unit 130 is wirelessly coupled to at least two accelerometers via a wireless transmission channel formed by the aforementioned boundary region, enabling the acceleration signals synchronously acquired by the at least two accelerometers to be wirelessly transmitted to the signal processing unit 130. In one example, the signal processing unit 130 is not limited to, but includes, a microcontroller, a digital signal processor, or an embedded processing chip.

[0050] Both the first and second adaptive filters refer to adaptive filtering algorithm modules that can automatically adjust the filter coefficients based on the characteristics of the input signal. In one example, both the first and second adaptive filters employ recursive least squares adaptive filters, which have a faster convergence speed and smaller steady-state error, making them suitable for scenarios where noise characteristics change rapidly under motion. The two filters can be configured with parameters independently; for example, the order of both the first and second adaptive filters can be set to 64, and the forgetting factor can be set to 0.998.

[0051] In one example, the motion noise estimation signal can represent the estimate of the motion noise component contained in the chest position acceleration signal, output by the first adaptive filter. Ideally, this motion noise estimation signal should closely match the actual motion noise in the chest position acceleration signal.

[0052] In this embodiment, the physiological signal after suppressing motion noise can represent the signal output after processing by the second adaptive filter. The motion noise component in this signal has been significantly eliminated, mainly including the heartbeat signal. This heartbeat signal can be further used for extracting physiological parameters such as heart rate calculation and heartbeat interval analysis.

[0053] Figure 2 The schematic diagram illustrates the signal processing flow of a motion noise suppression system according to an embodiment of this application.

[0054] like Figure 2 As shown, in this embodiment, the first adaptive filter 203 is configured to estimate the motion noise component in the chest position acceleration signal 202 using the abdominal position acceleration signal 201 as a reference signal. Through an adaptive filtering algorithm, the output of the first adaptive filter approximates the motion noise component in the chest position acceleration signal, thereby obtaining the motion noise estimation signal 204.

[0055] In this embodiment, since the motion noise characteristics in the abdominal position acceleration signal 201 and the chest position acceleration signal 202 are highly correlated, but the abdominal position acceleration signal 201 lacks the heartbeat impact component, the abdominal position acceleration signal 201 can be used as an ideal noise reference signal.

[0056] In this embodiment, the second adaptive filter 205 is configured to use the motion noise estimation signal 204 output by the first adaptive filter 203 as a reference signal to perform noise cancellation processing on the original chest position acceleration signal 202. Specifically, the chest position acceleration signal 202 is subtracted from the motion noise estimation signal 204 to extract the pure heartbeat signal component, which is then used as the heartbeat signal 206 after suppressing motion noise.

[0057] Based on this, embodiments of this application utilize boundary states formed between multiple functional modules at the physical layer to construct a low-loss, directional wireless transmission channel on the body surface. This frees data transmission between sensor nodes and the signal processing unit from dependence on free-space wireless links, reduces the absorption and scattering of wireless signals by human tissue, and ensures communication stability during movement. Furthermore, embodiments of this application, at the signal processing layer, employ a collaborative configuration of chest and abdominal accelerometers and a cascaded architecture of two-step adaptive filtering. Using the abdominal signal as a noise reference, they accurately estimate and eliminate motion noise components in the chest signal, thereby accurately extracting the heartbeat signal from a strong noise background. This solves the problem of signal distortion caused by motion artifacts in traditional wearable devices under dynamic conditions, providing fundamental hardware and algorithmic support for physiological signal monitoring during movement.

[0058] According to an embodiment of this application, the motion noise suppression system further includes: a third accelerometer for synchronously acquiring the lower abdominal position acceleration signal during motion; a signal processing unit wirelessly coupled to the third accelerometer via a wireless transmission channel, and the signal processing unit is further configured to: input the abdominal position acceleration signal and the lower abdominal position acceleration signal into a third adaptive filter, use the lower abdominal position acceleration signal as a reference signal to perform noise estimation on the motion noise component in the abdominal position acceleration signal to obtain a second motion noise estimation signal; input the abdominal position acceleration signal and the second motion noise estimation signal into a fourth adaptive filter, use the second motion noise estimation signal as a reference signal to perform noise cancellation processing on the abdominal position acceleration signal to obtain a physiological signal after motion noise suppression, the physiological signal including a respiratory signal.

[0059] Figure 3 A schematic diagram of a motion noise suppression system based on multiple body surface sensors according to another embodiment of this application is shown.

[0060] like Figure 3 As shown, in another embodiment, the motion noise suppression system further includes a third accelerometer 123, which is configured to synchronously acquire acceleration signals of the lower abdomen position during motion. The lower abdomen position corresponds to the lower abdominal region of the human body surface.

[0061] The third accelerometer 123 can be located in the lower abdominal region, below the abdomen and above the pubic symphysis. Since the mechanical vibrations caused by abdominal breathing are still quite noticeable in the lower abdomen, while the mechanical vibrations transmitted to the body surface by the heartbeat are extremely weak in the lower abdomen after attenuation by the abdominal tissues, the acceleration signal in the lower abdomen is mainly composed of abdominal breathing motion and whole-body motion noise, with a very low cardiac impact component, making it a cleaner reference signal for extracting respiratory signals.

[0062] In this embodiment, to achieve coordinated processing of signals from three accelerometers (chest, abdomen, and lower abdomen), the system employs a hardware-triggered synchronization mechanism. Specifically, the signal processing unit broadcasts a synchronization beacon to each sensor node via a wireless transmission channel. Upon receiving the synchronization beacon, each sensor node begins sampling using the same time reference. The sampling clock is calibrated with the synchronization beacon using a local crystal oscillator to ensure that the sampling time error of each channel is less than 1 millisecond. In one example, a higher-precision synchronization method can also be used, such as distributed clock synchronization via a flexible metamaterial transmission channel, further reducing synchronization errors by utilizing the low-latency characteristics of boundary state transmission. Time synchronization is fundamental to multi-sensor coordinated processing, ensuring precise time alignment between the reference signal and the target signal in adaptive filtering.

[0063] In this embodiment, wireless communication between the signal processing unit and the third accelerometer is achieved through a wireless transmission channel formed by the boundary region. Specifically, the acceleration signal of the lower abdomen position collected by the third accelerometer is transmitted to the signal processing unit via the wireless transmission channel, rather than relying on a traditional free-space wireless link. This ensures that all sensor nodes in the entire system, such as the first accelerometer, the second accelerometer, and the third accelerometer, share the same low-loss, interference-resistant surface wireless transmission channel, simplifying the system architecture and improving communication reliability.

[0064] The third and fourth adaptive filters can also refer to adaptive filtering algorithm modules that can automatically adjust the filter coefficients according to the characteristics of the input signal.

[0065] In this embodiment, the physiological signals after suppressing motion noise may also include respiratory signals.

[0066] Figure 4 A schematic diagram illustrating the signal processing flow of a motion noise suppression system according to another embodiment of this application is shown.

[0067] like Figure 4 As shown, in this embodiment, the third adaptive filter 403 is configured to use the lower abdominal position acceleration signal 401 as a reference signal to estimate the motion noise component in the abdominal position acceleration signal 402. Specifically, the noise component characteristics caused by whole-body motion in the lower abdominal position acceleration signal 401 and the abdominal position acceleration signal 402 are highly correlated, and the central impact component of the lower abdominal position acceleration signal 401 is extremely weak, thus it can be used as an ideal reference signal for the motion noise of the abdominal position acceleration signal 402. Through an adaptive filtering algorithm, the third adaptive filter 403 outputs a signal that approximates the true motion noise component in the abdominal position acceleration signal, thereby obtaining the second motion noise estimation signal 404.

[0068] In this embodiment, the fourth adaptive filter 405 is configured to use the second motion noise estimation signal 404 output by the third adaptive filter 403 as a reference signal to perform noise cancellation processing on the original abdominal position acceleration signal 402. Specifically, the abdominal position acceleration signal 402 is subtracted from the second motion noise estimation signal 404 to extract the pure respiratory signal component, which is then used as the noise-suppressed respiratory signal 406. This respiratory signal can be further used for physiological parameter extraction such as respiratory rate calculation and respiratory rhythm analysis.

[0069] In this embodiment, the estimation accuracy of the second motion noise estimation signal can be configured based on parameters such as the filter order and forgetting factor. Since the dominant frequency of the respiratory signal (e.g., 0.2Hz-0.5Hz) is significantly lower than that of the heartbeat signal, and the amplitude of the acceleration signal caused by respiratory motion is usually greater than that of the cardiac impact signal, the filter parameters of the respiratory signal extraction channel can be configured independently of the heartbeat signal extraction channel. In one example, the third and fourth adaptive filters for respiratory signal extraction employ a recursive least squares algorithm, and their order can be set to 64 (at a sampling rate of 256Hz), lower than the order of the heartbeat channel (e.g., 128), because the transmission path of respiratory noise is relatively simple; the forgetting factor can be set to 0.995, slightly lower than the 0.998 of the heartbeat channel, to accommodate the slower change characteristics of the respiratory signal. These parameters can be optimized through experimental calibration to obtain the optimal signal-to-noise ratio for respiratory signal extraction.

[0070] In this embodiment, the heartbeat signal extraction channels, such as the first and second adaptive filters, and the respiratory signal extraction channels, such as the third and fourth adaptive filters, operate independently in parallel within the signal processing unit. The two channels share the motion state recognition results, avoiding redundant calculations. When the heartbeat signal quality is extremely poor, leading to unreliable state recognition—for example, when the average signal-to-noise ratio is consistently below the third motion state threshold and the correlation between the heartbeat signal and the standard ECG signal is below 0.5—the system can switch to state recognition based on the respiratory signal signal-to-noise ratio. The calculation method for the respiratory signal signal-to-noise ratio is similar to that of the heartbeat signal, but the first signal frequency band is adjusted accordingly to the respiratory signal frequency band, such as 0.2Hz-0.5Hz. This dual-channel redundancy mechanism ensures that the system can maintain basic state recognition and filter parameter adjustment capabilities even when the signal quality of either channel deteriorates.

[0071] In this embodiment, since the human motion state (such as being stationary, walking, or running) changes rapidly, the statistical characteristics of motion noise also change rapidly. Therefore, the first, second, third, and fourth adaptive filters can all use recursive least squares adaptive filters to track noise changes in real time, thereby obtaining more accurate motion noise estimation results.

[0072] Based on this, the embodiments of this application, by adding a lower abdominal accelerometer and employing a two-step adaptive filtering architecture, construct an independent respiratory signal extraction channel, significantly improving the extraction accuracy of respiratory signals under dynamic conditions. This configuration fully utilizes the characteristics of lower abdominal signals—sensitivity to motion noise and insensitivity to cardiac impact—making it an ideal noise reference source when extracting abdominal respiratory signals. A hardware-triggered synchronization mechanism ensures precise alignment of multi-sensor signals, providing a reliable input basis for adaptive filtering. Independently configured filter parameters enable optimal performance for both respiratory and heartbeat signal extraction. A dual-channel redundancy mechanism enhances the system's robustness, maintaining basic functionality even when the signal quality of either channel deteriorates. Thus, the system can synchronously separate and extract high signal-to-noise ratio heartbeat and respiratory signals during motion, achieving coordinated monitoring of cardiopulmonary physiological signals.

[0073] According to an embodiment of this application, the signal processing unit is further configured to: divide the physiological signal after suppressing motion noise into multiple consecutive time windows based on a preset window step size; perform frequency domain transformation on the signal in each time window, and calculate the signal power in the first signal frequency band and the noise power in the second signal frequency band based on the frequency domain transformation result; calculate the signal-to-noise ratio in each time window based on the signal power and noise power; obtain the target motion state based on the comparison result of the average signal-to-noise ratio of multiple consecutive time windows and multiple motion state thresholds; and dynamically adjust the parameters of the first adaptive filter and the second adaptive filter according to the target motion state so that the first adaptive filter and the second adaptive filter adapt to the noise characteristics in the motion state.

[0074] In this embodiment, the signal processing unit can also perform motion state recognition and dynamic adjustment of filter parameters.

[0075] In this embodiment, physiological signals such as heartbeat or respiration signals, after suppressing motion noise, are divided into multiple continuous and non-overlapping or partially overlapping time windows according to a preset step size. Each window is an independent signal processing unit used to evaluate the dynamic changes in signal quality over different time periods, avoiding misjudgments of motion state caused by data bias in a single window. In one example, the preset step size can be adjusted according to actual monitoring needs to ensure both the capture of the temporal characteristics of the signal and real-time processing; for example, it can be configured to 30 seconds.

[0076] In this embodiment, the signal within each time window undergoes a frequency domain transformation to convert the time-domain signal into a frequency-domain signal, obtaining a frequency domain transformation result to facilitate the distinction between the frequency band distribution of the signal and noise. In one example, the frequency domain transformation method is not limited to Fast Fourier Transform (FFT).

[0077] The frequency domain transform result represents the frequency domain representation obtained after performing a Fourier transform on the time-domain signal within each time window. In one example, when considering a time window signal containing N sampling points... After performing a Fast Fourier Transform, a complex sequence is obtained. , where k corresponds to a discrete frequency point. The modulus of this complex sequence is... This indicates the amplitude of the frequency component. This represents the power of the frequency component. The mapping relationship between the discrete frequency point index k and the physical frequency f is as follows: Where fs is the sampling rate (in Hz) and N is the number of discrete frequency points in the FFT. The result of frequency domain transformation is essentially mapping the signal from the time domain to the frequency domain, so that the frequency components of the signal can be explicitly presented, making it easier to distinguish the energy distribution of different frequency bands.

[0078] The first signal frequency band represents the frequency range in which the heartbeat signal is located. In one example, for a human heart rate signal, the first signal frequency band can be between 0.8 Hz and 3.0 Hz, corresponding to between 48 beats / minute and 180 beats / minute. Signals within the first signal frequency band can be used as target physiological signals, and the signal strength of these target physiological signals can be expressed as signal power.

[0079] The second signal band refers to a frequency range other than the first signal band. In one example, the second signal band may include frequency bands below 0.8 Hz or above 3.0 Hz. Signals within the second signal band can serve as interference signals, such as motion artifacts, respiratory interference, or other non-target physiological signals, and the signal strength of such interference signals can be expressed as noise power.

[0080] In this embodiment, the signal power in the first signal frequency band and the noise power in the second signal frequency band are calculated based on the frequency domain transformation results, providing a basis for signal-to-noise ratio calculation.

[0081] In one specific implementation, in the frequency domain transformation result, a discrete frequency point index range corresponding to the first signal frequency band, such as the frequency band where the heartbeat signal is located, is located. The power values ​​of all frequency points within this range are accumulated or integrated. When the frequency band boundary is not aligned with the discrete frequency points, the boundary index can be determined by rounding down the nearest frequency point, or the contribution ratio of the boundary frequency points can be calculated by interpolation. The result obtained is the power estimate of the target physiological signal within this time window. The specific calculation formula is as follows:

[0082] (1);

[0083] Among them, K signal This represents the set of discrete frequency point indices that fall within the first signal frequency band.

[0084] In this specific implementation, in the frequency domain transformation result, the discrete frequency point index range corresponding to the second signal frequency band (excluding the first signal frequency band) is located. The power values ​​of all frequency points within this range are accumulated or integrated. The result obtained is the power estimate of noise such as motion artifacts and breathing interference within this time window. The specific calculation formula is as follows:

[0085] (2);

[0086] Among them, K noise This represents the set of discrete frequency point indices that fall within the second signal frequency band.

[0087] Signal-to-noise ratio (SNR) can be expressed as the ratio of signal power to noise power within each time window. In one example, SNR can be expressed in decibels (dB).

[0088] In this embodiment, for each time window, the signal-to-noise ratio (SNR) of that time window is obtained by calculating the ratio of signal power to noise power. The higher the SNR, the more prominent the target physiological signal and the less motion noise interference.

[0089] In one example, based on the signal power and noise power mentioned above, the signal-to-noise ratio (SNR) within each time window is calculated. This SNR directly reflects the intensity of the target physiological signal relative to the background noise. The formula is:

[0090] (3);

[0091] SNR stands for Signal-to-Noise Ratio.

[0092] The average signal-to-noise ratio (SNR) can represent the arithmetic mean of the SNR over multiple consecutive time windows. It is used to comprehensively evaluate the signal quality over a period of time and avoid misjudgment of motion state caused by drastic fluctuations in SNR due to accidental factors (such as instantaneous motion impact) in a single window.

[0093] In one example, the average signal-to-noise ratio (SNR) of the most recent L windows is calculated. The value of L can be determined by balancing real-time requirements and stability needs; a typical value is 5, corresponding to approximately 2.5 seconds of historical data (with a window step size of 30 seconds and an overlap rate of 50%) or a total duration of 150 seconds (with non-overlapping windows). The specific calculation formula is as follows:

[0094] (4);

[0095] Where t represents the current window index.

[0096] Motion state thresholds can represent preset signal-to-noise ratio (SNR) thresholds, and multiple motion state thresholds can be used to distinguish different levels of motion intensity. In one example, motion state thresholds can be obtained through experimental calibration, such as by collecting signals under different motion states, statistically analyzing the average SNR distribution under each state, and thus setting thresholds to distinguish states such as stationary, walking, jogging, and running.

[0097] The target motion state includes at least a stationary state, a walking state, a jogging state, and a running state. The intensity of motion noise generated by the human body varies in different motion states, and the corresponding signal-to-noise ratios also differ significantly. For example, the noise is lowest and the signal-to-noise ratio is highest in the stationary state, while the noise is highest and the signal-to-noise ratio is lowest in the running state.

[0098] In this embodiment, the parameters of the first and second adaptive filters are dynamically adjusted based on the identified target motion state to match the noise estimation performance of the filters with the noise characteristics of the current motion state. The filter parameters are not limited to filter order, forgetting factor, step size, etc. In one example, if the noise intensity is high and changes rapidly in a running state, the filter order can be increased to improve filtering capability; if the noise is low in a stationary state, the filter order can be decreased or a bypass mode can be used to simplify the filtering process. Parameter adjustment can employ a preset parameter switching strategy, i.e., a set of optimal filter parameters is pre-calibrated and stored for different motion states, and switching is performed after state identification; a parameter smoothing transition mechanism can be used during the switching process to avoid transient impacts on the output signal caused by parameter abrupt changes.

[0099] Based on this, embodiments of this application introduce motion state recognition and dynamic filter parameter adjustment mechanisms into the signal processing unit, achieving the system's adaptability to different motion intensities. By calculating the signal-to-noise ratio for each time window through frequency domain transformation, the system can automatically determine the current motion state and optimize the configuration parameters of the adaptive filter accordingly. The introduction of the hysteresis mechanism avoids frequent oscillations at the boundaries of the state, ensuring the stability of the filter parameter adjustment. This closed-loop control strategy ensures that the noise estimation performance of the filter always maintains optimal matching with the real-time motion noise characteristics, avoiding the performance degradation problem of using fixed parameters under different motion states, thereby maintaining high accuracy and robustness of heartbeat signal extraction in all motion scenarios.

[0100] According to an embodiment of this application, the signal processing unit is further configured to: identify the target motion state as stationary when the average signal-to-noise ratio is greater than or equal to a first motion state threshold; identify the target motion state as walking when the average signal-to-noise ratio is less than the first motion state threshold but greater than or equal to a second motion state threshold; identify the target motion state as jogging when the average signal-to-noise ratio is less than the second motion state threshold but greater than or equal to a third motion state threshold; and identify the target motion state as running when the average signal-to-noise ratio is less than the third motion state threshold.

[0101] In this embodiment, multiple motion state thresholds correspond to noise intensity under different motion states.

[0102] The first motion state threshold represents a preset upper limit for the signal-to-noise ratio (SNR). When the average SNR is greater than or equal to this threshold, it indicates excellent signal quality, minimal motion noise interference, and the state can be classified as stationary. In one example, a stationary state includes, but is not limited to, a person sitting or lying down. In this state, there is no significant human movement, and the acceleration signal mainly consists of heartbeat and respiratory components, with minimal motion noise. Heart rate detection can be performed directly based on the original signal or after simplified filtering.

[0103] The second motion state threshold represents a preset intermediate threshold, which is lower than the first motion state threshold. When the average signal-to-noise ratio is less than the first threshold but greater than or equal to the second threshold, it indicates that there is a certain degree of motion interference in the signal, but the signal-to-noise ratio remains high, and it can be determined as a walking state. In this state, human walking produces periodic swaying, the intensity of motion noise is moderate, and there is some overlap with the frequency band of the heartbeat signal, but it can be effectively suppressed by adaptive filtering.

[0104] The third motion state threshold represents a preset lower threshold, which is lower than the second motion state threshold. When the average signal-to-noise ratio is less than the second threshold but greater than or equal to the third threshold, it indicates a significant increase in motion noise, which can be identified as a jogging state. In this state, the human body vibrates violently up and down, the intensity of motion noise is significantly increased, and the frequency band overlaps severely with the heartbeat signal, requiring the use of a higher-order adaptive filter for noise estimation and elimination.

[0105] When the average signal-to-noise ratio is less than the threshold of the third motion state, it can be determined to be a running state. In this state, the human body vibration amplitude is large and the frequency is high, the motion noise intensity is the strongest, and the signal-to-noise ratio may drop below 0dB, that is, the noise power exceeds the signal power, the heartbeat signal is completely masked, and stronger filtering capabilities such as higher order and faster tracking speed are needed to achieve signal reconstruction.

[0106] In one specific implementation, the motion state threshold can be obtained through offline experimental calibration. The specific calibration process is as follows: For multiple target subjects, accelerometers are deployed on the chest, abdomen, and lower abdomen of the target subjects, while wearing standard electrocardiogram (ECG) monitoring equipment as a reference. The target subjects sequentially perform four states: stationary (sitting / lying down), walking (natural speed), jogging (approximately 6-8 km / h), and running (approximately 10-12 km / h), with each state lasting for no less than 5 minutes, and multi-sensor acceleration signals and standard ECG signals are collected simultaneously.

[0107] The signal-to-noise ratio (SNR) for each time window of the acquired acceleration signal is calculated. Simultaneously, using a standard electrocardiogram (ECG) signal as a benchmark, the error in heart rate detection within each time window (e.g., root mean square error) is calculated as an auxiliary evaluation metric for signal quality. Statistical analysis is performed on the SNR distribution for each motion state, and the 25th percentile of the SNR for each state is calculated as the lower bound of the state interval, or a Bayesian classifier is used to determine the optimal classification boundary. For example, by maximizing the state classification accuracy, SNR thresholds are determined to distinguish between stationary and walking, walking and jogging, and jogging and running. For instance, under conditions of a sampling rate of 256 Hz and a window length of 30 s, the calibrated SNR threshold range includes: approximately 15-20 dB for the first motion state, approximately 5-10 dB for the second motion state, and approximately -5 to 0 dB for the third motion state.

[0108] In this embodiment, the optimization objective of the threshold setting is to maximize heart rate detection accuracy while ensuring the accuracy of motion state classification. Specifically, a grid search method can be used to traverse possible threshold combinations and select the threshold combination that minimizes the root mean square error of heart rate detection on the test set as the final calibration result. To accommodate individual differences, the system can provide a personalized calibration mode, where data is automatically collected and a personalized threshold set is generated when the user performs a standard action sequence.

[0109] In this embodiment, to avoid frequent switching of motion states caused by signal-to-noise ratio (SNR) fluctuations near the threshold, a hysteresis mechanism can be introduced. Specifically, when the current state is stationary, the system switches to walking only when the average SNR drops below the difference between the first motion state threshold and the hysteresis value, rather than switching immediately when the SNR drops below the first motion state threshold. When the current state is walking, switching to stationary requires the average SNR to rise above the sum of the first motion state threshold and the hysteresis value; switching to jogging requires the average SNR to drop below the difference between the second motion state threshold and the hysteresis value. The hysteresis value can be determined based on the variance of the SNR measurement noise, for example, it can be set to 1-3 dB. Through the hysteresis mechanism, the system can maintain a stable output of the motion state when the SNR fluctuates, avoiding ineffective oscillations of the filter parameters.

[0110] Based on this, the embodiments of this application discretize the continuously changing signal-to-noise ratio (SNR) value into four physically meaningful motion state levels through explicit threshold comparison logic and a hysteresis mechanism. The thresholds, calibrated through offline experiments, ensure the accuracy and repeatability of state recognition, while the hysteresis mechanism avoids frequent oscillations at the boundaries of the states, guaranteeing the stability of filter parameter adjustments. Quantifying the correspondence between SNR and motion states ensures that the system can accurately adjust its operating mode according to the actual motion intensity, further improving the signal acquisition accuracy and system stability under different motion scenarios.

[0111] According to an embodiment of this application, the signal processing unit is further configured to: increase the filtering length of the first adaptive filter and the second adaptive filter when the target's motion state is sprinting or jogging; decrease the filtering length of the first adaptive filter and the second adaptive filter when the target's motion state is walking; and directly output the physiological signal after suppressing motion noise when the target's motion state is stationary.

[0112] The filter length or filter order refers to the number of tapped delay lines in an adaptive filter, i.e., the number of filter coefficients, denoted as M. In a recursive least squares adaptive filter, the order M determines the filter's ability to model noise propagation paths. A higher order allows the filter to model more complex noise channels, resulting in stronger tracking and suppression capabilities for high-intensity motion noise, but also increases computational complexity and signal processing delay. Conversely, a lower order results in faster processing speed and lower power consumption, but a corresponding decrease in the ability to suppress complex noise.

[0113] In this embodiment, during high-intensity exercise such as sprinting or jogging, the intensity of motion noise is high and changes rapidly. Therefore, the filter length of the adaptive filter can be increased to improve the filter's ability to track and suppress dynamic noise, ensuring that high-intensity motion noise can be effectively stripped away while retaining a clear heartbeat signal.

[0114] In this embodiment, during low-to-medium intensity motion, such as walking, the noise intensity is moderate, so a long filter length is not required. Therefore, the filter length of the adaptive filter can be reduced, which improves signal processing speed and reduces real-time monitoring delay while ensuring noise suppression.

[0115] In this embodiment, in a static state, since the motion noise is extremely small, the physiological signals such as heartbeat or respiration signals are already sufficiently pure after the two-stage adaptive filtering in the early stage, and there is no need to perform complex adaptive filtering processing. Therefore, the extracted physiological signals can be directly output as the physiological signals after noise suppression, which saves computing resources and avoids introducing unnecessary filtering errors in the absence of noise.

[0116] In one specific implementation, a preset parameter switching strategy can be adopted, that is, a set of optimal filter orders can be pre-calibrated and stored for different motion states.

[0117] When the target's motion state is sprinting, the order of the first and second adaptive filters is set to M. h For example, M h =128. This order is sufficient to model the complex motion noise transmission path during fast running, ensuring effective estimation and elimination of high-intensity motion noise.

[0118] When the target motion state is jogging, the order is set to M. mh For example, M mh =96. This order is between running and walking, balancing noise suppression and processing efficiency.

[0119] When the target's motion state is walking, the order is set to M. ml For example, M ml =64. The noise intensity during walking is moderate, and this order is sufficient to ensure noise suppression while reducing computational load.

[0120] When the target is stationary, a filter bypass mode can be used. This mode completely bypasses the first and second adaptive filters, directly outputting the original chest and abdominal position acceleration signals as physiological signals after necessary preprocessing. In a stationary state, motion noise is minimal, and introducing adaptive filters might introduce additional errors due to overfitting. The bypass mode ensures signal quality while maximizing computational resource conservation.

[0121] It should be noted that the order values ​​mentioned above are merely examples. In practical applications, the order can be determined through experimental calibration based on the sensor sampling rate, processor performance, and specific motion characteristics. In one example, when the sampling rate is 256Hz, the order can be adjusted between 32 and 256.

[0122] In this embodiment, to avoid transient impacts on the output signal caused by abrupt changes in filter parameters during motion state switching, the embodiments of this application can adopt a parameter smoothing transition strategy. When the target motion state changes, the signal processing unit does not immediately reset the filter coefficients to the initial state under the new order, but first stops updating the coefficients of the current filter and keeps the existing coefficients unchanged; creates a filter instance of the new order, and uses the existing signal to warm up the filter, so that its coefficients quickly converge to the current noise environment; within a preset transition window (e.g., 0.5 seconds), the outputs of the old and new filters are fused in a linear weighting manner, with the weight of the new filter gradually increasing from 0 to 1, and the weight of the old filter decreasing accordingly; after the transition is completed, the old filter instance is released, and the parameter switching is completed.

[0123] In this embodiment, in a stationary state, the signal processing unit performs filter bypass operation including: the coefficient updates of the first adaptive filter and the second adaptive filter are suspended, and the filter no longer outputs motion noise estimation signals; the chest position acceleration signal is directly output as a heartbeat signal, and the abdominal position acceleration signal is directly output as a respiratory signal; in this mode, the signal processing unit still monitors the signal-to-noise ratio, and once the signal-to-noise ratio drops below the walking state threshold, the system automatically exits the bypass mode and resumes adaptive filtering processing.

[0124] Based on this, embodiments of this application formulate differentiated filter parameter adjustment strategies according to motion intensity levels, achieving optimal matching between system computing resources and filtering performance. Under high-intensity motion, the preset parameter switching strategy avoids the complexity and instability of continuous parameter adjustment, ensuring the reliability of filter configuration under different motion states. The parameter smooth transition mechanism eliminates signal impact during state switching, ensuring the continuity of output physiological signals. The bypass mode minimizes system power consumption in a static state, while avoiding errors that may be introduced by filtering processing in a noise-free environment. This adaptive parameter configuration mechanism enables the system to maintain optimal operating state under different motion scenarios, balancing signal processing accuracy, response speed, and system resource efficiency.

[0125] According to an embodiment of this application, the signal processing unit is further configured to: input chest position acceleration signal, abdominal position acceleration signal and lower abdominal position acceleration signal into a pre-trained long short-term memory neural network model, and the long short-term memory neural network model outputs corresponding heartbeat signal and respiratory signal; wherein, the training data of the long short-term memory neural network model includes multiple sensor acceleration signals collected under multiple motion states as well as standard electrocardiogram signal and standard respiratory signal synchronized with the acceleration signal time.

[0126] Long Short-Term Memory (LSTM) neural networks are a variant of recurrent neural networks suitable for time-series data processing. By introducing memory units and gating mechanisms (input gate, forget gate, output gate), LSTM can effectively capture long-range dependencies in time series data, overcoming the vanishing gradient problem of traditional recurrent neural networks (RNNs). In this embodiment, LSTM is used to directly extract the temporal features of heartbeat and respiration from the time-series signals of three accelerometers, without the need for explicit design of filter structures and parameters.

[0127] The input signal refers to the acceleration signal synchronously collected by the first accelerometer (chest), the second accelerometer (abdomen), and the third accelerometer (lower abdomen).

[0128] In one specific implementation, the input signal can be preprocessed before being input into the LSTM model. Specifically, the DC component of the acceleration signal in each channel is removed to eliminate the influence of gravitational acceleration. The continuous signal is divided into multiple samples according to a fixed time window (e.g., 10 seconds), and the windows can overlap to increase the number of training samples. Each window contains N sampling points; at a sampling rate of 256Hz, a 10-second window contains 2560 sampling points. The window data from the three channels are concatenated with a shape equal to the window length of 3 to obtain a multidimensional tensor, which is used as the input feature of the LSTM model. This input feature contains mixed information of motion noise, heartbeat impact, and respiratory motion.

[0129] The output signals refer to the heartbeat and respiration signals output by the LSTM model. Through end-to-end learning, the LSTM model directly maps pure physiological signals from multi-channel mixed signals without the need for explicit design of filter structure and parameters.

[0130] In one example, the LSTM model uses the following network architecture: The input layer accepts input of shape [time step, feature dimension], where the time step is the window length, for example, 2560, and the feature dimension is the number of accelerometers, for example, 3. The first LSTM layer can have 128 memory units and returns a sequence output for extracting temporal features. The dropout rate of the first regularization layer is set to 0.3, randomly dropping 30% of the neurons to prevent overfitting. The second LSTM layer can have 64 memory units and returns a sequence output for further extraction of high-level temporal features. The dropout rate of the second regularization layer is also set to 0.3. The third LSTM layer can have 32 memory units and returns a sequence output, outputting the hidden state of the last time step. The fully connected layer contains two neurons, corresponding to the outputs of the heartbeat signal and the respiratory signal, respectively. The output layer is a linear activation function, outputting the time-domain waveforms of the heartbeat signal and the respiratory signal with the same length as the input window.

[0131] In another example, if only the heart rate and respiratory rate values ​​are needed instead of the complete waveform, a global average pooling layer can be added after the LSTM layer, and then connected to a fully connected layer to output the two values.

[0132] The training data includes multi-sensor acceleration signals collected under different motion states (such as rest, walking, jogging, and running), as well as standard electrocardiogram and standard respiratory signals synchronized with them in time.

[0133] In one example, data was collected from no fewer than 20 target subjects, covering different ages, genders, and body types. Data was collected for each motion state for at least 5 minutes to ensure the model learned sufficient motion noise features. Standard electrocardiogram (ECG) signals were used as the ground truth for heartbeat signals; standard respiratory signals were used as the ground truth for respiratory signals. The collected data was divided into training, validation, and test sets in an 8:1:1 ratio. The training set was used for model parameter learning, the validation set for hyperparameter tuning and early stopping monitoring, and the test set for final performance evaluation.

[0134] According to embodiments of this application, the following techniques can be used during model training to ensure training stability and prevent overfitting: Regularization layers are inserted between LSTM layers, and some neurons are randomly discarded to prevent the model from over-relying on specific features. The initial learning rate is set to 0.001, using exponential decay or cosine annealing strategies. When the validation set loss no longer decreases for 5 consecutive iterations, the learning rate is decayed to 0.5 times the current value. The validation set loss is monitored; when the validation set loss no longer decreases for 10 consecutive iterations, training is terminated early to avoid a decrease in generalization ability due to overtraining. Mean squared error is used as the loss function to calculate the error between the model's output heartbeat signal and the standard ECG signal, and the error between the respiratory signal and the standard respiratory signal. The weighted sum of these two errors is used as the total loss, for example, a weight of 0.5 for heartbeat loss and 0.5 for respiratory loss. An Adaptive Moment Estimation (Adam) optimizer is used, combined with gradient clipping, with a gradient threshold set to 1.0 to prevent gradient explosion.

[0135] According to an embodiment of this application, during the model deployment phase, the signal processing unit divides the three real-time acquired acceleration signals into sliding windows of the same size, such as 10 seconds, and inputs them into the trained LSTM model for forward inference. The model outputs the time-domain waveforms of the heartbeat and respiratory signals within the corresponding window. For heart rate or respiratory rate calculation, peak detection can be further performed on the output waveform to calculate the heart rate and respiratory rate values.

[0136] In this embodiment, the LSTM processing path and the adaptive filtering processing path are used as parallel signal processing schemes in the system. In one example, the LSTM path output can be preferentially used by default because it has higher accuracy under high-intensity motion. When the confidence level of the LSTM model output is low, such as when the output waveform deviates too much from the historical mean, the system automatically switches to the adaptive filtering path to ensure the continuity of monitoring. In another example, the system can also run both processing paths simultaneously, performing a weighted fusion of the LSTM output and the adaptive filtering output, dynamically adjusting the fusion weights according to the current motion state. For example, in a fast running state, the LSTM weight is 0.8 and the adaptive filtering weight is 0.2.

[0137] In one example, the multi-sensor system using the aforementioned LSTM model achieved a 3.7-fold improvement in heart rate detection accuracy compared to a traditional single-sensor system during high-noise activities such as running. The LSTM model can learn complex nonlinear mapping relationships from multi-sensor time-series signals, and especially under conditions of high-intensity exercise and drastic changes in noise characteristics, it can adaptively extract stable heartbeat and respiratory signals, providing powerful intelligent processing capabilities for physiological monitoring in complex dynamic environments.

[0138] Based on this, embodiments of this application introduce an LSTM neural network as a signal processing path parallel to adaptive filtering, providing the system with another dimension of motion noise suppression capability. The LSTM model can learn complex nonlinear mapping relationships from multi-sensor time-series signals, and can adaptively extract stable heartbeat and breathing signals, especially under conditions of high-intensity motion and drastic changes in noise characteristics.

[0139] Based on this, the embodiments of this application introduce an LSTM neural network as a signal processing path parallel to adaptive filtering, providing the system with another dimension of motion noise suppression capability. Explicit input preprocessing, network structure design, training data construction, and anti-overfitting strategies ensure the model's trainability and generalization ability. Through a collaborative fusion mechanism with the adaptive filtering path, the system can automatically select or fuse the optimal processing results under different motion scenarios, achieving high-precision and highly robust cardiopulmonary physiological signal monitoring.

[0140] According to an embodiment of this application, the signal processing unit is further configured to: calculate heart rate and respiratory rate values ​​based on physiological signals after suppressing motion noise; and transmit the heart rate, respiratory rate, and target motion state to an external terminal device via a wireless transmission channel.

[0141] Heart rate represents a physiological parameter extracted from the heartbeat signal after suppressing motion noise, indicating the number of heartbeats per minute. Since the heartbeat signal originates from a chest accelerometer, the heart rate value can be calculated by detecting the peak intervals of the heartbeat signal. In one example, an adaptive thresholding method is used, with an initial threshold of 1.5 times the signal mean, to detect positive peaks in the signal. Based on physiological constraints (heart rate range 30-220 beats / minute), spurious peaks with intervals less than 0.27 seconds (corresponding to 220 beats / minute) or greater than 2 seconds (corresponding to 30 beats / minute) are removed to ensure that the detected peaks are truly valid. The time interval between adjacent valid peaks is calculated and converted into instantaneous heart rate. To improve stability, the moving average of the five most recent instantaneous heart rates can be used as the final output heart rate value.

[0142] The respiratory rate value represents a physiological parameter extracted from the respiratory signal after suppressing motion noise, indicating the number of breaths per minute. Specifically, the respiratory rate value can be calculated by detecting the peaks, troughs, or zero-crossing points of the respiratory signal. In one example, the zero-crossing method can be used. First, the respiratory signal is low-pass filtered, then the position of the signal's zero-crossing point is detected. Each adjacent zero-crossing point corresponds to half a respiratory cycle. The time interval between two adjacent inspiratory or expiratory initiations is calculated, which is a single respiratory cycle. A single respiratory cycle is then converted into an instantaneous respiratory rate. To improve stability, the moving average of the most recent five respiratory cycles is taken as the final output respiratory rate value.

[0143] In this embodiment, the output frequency of heart rate and respiratory rate values ​​is determined according to the application scenario. In one example, the system employs a sliding window update strategy, updating the output value after each time window (e.g., 30 seconds) of signal processing is completed to ensure real-time performance. In continuous monitoring mode, the output update frequency can be set to once per second, allowing users to view real-time data on their terminal devices as needed.

[0144] In this embodiment, the signal processing unit transmits the calculated heart rate, respiratory rate, and identified target motion state (stationary, walking, jogging, running) to the external terminal device via the aforementioned boundary state wireless transmission channel. Specifically, the boundary state wireless transmission channel formed by the boundary between the first and second modules ensures stable data upload during motion, avoiding signal attenuation caused by human body obstruction or movement in traditional free-space wireless links. In one example, Bluetooth Low Energy (BLE) protocol is used for data transmission, with the signal processing unit acting as the Bluetooth master and the external terminal device as the slave. The data is encapsulated in a custom format, including a timestamp, heart rate, respiratory rate, motion state identifier, and checksum to ensure data integrity. The transmission interval is set according to the data update frequency, such as sending the latest data every 1 second. In power-sensitive scenarios, the transmission frequency can be reduced, such as sending every 5 seconds, with the terminal device performing data interpolation for display.

[0145] External terminal devices refer to data receiving and display terminals such as smartphones, tablets, personal computers, cloud servers, or medical monitoring centers. In one example, a smartphone with a dedicated application receives heart rate, respiratory rate, and activity status data via the BLE protocol, displays waveforms and values ​​in real time, and provides functions such as historical data storage, trend chart plotting, and anomaly alarms. In another example, data is uploaded to a cloud server via a cellular network or wireless LAN for remote access by medical personnel, enabling remote medical monitoring.

[0146] In this embodiment, target motion state information and physiological parameters are sent together to an external terminal to provide context for health assessment. In one example, the terminal application displays heart rate data in layers according to motion state. Resting heart rate is displayed in the resting state, and exercise heart rate is displayed in the exercise state, and heart rate recovery time is calculated as an indicator of cardiopulmonary function assessment.

[0147] Based on this, embodiments of this application wirelessly transmit extracted heart rate, respiratory rate, and exercise status information to an external terminal, enabling the system to achieve a complete workflow from signal acquisition, noise suppression, parameter calculation to data presentation. Users can obtain accurate physiological parameters during exercise in real time, providing comprehensive technical support for applications such as daily health management, exercise monitoring, and telemedicine.

[0148] Figure 5 The flowchart illustrating the synchronous data acquisition and processing of the heartbeat signal extraction channel and the respiratory signal extraction channel according to an embodiment of this application is shown.

[0149] like Figure 5 As shown in the flowchart, the flowchart illustrates two parallel signal processing channels: a heartbeat signal extraction channel and a respiratory signal extraction channel.

[0150] The first and second accelerometers synchronously acquire chest position acceleration signals 501 and abdominal position acceleration signals 502. The two sensors are aligned in sampling time via a synchronization beacon broadcast by the signal processing unit, ensuring that the sampling error of each channel is less than 1 millisecond. After preprocessing, the chest position acceleration signals 501 and 502 are input to the first adaptive filter 503. Using the abdominal position acceleration signal 502 as a reference signal, the motion noise component in the chest position acceleration signal 501 is estimated, and a motion noise estimation signal 504 is output. The chest position acceleration signals 501 and 504 are then input to the second adaptive filter 505. Using the motion noise estimation signal 504 output by the first adaptive filter 503 as a reference signal, the original chest position acceleration signal 501 undergoes noise cancellation processing, and a noise-suppressed heartbeat signal 506 is output.

[0151] The second and third accelerometers synchronously acquire abdominal position acceleration signals 502 and 507. The sampling times are aligned between them via a synchronization beacon broadcast by the signal processing unit, ensuring that the sampling error of each channel is less than 1 millisecond. After preprocessing, the abdominal position acceleration signals 502 and 507 are input to the third adaptive filter 508. Using the lower abdominal position acceleration signal 507 as a reference signal, the motion noise component in the abdominal position acceleration signal 502 is estimated, and a second motion noise estimation signal 509 is output. The abdominal position acceleration signal 502 and the second motion noise estimation signal 509 are input to the fourth adaptive filter 510. Using the second motion noise estimation signal 509 output by the third adaptive filter 508 as a reference signal, noise cancellation processing is performed on the original abdominal position acceleration signal 502, and a noise-suppressed respiratory signal 511 is output.

[0152] All four adaptive filters employ a recursive least squares algorithm. The filter order for the heartbeat channel can be set to 64, with a sampling rate of 256Hz. The filter order for the breathing channel can be set to 64 or configured independently according to the characteristics of the breathing signal. The forgetting factor is adapted to the rate of change of each signal.

[0153] The noise-suppressed heartbeat signal 506 is input to the signal processing unit 512, which performs the following motion state recognition operations: The signal is divided into continuous windows of 30 seconds; a fast Fourier transform is performed on each window to obtain a frequency domain representation; the signal power in the heartbeat signal frequency band (e.g., 0.8Hz-3.0Hz) is calculated compared to the noise power in other frequency bands to obtain the signal-to-noise ratio (SNR), and the average SNR of multiple continuous windows is taken; the average SNR is compared with a preset motion state threshold to identify the target motion state 513. Based on the identified target motion state 513, the parameters of the first adaptive filter 503, the second adaptive filter 505, the third adaptive filter 508, and the fourth adaptive filter 510 are dynamically adjusted, such as the order and forgetting factor. For example, the order is increased for fast / slow running states, decreased for walking states, and the filter bypass mode is enabled for stationary states.

[0154] When the quality of the noise-suppressed heartbeat signal 506 is extremely poor, for example, the average signal-to-noise ratio is consistently below the third threshold and the correlation with the standard electrocardiogram signal is less than 0.5, the system can switch to a state recognition mode based on the respiratory signal signal-to-noise ratio. The signal processing unit 512 adjusts the respiratory signal frequency band to 0.2Hz-0.5Hz based on the noise-suppressed respiratory signal 511 to achieve dual-channel redundancy.

[0155] The heart rate value is obtained by peak detection and RR interval calculation of the noise-suppressed heartbeat signal 506, and the moving average is taken to obtain the target heart rate value 514. The respiratory rate value is obtained by zero-crossing method or peak detection of the noise-suppressed respiratory signal 511, and the moving average is taken to obtain the target respiratory rate value 515. After adding timestamps and check bits to the target heart rate value 514, the target respiratory rate value 515 and the target motion state 513, they are encapsulated to obtain the encapsulated physiological data 516.

[0156] Physiological data 516 is transmitted to an external terminal device 517 via a wireless transmission channel. This channel utilizes the boundary state formed by the boundary between the first and second modules in the body surface wireless transmission structure to achieve low-loss, directional transmission, employing the Bluetooth Low Energy protocol. After receiving the data, the external terminal device 517, such as a smartphone or cloud server, performs real-time display, historical storage, trend analysis, and anomaly alarms.

[0157] Figure 5 In the illustrated process, the heartbeat and respiratory signal extraction channels operate in parallel and independently, sharing motion state recognition results to avoid redundant calculations. Hardware-triggered synchronization ensures precise alignment of multi-sensor signals, motion state recognition and dynamic adjustment of filter parameters achieve adaptive noise suppression across the entire motion scenario, and a dual-channel redundancy mechanism enhances system robustness, ultimately achieving coordinated monitoring and stable transmission of cardiopulmonary physiological signals.

[0158] According to an embodiment of this application, both the first module and the second module include periodically arranged conductive units; the conductive units are one of square metal patches, circular metal patches, or ring structures; the conductive units of the first module and the conductive units of the second module have different geometric parameters, so that the first module and the second module form a boundary state at the boundary region.

[0159] Periodically arranged conductive units refer to the basic structural units that constitute the first and second modules. These units are repeatedly arranged in a plane according to fixed periodic parameters, such as unit side length and unit spacing, forming a periodic artificial electromagnetic structure. The purpose of the periodic arrangement is to generate specific electromagnetic band gaps or dispersion characteristics through the coupling between units, thereby suppressing the propagation of wireless signals within the module. In one example, for the target operating frequency band of 2.4G-2.48GHz Bluetooth, the center-to-center spacing of adjacent conductive units can be set to 5mm-8mm, and the size of the conductive unit is 0.5-0.8 times the periodic parameter.

[0160] A conductive unit can represent a conductive material structure used to generate an electromagnetic response. Its specific form can include a square metal patch, a circular metal patch, or a ring structure. These conductive units can be formed on the surface of a flexible substrate by methods such as printing, electroplating, or attachment. In one example, the conductive unit can be made of a highly conductive metal material such as copper or silver, and the thickness can be selected according to the process conditions. Square metal patches are easy to design and fabricate, and are suitable for large-area fabrication; ring structures can generate higher equivalent inductance and are suitable for scenarios requiring a strong magnetic field response.

[0161] In this embodiment, the conductive units in the first module and the conductive units in the second module differ in geometric parameters such as patch side length, circular radius, and annular inner and outer diameters. The essence of this difference is breaking the rotational symmetry of the crystal lattice, thereby endowing the two modules with different topological phases or different topological invariants.

[0162] Figure 6 The diagram illustrates the topological patterns of the first and second modules according to a specific embodiment of this application.

[0163] like Figure 6 As shown, Figure 6 (a) and Figure 6 (b) The unit topology of the first module and the second module are shown respectively. Each module contains periodically arranged conductive units. The pattern of the conductive units can be floral, circular, triangular, etc. The example in the figure uses a circular pattern.

[0164] like Figure 6 (a) and Figure 6As shown in (b), within the conductive units of the parallelogram or hexagonal regions, the geometric dimensions and arrangement of the conductive units differ between the two types of modules, but they maintain the same periodic constant (e.g., the center-to-center spacing between adjacent units). For example, as... Figure 6 As shown in (a), the conductive unit within the first module is "a large circle as the center, with multiple smaller circles arranged in an equilateral triangle around the large circle." Figure 6 As shown in (b), the conductive unit in the second module is "centered on a large circle, with multiple smaller circles arranged in an inverted triangle around the large circle." This size difference breaks the rotational symmetry of the hexagonal lattice: originally, when all unit sizes are the same, the lattice has C6 symmetry (unchanged by a 60° rotation), but after introducing the size difference, the symmetry degenerates to C3 (unchanged by a 120° rotation), thus giving the two modules different topological phases. It should be noted that the hexagonal lattice is only an example; other lattice shapes (such as squares and parallelograms) and corresponding symmetry breaking methods (such as C4 degenerating into C2) also apply.

[0165] Within a single module, the electromagnetic coupling between adjacent conductive units is designed to be suppressed, causing the wireless signal to attenuate rapidly as it propagates along the planar direction within the module. When the first and second modules are arranged adjacent to each other on a flexible substrate, a topological interface state is formed at the boundary region due to their different topological phases.

[0166] Figure 7 The diagram illustrates the boundary state formed between the first module and the second module according to a specific embodiment of this application.

[0167] like Figure 7 As shown, when Figure 6 When the first and second modules are "joined" together, a boundary state is generated at the boundary. Efficient propagation of this boundary state requires a specific matching relationship between the units on both sides of the boundary, such as "large circle to large circle" or pattern pairs with complementary topological phases. In this case, the propagation of the wireless signal perpendicular to the boundary direction is suppressed, while a low-loss propagation path is formed along the boundary direction. Conversely, if the units on both sides of the boundary do not satisfy the matching relationship, such as "small circle to small circle," the transmission effect is poor, and it does not function as a normal signal channel. Based on the above principle, the directional guidance of the wireless signal by the boundary state can be achieved.

[0168] Based on this, the embodiments of this application define the functional modules as being composed of periodically arranged conductive units, and explicitly state that the first module and the second module achieve electromagnetic property differentiation through different geometric parameters, providing a specific physical implementation method for the formation of boundary states. This modular design based on a periodic unit structure allows the system to construct modules with different electromagnetic properties on a flexible substrate using standardized units, and then reconstruct the wireless transmission channel through the arrangement and combination of modules, providing stable and low-loss surface communication physical layer support for wireless interconnection of sensors.

[0169] According to an embodiment of this application, the first module and the second module are arranged in different ways in the plane to form at least one boundary state at the intersection of the first module and the second module.

[0170] In this embodiment, the relative position, arrangement direction, quantity ratio, and adjacency relationship of the first module and the second module in the plane can be changed according to application requirements. By changing the arrangement and combination of modules, wireless transmission channels with different quantities and topologies can be constructed to adapt to different sensor node layout requirements.

[0171] It is important to emphasize that not any two adjacent modules can automatically form a low-loss boundary state. Only boundaries that meet the topology matching condition (i.e., the units on both sides of the boundary are matched, such as "large circle to large circle") can serve as effective wireless transmission channels.

[0172] In this embodiment, when a single boundary line that meets the conditions is formed, a boundary channel is generated; when multiple mutually isolated boundary lines that meet the topology matching are formed, multiple parallel boundary channels are generated, thereby constructing a body surface sensor network with different numbers of wireless transmission channels and topology structures.

[0173] Figure 8 The diagram illustrates the number of wireless propagation channels corresponding to different permutations and combinations according to specific embodiments of this application.

[0174] exist Figure 8 In the specific example shown, the modules are arranged in a hexagonal honeycomb pattern, with each hexagon consisting of six triangular sub-regions, for example, categorized by orientation as: upper left, upper center, upper right, lower left, lower center, and lower right. Wireless propagation channels are formed by connecting effective boundary segments between the first and second modules that satisfy topological matching conditions (e.g., "large circle to large circle"). The number of channels equals the number of unconnected continuous paths formed by these effective boundary segments. When two modules of the same type are adjacent, no effective boundary is formed at their intersection; when modules of different types are adjacent but the units on both sides of the boundary do not satisfy topological matching, a low-loss boundary state cannot be formed either.

[0175] Figure 8Including sub Figure 8 (a) to 8(f) correspond to six exemplary arrangements of the number of wireless propagation channels n=0, 1, 2, 3, 4 and 6, respectively.

[0176] like Figure 8 As shown in (a), in this hexagonal cellular arrangement, all six triangular sub-regions are configured as first modules, and there are no second modules. Due to the lack of a topological phase difference interface between the two types of modules, there are no valid boundary state paths within the entire structure, therefore the number of wireless propagation channels is n=0.

[0177] like Figure 8 As shown in (b), in this hexagonal cellular arrangement, the upper right and lower right triangular regions are configured as the second module, and the remaining four triangular regions (upper left, upper center, lower left, and lower center) are configured as the first module. The boundaries of the first and second modules intersect near the center of the hexagon, forming a continuous topological interface path from the center to the upper right. This path is independent and open, and can serve as a wireless propagation channel; therefore, the number of wireless propagation channels is n=1.

[0178] like Figure 8 As shown in (c), in this hexagonal cellular arrangement, the upper left, upper middle, and upper right triangular regions are configured as the second module, and the lower left, lower middle, and lower right triangular regions are configured as the first module. The boundaries of the two types of modules form two mutually isolated topological interface paths near the center of the hexagon: one pointing to the left from the center and the other pointing to the right from the center. Therefore, the number of wireless propagation channels is n=2.

[0179] like Figure 8 As shown in (d), in this arrangement, the upper left, upper center, and lower right triangular regions are configured as the first module, and the lower left, lower center, and upper right triangular regions are configured as the second module. The boundary state path branches into three unconnected radial channels in the central region, pointing to the upper right, lower right, and right directions respectively. Each channel can independently guide the wireless signal, therefore the number of wireless propagation channels is n=3.

[0180] like Figure 8 As shown in (e), in this arrangement, the upper left and upper right triangular regions are configured as the second module, and the remaining four triangular regions (upper center, lower left, lower center, and lower right) are configured as the first module. The module boundaries form four radial independent channels in the central region, pointing to the upper left, upper right, left, and right directions respectively. Each channel is a valid topological interface state path, therefore the number of wireless propagation channels is n=4.

[0181] like Figure 8As shown in (f), in this arrangement, the first and second modules are alternately distributed: the six triangular regions are arranged alternately in sequence, for example, clockwise as second module, first module, second module, first module, second module, and first module. This completely alternating arrangement ensures that each side of the hexagon is composed of adjacent different modules and satisfies the topological matching condition, thus forming six radial independent channels in the central region, pointing to the upper left, upper right, left, right, lower left, and lower right directions respectively. Each channel is an independent wireless propagation path, therefore the number of wireless propagation channels is n=6.

[0182] It should be noted that, Figure 8 The arrangement shown is only for illustrating the possibility of reconfigurable topology; in actual implementation, simulation verification is required based on specific unit topology parameters (such as the shape and size of conductor units) to ensure that low-loss boundary states are obtained.

[0183] Furthermore, the module shape is not limited to triangles or hexagons; it can be freely topologically customized to meet specific needs, such as arbitrary polygons, curves, or irregular regions. Users can customize wireless networks with specific transmission channel structures by changing the arrangement and overall shape of the modules according to the number of sensors, their placement, and data transmission requirements in their actual application scenarios, without having to redesign the overall hardware structure.

[0184] By employing the six arrangement methods described above, and simply changing the relative positions and numbers of the first and second modules within the plane, multiple numbers of wireless transmission channels can be obtained without altering the structure of individual modules. This enables rapid reconfiguration of the surface sensor network topology, significantly improving the system's flexibility and applicability. Based on this, embodiments of this application, according to the number of sensors, their placement, and data transmission requirements in actual application scenarios, customize wireless networks with specific transmission channel structures by changing the arrangement of modules, without redesigning the overall hardware structure, thereby enhancing the system's flexibility and applicability.

[0185] The embodiments of this application also propose a motion noise suppression method based on multiple body surface sensors.

[0186] Figure 9 A flowchart illustrating a motion noise suppression method according to an embodiment of this application is shown schematically.

[0187] like Figure 9 As shown, the motion noise suppression method includes operations S910 to S930.

[0188] During operation of S910, chest position acceleration signals and abdominal position acceleration signals are acquired.

[0189] In operation S920, the chest position acceleration signal and the abdominal position acceleration signal are input into the first adaptive filter. The abdominal position acceleration signal is used as a reference signal to estimate the motion noise component in the chest position acceleration signal, thereby obtaining the motion noise estimation signal.

[0190] In operation S930, the chest position acceleration signal and motion noise estimation signal are input into the second adaptive filter. The motion noise estimation signal is used as a reference signal to perform noise cancellation processing on the chest position acceleration signal to obtain the physiological signal after suppressing motion noise.

[0191] In this embodiment, raw data is read from accelerometers deployed on the chest and abdomen of the human body. The acquisition process includes preprocessing operations such as analog-to-digital conversion of the sensor signals, time synchronization, and preliminary filtering to ensure the quality and synchronization of the input signals.

[0192] In this embodiment, an adaptive filtering algorithm is used to estimate the motion noise component in the chest signal by utilizing the motion noise-related components in the abdominal signal. The abdominal signal, as a reference signal, has the advantage of containing motion noise but lacking cardiac impact components, making it suitable as a noise reference source. The adaptive filter continuously adjusts its coefficients to make its output approximate the actual motion noise in the chest signal, thereby achieving accurate noise estimation.

[0193] In this embodiment, the original chest acceleration signal is subtracted from or canceled by the motion noise signal estimated by the first filter using a second adaptive filter. Since the motion noise estimation signal is highly correlated with the actual motion noise in the chest signal, subtraction effectively eliminates motion interference and preserves the pure heartbeat signal component.

[0194] The embodiments of this application leverage the advantages of multi-sensor collaborative acquisition, using abdominal signals as a noise reference. A cascaded adaptive filtering architecture sequentially performs noise estimation and cancellation, achieving effective suppression of motion noise. Compared to traditional single-sensor or single-step filtering schemes, this method significantly improves the signal-to-noise ratio of heartbeat signals under dynamic motion conditions, providing high-quality signal input for subsequent physiological parameter calculations.

[0195] It should be noted that the motion noise suppression method part in the embodiments of this application corresponds to the motion noise suppression system part in the embodiments of this application. The description of the motion noise suppression method part is specifically referred to in the motion noise suppression system part, and will not be repeated here.

[0196] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions. Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations are not explicitly described in this application. In particular, without departing from the spirit and teachings of this application, the features described in the various embodiments of this application can be combined and / or combined in various ways. All such combinations and / or combinations fall within the scope of this application.

[0197] The embodiments of this application have been described above. However, these embodiments are merely illustrative and not intended to limit the scope of this application. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. Without departing from the scope of this application, those skilled in the art can make various substitutions and modifications, all of which should fall within the scope of this application.

Claims

1. A motion noise suppression system based on multiple body surface sensors, wherein, The system includes: A surface-mount wireless transmission structure is designed to be attached to the surface of the human body. The surface-mount wireless transmission structure includes multiple functional modules disposed on a flexible substrate. The multiple functional modules include at least a first module and a second module. The first module and the second module have different equivalent electromagnetic characteristics in the target operating frequency band to form a boundary state at the boundary region formed by the contact between the first module and the second module. The boundary state is used to guide the wireless signal to propagate along the boundary region. At least two accelerometers are used to synchronously acquire chest position acceleration signals and abdominal position acceleration signals during motion. The signal processing unit is wirelessly coupled to the at least two accelerometers via a wireless transmission channel formed in the boundary region, and the signal processing unit is configured to: The chest position acceleration signal and the abdominal position acceleration signal are input into the first adaptive filter. The abdominal position acceleration signal is used as a reference signal to perform noise estimation on the motion noise component in the chest position acceleration signal to obtain the motion noise estimation signal. The chest position acceleration signal and the motion noise estimation signal are input into a second adaptive filter. The motion noise estimation signal is used as a reference signal to perform noise cancellation processing on the chest position acceleration signal to obtain a physiological signal after suppressing motion noise. The physiological signal includes a heartbeat signal.

2. The system according to claim 1, wherein, The system also includes: The third accelerometer is used to synchronously collect acceleration signals of the lower abdomen during movement. The signal processing unit is wirelessly coupled to the third accelerometer via the wireless transmission channel, and the signal processing unit is further configured to: The abdominal position acceleration signal and the lower abdominal position acceleration signal are input into the third adaptive filter. The lower abdominal position acceleration signal is used as a reference signal to estimate the motion noise component in the abdominal position acceleration signal, thereby obtaining the second motion noise estimation signal. The abdominal position acceleration signal and the second motion noise estimation signal are input into a fourth adaptive filter. The second motion noise estimation signal is used as a reference signal to perform noise cancellation processing on the abdominal position acceleration signal to obtain the physiological signal after suppressing motion noise. The physiological signal includes a respiratory signal.

3. The system according to claim 2, wherein, The first adaptive filter, the second adaptive filter, the third adaptive filter, and the fourth adaptive filter are all recursive least squares adaptive filters.

4. The system according to claim 1, wherein, The signal processing unit is further configured to: The physiological signal after suppressing motion noise is divided into multiple consecutive time windows based on a preset step size; The signal within each time window is subjected to frequency domain transformation, and based on the frequency domain transformation results, the signal power in the first signal frequency band and the noise power in the second signal frequency band are calculated; wherein, the first signal frequency band is the frequency band range in which the heartbeat signal is located, and the second signal frequency band is the frequency band range that does not include the first signal frequency band; Based on the signal power and the noise power, the signal-to-noise ratio within each time window is calculated; The target motion state is obtained by comparing the average signal-to-noise ratio of multiple consecutive time windows with multiple motion state thresholds. The target motion state includes at least a stationary state, a walking state, a jogging state, and a running state. Based on the target motion state, the parameters of the first adaptive filter and the second adaptive filter are dynamically adjusted so that the first adaptive filter and the second adaptive filter adapt to the noise characteristics under the motion state.

5. The system according to claim 4, wherein, The signal processing unit is further configured to: When the average signal-to-noise ratio is greater than or equal to the first motion state threshold, the target motion state is identified as the stationary state. When the average signal-to-noise ratio is less than the first motion state threshold and greater than or equal to the second motion state threshold, the target motion state is identified as the walking state. When the average signal-to-noise ratio is less than the second motion state threshold and greater than or equal to the third motion state threshold, the target motion state is identified as the jogging state. When the average signal-to-noise ratio is less than the third motion state threshold, the target motion state is identified as the sprinting state.

6. The system according to claim 4, wherein, The signal processing unit is further configured to: When the target motion state is the sprinting state or the jogging state, increase the filter length of the first adaptive filter and the second adaptive filter; When the target's motion state is the walking state, the filter lengths of the first adaptive filter and the second adaptive filter are reduced; When the target motion state is the stationary state, the physiological signal after suppressing motion noise is directly output.

7. The system according to claim 1, wherein, Both the first module and the second module include periodically arranged conductive units; The conductive unit is one of a square metal patch, a circular metal patch, or a ring structure. The conductive units of the first module and the conductive units of the second module have different geometric parameters, so that the first module and the second module form the boundary state at the boundary region.

8. The system according to claim 7, wherein, The first module and the second module are arranged in different ways in the plane, so as to form at least one boundary state at the intersection of the first module and the second module.

9. The system according to claim 1, wherein, The flexible substrate is a fabric substrate or a polymer film substrate, and the multiple functional modules are fixed on the flexible substrate based on at least one of the following methods: printing, laser cutting, or hot pressing.

10. A motion noise suppression method based on multiple body surface sensors, applied to the system according to any one of claims 1 to 9, wherein, The method includes: Acquire chest position acceleration signals and abdominal position acceleration signals; The chest position acceleration signal and the abdominal position acceleration signal are input into the first adaptive filter. The abdominal position acceleration signal is used as a reference signal to perform noise estimation on the motion noise component in the chest position acceleration signal to obtain the motion noise estimation signal. The chest position acceleration signal and the motion noise estimation signal are input into a second adaptive filter. The motion noise estimation signal is used as a reference signal to perform noise cancellation processing on the chest position acceleration signal to obtain a physiological signal after suppressing motion noise. The physiological signal includes a heartbeat signal.

Images