Percutaneous hypoglossal nerve stimulation system and control methods, devices, and storage media therefor

Abstract

The application discloses a transdermal hypoglossal nerve stimulation system and a control method, equipment and storage medium thereof, the system comprising a sound sensor, a vibration sensor, a stimulation electrode and a processor; the processor is configured to: synchronously acquire acoustic signals and mandibular vibration signals, filter and verify the acoustic signals in a progressive manner to identify snoring sounds, filter and verify the mandibular vibration signals in the same manner to identify snoring sound associated vibrations; when both indicate the presence of snoring sounds, control the stimulation electrode to synchronously output two high-frequency sinusoidal carriers with different frequencies, and form a low-frequency envelope stimulation through time interference in the genioglossus target area; and stop outputting when the snoring sound characteristics disappear. The application reduces the algorithm power requirement through progressive filtering and verification to adapt to a wearable and light equipment, realizes deep target stimulation without pain on the epidermis through time interference, and realizes low-power-consumption high-precision closed-loop intervention.

Description

Technical Field

[0001] This invention relates to the field of medical device technology, specifically to a percutaneous hypoglossal nerve stimulation system and its control method, device, and storage medium. Background Technology

[0002] Obstructive sleep apnea (OSA) is a common chronic sleep disorder. Its pathological root cause is the relaxation of the genioglossus muscle and the collapse of soft tissues in the upper airway during sleep, which leads to repeated snoring and apnea. Long-term OSA can easily lead to secondary cardiovascular and metabolic complications.

[0003] Existing clinical protocols for OSA have inherent shortcomings: CPAP ventilators cause stuffiness, claustrophobia, and noise, with long-term clinical compliance rates below 50%; implantable hypoglossal nerve stimulation requires surgical implantation of a pulse generator and electrodes, which carries drawbacks such as surgical infection, secondary surgery for battery replacement, and high cost, making it suitable only for a small number of severely affected patients; traditional transcutaneous electrical stimulation uses a single low-frequency direct stimulation of the skin surface, resulting in shallow penetration depth, requiring increased current to be effective, and is prone to skin burning and abnormal facial muscle twitching, and often uses a single microphone sampling, making it susceptible to accidental triggering by environmental noise, and ineffective stimulation disrupting sleep; wearable snoring stimulators on the market generally use a hardware solution with separate microcontroller and Bluetooth module designs, resulting in numerous components, bulky size, high power consumption, frequent charging, and poor comfort when worn at night.

[0004] In recent years, temporal interference (TI) stimulation technology has provided a new approach for non-invasive deep neural modulation. Published technical solutions (such as Chinese invention patent application CN122377011A) propose a closed-loop neural modulation system based on temporal interference, using respiratory signals and surface electromyography (EMG) signals as trigger sources to perform TI stimulation on the mandibular region for obstructive sleep apnea (OSA). This solution applies TI technology to OSA scenarios, but the trigger signals used are respiratory and EMG signals, requiring additional physiological sensors, and it does not involve trigger recognition based on snoring itself. Regarding snoring recognition, published technical solutions (such as Chinese invention patent CN119867667A) propose a snoring detection method based on a smart bed, combining sleep sounds and bed vibration signals for snoring detection. The detection method involves randomly combining the collected signal features for model training, and then using the trained model for feature matching-based detection. However, it still has the following problems: First, it requires a long data acquisition and training period for model matching training, resulting in a poor user experience; second, the effectiveness and sensitivity of the signal used for snoring-assisted detection are questionable. Other published technical solutions (such as Chinese invention patent CN111374819A) propose a dual-modal snoring recognition method that combines sound and laryngeal acceleration data; however, this solution only uses simple dual-threshold AND gate logic, which is easily affected by external noise interference, and there is still room for improvement in the accuracy of snoring recognition. On the other hand, this solution requires continuous signal acquisition and a large amount of computation, which places high demands on the performance and battery life of the hardware, making it difficult to achieve the goals of miniaturization and long battery life. Summary of the Invention

[0005] Based on the above background, this invention provides a percutaneous hypolingual nerve stimulation system and its control method to achieve more accurate snoring detection, and optimizes for the low computing power constraints of wearable devices. Specifically, the following technical solution is adopted:

[0006] The first aspect of the present invention provides a percutaneous hypolingual nerve stimulation system, comprising at least one stimulator that can be applied to the mandible of a target subject, said stimulator being configured with at least a sound sensor, a vibration sensor, and stimulation electrodes; and a processor integrated with or communicatively connected to the stimulator, and configured to:

[0007] Simultaneously acquire acoustic signals collected by the sound sensor and mandibular vibration signals collected by the vibration sensor;

[0008] The acoustic signal and the mandibular vibration signal are preprocessed and feature extracted respectively. Within a set sampling period, the acoustic signal is filtered by decibel value, time / frequency domain feature verification and time sequence continuity verification are performed step by step. At the same time, the mandibular vibration signal is filtered by vibration amplitude, time / frequency domain feature verification and period continuity verification are performed step by step.

[0009] When the acoustic signal or mandibular vibration signal fails any level of filtering or verification, the determination of the current sampling period is stopped; only when both the acoustic signal that passes the temporal continuity verification and the mandibular vibration signal that passes the periodic continuity verification are obtained simultaneously, the triggering condition is determined to be met and the stimulation electrode is controlled to output a stimulation signal.

[0010] Furthermore, the time / frequency domain feature verification of the acoustic signal includes at least one or all of the following:

[0011] The duration of a single effective waveform of the time-domain signal is greater than a preset time threshold;

[0012] The rising and / or falling edge slope of the time-domain signal is less than a preset slope threshold;

[0013] The dominant frequency of the frequency domain signal is located within a preset frequency range;

[0014] The frequency of a frequency domain signal conforms to harmonic laws.

[0015] Furthermore, the time / frequency domain feature verification of the mandibular vibration signal includes at least one or all of the following:

[0016] The duration of a single effective waveform of the time-domain signal is greater than a preset time threshold;

[0017] The rising and / or falling edge slope of the time-domain signal is less than a preset slope threshold;

[0018] The waveform duty cycle of the time-domain signal is greater than a preset threshold;

[0019] The dominant frequency of the frequency domain signal is located within a preset frequency range;

[0020] The frequency of a frequency domain signal conforms to harmonic laws.

[0021] Furthermore, the temporal continuity verification of the acoustic signal and the periodic continuity verification of the mandibular vibration signal each include:

[0022] The signal satisfies the time / frequency domain feature verification condition for n consecutive frames, where n≥3;

[0023] The signal repeats periodically at preset intervals.

[0024] Furthermore, the processor is also configured to:

[0025] Waveform template matching is performed on acoustic signals that pass the temporal continuity check and / or mandibular vibration signals that pass the periodic continuity check.

[0026] When the acoustic signal or mandibular vibration signal fails to match the waveform template, the determination of the current sampling cycle is stopped; only when both the acoustic signal and the mandibular vibration signal that pass the waveform template matching are acquired simultaneously, it is determined that the triggering condition is met and the stimulation electrode is controlled to output a stimulation signal.

[0027] Furthermore, the processor is also configured to:

[0028] During the output of the stimulus signal, if either the acoustic signal or the mandibular vibration signal fails to pass the step-by-step filtering, verification, or waveform template matching, causing the triggering condition to no longer be met, the output of the stimulus signal shall be stopped.

[0029] Furthermore, the processor is also configured to:

[0030] During the routine monitoring phase, low-power sampling is performed at the first sampling frequency, and only coarse detection of decibel threshold and / or vibration amplitude is performed.

[0031] When both the instantaneous decibel level and the vibration amplitude reach the preset threshold, the sampling frequency is switched to a second sampling frequency higher than the first sampling frequency for full-speed sampling, and the complete trigger condition determination process is initiated.

[0032] Furthermore, the stimulator includes two stimulators that can be applied to the left and right sides of the target subject's mandible, and the processor controls the stimulation electrodes to output stimulation signals including:

[0033] The stimulation electrodes on the two stimulators synchronously output two high-frequency sinusoidal carriers with different frequencies. The frequency difference between the two high-frequency sinusoidal carriers is within a preset frequency difference range, so that the two high-frequency sinusoidal carriers form a low-frequency envelope stimulation at the target tissue of the target object through time interference.

[0034] Furthermore, the sound sensor and the vibration sensor share the same sampling clock and are synchronously initiated by the same hardware trigger signal. The processor is also configured to:

[0035] Based on the time delay characteristics between the acoustic signal and the mandibular vibration signal, signal homology verification is performed. When the signal homology verification also passes, it is determined that the triggering condition is met and the stimulation electrode is controlled to output a stimulation signal.

[0036] Furthermore, the signal homology verification based on the time delay characteristics between the acoustic signal and the mandibular vibration signal includes:

[0037] For each frame of the acoustic signal and the mandibular vibration signal, a sampling timestamp is recorded, and the original time difference between the two is calculated.

[0038] Calculate the single-frame delay fluctuation variance of the acoustic signal and the mandibular vibration signal respectively;

[0039] When the original time difference is greater than zero and the variance of the single-frame delay fluctuation is less than a preset threshold, the signals are determined to be from the same source.

[0040] Furthermore, the signal homology verification based on the time delay characteristics between the acoustic signal and the mandibular vibration signal also includes:

[0041] Based on the original time difference, the true relative delay of the acoustic vibration signal is obtained by subtracting the pre-calibrated inherent hardware delay.

[0042] When the actual relative delay of the acoustic vibration signal is also within a preset range, it is determined that the acoustic vibration originates from the same source.

[0043] Furthermore, the vibration sensor is a triaxial accelerometer, whose sensitive axis is approximately perpendicular to the surface of the target object's chin skin during application, and the processor is further configured to:

[0044] The sensitive axis is the main vibration axis, and the other two axes are auxiliary verification axes.

[0045] When the main vibration axis exhibits characteristic vibration and the amplitude of the auxiliary verification axis is lower than a preset threshold, it is determined to be local snoring vibration of the mandible; when the auxiliary verification axis simultaneously exhibits large fluctuations higher than a preset threshold, it is determined to be a whole-body movement and is filtered out.

[0046] A second aspect of the present invention provides a method for controlling a percutaneous hypoglossal nerve stimulation system, comprising:

[0047] Simultaneously acquire acoustic signals collected by the sound sensor and mandibular vibration signals collected by the vibration sensor;

[0048] The acoustic signal and mandibular vibration signal are preprocessed and feature extracted respectively. Feature filtering and verification are performed step by step based on a preset algorithm. The acoustic signal and mandibular vibration signal that pass the feature filtering and verification are respectively determined as valid snoring signal and valid snoring-related vibration.

[0049] The acoustic signal feature filtering verification includes, in sequence, decibel value filtering, time / frequency domain feature filtering, and time sequence continuity verification; the mandibular vibration signal feature filtering verification includes, in sequence, vibration amplitude filtering, time / frequency domain feature verification, and period continuity verification.

[0050] A stimulation control signal is generated to control the output stimulation signal of the stimulation electrode only when the effective snoring signal and the effective snoring-related vibration are acquired simultaneously.

[0051] Furthermore, the control method also includes:

[0052] Waveform template matching is performed on the acoustic signal and mandibular vibration signal that pass the feature filtering verification, and the acoustic signal and mandibular vibration signal that pass the waveform template matching are determined as valid snoring signal and valid snoring associated vibration;

[0053] The waveform template matching of the acoustic signal and the mandibular vibration signal respectively includes waveform similarity matching with a preset sound / vibration feature template.

[0054] A third aspect of the present invention provides a percutaneous hypolingual nerve stimulation device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the control method described in the second aspect above.

[0055] A fourth aspect of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the control method described in the second aspect above.

[0056] Compared with the prior art, the beneficial effects of the present invention include at least the following:

[0057] 1) Using dual-modal cross-validation of acoustic signals and mandibular vibration signals as the stimulus triggering condition, the acoustic and vibration signals both originate from the same target area and are triggered by the same physiological event, with strong signal homology. Stimulation is only triggered when both signals are simultaneously determined to be effective snoring signals and effective snoring-related vibrations, effectively avoiding false triggering caused by single-source interference such as environmental noise and heterogeneous vibrations, reducing the probability of ineffective stimulation, and improving sleep quality and patient compliance.

[0058] 2) The identification method adopts a step-by-step approach, which involves coarse filtering of decibel / vibration amplitude, time / frequency domain feature verification, time sequence / period continuity verification, and matching with waveform templates. Most non-target signals are intercepted at the first level of decibel or amplitude. This method can accurately distinguish pathological snoring from human voices, coughs, turning over, friction and other interference, while significantly reducing the computing power requirements and power consumption of the device. This allows high-precision snoring recognition to be achieved on a single chip without the need for additional dedicated computing hardware, making wearable and lightweight devices possible at the algorithm level.

[0059] 3) By employing a tiered sampling strategy that uses a first sampling frequency for low-power coarse detection during the normal monitoring phase and switches to a second sampling frequency for full-speed sampling only when snoring is suspected, the system balances detection sensitivity and battery life, extending the continuous working time of the device.

[0060] 4) By synchronizing timestamps with hardware, calculating and compensating for relative delay, combining the dual judgment of the real relative delay interval and the variance of single-frame delay fluctuation, and the physiological transmission timing of vibration leading sound, the same source verification of sound and vibration signals is realized. This can effectively filter the combined interference of environmental noise and heterogeneous vibration, and further reduce the false trigger rate.

[0061] 5) By using two high-frequency sinusoidal carrier waves with different frequencies to form a low-frequency envelope stimulation at the target tissue through time interference, the effective stimulation is located in the deep target area, realizing painless deep targeted stimulation of the epidermis, avoiding the skin burning pain and abnormal facial muscle twitching of traditional transcutaneous electrical stimulation; combined with closed-loop control that stops output when snoring characteristics disappear, sleep continuity is protected and the safety of long-term wear is improved. Attached Figure Description

[0062] Figure 1 This is a schematic diagram of the working logic of the percutaneous hypoglossal nerve stimulation system provided in an embodiment of the present invention.

[0063] Figure 2 This is a schematic diagram of the external structure of the stimulator in an embodiment of the present invention.

[0064] Figure 3 This is an exploded view of the stimulator in an embodiment of the present invention.

[0065] Figure 4 This is a schematic diagram illustrating the application effect of the stimulator in an embodiment of the present invention. Detailed Implementation

[0066] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0067] See Figure 1 This invention discloses a percutaneous hypolingual nerve stimulation system, comprising at least one stimulator that can be applied to the mandible of a target subject, which is equipped with at least a sound sensor, a vibration sensor, and stimulation electrodes; and a processor, integrated or communicatively connected to the stimulator, and configured to:

[0068] Simultaneously acquire acoustic signals collected by the sound sensor and mandibular vibration signals collected by the vibration sensor;

[0069] The acoustic signal and mandibular vibration signal are preprocessed and feature extraction is performed within a set sampling period. The extracted acoustic signal features are filtered by decibel value, verified by time / frequency domain features and verified by temporal continuity step by step. Simultaneously, the extracted mandibular vibration signal features are filtered by vibration amplitude, verified by time / frequency domain features and verified by periodic continuity step by step.

[0070] If the acoustic signal feature or the mandibular vibration signal feature fails any level of filtering or verification, the determination of the current sampling period is stopped; only when both the acoustic signal that passes the temporal continuity verification and the mandibular vibration signal that passes the periodic continuity verification are acquired simultaneously, the triggering condition is determined to be met and the stimulation electrode is controlled to output a stimulation signal.

[0071] The technical solution of the present invention will be further described in detail below with reference to specific embodiments.

[0072] Example 1

[0073] This embodiment discloses an illustrated example of the stimulator in this invention. See also Figure 2 and Figure 3 The stimulator shown in this embodiment includes a housing 1 composed of upper and lower shells, and a circuit board 15 and a battery 16 installed inside the housing. Electrode holes are provided on the application surface of the housing 1, through which two stimulation electrodes 11 mounted on the circuit board 15 protrude. For ease of application, a magnetically attachable replaceable hydrogel patch 14 is also provided on the application surface of the housing 1. Correspondingly, magnetic rings 12 are provided at each of the two stimulation electrodes, thereby attracting the replaceable hydrogel patch 14 to the application surface of the housing 1. In addition, the housing 1 also has charging contacts 13 for connecting to the battery 16. See also... Figure 3 The circuit board also includes a sound sensor 17 and a vibration sensor 18, and the housing 1 has a sound-receiving hole 19 adapted to the sound sensor 17. The entire stimulator is lightweight, with a net weight of no more than 20g.

[0074] In a preferred embodiment, the sound sensor 17 uses a MEMS microphone as the sound sensor, a triaxial MEMS accelerometer as the vibration sensor, two isolated high-frequency pulse output electrodes as stimulation electrodes, and an nRF52840 Bluetooth system-on-a-chip (SoC) as the processor. The SoC integrates a 12-bit analog-to-digital converter (ADC), a 1MHz hardware timer, an ARM Cortex-M4 arithmetic unit, and a Bluetooth 5.0 communication module. In this embodiment, replacing the traditional architecture of a separate microcontroller and Bluetooth module with a single chip further simplifies the peripheral circuitry, achieving miniaturization and ultra-low power consumption. The printed circuit board adopts a partitioned layout, physically separating the vibration acquisition area from the radio frequency area and the stimulation driving area, and adding grounding shielding, which effectively suppresses electromagnetic crosstalk from the electrical stimulation circuit to the sensor signal.

[0075] In a further preferred embodiment, the sound sensor uses an analog output MEMS microphone, which has the advantages of low power consumption, miniaturization, and suitability for embedded installation. Preferred microphone parameters are: frequency response range 20Hz~10kHz (fully covering the human voice and snoring frequency bands), sensitivity -42dB, signal-to-noise ratio ≥62dB, low power consumption (operating current <150μA), and compatibility with four levels of low power consumption modes in compatible devices.

[0076] It should be noted that this embodiment only illustrates a preferred structure and implementation of the stimulator. The stimulator can be implemented using various different structures and devices without departing from the inventive concept, and this does not constitute a limitation.

[0077] Example 2

[0078] This embodiment illustrates a specific implementation scheme for preprocessing acoustic signals and mandibular vibration signals and extracting features within a set sampling period, and for progressively filtering the extracted acoustic signal features using decibel values, verifying time / frequency domain features, and verifying temporal continuity, as well as simultaneously performing progressively filtering vibration amplitude, verifying time / frequency domain features, and verifying periodic continuity on the extracted mandibular vibration signal features.

[0079] In this embodiment, the preprocessing of the acoustic signals collected by the microphone includes:

[0080] 1) Use a front-end signal conditioning circuit for signal amplification and filtering. The original analog signal of the microphone has a weak amplitude and is easily affected by power frequency interference (50Hz AC) and high-frequency electromagnetic interference. Therefore, two stages of circuitry are added: the first stage is a preamplifier with a fixed gain of 20 to amplify the weak sound signal; the second stage is a second-order active bandpass filter with a passband setting of 30Hz to 3kHz.

[0081] The design logic is as follows: the main frequency of human snoring is concentrated in the range of 80 to 800 Hz, and low-frequency vibration noise <30 Hz, sharp environmental noise >3 kHz, and power frequency interference 50 Hz are filtered out.

[0082] 2) Analog-to-digital conversion is performed using an ADC (Analog-to-Digital Converter). The ADC used has a 12-bit bit depth and a voltage sampling range of 0–3.3V, which meets the dynamic range requirements for medical sound signals.

[0083] As a preferred embodiment, in this example, feature extraction of the preprocessed digital audio signal within a set sampling period includes:

[0084] 1) Basic sampling:

[0085] A. Sampling frequency: Fs = 1kHz

[0086] Basis: The fundamental frequency of snoring is 80~800Hz. According to the Nyquist sampling theorem, the sampling frequency should be greater than twice the highest signal frequency. 1kHz balances accuracy and low power consumption. A frequency higher than 1kHz will significantly increase the amount of computation and increase the power consumption of the whole device, which violates the design goal of long battery life of the device.

[0087] B. Sampling bit depth: 12 bits, single sampling point value range: 0 ~ 4095.

[0088] C. Sampling timing mode: synchronous sampling (sharing the sampling clock with the vibration sensor).

[0089] 2) Divide the signal into periods and calculate the total number of sampling points per period:

[0090] A. Definition of the Physiological Cycle of Snoring

[0091] Clinical measurements show that the duration of a single complete snoring cycle in adults is T = 0.4s ~ 1.2s, with an average cycle of 0.7s.

[0092] A standard snoring sound can be broken down into three stages: the inhalation stage → the main snoring stage (core characteristic stage) → the expiratory stage.

[0093] B. Perform windowing (combination of sliding window and fixed frame), construct a frame sampling mechanism, and define:

[0094] Single frame duration: T frame = 100ms (100ms / frame);

[0095] Number of sampling points per frame: N=F s x T frame = 100 points / frame

[0096] Sliding step size: 50ms (50% overlap between frames), function: to prevent snoring signals from being truncated by frame boundaries and to ensure feature integrity.

[0097] C. Calculation of the total number of sampling points per snoring cycle

[0098] With an average period of 0.7s: Total number of frames per period: 0.7 / 0.05 = 14 frames, Total number of sampling points per period: 0.7 x 1000 = 700 sampling points.

[0099] Summary parameters: Sampling frequency: 1000Hz (1000 points / second); Single frame: 100ms, 100 sampling points; Sliding step: 50ms (50 sampling points); Single complete snoring sound (0.7s): Total 700 sampling points.

[0100] 3) Perform sound data decoding and basic feature calculation.

[0101] The original ADC sampling data is a digital voltage quantity, which needs to be decoded and converted into three core indicators: sound pressure level (dB (A)), time domain waveform, and frequency domain characteristics.

[0102] ①Decibel (Sound Pressure Level) Decoding (Basic Threshold Judgment)

[0103] A. Conversion logic: 12-bit ADC sampled value → analog voltage → sound pressure → A-weighted decibels (medical standard dB(A)).

[0104] B. Sample value converted to voltage: ;

[0105] C. Voltage corresponds to microphone sound pressure level, which, combined with device sensitivity, is converted into A-weighted sound pressure level dB(A) (human hearing weighted, conforming to medical snoring detection standards).

[0106] ② Decibels detection threshold: Preset trigger threshold: Instantaneous decibels > 60 dB (A)

[0107] A < 60dB: Determined as ambient background noise, skipping subsequent feature analysis;

[0108] B > 60dB: Proceed to the snoring feature decoding process.

[0109] ③ Temporal feature decoding: Extract 4 core temporal features from 100 sampling points in a 100ms single frame:

[0110] A. Peak amplitude: The maximum sampled value within a single frame, reflecting the intensity of the sound;

[0111] B. Effective value (RMS): Characterizes the average energy of the signal. Snoring energy distribution is stable, but noise fluctuations are large.

[0112] C. Waveform duty cycle: Valid signal duration / total frame length. Snoring is a continuous long waveform, while sudden noise is a narrow pulse.

[0113] D. Waveform rising / falling edge slope: Snoring starts smoothly and decays slowly; impact / friction noise has steep edges.

[0114] ④ Frequency domain feature decoding: Fast Fourier Transform (FFT) is used to perform spectral analysis on single-frame data.

[0115] A. Effective analysis frequency band: limited to 30Hz ~ 3kHz (out-of-band signals have been removed by front-end filtering);

[0116] B. Extracting core frequency domain features:

[0117] Fundamental frequency (FIN): The frequency point where the signal energy is strongest;

[0118] Main frequency bandwidth: the range of main energy distribution;

[0119] Harmonic distribution: Snoring has regular harmonics, while environmental noise has chaotic harmonics.

[0120] For example, the clinical standard snoring frequency domain characteristics are as follows: standard pathological snoring main frequency: 80Hz ~ 800Hz, main energy concentration range: 100 ~ 500Hz; harmonic characteristics: regular distribution of integer multiples of the fundamental frequency, and stable spectrum morphology.

[0121] In this embodiment, the extracted acoustic signal features are filtered by decibel value, verified by time / frequency domain features, and verified by timing continuity step by step. Through step-by-step layered filtering and verification, external sound interference sources can be eliminated in turn, and valid snoring signals can be accurately identified.

[0122] For example, external sound interference sources may include:

[0123] A. Steady-state environmental noise: air conditioner, fan, outdoor traffic (continuous low volume);

[0124] B. Sudden impulse noise: slamming when turning over, unusual noises from the bed frame, objects falling (instantaneous high decibels);

[0125] C. Human voice: speaking, coughing, clearing the throat (special spectrum / waveform);

[0126] D. Friction noise: friction between the equipment and skin / fabric (mainly high frequency);

[0127] E. Breathing airflow noise: normal shallow breathing (low decibel, extremely low dominant frequency).

[0128] As a preferred implementation, in this embodiment, the extraction of acoustic signal features through step-by-step decibel filtering, time / frequency domain feature verification, and timing continuity verification specifically includes:

[0129] First level: Decibel coarse filtering (fastest interception, effectively reducing power consumption)

[0130] Rule: Instantaneous decibel ≤ 60dB (A) → directly judged as background noise, and subsequent calculations terminated.

[0131] This level can block low-volume steady-state noise such as normal breathing, fans, and air conditioners.

[0132] Second level: Time domain feature verification (for signals > 60dB, verify time domain features to distinguish between impulse noise and continuous snoring)

[0133] Rules: The duration of a single valid waveform (a single band of the acquired snoring waveform) is ≥300ms (the duration of normal snoring); the rising / falling edge is smooth (below the preset threshold after smoothing, which can be determined based on experience data and the snoring data of the target object).

[0134] This level can intercept sudden noises (narrow pulses with a waveform duration of <300ms) caused by turning over, bumping, or instantaneous impacts, as well as sharp impact sounds (with steep edge slopes).

[0135] Third level: Frequency domain feature filtering (core distinction between human voice, friction sound and snoring sound)

[0136] Rule: If the main frequency of the acoustic signal falls between 80Hz and 800Hz and the harmonics are regular, then the candidate snoring sound will be retained.

[0137] This level can intercept:

[0138] Human voice (speaking / coughing): The dominant frequency of speaking is >800Hz, the spectrum is messy and there are no fixed harmonics → filter.

[0139] Cough: Pulsating wide spectrum, deviating from the 80~800Hz range of the main snoring frequency band → filter.

[0140] Fabric / skin friction noise: concentrated in the high frequency range >1kHz, deviating from the main frequency of snoring → filtered.

[0141] Fourth level: Timing continuity check (final filtering of intermittent noise)

[0142] Rule: If three or more consecutive frames meet all the above characteristics and the interval period is 0.4~1.2s, the snoring is considered valid.

[0143] This level can intercept sounds that occasionally meet the criteria in a single frame but are not periodic → these are judged as occasional interference and filtered out.

[0144] As a preferred embodiment, in this example, a triaxial MEMS accelerometer is selected as the vibration sensor to meet the requirements of low power consumption, miniaturization, and embedded installation. Preferred performance parameters are:

[0145] Measurement range: ±2g (human mandibular vibration amplitude falls entirely within this range, with reasonable range redundancy); Power consumption: Ultra-low power mode <10μA, compatible with the stimulator's four-level sleep architecture; Output interface: Analog voltage output; Frequency response range: 0.5Hz ~ 1kHz, fully covering the mandibular vibration frequency band corresponding to human breathing and snoring; Resolution: 12bit, consistent with the acoustic ADC bit depth, ensuring data accuracy matching.

[0146] As a preferred embodiment, in this example, a front-end signal conditioning circuit is used to preprocess the mandibular vibration signal collected by the vibration sensor. Mandibular vibration is a low-frequency, weak mechanical signal, easily affected by PCB electromagnetic interference, human static electricity, and slight friction interference; therefore, a two-stage conditioning circuit is designed.

[0147] A. First-order RC low-pass filter circuit, cutoff frequency: 1kHz, used to filter out high-frequency electromagnetic interference and device self-excitation noise, while retaining the effective vibration frequency band.

[0148] B. DC bias circuit, which boosts the unipolar vibration signal to the effective input range of the ADC (0~3.3V) to ensure complete sampling of positive and negative vibration waveforms without distortion.

[0149] As a preferred embodiment, in this example, feature extraction of the preprocessed digital vibration signal within a set sampling period includes:

[0150] 1) Basic sampling parameters:

[0151] A. Main sampling frequency: 1kHz (effective operating mode);

[0152] B. Sampling bit depth: 12 bits, ADC value range 0 ~ 4095;

[0153] C. Three-axis acquisition rules: The Z-axis (mandibular main vibration axis) is used first, and the X / Y axes are used as auxiliary verification axes (body position, sway judgment).

[0154] 2) Frame structure design (sliding frame, to prevent signal truncation):

[0155] A. Single frame duration: 100ms;

[0156] B. Number of sampling points per frame: 1000Hz x 0.1s = 100 points / frame;

[0157] C. Frame sliding step: 50ms (50% frame overlap) to avoid vibration features being cut off by frame boundaries.

[0158] 3) Determine the snoring vibration cycle and the number of sampling points per cycle by combining clinical human data:

[0159] A. The mandibular vibration cycle triggered by a single complete snoring sound: 0.4s ~ 1.2s, with an average cycle of 0.7s;

[0160] B. Total number of sampling points per average period: 0.7s x 1000Hz = 700 points / period;

[0161] C. Segmentation within the cycle: Inspiratory vibration segment → Main snoring vibration segment (core feature) → Expiratory decay segment.

[0162] 4) Decode vibration data and extract features:

[0163] ①Basic numerical decoding:

[0164] A. Converting ADC values ​​to analog voltage: .

[0165] B. Voltage to acceleration value conversion: Based on the sensor's factory sensitivity parameters, the analog voltage is converted into standard acceleration units (g) to obtain the instantaneous vibration acceleration.

[0166] C. Amplitude determination criteria:

[0167] Human resting mandibular baseline vibration: <0.05g;

[0168] Effective vibration associated with snoring: 0.08g ~ 0.4g (clinically calibrated range).

[0169] ② Temporal feature extraction (calculation of 100 points per frame): Temporal features are the core of distinguishing vibration types. Five key indicators are extracted per frame:

[0170] A. Peak acceleration: The maximum instantaneous acceleration within a single frame, characterizing the intensity of vibration;

[0171] B. Root Mean Square (RMS): Reflects the average energy of vibration; the vibration energy of snoring is continuous and stable.

[0172] C. Waveform duration: The continuous existence time of the effective vibration signal;

[0173] D. Waveform rising / falling edge slope: Snoring vibrations have a gentle onset and decay; Impact / rolling vibrations have steep edges.

[0174] E. Waveform duty cycle: Effective vibration duration / total frame duration, distinguishing between continuous vibration and instantaneous pulse vibration.

[0175] ③ Frequency domain feature extraction: A lightweight FFT algorithm (fixed-point operation, low computational cost) is used to analyze the spectral distribution of single-frame vibration data.

[0176] A. Effective analysis frequency band: 0.5Hz ~ 800Hz (excluding UHF interference);

[0177] B. Core frequency domain indicators: main frequency, main power bandwidth, harmonic distribution;

[0178] C. Clinical gold standard (pathological snoring vibration frequency domain characteristics):

[0179] Main frequency range: 10Hz ~ 150Hz (specific frequency band for snoring jaw vibration)

[0180] Energy distribution: Energy is concentrated in the mid-to-low frequency band, with a smooth spectrum and regular harmonics;

[0181] Distinguishing features: Normal respiratory frequency <10Hz, rolling / friction vibration spectrum is disordered and has no fixed frequency.

[0182] For example, based on the patient's nighttime sleep state, mandibular vibrations can be divided into two main categories: target characteristic vibrations and non-characteristic interference vibrations.

[0183] A. Target vibration: Pathological snoring associated with jaw vibration (effective trigger source):

[0184] Synchronous, regular vibrations of the mandible, caused by upper airway collapse and posterior displacement of the tongue base, occur periodically.

[0185] B. Non-characteristic interference vibrations (all filtered, no stimulus triggered):

[0186] Normal sleep breathing vibrations: low amplitude, ultra-low frequency, and stable rhythm;

[0187] Static body position deviation: slow, gradual vibration, without sudden fluctuations;

[0188] Rolling over / limb tremors: transient large impulses, steep edges, and no periodicity;

[0189] Skin / fabric friction: high-frequency, fine vibrations, energy dispersion;

[0190] Swallowing, coughing, teeth grinding: short-duration pulsed vibrations with spectral distortion;

[0191] External environmental vibrations: ground vibrations, unusual noises from doors and windows, causing simultaneous interference throughout the entire area.

[0192] As a preferred embodiment, in this example, the extraction of the mandibular vibration signal is subjected to vibration amplitude filtering, time / frequency domain feature verification, and periodic continuity verification step by step, specifically including:

[0193] First level: coarse amplitude filtering (lowest power consumption, first-level interception)

[0194] Rule: 0.08g ≤ instantaneous acceleration ≤ 0.4g: Proceed to the next level of feature verification;

[0195] Instantaneous acceleration <0.08g: judged as normal breathing, static baseline vibration, direct filtering;

[0196] Instantaneous acceleration > 0.4g: judged as large-amplitude pulse vibration such as rolling over or impact, and directly filtered out;

[0197] It can intercept invalid vibrations of high and low amplitude, reducing the amount of subsequent calculations.

[0198] Second level: Time domain feature filtering (for signals with acceptable amplitude, verify time domain features to distinguish between continuous vibration and impulse vibration).

[0199] Rules: Effective vibration duration ≥ 300ms; smooth rise / fall slope of waveform; waveform duty cycle ≥ 60% (determined as continuous vibration, proceed to the next level).

[0200] It can intercept coughing, swallowing, momentary shaking (duration <300ms); and steep-edge pulse vibrations such as bumps and friction.

[0201] Third level: Frequency domain feature filtering (core differentiation, separating breathing vibrations and snoring vibrations).

[0202] Rule: Vibration frequency falls within the range of 10Hz to 150Hz, harmonic patterns → retain candidate snoring vibrations.

[0203] It can block normal sleep breathing vibrations with a main frequency of <10Hz, and filter them directly; as well as friction and external vibrations with chaotic spectrum and no fixed harmonics, and filter them directly.

[0204] Level 4: Periodic continuity verification (identifying regular snoring vibrations)

[0205] Rule: If the feature remains stable for more than 3 consecutive frames and the vibration repetition period falls within the range of 0.4~1.2s and occurs periodically (pathological snoring vibration has a fixed periodicity) → candidate valid vibration.

[0206] It can intercept and filter occasional interference that occasionally meets the standard in a single frame and does not repeat periodically.

[0207] In this embodiment, if the acoustic signal feature or mandibular vibration signal feature fails any level of filtering or verification, the determination of the current sampling cycle is stopped; only when the acoustic signal and mandibular vibration signal that have passed the final verification are obtained simultaneously is it determined that the triggering condition is met and the stimulation electrode is controlled to output a stimulation signal.

[0208] By adopting the above-mentioned hierarchical filtering and verification scheme, the amount of computation and storage space can be effectively reduced, as well as the time for triggering and issuing instructions; and energy consumption can be effectively reduced, extending the working time of the equipment.

[0209] Furthermore, this embodiment also includes a stop stimulation logic (immediate shutdown to protect sleep). When either the sound or vibration index deviates from the snoring standard (decibels drop, vibration amplitude deviates from the range, characteristic shift / disappearance), the stimulation signal output is immediately stopped, and the system returns to the low-power monitoring mode to protect sleep continuity.

[0210] Example 3

[0211] Based on the scheme shown in Embodiment 2 above, this embodiment further performs waveform template matching on the acoustic signal that has been filtered by decibel value, verified by time / frequency domain features and verified by time sequence continuity, and on the mandibular vibration signal that has been filtered by vibration amplitude, verified by time / frequency domain features and verified by periodic continuity, in order to obtain a more accurate and effective snoring determination result.

[0212] Specifically, for acoustic signals and mandibular vibration signals, after passing the first four levels of filtering and verification shown in Example 2, the following fifth level of waveform model matching is also included:

[0213] For acoustic signals, the fifth level of waveform template matching includes:

[0214] Pre-store clinically calibrated standard feature templates for pathological snoring (time-domain waveform template + frequency-domain template).

[0215] For the selected candidate acoustic feature signals, perform waveform similarity calculation:

[0216] If the similarity to the standard feature template of pathological snoring is ≥95%, it is confirmed as clinical pathological snoring.

[0217] If the similarity to the standard feature template of pathological snoring is <95%, it is judged as a non-target sound.

[0218] For the mandibular vibration signal, the fifth-level waveform template matching includes:

[0219] Multiple clinically calibrated standard templates for snoring mandibular vibration (time-domain waveform + spectrum template) are pre-stored.

[0220] Calculate the similarity between the current vibration waveform and the standard template:

[0221] If the waveform similarity is ≥ 95%, it is determined to be a pathological snoring characteristic vibration;

[0222] If the waveform similarity is less than 95%, it is determined to be other complex vibrations and filtered out.

[0223] In this embodiment, the acoustic-vibration dual-mode joint decision logic is as follows:

[0224] ① Prerequisites for dual-modal synchronization: acoustic and vibration feature data must be from the same clock source, frame synchronized, and period aligned.

[0225] ②Final triggering condition (logical AND, stimulus is output only if both conditions are met):

[0226] A. Acoustic characteristic conditions: Passes all levels of filtering and verification, and the acoustic waveform template matching similarity is ≥95%;

[0227] B. Vibration side conditions: Pass all levels of filtering verification (vibration amplitude 0.08~0.4g + main frequency 10-150Hz + period 0.4~1.2s) and vibration waveform template matching similarity ≥95%;

[0228] Only when both conditions are met will the stimulation electrode be controlled to output a stimulation signal.

[0229] Furthermore, this embodiment also includes a stop stimulation logic (immediate shutdown to protect sleep). When either the sound or vibration index deviates from the snoring standard (decibels drop, vibration amplitude deviates from the range, feature shift / disappearance, waveform template matching similarity is less than 95%), the stimulation signal output is immediately stopped, and the system returns to the low-power monitoring mode to protect sleep continuity.

[0230] Example 4

[0231] Based on the schemes shown in Embodiment 1 or Embodiment 2, this embodiment further incorporates an auxiliary shaft calibration based on a vibration sensor to obtain a more accurate and effective snoring determination result.

[0232] Specifically, in this embodiment, the vibration sensor is a triaxial accelerometer, and its sensitive axis (i.e., the main vibration axis Z-axis) is approximately perpendicular to the surface of the target object's chin skin during application. Auxiliary axis calibration specifically includes:

[0233] Using two axes of a triaxial accelerometer for auxiliary determination:

[0234] A. If characteristic vibrations only occur on the Z-axis (main vibration axis), and the X / Y-axis amplitudes are extremely low, it can be confirmed as localized snoring vibrations in the mandible;

[0235] B. If the X / Y axes fluctuate significantly in sync, it is determined to be a full-body rollover or body position change, and the jaw vibration signal is directly filtered out.

[0236] Example 5

[0237] To improve the accuracy of effective snoring signal recognition, in this embodiment, the sound sensor and vibration sensor share the same sampling clock and are synchronously initiated by the same hardware trigger signal. In a preferred embodiment, the following aspects are included:

[0238] 1) Global Synchronization Clock Design

[0239] The sound sensor and vibration sensor share the 1kHz system sampling clock of the SoC (System-on-a-Chip). The two ADC conversions are synchronously initiated by the same GPIO trigger signal on the SoC, achieving point-to-point timing alignment of the sound and vibration data and eliminating scheduling errors of tens of microseconds introduced by time-division multiplexing. Specifically, this includes:

[0240] A. Unified system clock source: Includes an acoustic acquisition circuit with a microphone, a vibration acquisition circuit with a triaxial accelerometer, an acoustic / vibration ADC analog-to-digital conversion unit, and a shared system chip (nRF52840) for timestamp counters. The internal high-precision system clock has a clock frequency of 1MHz and a clock accuracy of <1μs, which can completely eliminate the inherent timing deviations caused by different crystal oscillators.

[0241] B. Hardware Synchronous Trigger Mechanism: A trigger signal from the same GPIO of the SoC is simultaneously sent to the microphone conditioning circuit and the accelerometer conditioning circuit, causing both ADCs to start conversion at the same time, ensuring that the sampling start time is completely consistent. Compared to software triggering, which has a scheduling error of tens of μs, the hardware triggering in this embodiment can achieve zero start-up latency.

[0242] In this embodiment, the microphone and the triaxial accelerometer share a 1kHz sampling clock, the acoustic and vibration signals are strictly time-aligned, the frame length and sliding step size are completely unified, and the acoustic-vibration sampling data are time-aligned point by point to ensure the synchronization of dual-modal data.

[0243] Preferably, the installation requirements for the microphone and the triaxial accelerometer are as follows:

[0244] The microphone and triaxial accelerometer are embedded inside the medical-grade silicone substrate, with no external protrusions;

[0245] The microphone and the triaxial accelerometer are coaxially and integrated at the same point to eliminate spatial timing deviations.

[0246] The sensitive axis (Z-axis) of the triaxial accelerometer is perpendicular to the skin of the mandible, and prioritizes the acquisition of mechanical vibrations of the mandible in the vertical / front-back directions;

[0247] In this embodiment, time difference acquisition, delay analysis, and source identification are also performed on the acoustic vibration signal. Specifically, this includes the following aspects:

[0248] 1) Hardware configuration:

[0249] Counter Configuration: The System-on-a-Chip (SoC) has a built-in general-purpose hardware timer configured for free-counting mode.

[0250] Counting clock: 1MHz (minimum timing unit: 1μs)

[0251] Counting range: 0 ~ 65535 (maximum single count 65.535s, fully covering the breathing and snoring cycles)

[0252] Function: After each sampling is completed, the current count value is automatically latched as a sampling timestamp.

[0253] 2) Timestamp sampling

[0254] A. Dual-signal hardware synchronous sampling, generating timestamps frame by frame:

[0255] Acoustic signal: After each frame is sampled, a timestamp is recorded;

[0256] Vibration signal: Timestamps are recorded synchronously on the same trigger edge.

[0257] B. Frame Structure and Delay Statistics Rules: The existing 100ms / frame, 50ms sliding step structure will be used.

[0258] Each frame has 100 sampling points, and each sampling point carries a timestamp;

[0259] Calculate two statistics within a single frame: average latency and latency fluctuation variance.

[0260] Average delay: characterizes the overall time series offset

[0261] Delay variance: Characterizes the timing stability of a signal (smaller variance for signals originating from the same source, and larger variance for interfering signals).

[0262] 3) Delay calculation

[0263] A. Relative delay calculation: ΔT = T A -T V Where ΔT is the time difference (unit: μs) between the sound signal and the vibration signal, T A T is the timestamp of the sound signal. V ΔT is the timestamp of the vibration signal; when ΔT > 0, the sound lags behind the vibration; when ΔT < 0, the vibration lags behind the sound signal.

[0264] The above delay calculation can be performed in parallel with time-domain and frequency-domain features.

[0265] 4) Division of acoustic-vibration time delay intervals and clinical benchmarks

[0266] ① Explanation of the physiological timing principle of pathological snoring (effective target signal):

[0267] During sleep, the relaxation and collapse of the tongue base / genioglossus muscles in patients with A.OSA cause mechanical vibration of the jaw, which in turn leads to airway turbulence and snoring.

[0268] B. Physiological conduction sequence: muscle vibration → airflow and sound production;

[0269] C. Timing conclusion: Vibration precedes sound, i.e.: T V <T A .

[0270] ② Clinical standard time delay parameters (calibrated through multiple clinical trials). For adult OSA patients, the relative delay (original time difference) of effective homologous snoring was statistically analyzed:

[0271] A. Standard delay interval: ΔT = 200-1500μs;

[0272] B. Optimal central interval: ΔT = 500-1000μs (the distribution interval of most pathological snoring);

[0273] C. Single-frame delay fluctuation variance: <50μs (time sequence of signals from the same source is stable and fluctuation is small).

[0274] ③ Based on the type of interference in the sleep scenario, time delay characteristics are defined for source attribution determination (all heterogeneous signals are filtered out):

[0275] Type 1: Environmental noise + normal human body vibration (external source)

[0276] For example: the sound of air conditioning + normal breathing vibrations, traffic outside the window + slight body movements;

[0277] Temporal characteristics: Sounds and vibrations have no fixed order and are random in both positive and negative directions;

[0278] Time delay fluctuation variance: >200μs (time sequence disorder, extremely large fluctuation);

[0279] Judgment result: Not from the same source, directly filtered.

[0280] Type 2: Sound precedes vibration (heterogeneous source)

[0281] For example: an external impact sound → the human body is startled and shakes, and limb movements occur after speaking;

[0282] Timing characteristics: The sound vibrates ahead of the time, ΔT < 0;

[0283] Judgment result: Violates the physiological sequence of snoring, not the target signal, filtered.

[0284] Type 3: Vibration delay exceeds the normal range

[0285] The vibration precedes the sound, but ΔT >1500μs;

[0286] For example, ambient noise appears long after the rolling over action;

[0287] Judgment result: Severe temporal disconnect, no physiological correlation, filtered.

[0288] Type 4: Transient Pulse Interference

[0289] For example: collision, cough;

[0290] Timing characteristics: abrupt changes in short-time signal delay, and extremely large variance in single-frame delay;

[0291] Judgment result: The timing is unstable, so it is filtered out.

[0292] 6) Latency Dimension Judgment Rules

[0293] Snoring is considered to be pathological if it meets either of the following two criteria:

[0294] A. Original time difference: 200μs≤ΔT≤1500μs;

[0295] B. Single-frame delay fluctuation variance: ≤50 μs;

[0296] If any of the above conditions are not met, it is determined to be interference from a different source.

[0297] To further conserve computing power and reduce operating power consumption and computing requirements, this embodiment configures a hierarchical source tracing and verification method based on signal source attribution for the above-mentioned latency determination method, specifically including:

[0298] First layer: Initial delay screening (lowest power consumption, executed first).

[0299] A. ΔT <0 (sound vibration ahead of time), categorized as: external sound inducing human movement, interference;

[0300] B. ΔT > 1500 μs or ΔT < 200 μs, classified as: time series out of physiological range, interference;

[0301] C. If 200μs≤ΔT≤1500μs, the sample passes the initial screening and proceeds to the second layer of verification.

[0302] Second layer: Delay stability verification

[0303] A. Delay variance > 50μs, timing disorder, classified as asynchronous mixed interference, filtered;

[0304] B. If the time delay variance is ≤50μs and the timing is stable, it is determined to be a signal from the same source as the sound and vibration, which triggers the stimulus signal.

[0305] As a further preferred implementation, in this embodiment, the determination of the original time difference also includes hardware inherent delay calibration and calculation of the true relative delay. Specifically, this includes:

[0306] A. Hardware inherent delay calibration. Different components and circuits have fixed hardware delays, which must be calibrated and compensated for in advance:

[0307] Calibration method: Standard homogeneous snoring signal (sound + mechanical synchronous vibration) was played manually, 1000 sets of samples were collected continuously, and the inherent average delay of the hardware was statistically analyzed.

[0308] For example, hardware latency can be divided into:

[0309] Device response delay: Inherent response time of MEMS microphone / accelerometer (<50μs);

[0310] Circuit propagation delay: Filtering and op-amp routing delay (<30μs);

[0311] ADC conversion delay: Synchronous ADC conversion difference (<20μs);

[0312] The total inherent latency of the hardware is less than 100μs, which can be completely eliminated after calibration and will not affect the judgment of physiological latency.

[0313] Calibration parameter storage: The hardware reference delay is written into the non-volatile Flash memory of the system chip (SoC), which is automatically loaded upon power-on. Each device is calibrated independently to ensure consistency.

[0314] B. Calculate the true relative delay frame by frame: ΔT real =ΔT-ΔT0, where ΔT0 is the inherent average latency of the hardware.

[0315] After the above calculations, when the actual relative delay is also within the preset range (instead of the original time difference for the aforementioned determination), it is determined that the sound and vibration are from the same source, thereby further improving the accuracy of the judgment.

[0316] Example 6

[0317] To further improve the accuracy of effective snoring signal recognition, this embodiment adopts a joint judgment mode of sound, vibration and sound vibration time delay, combined with the judgment process in the aforementioned embodiments 1-5, to form a multi-dimensional joint judgment system. Electrical stimulation is triggered only when all conditions are met simultaneously.

[0318] Specifically, based on the step-by-step filtering, time / frequency domain feature verification, timing continuity verification, and waveform template matching of sound and vibration signals, a further sound and vibration delay determination is added to achieve accurate identification of valid snoring signals. Exemplary determination conditions include:

[0319] ①Acoustic dimension: Instantaneous decibel > 60dB (A) + acoustic waveform / spectrum matching for pathological snoring + sound waveform template matching similarity ≥ 95%;

[0320] ② Vibration dimensions: vibration amplitude 0.08~0.4g + vibration dominant frequency 10~150Hz + vibration waveform template matching similarity ≥95%;

[0321] ③ Timing and delay dimension:

[0322] A. True relative delay: 200μs≤ΔTreal≤1500μs;

[0323] B. Single-frame delay fluctuation variance: ≤50 μs;

[0324] ④ Periodic continuity: The signal period is 0.4~1.2s, and the characteristics are stable for 3 consecutive frames.

[0325] Further preferred embodiments may include:

[0326] ⑤ Auxiliary axis verification: The X / Y axes do not wobble significantly, ruling out the possibility of the whole body turning over.

[0327] If all the above conditions are met, a valid snoring sound is detected, and stimulation is initiated. Otherwise, if any condition is not met, stimulation is immediately stopped, and monitoring continues in low-power mode.

[0328] Example 7

[0329] To further reduce power consumption and extend the stimulator's operating time, this implementation employs a high-low power switching mode during stimulator operation. In normal sleep monitoring mode, the SoC samples at a low frequency of 200Hz, performing only coarse detection of decibel thresholds and vibration amplitudes, without enabling high-precision timestamps and delay calculations to reduce power consumption. When an instantaneous decibel level approaches the 60dB threshold, the sampling frequency automatically switches to full-speed sampling at 1kHz, simultaneously initiating the delay analysis function. Once stimulation output ends and snoring characteristics disappear, the device returns to the 200Hz low-power monitoring mode.

[0330] Specifically, the stimulator can operate in the following different power consumption modes:

[0331] A. Normal sleep monitoring mode (low power consumption)

[0332] The sound sensor sampling frequency is reduced to 200Hz (20 points / frame, 500ms frame length), and only coarse detection of decibel threshold is performed; when there is no obvious sound, the SoC maintains shallow sleep and has the lowest power consumption.

[0333] At the same time, the sampling frequency of the vibration sensor is reduced to 200Hz, only making a rough judgment on the vibration amplitude without complex calculations; when there is no obvious vibration, the sensor enters a wake-up standby state.

[0334] B. Pre-wake-up mode

[0335] When the sound sensor detects that the instantaneous decibel level is close to the 60dB threshold, the sound sensor and vibration sensor automatically switch to standard 1kHz full sampling and initiate complete feature decoding and comparison.

[0336] C. Stimulation mode: The sound sensor and vibration sensor maintain full-speed sampling at 1kHz to monitor the disappearance of snoring in real time and shut off the stimulation in a timely manner.

[0337] This device operates under the aforementioned mode: during routine monitoring, the SoC enters a sleep sampling mode, with only the sensors intermittently powered; the chip operates at full power only when a stimulus is triggered. In actual testing, a 300mAh lithium battery provided 7.5 days of continuous use after 8 hours of nighttime operation, representing a 102% improvement in battery life compared to similar discrete architecture products.

[0338] Example 8

[0339] This embodiment illustrates a specific implementation scheme for the output of stimulation signals by the stimulation electrode.

[0340] As a preferred embodiment, this example abandons the traditional single-frequency direct stimulation method and instead uses dual-channel high-frequency carrier waves introduced percutaneously. The interference waves formed by the superposition of these waves, based on the principle of time interference, generate a therapeutic low-frequency envelope waveform at the target depth under the skin. Because the high-frequency current only penetrates the epidermis and has no neuromuscular excitatory effect, it effectively stimulates only the deep target area forming the low-frequency envelope, thus achieving painless epidermal stimulation and precise activation of the genioglossus muscle. Furthermore, the pure sinusoidal waveform has low harmonic interference, improving the safety of the device.

[0341] As a preferred embodiment, the following waveform design is adopted in this example:

[0342] ① Waveform shape: A continuous symmetrical sine wave is selected as the high-frequency carrier wave. Compared with square wave and pulse wave, the harmonic distortion rate is extremely low (<0.14%), which can significantly reduce skin electrochemical stimulation and soreness, and improve long-term wearing comfort.

[0343] ② Waveform matching: The carrier frequency range (1200Hz~8500Hz) and frequency difference range (1~95Hz) are customized according to the electrophysiological characteristics of the human genioglossus muscle. The low-frequency envelope generated by the interference falls in the optimal activation frequency band of the genioglossus muscle (30~50Hz), and the waveform is highly matched with the physiological characteristics of the target tissue.

[0344] ③ Waveform control: Equipped with 3s soft start and 3s soft stop waveform gradual change logic, without waveform sudden spikes, further avoiding the discomfort caused by instantaneous current surges, and complies with medical electrical safety standard IEC60601-2-10.

[0345] See Figure 4In one example, upon receiving a stimulation command, two stimulators 31 and 32 control the stimulation electrodes to synchronously output two high-frequency sinusoidal carrier waves with different frequencies. Carrier 1 is set to 5000Hz, and carrier 2 to 5040Hz. Both are continuous symmetrical sinusoidal waves with a harmonic distortion rate of less than 0.14% and no DC component. After percutaneous introduction, the two carrier waves undergo temporal interference at a depth of 5mm to 10mm in the genioglossus muscle target area. The 40Hz frequency difference automatically synthesizes a 40Hz low-frequency envelope waveform, which falls within the optimal activation frequency range of 30Hz to 50Hz for the genioglossus muscle. Since the high-frequency sinusoidal wave only penetrates the skin surface and cannot activate epidermal nerves, the low-frequency envelope generated by the interference precisely targets the deep target area of ​​the genioglossus muscle, eliminating the skin burning problem associated with traditional percutaneous electrical stimulation from the waveform source. Preferably, the carrier wave smoothly increases from zero to the target amplitude using a 3-second soft-start gradual increase, and smoothly decreases from the target amplitude to zero using a 3-second soft-shutdown method when output stops. This eliminates waveform spikes and avoids discomfort caused by instantaneous current surges, complying with the IEC 60601-2-10 medical electrical safety standard. More generally, the carrier frequency can be selected within the range of 1200Hz to 8500Hz, and the frequency difference can be selected within the range of 30Hz to 50Hz, calibrated according to the electrophysiological characteristics of the genioglossus muscle of the target subject.

[0346] During the stimulation output process, the stimulator maintains continuous sampling at 1kHz full speed, continuously cycling through the sampling and analysis process. When the acoustic signal decibel drops below 60dB, or the vibration characteristics disappear, or the actual relative delay deviates from the standard time delay range, the snoring event is determined to have ended. The stimulator immediately cuts off the two high-frequency carrier outputs simultaneously in a 3s soft shutdown mode, and the device returns to low-power monitoring mode.

[0347] Clinical testing of the scheme shown in the above embodiments showed that, under the combined interference of environmental noise (continuous low-volume steady-state noise such as air conditioner, fan, outdoor traffic, etc.) and heterogeneous vibration (instantaneous pulse vibration such as turning over, bumping, coughing, swallowing, etc.), the system's false trigger rate was no higher than 2%, which is a significant improvement over the traditional single-microphone detection scheme (false trigger rate higher than 30%).

[0348] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A transcutaneous hypoglossal nerve stimulation system, characterized in that, The device includes at least one stimulator that can be applied to the jaw of a target subject, the stimulator being equipped with at least a sound sensor, a vibration sensor, and stimulation electrodes; and a processor, integrated or communicatively connected to the stimulator, and configured to: Simultaneously acquire acoustic signals collected by the sound sensor and mandibular vibration signals collected by the vibration sensor; The acoustic signal and the mandibular vibration signal are preprocessed and feature extracted respectively. Within a set sampling period, the acoustic signal is filtered by decibel value, time / frequency domain feature verification and time sequence continuity verification are performed step by step. At the same time, the mandibular vibration signal is filtered by vibration amplitude, time / frequency domain feature verification and period continuity verification are performed step by step. When the acoustic signal or mandibular vibration signal fails any level of filtering or verification, the determination of the current sampling period is stopped; only when both the acoustic signal that passes the temporal continuity verification and the mandibular vibration signal that passes the periodic continuity verification are obtained simultaneously, the triggering condition is determined to be met and the stimulation electrode is controlled to output a stimulation signal.

2. The system according to claim 1, characterized in that, The time / frequency domain feature verification of the acoustic signal includes at least one or all of the following: The duration of a single effective waveform of the time-domain signal is greater than a preset time threshold; The rising and / or falling edge slope of the time-domain signal is less than a preset slope threshold; The dominant frequency of the frequency domain signal is located within a preset frequency range; The frequency of a frequency domain signal conforms to harmonic laws.

3. The system according to claim 1, characterized in that, The time / frequency domain feature verification of the mandibular vibration signal includes at least one or all of the following: The duration of a single effective waveform of the time-domain signal is greater than a preset time threshold; The rising and / or falling edge slope of the time-domain signal is less than a preset slope threshold; The waveform duty cycle of the time-domain signal is greater than a preset threshold; The dominant frequency of the frequency domain signal is located within a preset frequency range; The frequency of a frequency domain signal conforms to harmonic laws.

4. The system according to claim 1, characterized in that, The temporal continuity verification of the acoustic signal and the periodic continuity verification of the mandibular vibration signal each include: The signal satisfies the time / frequency domain feature verification condition for n consecutive frames, where n≥3; The signal repeats periodically at preset intervals.

5. The system according to claim 1, characterized in that, The processor is also configured to: Waveform template matching is performed on acoustic signals that pass the temporal continuity check and / or mandibular vibration signals that pass the periodic continuity check. When the acoustic signal or mandibular vibration signal fails to match the waveform template, the determination of the current sampling cycle is stopped; only when both the acoustic signal and the mandibular vibration signal that pass the waveform template matching are acquired simultaneously, it is determined that the triggering condition is met and the stimulation electrode is controlled to output a stimulation signal.

6. The system according to claim 1 or 5, characterized in that, The processor is also configured to: During the output of the stimulus signal, if either the acoustic signal or the mandibular vibration signal fails to pass the step-by-step filtering, verification, or waveform template matching, causing the triggering condition to no longer be met, the output of the stimulus signal shall be stopped.

7. The system according to claim 1 or 5, characterized in that, The processor is also configured to: During the routine monitoring phase, low-power sampling is performed at the first sampling frequency, and only coarse detection of decibel threshold and / or vibration amplitude is performed. When both the instantaneous decibel level and the vibration amplitude reach the preset threshold, the sampling frequency is switched to a second sampling frequency higher than the first sampling frequency for full-speed sampling, and the complete trigger condition determination process is initiated.

8. The system according to claim 1, characterized in that, The stimulator includes two stimulators that can be applied to the left and right sides of the target's mandible, and the processor controls the output of stimulation signals from the stimulation electrodes, including: The stimulation electrodes on the two stimulators synchronously output two high-frequency sinusoidal carriers with different frequencies. The frequency difference between the two high-frequency sinusoidal carriers is within a preset frequency difference range, so that the two high-frequency sinusoidal carriers form a low-frequency envelope stimulation at the target tissue of the target object through time interference.

9. The system according to claim 1 or 5, characterized in that, The sound sensor and the vibration sensor share the same sampling clock and are synchronously initiated by the same hardware trigger signal. The processor is further configured to: Based on the time delay characteristics between the acoustic signal and the mandibular vibration signal, signal homology verification is performed. When the signal homology verification also passes, it is determined that the triggering condition is met and the stimulation electrode is controlled to output a stimulation signal.

10. A method for controlling a percutaneous hypoglossal nerve stimulation system, characterized in that, include: Simultaneously acquire acoustic signals collected by the sound sensor and mandibular vibration signals collected by the vibration sensor; The acoustic signal and mandibular vibration signal are preprocessed and feature extracted respectively. Feature filtering and verification are performed step by step based on a preset algorithm. The acoustic signal and mandibular vibration signal that pass the feature filtering and verification are respectively determined as valid snoring signal and valid snoring-related vibration. The acoustic signal feature filtering verification includes, in sequence, decibel value filtering, time / frequency domain feature filtering, and time sequence continuity verification; the mandibular vibration signal feature filtering verification includes, in sequence, vibration amplitude filtering, time / frequency domain feature verification, and period continuity verification. A stimulation control signal is generated to control the output stimulation signal of the stimulation electrode only when the effective snoring signal and the effective snoring-related vibration are acquired simultaneously.

11. The method according to claim 10, characterized in that, Also includes: Waveform template matching is performed on the acoustic signal and mandibular vibration signal that pass the feature filtering verification, and the acoustic signal and mandibular vibration signal that pass the waveform template matching are determined as valid snoring signal and valid snoring associated vibration; The waveform template matching of the acoustic signal and the mandibular vibration signal respectively includes waveform similarity matching with a preset sound / vibration feature template.

12. A percutaneous hypoglossal nerve stimulation device, characterized in that, It includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the control method as described in claim 10 or 11.

13. A computer-readable storage medium, characterized in that, The system contains a computer program that, when executed by a processor, implements the steps of the control method as described in claim 10 or 11.

Images