Special MEMS (Micro Electro Mechanical System) sensor for heart sound and signal processing method thereof

Abstract

The invention relates to the technical field of biomedical electronic engineering and physiological signal processing, and discloses a special MEMS sensor for heart sound, which comprises an MEMS sensor shell, a silicon substrate is fixedly connected to the inner bottom of the MEMS sensor shell, and a central acoustic unit and a peripheral inertial unit are integrated on the silicon substrate in a monolithic manner; the central acoustic unit is used for collecting sound pressure signals; the central acoustic unit is arranged in the MEMS sensor shell, the peripheral inertial unit is arranged around the central acoustic unit and is used for collecting three-axis acceleration signals, a processor is arranged in the MEMS sensor shell and is electrically connected with the central acoustic unit and the peripheral inertial unit, the central acoustic unit is of a capacitive MEMS microphone structure, and the central acoustic unit and the peripheral inertial unit are arranged in the MEMS sensor shell. The peripheral inertia unit is of a three-axis MEMS accelerometer structure. Through multi-modal signal fusion and closed-loop adaptive adjustment, the problems that traditional heart sound collection is low in signal-to-noise ratio, unstable in signal and lack of comparability are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of biomedical electronic engineering and physiological signal processing, in particular to a MEMS sensor special for heart sound and a signal processing method thereof. BACKGROUND

[0002] Heart sound (PCG) is the acoustic manifestation of the mechanical activity of the heart on the body surface, which contains a wealth of pathophysiological information about the function of heart valves and hemodynamics. Using electronic stethoscopes and other devices to collect and analyze heart sound signals is an important non-invasive means for diagnosing cardiovascular diseases such as heart valve disease and heart failure.

[0003] However, the existing heart sound collection devices in the prior art face serious limitations in clinical and home applications. The existing electronic auscultation devices mainly rely on a single acoustic sensor (such as a microphone) for signal collection, which is extremely sensitive to noise. The collected sound pressure signals often contain strong interference, including high-frequency noise generated by the friction between the sensor and the skin or clothing, and low-frequency body motion noise caused by the patient's breathing and body movement. These noises severely drown out the key heart sound components such as the first and second heart sounds and the weak heart murmurs, resulting in extremely low signal-to-noise ratio and greatly limiting the accuracy of diagnosis.

[0004] In addition, the signal quality of existing devices is heavily dependent on the use of the operator and the physiological conditions of the patient. For example, the pressing force applied by the operator on the sensor can significantly change the acoustic coupling characteristics of the sensor and the chest wall, resulting in unstable signal amplitude and frequency response. At the same time, individual differences in the chest wall tissue (such as fat and muscle thickness) of different patients can produce unique and strong filtering and attenuation effects on the heart sound signals. This uncertainty caused by the operation method and individual physiological differences leads to two key problems: first, the operator has difficulty obtaining stable and reproducible signals; second, the signals lack comparability between different individuals, making it difficult for doctors to perform objective quantitative analysis and severely hindering the development and application of artificial intelligence-based automated diagnosis algorithms.

[0005] Therefore, the present application provides a MEMS sensor special for heart sound and a signal processing method thereof to solve the deficiencies in the prior art. SUMMARY

[0006] In view of the deficiencies of the prior art, the present application provides a MEMS sensor special for heart sound and a signal processing method thereof, which solves the problems of low signal-to-noise ratio caused by friction and body motion noise interference in traditional heart sound collection technology, and the instability and lack of comparability of signals caused by operation method and individual chest physiological differences.

[0007] In order to achieve the above object, the application is implemented by the following technical scheme: a heart sound special MEMS sensor, comprising a MEMS sensor shell, a silicon substrate is fixedly connected to the inner bottom of the MEMS sensor shell, and a central acoustic unit and a peripheral inertial unit are monolithically integrated on the silicon substrate; the central acoustic unit is used for collecting sound pressure signals; the peripheral inertial unit is arranged around the central acoustic unit and is used for collecting three-axis acceleration signals; a processor is arranged in the MEMS sensor shell, and the processor is electrically connected with the central acoustic unit and the peripheral inertial unit.

[0008] Preferably, the central acoustic unit is a capacitive MEMS microphone structure, and the peripheral inertial unit is a three-axis MEMS accelerometer structure.

[0009] Preferably, a bias voltage adjusting circuit is further included, which is electrically connected with the capacitive MEMS microphone structure and the processor, is used for receiving an external control signal from the processor, and dynamically adjusts a direct current bias voltage of the capacitive MEMS microphone structure according to the external control signal.

[0010] A signal processing method of a heart sound special MEMS sensor, characterized by comprising the following steps: synchronously collecting sound pressure signals of the central acoustic unit and three-axis acceleration signals of the peripheral inertial unit, wherein the three-axis acceleration signals include tangential acceleration signals and normal acceleration signals; performing noise separation processing on the collected sound pressure signals by using the tangential acceleration signals and the normal acceleration signals to obtain purified heart sound signals; identifying a system transfer function by using the obtained purified heart sound signals and the normal acceleration signals; solving contact state parameters by using the tangential acceleration signals and the normal acceleration signals through short-time window statistical analysis; generating a bias voltage control signal according to the identified system transfer function and the solved contact state parameters; applying the generated bias voltage control signal to a bias voltage adjusting circuit to dynamically adjust a direct current bias voltage of the central acoustic unit.

[0011] Preferably, the noise separation processing step comprises: based on the tangential acceleration signals, as a reference signal, adaptively eliminating tangential friction noise in the sound pressure signals; based on a low-frequency component of the normal acceleration signals, as a reference signal, adaptively eliminating breathing and body movement noise in the sound pressure signals.

[0012] Preferably, the method further comprises: extract a ballistocardiogram signal based on the normal acceleration signal; perform heart cycle segmentation on the purified heart sound signal according to peak features of the ballistocardiogram signal.

[0013] Preferably, the step of identifying the system transfer function comprises: taking the ballistocardiogram signal represented by the normal acceleration signal as a system input excitation; taking the purified heart sound signal as a system output response; calculating a transfer function between the system input excitation and the system output response to represent the acoustic transfer characteristics.

[0014] Preferably, the contact state parameters comprise: a contact instability parameter obtained by analyzing the tangential acceleration signal and a contact pressure parameter obtained by analyzing the normal acceleration signal.

[0015] Preferably, the step of generating the bias voltage control signal comprises: generating a pressure compensation adjustment amount according to the contact pressure parameter in the contact state parameters; generating a friction suppression adjustment amount according to the contact instability parameter in the contact state parameters; generating a frequency response compensation adjustment amount according to the system transfer function; combining the pressure compensation adjustment amount, the friction suppression adjustment amount and the frequency response compensation adjustment amount to generate the bias voltage control signal.

[0016] Preferably, the method further comprises: performing deconvolution processing on the purified heart sound signal using an inverse function of the system transfer function to generate a standardized heart sound signal.

[0017] The present application provides a heart sound dedicated MEMS sensor and a signal processing method thereof. The present application has the following beneficial effects: 1. The present application integrates a central acoustic unit and a peripheral inertial unit in a single chip to ensure synchronization of the sound pressure signal and the acceleration signal in time and space. By decomposing the acceleration into tangential and normal components and using cascaded adaptive filtering processing, the present application can accurately eliminate the friction noise related to the tangential acceleration and the breathing and body movement noise related to the low-frequency component of the normal acceleration, respectively, to obtain a purified heart sound signal with high signal-to-noise ratio and high-fidelity morphological features.

[0018] 2、The application obtains the contact instability and contact pressure parameters by real-time solving the tangential and normal acceleration signals, and combines the identified individual system transfer function, so that the processor can dynamically generate a bias voltage control signal. The signal actively adjusts the sensitivity of the center acoustic unit, realizes compensation for the change in pressing, instantaneous inhibition of severe friction and frequency response compensation for individual chest cavity attenuation, and ensures that the optimal signal can be collected under different users and dynamic scenes.

[0019] 3、The application solves the key problem that the traditional heart sound signal lacks comparability due to individual physiological differences (such as chest wall thickness). By associating the normal acceleration signal (as the system excitation) with the purified heart sound signal (as the system response), the application can identify the system transfer function representing the acoustic characteristics of the individual chest cavity. Then, through deconvolution processing, the individual filtering effect is eliminated, thereby generating a standardized heart sound signal, which provides a highly consistent data basis for cross-population objective diagnosis and AI model training. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 is a perspective view of the application; Figure 2 is a schematic view of the internal structure of the application; Figure 3 is a schematic view of the signal processing method flow of the application; Figure 4 is a schematic view of the overall structure of the sensor system of the application; Figure 5 is a schematic view of the noise separation processing flow of the application; Figure 6 is a schematic view of the closed-loop adaptive adjustment logic of the application.

[0021] Among them, 100, MEMS sensor shell; 200, silicon substrate; 300, capacitive MEMS microphone structure; 400, three-axis MEMS accelerometer structure; 500, processor. DETAILED DESCRIPTION

[0022] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the specification of the application. Obviously, the described embodiments are only some of the embodiments of the application, not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the application.

[0023] With reference to the drawings Figure 1 and the drawings Figure 2 , the embodiments of the application provide a heart sound dedicated MEMS sensor.

[0024] The sensor includes a MEMS sensor housing 100. The MEMS sensor housing 100 is injection molded, for example, from medical-grade polycarbonate (PC) or ABS engineering plastic, and provides structural support, physical protection, and a certain degree of electromagnetic shielding for all internal electronic and sensing elements.

[0025] A silicon substrate 200 is fixedly connected to the inner bottom of the MEMS sensor housing 100. The silicon substrate 200 is a single-crystal silicon substrate, which is fixed to the inner bottom wall of the MEMS sensor housing 100 by means of epoxy resin adhesive with high thermal conductivity and high mechanical strength. This fixing method ensures that when the MEMS sensor housing 100 comes into contact with human skin, the mechanical vibration of the chest cavity can be transmitted to the silicon substrate 200 with high fidelity.

[0026] A core technical feature of this invention is that the central acoustic unit and the peripheral inertial unit are monolithically integrated on the same silicon substrate 200 using microelectromechanical systems (MEMS) technology. Physically, the peripheral inertial units are positioned on the silicon substrate 200, surrounding the central acoustic unit. This monolithically integrated, coplanar, and common-substrate layout ensures that the sound pressure signal and the triaxial acceleration signal are synchronously acquired at exactly the same spatial point, fundamentally eliminating the spatial sampling asynchrony and phase mismatch problems caused by the separate placement of multiple sensors.

[0027] In one specific embodiment, the central acoustic unit is specifically a capacitive MEMS microphone structure 300. (See attached...) Figure 3 As shown, the structure includes a flexible diaphragm made of silicon nitride or doped polycrystalline silicon, and a corresponding rigid porous backplate. An air gap is formed between the diaphragm and the backplate, constituting a parallel-plate capacitor. An external sound pressure signal causes the diaphragm to vibrate, thereby changing the capacitance value and realizing the conversion of acoustic signals into electrical signals.

[0028] The peripheral inertial unit is specifically a triaxial MEMS accelerometer structure 400. This structure employs a comb-like differential capacitance sensing principle, comprising a central mass block fixed by a cantilever beam and multiple sets of fixed comb-like electrodes arranged around the mass block. When the sensor... , , When acceleration occurs in the three axes, the central mass block will undergo a corresponding small displacement due to inertia, resulting in a differential change in the capacitance between the movable comb teeth and the fixed comb teeth. The triaxial acceleration signal can be calculated by detecting this capacitance change.

[0029] The MEMS sensor housing 100 also houses a processor 500. The processor 500 is a low-power microcontroller (MCU) with digital signal processing (DSP) capabilities, such as a chip based on an ARM Cortex-M core. The processor 500 is electrically connected to the input terminals of the readout circuits of the central acoustic unit and the peripheral inertial unit, respectively, through its internally integrated multi-channel analog-to-digital converter (ADC). The processor 500 functions to receive and store sound pressure signals and triaxial acceleration signals in real time, and acts as the execution core to run the signal processing methods described in the foregoing embodiments of this invention.

[0030] This embodiment also includes a bias voltage adjustment circuit. The bias voltage adjustment circuit is electrically connected to both the processor 500 and the condenser MEMS microphone structure 300. Specifically, the control input terminal of the bias voltage adjustment circuit is connected to a digital control output terminal of the processor 500, such as the output terminal of a digital-to-analog converter (DAC) or a pulse-width modulation (PWM) output terminal. The voltage output terminal of the bias voltage adjustment circuit is connected to the backplate electrode or diaphragm electrode of the condenser MEMS microphone structure 300.

[0031] The bias voltage adjustment circuit, for example, consisting of a DAC chip and a subsequent voltage follower, receives the digitized external control signal from the processor 500—that is, the bias voltage control signal generated by the aforementioned method—and converts it into a precise, stable, and dynamically adjustable DC bias voltage, which is then applied to the capacitive MEMS microphone structure 300. By adjusting this bias voltage, the microphone's sensitivity and frequency response characteristics can be actively changed, thus forming a complete closed-loop adaptive adjustment system.

[0032] See attached document Figure 4 The diagram illustrates the overall structure of the sensor system in this embodiment, clearly showing the signal flow and control relationships between the various hardware units.

[0033] See attached document Figure 1 This invention provides a signal processing method for a dedicated MEMS sensor for heart sounds. This method is executed by a processor 500 inside the sensor and specifically includes the following steps: The sound pressure signal of the central acoustic unit and the triaxial acceleration signal of the peripheral inertial unit are acquired synchronously, and the triaxial acceleration signal is decomposed into tangential acceleration signal and normal acceleration signal; The collected sound pressure signals are processed by noise separation using tangential acceleration signals and normal acceleration signals to obtain purified heart sound signals; The system transfer function was identified using the obtained purified heart sound signal and normal acceleration signal, and the contact state parameters were calculated using the tangential acceleration signal and normal acceleration signal. Based on the identified system transfer function and the calculated contact state parameters, a bias voltage control signal is generated and output to the bias voltage adjustment circuit to dynamically adjust the DC bias voltage of the central acoustic unit.

[0034] Cardiac cycle segmentation is performed using the normal acceleration signal and the purified heart sound signal; and deconvolution processing is performed using the inverse function of the system transfer function and the purified heart sound signal to generate a standardized heart sound signal.

[0035] In the initial steps of the signal processing method, the processor 500 is responsible for the synchronous acquisition and decomposition of multimodal signals. Synchronous acquisition is a technical prerequisite for ensuring accurate results in subsequent fusion processing steps such as noise separation and system identification. Specifically, the processor 500 uses its internally integrated multi-channel analog-to-digital converter and a unified system clock source to synchronously sample and quantize the analog sound pressure signal from the central acoustic unit and the three analog acceleration signals from the peripheral inertial units. This process ensures that the final generated digital sound pressure signal data stream and the three-axis acceleration signal data stream are strictly aligned at every discrete time point index, eliminating uncertainties caused by sampling phase deviations.

[0036] After obtaining the synchronized triaxial acceleration signal, the processor 500 decomposes the signal in real time according to the sensor coordinate system definition preset in the internal memory, extracting signal components with clear physical meaning. This coordinate system uses the contact surface between the sensor and human skin as the reference plane, defining the direction perpendicular to the contact surface outwards as the normal. The acceleration in this direction primarily reflects the chest wall vibration caused by heartbeats, as well as low-frequency motion caused by breathing or body movement. A plane parallel to the contact surface is defined as the tangential plane. The acceleration on the plane primarily reflects the relative sliding or friction between the sensor and the skin.

[0037] Based on this coordinate system definition, the triaxial accelerometer directly outputs... The axial acceleration component is specified as the normal acceleration signal. .at the same time, shaft and The acceleration components of the axis are then used to calculate the tangential acceleration signal. The tangential acceleration signal is defined as the magnitude of the acceleration vector on the tangential plane, and its calculation method is as follows: ; in, Indicates a discrete time point index; Represents at discrete time points The calculated tangential acceleration signal; Represents the triaxial acceleration signal at discrete time points of Axial components; Represents the triaxial acceleration signal at discrete time points of Axial components.

[0038] By performing the aforementioned synchronous acquisition and signal decomposition steps, the processor 500 successfully converted the original sensor data stream into one sound pressure signal, one normal acceleration signal, and one tangential acceleration signal. These three data streams are strictly synchronized in time and orthogonal to each other in a physical sense, providing high-quality and information-rich input for subsequent core algorithms such as noise separation, system characteristic identification, and contact state calculation.

[0039] See attached document Figure 5 During noise separation processing, the processor 500 first performs a first-stage adaptive noise cancellation, the purpose of which is to eliminate tangential friction noise introduced by the relative sliding between the sensor and the skin surface. This step utilizes the tangential acceleration signal obtained in the previous processing stage. .

[0040] The physical principle behind this processing lies in the tangential acceleration signal. The vibrations on the sensor's tangential plane, as characterized, are highly correlated with the frictional noise component contained in the original sound pressure signal, but not with the actual heart sound signal component transmitted via the normal (vertical) path.

[0041] In one specific embodiment, the processor 500 employs an adaptive filter based on the Normalized Least Mean Square (NLMS) algorithm to implement this function, thereby improving the convergence stability of the algorithm under different signal amplitudes. The raw sound pressure signal acquired from the center acoustic unit is defined as the first main input signal. .

[0042] Processor 500 will receive the tangential acceleration signal As the first reference input signal ,Right now .

[0043] An adaptive filter internally maintains a set of adjustable filter coefficient vectors. The order (or length) of this vector is preset to M. At each discrete time point The filter is based on the reference input vector (i.e., from) (and its M-1 historical values ​​forming a vector) and coefficient vector An estimated friction noise signal is calculated. .

[0044] Subsequently, the processor 500 receives the first main input signal. Subtract the estimated friction noise signal from the middle. The first-level purification signal was obtained. The calculation formula is as follows: ; in, It is a coefficient vector Transpose of; It is the reference input vector.

[0045] At the same time, the processor 500 utilizes this first-stage purification signal The value of is used to update the filter coefficient vector according to the iteration rules of the NLMS algorithm. For the next time point Use it. Its update formula is as follows: ; in, It is a preset step size factor used to control the convergence speed of the algorithm; It is the reference input vector The energy (square of the L2 norm); It is a very small positive constant used to prevent the denominator from being zero when the input energy is close to zero, ensuring the stability of the calculation. Through this step, the tangential friction noise in the sound pressure signal is adaptively eliminated. First-stage purified signal. It will be used as the main input signal for the next stage of noise cancellation (i.e., cancellation of breathing and body movement noise).

[0046] After completing the first stage of tangential friction noise elimination, the processor 500 continues to perform the second stage of noise separation, which aims to eliminate low-frequency body motion noise introduced by the subject's breathing and body swaying.

[0047] This second-stage processing will process the first-level purification signal output from the previous stage. As the second main input signal, denoted as .

[0048] The physical principle behind this processing is that low-frequency bodily noise, such as breathing and body swaying, has its main energy concentrated in a lower frequency range (e.g., below 2Hz), and these movements are primarily manifested in normal acceleration signals. In the low-frequency components. Therefore, low-frequency components and The low-frequency body motion noise components contained therein are highly correlated.

[0049] To obtain this noise reference, the processor 500 first processes the normal acceleration signal. A digital low-pass filter is performed. In one specific embodiment, the low-pass filter is an IIR (Infinite Impulse Response) or FIR (Finite Impulse Response) filter with a preset cutoff frequency of 2Hz. The signal obtained after filtering is the second reference input signal. .

[0050] Subsequently, the processor 500 employs a separate, second adaptive filter based on the Normalized Least Mean Square (NLMS) algorithm. This filter internally maintains a set of adjustable filter coefficient vectors. Its order (or length) is preset to N.

[0051] At each discrete time point The filter is based on the second reference input vector. (i.e., from) (and its N-1 historical values ​​forming a vector) and coefficient vector An estimated low-frequency body motion noise signal is calculated. Processor 500 receives the second main input signal. Subtract the estimated low-frequency body motion noise signal from the middle. The final purified heart sound signal is obtained. The calculation formula is as follows: ; in, It is a coefficient vector Transpose of; This is the second reference input vector. This is the purified heart sound signal sought by this invention.

[0052] At the same time, the processor 500 utilizes this final purified heart sound signal. The value of is used to update the filter coefficient vector according to the iteration rules of the NLMS algorithm. For the next time point Use it. Its update formula is as follows: ; in, It is a preset second step size factor used to control the convergence speed of the second-stage filter; It is the second reference input vector Energy; It is a positive number used to prevent the denominator from being zero.

[0053] Through the cascaded adaptive filtering process described in the first and second stages, the tangential friction noise and normal low-frequency body motion noise in the original sound pressure signal are effectively eliminated, and the processor 500 finally obtains a high-quality purified heart sound signal. This signal will be used for subsequent system identification, state calculation, and cardiac cycle segmentation.

[0054] High-quality purified heart sound signal This represents the processor 500's best estimate of the actual heart sound signals collected by the central acoustic unit.

[0055] In this signal, the tangential acceleration signal Strongly correlated high-frequency friction noise components, and their correlation with the normal acceleration signal. The low-frequency components of breathing and body movement noise that are strongly correlated have been effectively separated and eliminated by the above dual-path adaptive filtering process.

[0056] Therefore, the obtained high-quality purified heart sound signal Compared to the original acquired sound pressure signal It exhibits a significantly improved signal-to-noise ratio. Key morphological features of the heart sound signal, such as the pacemaker, amplitude, and spectral structure of the first and second heart sounds, as well as details of potential heart murmurs or additional heart sounds, are preserved with high fidelity.

[0057] This high-quality purified heart sound signal It is a key foundation for realizing the subsequent advanced functions of this invention. It will be used as a highly reliable input for parallel execution of system transfer function identification, cardiac cycle segmentation, and generation of standardized heart sound signals.

[0058] High-quality purified heart sound signals were obtained from the processor 500. The method then enters a parallel processing phase. In this phase, the processor 500 uses the acquired signals to simultaneously initiate two computationally independent key tasks: the first task is to identify the system transmission characteristics of the sensor coupling with the human chest cavity; the second task is to calculate the current contact state parameters of the sensor.

[0059] The purpose of this parallel processing architecture is to efficiently provide real-time and complete decision-making support for the two core functions that follow (i.e., closed-loop adaptive adjustment and advanced analysis).

[0060] Specifically, the identified system transmission characteristics and calculated contact state parameters will serve as the core inputs for generating the bias voltage control signal, enabling dynamic closed-loop adaptive adjustment of the central acoustic unit. Simultaneously, these identified and calculated parameters are also essential for subsequent advanced analyses such as heart sound signal standardization and cardiac cycle segmentation.

[0061] One of the parallel processing tasks is to identify and quantify the transmission characteristics of the coupled acoustic system formed by the sensor and the subject's chest cavity. The physical model for this step considers the subject's chest tissue, the sensor-skin interface, and the sensor's own mechanical structure as a linear force-sound transmission system. This system describes how the internal mechanical vibrations generated by the heartbeat propagate through the chest cavity medium and are ultimately converted into sound pressure signals by the central acoustic unit.

[0062] In this model, the normal acceleration signal It is used as the system input stimulus. This is because... Essentially, it's a seismocardiogram (SCG) signal, which directly reflects the mechanical impact force on the chest wall during cardiac contraction and relaxation, and is the original driving source of the entire acoustic event. Therefore, the normal acceleration signal... Specify as system input .

[0063] Correspondingly, high-quality purified heart sound signals are obtained in the previous processing stage. This is used as the system output response. This signal is the final manifestation of the acoustic signal formed on the body surface after the mechanical vibrations of the heart propagate through the pleural cavity. Therefore, a high-quality purified heart sound signal is generated. Designated as system output signal .

[0064] Because real biological signals contain noise and are not perfectly stationary, processor 500 employs a power spectrum-based calculation method to obtain robust system transfer function estimation. Specifically, processor 500 estimates the input signal within a certain time window. and output signal Segment the system (e.g., using the Welch method), and then calculate its power spectral density. System transfer function. In the frequency domain, it is given by the following formula: ; in, The desired system transfer function is a complex function whose magnitude represents the system's transfer function at a given frequency. Gain at a given frequency, phase indicates the system's gain at a given frequency. The phase delay at that point. This function characterizes the unique attenuation or enhancement properties of an individual's chest wall (including fat and muscle thickness) to different frequency heart sound components. Input signal and output signal The cross-power spectral density. It measures the cross-power spectral density at frequency. Output signal and input signal The part that is linearly related, Input signal The self-power spectral density. It measures the power at a frequency of [missing information]. The energy distribution of the input signal itself.

[0065] The system transfer function calculated using this method This study precisely quantifies the personalized acoustic channel characteristics between the cardiac mechanical vibration source and the acoustic signal on the body surface, providing a crucial mathematical basis for subsequent frequency response compensation and standardized processing of heart sound signals in closed-loop regulation.

[0066] Executed in parallel with the system transfer function identification task is the real-time calculation of the sensor's current physical contact state by the processor 500. This task is achieved by performing short-time window statistical analysis on the acceleration signal stream, thereby converting dynamic signal characteristics into quantified state parameters.

[0067] Specifically, processor 500 will transmit continuous tangential acceleration signals and normal acceleration signal The data stream is divided into segments of length [length missing]. A series of data frames from multiple sampling points. These data frames can overlap to ensure smooth parameter updates. Within each data frame, the processor 500 calculates specific statistics to extract contact state information.

[0068] The first contact state parameter is contact instability, which quantifies the relative sliding or friction occurring between the sensor and the skin. This parameter is obtained by analyzing the tangential acceleration signal. The energy is obtained within a short time window (i.e., one data frame). When sliding friction occurs between the sensor and the skin, high-frequency vibrations are generated, leading to a significant increase in the amplitude and energy of the tangential acceleration signal. Within one data frame, the contact instability parameter... It can be calculated using the following formula: ; in, This is the length of the data frame; For the first in this frame One tangential acceleration sample value. The calculated The magnitude of the value directly reflects the intensity of tangential friction within that time window.

[0069] The second contact state parameter is the contact pressure, which is used to estimate the magnitude of the static pressing force applied to the sensor. This parameter is obtained by analyzing the normal acceleration signal. The average (DC component) over a short time window is obtained. This is because the static pressing force applied to the sensor will generate a DC or quasi-DC bias in the normal acceleration signal that is proportional to the magnitude of the pressing force. Within a data frame, the contact pressure parameter... It can be calculated using the following formula: ; in, The length of the data frame. For the first in this frame Each normal acceleration sample value. The calculated... The value provides a quantitative indicator that is linearly related to the actual contact pressure.

[0070] Contact instability parameters are calculated in parallel within each time window. and contact pressure parameters The processor 500 obtains a real-time, quantitative description of the physical contact quality of the sensor. These two parameters will be compared with the previously identified system transfer function. Together, they serve as the core decision-making basis for generating the closed-loop adaptive adjustment control signal in the next processing step.

[0071] See attached document Figure 6 Then it enters its core closed-loop adaptive adjustment stage. In this stage, the processor 500 generates a bias voltage control signal based on the analysis results obtained from the aforementioned parallel processing tasks. The ultimate goal of this signal is to achieve dynamic and intelligent adjustment of the DC bias voltage of the central acoustic unit.

[0072] The decision-making logic of this adjustment process integrates information extracted from both the frequency and time domains. Specifically, the input to this decision-making logic consists of all the parameters identified and calculated in the previous section: System transfer function This parameter represents the individualized characteristics of the chest acoustic channel in the test subject.

[0073] Contact condition parameters: This group of parameters includes contact instability parameters. and contact pressure parameters Both of these factors together quantify the quality of the physical coupling between the sensor and the skin.

[0074] The processor 500 will analyze these three inputs to generate an optimal bias voltage adjustment instruction.

[0075] After obtaining the system transfer function and contact state parameters , Following this, the processor 500 generates three independent adjustment quantities with clear physical meanings in parallel based on a multi-dimensional decision-making logic. These three adjustment quantities address different physical problems and together form the basis for achieving closed-loop adaptive adjustment.

[0076] First, processor 500 generates pressure compensation adjustment amount. The purpose of this adjustment is to compensate for the sensitivity drift of the central acoustic unit caused by changes in the sensor's pressure depth. In one specific embodiment, the processor 500 internally stores a pressure bias lookup table. This lookup table, obtained through experimental calibration during the sensor design phase, accurately maps different contact pressure parameters. The value corresponds to the bias voltage correction value that enables the microphone diaphragm system to operate in the optimal linear region and sensitivity point. The processor 500 will calculate the value in real time. Using the value as an index, a corresponding voltage adjustment value is obtained by querying the lookup table or by interpolation; this is the pressure compensation adjustment amount. This adjustment ensures that the acoustic sensor core always maintains optimal operating conditions regardless of changes in the user's pressing pressure.

[0077] Secondly, the processor 500 generates friction suppression adjustment amount. The purpose of this adjustment is to proactively and instantaneously reduce the system sensitivity when severe friction is detected, to prevent signal clipping distortion caused by friction noise, and to reduce the burden on subsequent adaptive noise cancellation algorithms. The processor 500 will calculate the contact instability parameters in real time. With a preset trigger threshold Comparison. When If the threshold is not exceeded, it indicates stable contact, and the friction suppression adjustment amount is adjusted accordingly. It is set to zero. Once Exceeding the threshold The processor 500 then generates a significant negative friction suppression adjustment amount as... The absolute value of this negative friction suppression adjustment amount can be compared with... The degree to which the threshold is exceeded is proportional to the sensitivity, thus achieving proportional suppression. This adjustment is a dynamic, protective, instantaneous response.

[0078] Finally, the processor 500 generates the frequency response compensation adjustment. The purpose of this adjustment is to compensate for the overall energy of the signal based on the individual subject's chest acoustic characteristics. The processor 500 analyzes the identified system transfer function. amplitude response Specifically, it is calculated within the critical energy frequency band of the heart sound signal (e.g., a preset band from 20Hz to 400Hz). The average value or energy integral is used to quantify the overall attenuation of heart sound signals in the individual's chest cavity. If the processor 500 determines that the attenuation exceeds a preset nominal attenuation value (indicating that the subject has, for example, a high body fat percentage), it will generate a positive adjustment amount as... This positive adjustment will be used to increase the bias voltage, thereby improving the overall sensitivity of the acoustic unit to compensate for signal energy loss caused by human tissue, and thus improve the signal-to-noise ratio of weak heart sound signals.

[0079] Pressure compensation adjustment was generated in parallel. Friction suppression adjustment amount and frequency response compensation adjustment amount The processor 500 then merges these three independent adjustment values ​​to form a final, comprehensive bias voltage adjustment instruction.

[0080] Specifically, the processor 500 algebraically sums these three adjustment values ​​to obtain a total bias voltage change. It is worth noting that this is an algebraic sum because the pressure compensation adjustment and frequency response compensation adjustment are typically positive or zero, while the friction suppression adjustment is negative when triggered. This combination allows the system to respond simultaneously to multiple physical changes, such as instantaneously suppressing a severe friction event (requiring a reduction in bias pressure) while compensating for individual thoracic cavity attenuation (which requires increasing bias pressure).

[0081] Subsequently, the processor 500 will calculate the total change. Applying the current bias voltage setting to obtain a new target bias voltage. To ensure a smooth adjustment process and prevent sudden voltage changes from interfering with signal acquisition, the processor 500 will check the total change before updating the voltage. Apply a low-pass filter or rate limiter.

[0082] Ultimately, processor 500 will represent this new target bias voltage. A digital value is converted into a stable analog DC voltage via an internal or external digital-to-analog converter (DAC). This analog voltage is then applied to the sensing core of the central acoustic unit (e.g., an electret microphone or MEMS microphone), thus completing a closed-loop adjustment of the sensor's operating point.

[0083] The entire process repeats itself at set time intervals (e.g., after processing each data frame), forming a continuous adaptive adjustment closed loop that ensures that the sensor can always be intelligently maintained in the best data acquisition state under dynamically changing usage scenarios.

[0084] After completing the closed-loop adaptive adjustment of the acquisition system, high-quality purified heart sound signals were obtained. Subsequently, the method of the present invention can further perform a series of optional enhancement processing steps. The aim is to further improve the diagnostic value and analytical consistency of heart sound signals.

[0085] These processes are not intended to eliminate external noise (this task has been accomplished by the noise separation module in Section 2), but rather to focus on optimizing and standardizing the intrinsic characteristics of the heart sound signal itself. This ensures that the final output heart sound signal is comparable regardless of the subject's physiological conditions (such as chest wall thickness), and that key pathophysiological features are highlighted.

[0086] This stage of treatment will primarily utilize high-quality purified heart sound signals. As input, and combined with the individualized system transfer function identified in Section 3 This is to achieve in-depth signal processing. These enhancement processing steps include, but are not limited to: standardizing the signal to eliminate the influence of individual chest cavity differences, and sharpening key components of heart sounds to assist in subsequent automatic segmentation and recognition.

[0087] As a core enhancement, the processor 500 utilizes the fusion of multimodal signals to achieve precise segmentation of the cardiac cycle. This function overcomes the limitations of traditional segmentation that relies solely on acoustic signals, thereby resolving the inherent ambiguity of heart sound signals in locating the cardiac cycle reference point.

[0088] Its basic principle lies in the normal acceleration signal. The cardiac impulse map (SCG), as a signal reflecting the mechanical motion of the heart, contains a more acute and stable time reference point than acoustic signals. Specifically, the processor 500 utilizes the peak value in the SCG signal that precisely corresponds to the aortic valve opening event. Compared to the broad, low-frequency, and morphologically variable characteristics of the first heart sound acoustically, the peak value generated by mechanical vibration is more concentrated in time, more morphologically stable, and less susceptible to interference from respiration and noise, thus making it an ideal anchor point for locating the beginning of the cardiac cycle.

[0089] In the specific implementation, the processor 500 pairs the normal acceleration signal. The data stream is processed using a peak detection algorithm. This algorithm is specifically configured to find peaks that conform to the morphological characteristics of the J wave; for example, it will look for the most significant positive peak that meets specific amplitude and sharpness thresholds within a predicted time range of a cardiac cycle.

[0090] Once the processor 500 is in continuous By locating a series of J-wave peak timestamps in the signal, it possesses a precise mechanical starting marker for each cardiac cycle.

[0091] Subsequently, the processor 500 uses these high-precision wave timestamps as anchor points to trace back and align with the synchronously acquired high-quality purified heart sound signals. Due to physiological connections, the timing of the wave's occurrence must fall within the composite waveform of the first heart sound. Therefore, the processor 500 can resonate around each return-to-wave anchor point. A time window is opened to accurately separate the complete waveform of S1.

[0092] Once the first heart sound is accurately framed, locating the next major acoustic event (i.e., the second heart sound) becomes more reliable. The processor 500 can search for the next energy-concentrated acoustic event within the time region following the first heart sound to determine the second heart sound. By using the wave-to-wave time interval, the processor 500 can also obtain a highly reliable heart rate estimate, which helps to further constrain the search range for the second heart sound.

[0093] Ultimately, through this method based on mechanical signal anchoring, the processor 500 is able to transmit high-quality purified heart sound signals. It accurately segments the heart into a series of physiologically significant segments, including: the first heart sound, the second heart sound, the systolic phase (between the first and second heart sounds), and the diastolic phase (between the second and the next first heart sound). This level of segmentation precision is unmatched by traditional methods that rely solely on heart sound signal segmentation.

[0094] Another core enhancement process aims to generate a standardized heart sound signal to eliminate the acoustic effects caused by physiological differences (such as chest wall thickness, tissue density, etc.) between different subjects, thereby improving the comparability of the signal between different individuals.

[0095] The physical principle lies in the system transfer function identified in Section 3. The filtering effect produced by the individual chest cavity as an acoustic channel has been quantified. Therefore, by performing an inverse operation on this filtering effect, i.e., mathematical deconvolution, it is theoretically possible to purify high-quality heart sound signals that have been attenuated and distorted. In the process, the original heart sound signal, which is closer to that produced by the sound source (i.e., the heart valve itself), is recovered.

[0096] In its implementation, processor 500 performs this inverse operation in the frequency domain, because deconvolution in the frequency domain can be simplified to a direct division operation. Processor 500 first acquires high-quality purified heart sound signals. The spectrum, denoted as Subsequently, the obtained individualized system transfer function was used. Calculate the spectrum of the standardized heart sound signal. The calculation formula is as follows: ; in, It is the frequency domain representation of the standardized heart sound signal, which represents an idealized heart sound signal, that is, an estimate of the signal that has not been affected by the filtering effect of the thoracic tissue of a specific individual during propagation; It is a high-quality purified heart sound signal The spectrum, It is the previously identified system transfer function that characterizes the acoustic properties of an individual's chest cavity.

[0097] To ensure the stability and robustness of the computation, especially At frequency points where the amplitude is close to zero (i.e., the frequency band where the signal is severely attenuated by the chest cavity), direct division would lead to excessive amplification of noise. Therefore, in practical embodiments, the processor 500 employs a regularized inverse filtering method to avoid numerical instability at these frequency points and ensure that the final standardized signal has a good signal-to-noise ratio.

[0098] After calculating the standardized spectrum Then, the processor 500 performs an inverse Fourier transform on it to obtain the final time-domain standardized heart sound signal. This signal provides doctors with more consistent and comparable auscultation evidence, and also provides high-quality, standardized input data for training machine learning diagnostic models across populations.

Claims

1. A dedicated MEMS sensor for heart sounds, comprising a MEMS sensor housing (100), characterized in that, A silicon substrate (200) is fixedly connected to the inner bottom of the MEMS sensor housing (100). A central acoustic unit and a peripheral inertial unit are monolithically integrated on the silicon substrate (200). The central acoustic unit is used to collect sound pressure signals. The peripheral inertial unit is arranged around the central acoustic unit and is used to collect triaxial acceleration signals. A processor (500) is arranged inside the MEMS sensor housing (100). The processor (500) is electrically connected to the central acoustic unit and the peripheral inertial unit.

2. The MEMS sensor for heart sounds according to claim 1, characterized in that, The central acoustic unit is a capacitive MEMS microphone structure (300), and the peripheral inertial unit is a triaxial MEMS accelerometer structure (400).

3. A dedicated MEMS sensor for heart sounds according to claim 2, characterized in that, It also includes a bias voltage adjustment circuit, which is electrically connected to the capacitive MEMS microphone structure (300) and the processor (500), for receiving external control signals from the processor (500) and dynamically adjusting the DC bias voltage of the capacitive MEMS microphone structure (300) according to the external control signals.

4. A signal processing method for a dedicated MEMS sensor for heart sounds, applied to a dedicated MEMS sensor for heart sounds as described in any one of claims 1-3, characterized in that, Includes the following steps: The sound pressure signal of the central acoustic unit and the triaxial acceleration signal of the peripheral inertial unit are acquired synchronously. The triaxial acceleration signal includes tangential acceleration signal and normal acceleration signal. The acquired sound pressure signal is processed by noise separation using the tangential acceleration signal and the normal acceleration signal to obtain a purified heart sound signal; The system transfer function is identified using the obtained purified heart sound signal and the normal acceleration signal. The contact state parameters are calculated using the tangential acceleration signal and the normal acceleration signal through short-time window statistical analysis. Based on the identified system transfer function and the calculated contact state parameters, a bias voltage control signal is generated; The generated bias voltage control signal is applied to the bias voltage adjustment circuit to dynamically adjust the DC bias voltage of the central acoustic unit.

5. The signal processing method for a dedicated MEMS sensor for heart sounds according to claim 4, characterized in that, The noise separation process includes the following steps: Based on the tangential acceleration signal as a reference signal, the tangential friction noise in the sound pressure signal is adaptively eliminated; Based on the low-frequency component of the normal acceleration signal, as a reference signal, respiratory and body movement noise in the sound pressure signal is adaptively eliminated.

6. The signal processing method for a dedicated MEMS sensor for heart sounds according to claim 4, characterized in that, The method also includes: Based on the normal acceleration signal, the cardiac impact map signal is extracted; Based on the peak characteristics of the cardiac impact signal, the purified heart sound signal is segmented into cardiac cycles.

7. The signal processing method for a dedicated MEMS sensor for heart sounds according to claim 4, characterized in that, The steps for identifying the system transfer function include: The cardiac impact signal, represented by the normal acceleration signal, is used as the system input excitation. The purified heart sound signal is used as the system output response; Calculate the transfer function between the system input excitation and the system output response to characterize the acoustic transfer characteristics.

8. The signal processing method for a dedicated MEMS sensor for heart sounds according to claim 4, characterized in that, The contact state parameters include: The contact instability parameters are obtained by analyzing the tangential acceleration signal and the contact pressure parameters are obtained by analyzing the normal acceleration signal.

9. The signal processing method for a dedicated MEMS sensor for heart sounds according to claim 4, characterized in that, The step of generating the bias voltage control signal includes: Based on the contact pressure parameter in the contact state parameters, a pressure compensation adjustment amount is generated; Based on the contact instability parameter in the contact state parameters, a friction suppression adjustment amount is generated; Based on the system transfer function, generate the frequency response compensation adjustment amount; The bias voltage control signal is generated by combining the pressure compensation adjustment, the friction suppression adjustment, and the frequency response compensation adjustment.

10. The signal processing method for a dedicated MEMS sensor for heart sounds according to claim 7, characterized in that, The method also includes: using the inverse function of the system transfer function to perform deconvolution processing on the purified heart sound signal to generate a standardized heart sound signal.

Images