Snore stopping method and device based on closed loop control of interfering electrical stimulation

Abstract

This invention discloses an anti-snoring method and device based on closed-loop control of interference electrical stimulation. The method determines the occurrence of snoring events based on sound signals and skin vibration signals collected by sensors. When the determination indicates the occurrence of a snoring event, first and second high-frequency stimulation waves with similar frequencies are generated. These first and second high-frequency stimulation waves are applied as first and second high-frequency stimulation pulses to the genioglossus muscle of the chin via first and second sets of stimulation electrodes, respectively. The interference of the first and second high-frequency stimulation pulses activates motor neurons in the target area. Based on the collected sound signals and skin vibration signals after intervention, as well as historical intervention effect data, the stimulation parameters are continuously updated using a Bayesian optimization algorithm to output personalized third and fourth high-frequency stimulation waves for anti-snoring. This improves the safety and comfort of overnight use, achieving efficient, comfortable, and personalized anti-snoring intervention.

Description

Technical Field

[0001] This invention relates to the fields of sleep medicine and neuromodulation technology, specifically to an anti-snoring method and device based on closed-loop control of interference electrical stimulation, belonging to the interdisciplinary technical field of wearable medical devices and non-invasive neuromuscular electrical stimulation. Background Technology

[0002] Snoring is a very common sleep-related disorder. Its core pathological mechanism is that during sleep, the muscle tone of the muscles surrounding the upper airway, especially the genioglossus muscle, decreases, causing airway collapse and narrowing. As airflow passes through, it causes vibrations in soft tissues such as the soft palate and pharyngeal wall, thus producing snoring. In severe cases, it can progress to obstructive sleep apnea-hypopnea syndrome (OSAHS), leading to intermittent nocturnal hypoxia, disrupted sleep structure, and an increased risk of cardiovascular and cerebrovascular complications. With the continuous maturation of wearable medical devices and neuromodulation technology, actively modulating upper airway muscle function and maintaining airway patency during sleep using non-invasive electrical stimulation has become an important research and commercialization direction in this field.

[0003] Current electrical stimulation intervention techniques for snoring and sleep apnea mainly fall into two categories: one is implantable hypoglossal nerve stimulation systems, such as the Inspire system, which require surgical implantation of electrodes into the chest cavity and matching them with the hypoglossal nerve. Triggered by respiratory signals, these electrodes precisely drive the contraction of the genioglossus muscle. The other is percutaneous neuromuscular electrical stimulation (TENS / NMES) patch technology, which places low-frequency electrical stimulation electrodes on the skin surface of the neck or jaw, directly outputting low-frequency pulsed currents to activate superficial muscle tissue. Both techniques have demonstrated certain interventional effects in clinical studies.

[0004] However, existing transcutaneous electrical stimulation (TES) protocols still suffer from two key unresolved drawbacks: First, low-frequency currents can cause significant stinging and burning sensations when passing through the skin and subcutaneous tissue. This is because pain and tactile receptors in the superficial layers of the skin are highly sensitive to low-frequency electrical signals, easily causing awakenings during sleep and significantly reducing long-term user compliance. Second, the effective depth of low-frequency currents is insufficient. Tissue impedance increases with depth, and the current input from the skin surface significantly attenuates by the time it reaches the core belly of the genioglossus muscle, making it difficult to achieve sufficient neuromuscular activation at the target depth. Simply increasing the output current intensity exacerbates the discomfort caused by skin surface stimulation, creating a technical contradiction that makes it difficult to balance effectiveness and comfort. Summary of the Invention

[0005] This invention aims to overcome the technical shortcomings of existing transcutaneous electrical stimulation (TES) anti-snoring programs, such as strong skin surface stimulation, low deep target muscle activation efficiency, and lack of individualized closed-loop control. It provides a closed-loop control anti-snoring method and device based on the principle of interference electrical stimulation, which achieves deep and precise driving of the genioglossus muscle without obvious skin stimulation. At the same time, it combines sleep stage perception and parameter adaptive optimization to achieve individualized and efficient anti-snoring intervention while protecting the integrity of sleep structure.

[0006] The first aspect of this invention provides an anti-snoring method based on closed-loop control of interference electrical stimulation, comprising the following steps:

[0007] Based on the acquisition of sound signals and skin vibration signals by sensors, the sound signals and skin vibration signals are preprocessed, and the occurrence of snoring events is determined based on the preprocessed signals; If the determination result indicates that the snoring event has occurred, a first high-frequency stimulation wave and a second high-frequency stimulation wave with similar frequencies are generated respectively; wherein the frequency range of the first high-frequency stimulation wave and the second high-frequency stimulation wave is 200Hz to 20kHz, and the interference difference frequency is between 7 and 15Hz. The first high-frequency stimulation wave and the second high-frequency stimulation wave are applied to the genioglossus muscle of the chin via the first group of stimulation electrodes and the second group of stimulation electrodes, respectively. The target area motor neurons are activated by the interference of the first high-frequency stimulation pulse and the second high-frequency stimulation pulse, causing the genioglossus muscle to contract rhythmically to expand the airway cross-sectional area. Based on the collected sound signals and skin vibration signals after intervention, as well as historical intervention effect data, the stimulation parameters are continuously updated through a Bayesian optimization algorithm to output personalized third and fourth high-frequency stimulation waves for snoring relief.

[0008] A second aspect of the present invention provides an anti-snoring method based on closed-loop control of interference electrical stimulation, comprising the following steps: Based on the acquisition of sound signals and skin vibration signals by sensors, the sound signals and skin vibration signals are preprocessed, and the occurrence of snoring events is determined based on the preprocessed signals; If the determination result indicates that the snoring event has occurred, pulse wave signal and body movement signal are collected. The adjacent heartbeat RR interval sequence is extracted from the pulse wave signal, and the time domain index and frequency domain index of heart rate variability are calculated. The mean and peak value of body movement energy per unit time are extracted from the body movement signal to obtain a multidimensional sleep state characterization sequence containing the time domain index and frequency domain index of heart rate variability, and the mean and peak value of body movement energy per unit time. The multidimensional sleep state representation sequence is used to determine the current sleep stage through a sleep stage estimation algorithm, and a corresponding graded high-frequency stimulation wave parameter configuration is generated based on the determined current sleep stage; the interference difference frequency of the graded high-frequency stimulation wave parameter configuration is between 7Hz and 15Hz. The first high-frequency stimulation wave and the second high-frequency stimulation wave are generated according to the configuration of the graded high-frequency stimulation wave parameters. The first high-frequency stimulation wave and the second high-frequency stimulation wave are applied to the genioglossus muscle of the chin through the first group of stimulation electrodes and the second group of stimulation electrodes, respectively. The target area motor neurons are activated by the interference of the first high-frequency stimulation pulse and the second high-frequency stimulation pulse, causing the genioglossus muscle to contract rhythmically to expand the airway cross-sectional area. Based on the collected sound signals and skin vibration signals after intervention, as well as historical intervention effect data, the stimulation parameters are continuously updated through a Bayesian optimization algorithm to output personalized third and fourth high-frequency stimulation waves for snoring relief.

[0009] A third aspect of the present invention provides an anti-snoring device based on closed-loop control of interference electrical stimulation, the device comprising: The acquisition and extraction module is used to acquire sound signals and skin vibration signals, preprocess the sound signals and skin vibration signals, and determine the occurrence of snoring events based on the preprocessed signals; The stimulation circuit is used to generate a first high-frequency stimulation wave and a second high-frequency stimulation wave with similar frequencies if a snoring event is detected; wherein the interference difference frequency between the first high-frequency stimulation wave and the second high-frequency stimulation wave is between 7 Hz and 15 Hz. The electrode module is used to apply a first high-frequency stimulation wave and a second high-frequency stimulation wave to the genioglossus muscle of the chin via a first set of stimulation electrodes and a second set of stimulation electrodes, respectively; under the interference of the first high-frequency stimulation pulse and the second high-frequency stimulation pulse, the target area motor neurons are activated, causing the genioglossus muscle to contract rhythmically to expand the airway cross-sectional area. The correction module is used to continuously update the stimulation parameters based on the collected sound signals and skin vibration signals after intervention and historical intervention effect data, and output personalized anti-snoring third and fourth high-frequency stimulation waves through a Bayesian optimization algorithm.

[0010] A fourth aspect of the present invention provides an anti-snoring device based on closed-loop control of interference electrical stimulation, the device comprising: The acquisition and extraction module is used to acquire sound signals and skin vibration signals, preprocess the sound signals and skin vibration signals, and determine the occurrence of snoring events based on the preprocessed signals. If a snoring event is detected, pulse wave signals and body movement signals are acquired. The adjacent heartbeat RR interval sequence is extracted from the pulse wave signal, and the time-domain and frequency-domain indices of heart rate variability are calculated. The mean and peak values ​​of body movement energy per unit time are extracted from the body movement signal to obtain a multidimensional sleep state representation sequence containing the time-domain and frequency-domain indices of heart rate variability and the mean and peak values ​​of body movement energy per unit time. The current sleep stage is determined based on the multidimensional sleep state representation sequence using a sleep staging estimation algorithm. The stimulation circuit is used to generate corresponding graded high-frequency stimulation wave parameter configurations based on the sleep stage, wherein the interference difference frequency of the graded high-frequency stimulation wave parameter configurations is between 7Hz and 15Hz; and to generate a first high-frequency stimulation wave and a second high-frequency stimulation wave according to the graded high-frequency stimulation wave parameter configurations. The electrode module is used to apply a first high-frequency stimulation wave and a second high-frequency stimulation wave to the genioglossus muscle of the chin via a first set of stimulation electrodes and a second set of stimulation electrodes, respectively; under the interference of the first high-frequency stimulation pulse and the second high-frequency stimulation pulse, the target area motor neurons are activated, causing the genioglossus muscle to contract rhythmically to expand the airway cross-sectional area. The correction module is used to continuously update the stimulation parameters based on the collected sound signals and skin vibration signals after intervention and historical intervention effect data, and output personalized anti-snoring third and fourth high-frequency stimulation waves through a Bayesian optimization algorithm.

[0011] Based on the above embodiments of the present invention, it can be seen that the present invention employs an interference electric stimulation method, inputting two high-frequency currents (such as 2000Hz and 2010Hz) of similar frequencies into human tissue through surface electrodes, forming a 10Hz equivalent low-frequency stimulation electric field deep within the muscles. The skin surface only experiences the high-frequency current, resulting in low sensitivity of nerve receptors and significantly reducing discomfort such as stinging and burning. Under the same stimulation intensity, skin discomfort is reduced to less than 20% of that of traditional low-frequency electric stimulation. Simultaneously, the interference electric field can be precisely focused on the target muscle location at a depth of 15mm–30mm under the skin, solving the problems of insufficient penetration depth and low activation efficiency of traditional low-frequency stimulation.

[0012] This invention utilizes a Bayesian optimization algorithm to adaptively and iteratively optimize key parameters such as output current, stimulation duration, frequency difference, and channel ratio. It achieves individualized parameter convergence within five interventions, precisely adapting to differences in jaw anatomy and muscle activation thresholds among different users, thus continuously improving the anti-snoring effect. A multi-channel fatigue protection mechanism limits the cumulative stimulation time of a single channel to no more than 20 minutes within 60 minutes. Combined with adaptive parameter adjustment, this further enhances the safety and comfort of overnight use.

[0013] The above description is merely an overview of the technical solution disclosed herein. In order to better understand the technical means of this disclosure and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this disclosure more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0014] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 A flowchart of an anti-snoring method based on closed-loop control of interference electrical stimulation provided in an embodiment of this disclosure.

[0016] Figure 2 A flowchart illustrating a method for determining whether to initiate electrical stimulation based on snoring, as provided in an embodiment of this disclosure.

[0017] Figure 3 A flowchart illustrating the adaptive anti-snoring control parameter determination method provided in this embodiment.

[0018] Figure 4 This is a structural diagram of an anti-snoring device based on closed-loop control of interference electrical stimulation, provided in an embodiment of this disclosure.

[0019] Figure 5 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present disclosure. Detailed Implementation

[0020] The embodiments of this disclosure will now be described in detail with reference to the accompanying drawings.

[0021] It should be understood that the following specific examples illustrate the implementation of this disclosure, and those skilled in the art can easily understand other advantages and effects of this disclosure from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. This disclosure can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this disclosure. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0022] It should be noted that various aspects of embodiments within the scope of the appended claims are described below. It will be apparent that the aspects described herein can be embodied in a wide variety of forms, and any particular structure and / or function described herein is merely illustrative. Based on this disclosure, those skilled in the art will understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects set forth herein can be used to implement the device and / or practice the method. Additionally, this device and / or method can be implemented using other structures and / or functionalities besides one or more of the aspects set forth herein.

[0023] It should also be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of this disclosure. The drawings only show the components related to this disclosure and are not drawn according to the number, shape and size of the components in actual implementation. In actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0024] Furthermore, specific details are provided in the following description to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that the described aspects can be practiced without these specific details.

[0025] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the present invention. Various corresponding modifications and alterations can be made by those skilled in the art without departing from the spirit and essence of the present invention. See the appendix for specific implementation processes. Figure 1 A method for stopping snoring based on closed-loop control of interference electrical stimulation.

[0026] Step S1: Simultaneously acquire sound and throat vibration signals using a microphone and piezoelectric vibration sensor. After preprocessing and feature extraction, determine whether snoring occurs and decide whether to initiate electrical stimulation. See the appendix for detailed implementation steps. Figure 2 As shown: Step S101: Using the original acoustic and vibration information of the user's current sleep environment as input, after the device is applied, the sound signal of the onboard microphone is continuously collected. At the same time, the piezoelectric vibration sensor applied to the skin obtains the snoring spectrum characteristics and throat tissue vibration characteristics with the main energy concentrated in the 100Hz to 500Hz frequency band. The original acoustic and vibration information constitutes a dual-channel original time domain signal.

[0027] Step S102: Using the original dual-channel time-domain signal as input, feature extraction is performed. First, a 4th-order Butterworth 100Hz high-pass filter is applied to the audio signal to remove low-frequency environmental noise and DC offset. Then, a Hamming window is used to perform frame segmentation processing on the signal. In this embodiment, a frame length of 25ms corresponds to 200 sampling points, a frame shift of 10ms corresponds to 80 sampling points, and the inter-frame overlap rate is 60%. The short-time energy (STE) and zero-crossing rate (ZCR) are calculated frame by frame.

[0028] Extracting the instantaneous envelope by performing Hilbert transform on the vibration signal ,in The result is the Hilbert transform of the original signal. The power spectral density (PSD) is estimated using the Welch method with a window size of 256 points and 50% overlap. To ensure time alignment with the audio signal, the vibration analysis window is synchronized with the audio frame, and an eigenvalue is output every 10ms.

[0029] In the frequency domain, a 512-point Fast Fourier Transform (FFT) is performed on each frame of the audio signal, focusing on extracting the energy proportion of the 100Hz to 500Hz frequency band as the core feature of snoring. The extraction formula is as follows: The average vibration power in the frequency band from 50Hz to 300Hz is extracted synchronously from the vibration signal.

[0030] Finally, the short-time energy (STE), zero-crossing rate (ZCR), frequency band energy ratio, and average vibration power are concatenated into a 4-dimensional feature vector, which forms a sliding feature vector sequence with a step size of 10ms and is output to the subsequent decision module.

[0031] Step S103: Using the sliding feature vector sequence as input, a two-level judgment logic is used to confirm snoring events: The first level is dynamic hard threshold judgment, which updates the background noise baseline every 30 seconds and takes the lowest 5th percentile of the STE. The sound signal STE must be at least 3dB higher than the current baseline, and the energy proportion of 100–500Hz must exceed 40%. The second level is time continuity verification: the above conditions must be continuously met within a time window of at least 500ms to exclude false triggers from short-lived noise events such as turning over or coughing. When both conditions are met simultaneously, the snoring trigger judgment is output as "positive," and the trigger timestamp is recorded to obtain a reliable snoring event trigger signal.

[0032] Step S2: Based on the snoring trigger determination result, two high-frequency sinusoidal currents with similar frequencies are generated by the dual-channel independent high-frequency electrical signal generation unit to obtain two sets of structured high-frequency stimulation waveforms for subsequent electrode output.

[0033] Using a snoring-triggered positive signal as the enable input, the dual-channel digital direct frequency synthesis (DDS) circuit is activated, and channel A outputs a frequency. Channel B output frequency The basic sinusoidal waveform is used, with the initial phase values ​​of both channels set to 0°. The amplitude can be independently programmed and controlled. The DDS clock accuracy is better than ±1ppm to ensure interference difference frequency. To achieve long-term stability, two precise and phase-controllable digital sine wave sequences are obtained.

[0034] Using two digital sine wave sequences as input, a digital signal processor (DSP) applies parametric modulation to the waveforms: Regarding current amplitude, the output peak current is set within the range of 0.1mA to 20mA based on the current adaptive parameter configuration, with a default initial value of 2mA. The amplitudes of the two digital sine waves can be the same or different. Regarding pulse width constraints, gating is applied to the sine waves to control the effective pulse width of each half-cycle within the range of 10μs to 500μs, preventing heat accumulation caused by long pulse widths at high frequencies. The above processing is performed on both channels to obtain a dual-channel modulated high-frequency waveform that conforms to safe output specifications.

[0035] Using a dual-channel modulated high-frequency waveform as input, the voltage waveform is converted into a constant current output through a voltage-controlled constant current source (VCCS) circuit configured independently for each channel. The output impedance of each channel is designed to be higher than 100kΩ, so that the output current can remain accurate even if the skin contact impedance varies between 500Ω and 5kΩ. Simultaneously, overcurrent protection and electrode detachment detection circuits are configured. When the loop impedance is detected to exceed 20kΩ, the output is automatically paused and an alarm is triggered, obtaining a safe and stable dual-channel high-frequency constant current stimulation signal, ready to be delivered to the electrode array.

[0036] Step S3: Construction of deep interference field of genioglossus muscle.

[0037] Based on two high-frequency stimulation waveforms, by applying a 4-channel surface electrode to the corresponding body surface area of ​​the chin-genioglossus muscle and applying current, the two sets of spherical electric fields are spatially superimposed and interfered in the deep muscle, resulting in a deep interference low-frequency driving field with a frequency of 10Hz.

[0038] Based on the anatomical location of the genioglossus muscle (originating at the mental spine of the mandible, with the muscle belly extending posterosuperiorly to the hyoid bone and tongue body), four surface gel electrodes were arranged in a 2×2 matrix and applied to the midline region of the chin below the mandible. The two electrodes (A+, A-) in channel A were symmetrically distributed along the coronal plane, with a center-to-center distance of 25 mm to 35 mm. The two electrodes (B+, B-) in channel B were distributed along the sagittal plane, also with a center-to-center distance of 25 mm to 35 mm. The four electrodes formed an approximately square arrangement, covering a skin area of ​​approximately 40 mm × 40 mm. This electrode geometry allows the equipotential lines of the two sets of electric fields to effectively intersect at a depth of approximately 15 mm to 30 mm subcutaneously within the main muscle belly of the genioglossus muscle.

[0039] To demonstrate that the above parameters can achieve the purpose of this invention and are safe and beneficial to individuals, quantitative verification is performed using finite element simulation, as detailed below: Using the above electrode geometry and dual-channel high-frequency constant current output as input, when channel A applies... Current, channel B applied When an electric current is applied, the two sets of currents establish their own quasi-spherical electric field distributions in the tissue; in the region where the two sets of electric fields overlap (i.e., deep within the genioglossus muscle), according to the principle of interference, the instantaneous amplitude of the synthesized electric field is... satisfy:

[0040] Using product-to-sum expansion: .

[0041] Because neural tissue exhibits low-pass filtering characteristics to high-frequency carrier waves, the actual response is the peak frequency of the envelope: the envelope function. A peak is generated every half cycle (i.e., 100ms), which is equivalent to a 10Hz periodic driving signal. Therefore, an interference low-frequency driving field with an equivalent effect frequency of 10Hz is obtained in the deep part of the genioglossus muscle, while the skin surface is subjected to a high-frequency current of 2000Hz / 2010Hz. The receptor response threshold is high, and there is no obvious stimulation.

[0042] Using the aforementioned interference field construction parameters as input, a multi-layer tissue model was established with reference to the CT cross-section of the human mandible. The electric field distribution was numerically calculated using a finite element simulation model, as follows: The geometric model was constructed by referencing CT cross-sectional data of the human mandible, establishing a three-dimensional finite element geometric model containing five tissue layers: skin layer (approximately 1.5 mm thick), subcutaneous fat layer (approximately 5 mm thick), mandibular cortical bone layer (approximately 2 mm thick), cancellous bone layer (approximately 8 mm thick), and genioglossus muscle layer (approximately 15 mm to 25 mm thick). The overall size of the model was set to 80 mm × 80 mm × 60 mm (width × length × depth) to fully encompass the coverage area of ​​the four electrodes and the distribution range of their deep interference fields. The boundaries of each tissue layer were geometrically fitted using anatomical thickness data, and the morphology of the muscle layer was established as an arc-shaped muscle belly geometry based on the anatomy of the origin and insertion points of the genioglossus muscle.

[0043] Electrical conductivity of each tissue layer ( (unit: S / m) and relative permittivity ( Based on the human tissue electrical parameter database of Gabriel et al. (1996), the following values ​​are assigned at a frequency of 2000 Hz:

[0044] Under the quasi-static electric field assumption, the electric field distribution within each tissue layer satisfies the Laplace equation: ,in For a potential scalar field, Let be the electrical conductivity. The electric field strength is obtained from the potential gradient. Boundary conditions: A first-type Dirichlet boundary condition is applied to the electrode attachment area, and a boundary condition is applied to the A+ electrode of channel A. A-electrode applied Channel B is similar; a second-type Neumann boundary condition (normal current density is zero) is applied to the outer surface of the model (non-electrode region).

[0045] An adaptive unstructured tetrahedral mesh is used, with local refinement in the electrode region and deep muscle region. The total number of mesh elements is approximately 1.2 million to 1.8 million, and the number of nodes is approximately 250,000 to 350,000. The mesh quality index (minimum orthogonal quality) is not less than 0.3 to ensure the convergence of the numerical solution.

[0046] The peak electric field intensity of the interference envelope in the central region of the genioglossus muscle belly (approximately 20 mm from the skin surface) was calculated using simulation, and it was required to be no less than 6 V / m (the lower reference limit for the action potential triggering threshold of motor neurons, corresponding to Aα class motor nerve fibers). If the simulation result was lower than the threshold, the output current amplitude was increased proportionally until the requirement was met. Simultaneously, the skin surface current density was verified. (IEC 60601-2-10 standard safety upper limit). Obtain theoretically verified spatial distribution parameters of the interference field as a physical constraint benchmark for subsequent stimulation parameter optimization.

[0047] Step S4: Based on the deep 10Hz interference low-frequency driving field, the motor nerves and muscle fibers of the genioglossus muscle are periodically activated by this frequency, causing the genioglossus muscle to contract rhythmically, thereby achieving the physiological effects of forward expansion of the upper airway and reduction of airway resistance, and thus inhibiting snoring.

[0048] When a 10Hz low-frequency interference driving field is formed deep within the genioglossus muscle, if the equivalent electric field strength of this driving field exceeds the action potential trigger threshold of the genioglossus motor neurons (typically ranging from 6V / m to 10V, varying depending on the diameter of the nerve fiber), the terminal branches of the hypoglossal nerve (the 12th cranial nerve) innervating the genioglossus muscle will depolarize. Following the Henneman size principle, the nerve will recruit motor units sequentially from smallest to largest, first activating the small-diameter units corresponding to type I slow-twitch muscle fibers, and then activating the large-diameter units corresponding to type II fast-twitch muscle fibers. At a 10Hz stimulation frequency, each motor unit will form an incomplete tetanic contraction, with a single contraction and relaxation cycle of approximately 100ms, maintaining a regular and stable rhythmic contraction state of the genioglossus muscle.

[0049] Under the rhythmic contraction of the genioglossus muscle, the muscle exerts a forward and downward traction on the tongue, causing the tongue root to shift forward by 3mm to 8mm (the magnitude of the shift varies depending on the intensity of the stimulus and individual anatomical differences). The cross-sectional area of ​​the glossopharyngeal space can correspondingly increase by 20% to 40%. Simultaneously, the contraction of the genioglossus muscle causes a slight forward shift of the hyoid bone, increasing the overall tension of the suprahyoid muscle group and synergistically expanding the oropharyngeal lateral walls, allowing for a quantifiable expansion of the cross-sectional area at the narrowest point of the airway (mostly located at the palatopharyngeal and glossopharyngeal planes). According to Poiseuille's law, airway resistance decreases significantly with increasing airway diameter, and the resistance is inversely proportional to the fourth power of the airway diameter.

[0050] As the airway cross-sectional area expands, the airflow velocity and local aerodynamic negative pressure decrease accordingly. This weakens the vibrational excitation experienced by soft tissues such as the soft palate, uvula, and pharyngeal walls, resulting in a significant decrease in vibration amplitude and frequency. When the Reynolds number of the airflow at the narrowest cross-section of the airway falls below the critical value of 2300, the main source of snoring, turbulent vibration, can be effectively suppressed, ultimately achieving a significant reduction in snoring sound pressure level (a target reduction of no less than 10 dB), thus fulfilling the expected physiological goal of a single intervention.

[0051] Step S5: Based on continuous sensor monitoring data after intervention and historical intervention effect records, an adaptive anti-snoring control output that balances intervention effectiveness and user comfort is obtained through a closed-loop scheduling logic of fixed-duration observation and judgment, and a parameter optimization strategy. See the appendix for detailed steps. Figure 3 As shown: Step S501: Single intervention timing control.

[0052] When the system determines that snoring trigger is positive, an intervention begins according to the preset timing template. First, electrical stimulation is output, and the duration is recorded as [duration missing]. The default setting is 30 seconds, but it can be adjusted from 10 to 120 seconds depending on the actual situation. Output is forcibly paused after the stimulus ends, and the user enters the observation window; the duration is recorded as [duration not specified]. The default observation period is 20 seconds. During this period, the microphone and vibration sensor continue to monitor sound and vibration signals. If snoring is not detected again before the end of the observation window, the system remains silent until the next independent trigger signal appears. If snoring is again detected as positive during the observation period, a new round of intervention begins immediately.

[0053] Step S502: The balance mechanism between muscle fatigue protection and intervention effects.

[0054] The system will record in real time the cumulative number of interventions (N) and the total stimulation duration within the current sleep cycle. Furthermore, a fatigue protection mechanism is employed to limit continuous intervention. Specifically, within any 60-minute sliding time window, if the cumulative stimulation duration exceeds [a certain limit], [the intervention will be restricted]. After a few minutes, the system will automatically enter forced rest mode, with a rest duration of [duration missing]. The default duration is 10 minutes. Electrical stimulation is stopped during the rest period, but the sensor continues to monitor.

[0055] To prevent forced rest from causing insufficient intervention for heavy users throughout the night, the system also performs adaptive intensity adjustments. For example, if the intervention is interrupted more than three times due to fatigue protection within two consecutive sleep cycles, the system will automatically shorten the duration of each stimulation session. For example, the time might be reduced from 30 seconds to 20 seconds, while the output current amplitude is appropriately increased, but not exceeding the safety limit of 20mA, to maintain the effectiveness of a single intervention. Additionally, the observation window will be shortened. For example, reducing the duration from 20 seconds to 10 seconds. In this way, the frequency of intervention can be increased while keeping the total stimulus duration constraint unchanged.

[0056] In addition, the system dynamically monitors the output current amplitude and muscle impedance changes during each intervention. If the output has reached the preset maximum intensity, but snoring has not been effectively improved after three consecutive interventions, the system will automatically initiate a parameter optimization process. Through these multiple mechanisms, an overall balance is achieved between intervention effectiveness, muscle safety, and the sustainability of use.

[0057] Step S503: Adaptive optimization of stimulation parameters.

[0058] The system dynamically adjusts stimulation parameters based on historical intervention effects. The quantifiable indicator of the effect is the result displayed in the observation window after each intervention. The relative decrease in snoring sound pressure level measured internally To match the physiological characteristics of different users, the system uses a Bayesian optimization algorithm to automatically find the optimal parameters in a multi-dimensional parameter space. The parameters to be optimized form a vector. This includes: the amplitude of the two output current channels. and The range is 0.1 mA to 20 mA; the duration of a single intervention is... The range is 10 to 120 seconds; dual-channel carrier frequency difference. The range is from 7 Hz to 15 Hz.

[0059] The system updates its parameters every 5 interventions. The optimization objective is to maximize the expected value of the anti-snoring strategy while strictly adhering to safety constraints: the single-channel output current should not exceed 20mA; the cumulative stimulation time within any 60-minute sliding window should not exceed 20 minutes. This allows for continuous optimization of the anti-snoring strategy based on individual differences. The Bayesian optimization implementation method is explained in detail below.

[0060] The system employs Gaussian process regression as a surrogate model for Bayesian optimization to fit the mapping relationship between stimulus parameters and intervention effects. The model uses the Matérn 5 / 2 kernel function, and automatically updates the signal variance after each iteration by maximizing the log-marginal likelihood. Two types of hyperparameters. The kernel function takes the following form:

[0061] in, and This represents two arbitrary sample points in the parameter space (e.g., two different combinations of stimulus parameters). It is the Euclidean distance between these two points; the closer the distance, the more similar the two points are. This is called the signal variance, which controls the overall variation of the function's output value. The larger the value, the better the objective function (e.g., the reduction in snoring). The wider the fluctuation range of ), the better; It is a length scale hyperparameter that determines how quickly the function value decays as the distance to the input point increases. The larger the value, the less sensitive the function is to changes in input, and the smoother the overall trend. The smaller the kernel size, the more drastic the function changes and the richer the details; the kernel function output... This represents the covariance between two points, when the two points completely coincide. Take the maximum value As the distance increases, Gradually approaching 0. These three hyperparameters ( Typically, it learns automatically from the data by maximizing the log marginal likelihood.

[0062] The optimization process uses the desired improvement (EI) as the acquisition function to determine the next set of test parameters, which is defined as follows: ,in This indicates the parameter points to be evaluated (such as current amplitude, stimulation duration, etc.). The Gaussian process surrogate model represents the objective function value at that parameter point (e.g., the decrease in snoring sound pressure level). The prediction of ) (which is a random variable that follows a normal distribution). It is the currently observed historical best objective function value. This indicates taking the expected value of the random variable within the parentheses, while This means that a positive improvement is only counted when the predicted value exceeds the current best value; otherwise, it is zero. EI measures: at parameter points... This value represents the average improvement in the target value predicted by the model compared to the current best results. A larger value indicates greater potential for finding better parameters at that point. By maximizing EI, the algorithm automatically balances "utilizing" the currently predicted high-mean regions with "exploring" regions with high prediction uncertainty, thus efficiently searching for optimal stimulus parameters.

[0063] In practice, after every 5 interventions, the system adds the newly measured ΔSPL to the dataset and updates the Gaussian process model. Then, it performs multi-starting point optimization using the L-BFGS-B algorithm and outputs the parameters for the next round of intervention.

[0064] In the Bayesian optimization framework, a mandatory safety constraint is added. For any parameter point that violates any of the following constraints, it is directly given an extremely low EI value, making it almost impossible to be selected. The constraint conditions can be: (1) Upper limit of current amplitude: This is the safe threshold for body surface current specified in IEC 60601-1 standard and cannot be exceeded. (2) Cumulative stimulation duration: Within any 60-minute window, the total stimulation time cannot exceed 20 minutes to prevent excessive fatigue of the genioglossus muscle. (3) Upper limit of arousal rate: The arousal rate in each sleep stage must be ≤ 5%. The arousal rate is estimated by real-time body movement and heart rate mutation detection. If a certain parameter combination has historically caused the arousal rate to exceed the limit, the system will mark this region as an infeasible region in the surrogate model.

[0065] For new users, there is initially no historical data. In this case, the system uses Latin hypercube sampling to uniformly generate 5 initial parameter points in the parameter space as a cold start dataset. These 5 initial interventions are then completed and the corresponding data are collected. After the value is determined, the system officially begins Bayesian optimization iteration. In this way, the anti-snoring intervention strategy can be continuously optimized based on the individual differences of each user.

[0066] In the second embodiment of the present invention, after snoring detection is triggered, the system does not directly stimulate with a fixed intensity. Instead, it first determines which sleep stage the user is currently in and then sets appropriate stimulation parameters accordingly. This approach ensures the anti-snoring effect while minimizing disruption to the sleep structure.

[0067] The system monitors the user's sleep status in real time through two sensors: one is a wrist-worn photoplethysmography (PPG) sensor with a sampling rate of no less than 100Hz, used to collect pulse signals; the other is a triaxial accelerometer attached to the chest with a sampling rate of no less than 50Hz, used to collect body movement signals.

[0068] Based on PPG signals, the inter-heart rate sequence can be extracted, and further time-domain indices (such as SDNN, RMSSD) and frequency-domain indices (low-frequency / high-frequency power ratio LF / HF) of heart rate variability (HRV) can be calculated. Simultaneously, with a 30-second time window, the mean and peak body kinetic energy of the accelerometer signals are calculated, ultimately forming a multidimensional sleep state feature sequence containing both HRV and body kinetic characteristics.

[0069] The aforementioned features are input into a random forest classification model mounted on the local microcontroller. This model is trained on standard polysomnography (PSG) data and lightweight compressed, with fewer than 50KB of parameters, enabling stable operation on embedded platforms. The model outputs the current sleep stage in 30-second epochs: light sleep (N1 / N2), deep sleep (N3), or rapid eye movement (REM) sleep. To avoid inconsistent classification results, the system performs majority voting smoothing on results from five consecutive epochs (150 seconds in total), outputting stable sleep stage labels and confidence scores.

[0070] After identifying the sleep stage, the system configures stimulation parameters according to a three-level strategy to minimize sleep disturbance while ensuring anti-snoring effectiveness: Light sleep stage (N1 / N2): Output current 0.5mA–3mA, duration of single stimulation Set to 15 seconds; Deep sleep stage (N3): Output current 3mA–8mA, duration of single stimulation Set to 30 seconds; Rapid eye movement (REM) stimulation: Output current 0.3mA–1mA, duration of single stimulation Set to 10 seconds.

[0071] Through the above-mentioned hierarchical matching, the system obtains personalized stimulation parameters that are adapted to the current sleep stage, which are then used by the waveform generation module to output stimulation signals.

[0072] In the third embodiment of the present invention, we can design a multi-site synergistic interference stimulation array, such as three sets of targeted application schemes for eight-channel electrodes.

[0073] Electrodes are placed according to the anatomical location of each muscle. A total of eight surface gel electrodes are used, divided into three groups: The first group, the main channel group (channels A / B, targeting the genioglossus muscle): four electrodes are arranged in a 2×2 matrix and applied to the area below the mandible and in the midline of the chin. The two electrodes (A+, A-) in channel A are placed symmetrically in the left-right direction (coronal plane), with a center-to-center distance between the electrodes between 25 mm and 35 mm; the two electrodes (B+, B-) in channel B are arranged in the anteroposterior direction (sagittal plane), also with a distance between them between 25 mm and 35 mm. The skin area covered by the four electrodes is approximately 40 mm × 40 mm.

[0074] The second group, auxiliary channel group C (for the mylohyoid muscle): two electrodes are attached to the inner side of the bilateral mandibular angles, symmetrically distributed along the lower edge of the mandible, with a center-to-center distance of 40 mm to 50 mm.

[0075] The third group, auxiliary channel group D (for the stylohyoid muscle): two electrodes are attached to the midline of the anterior neck, above the hyoid bone, arranged vertically along the anterior-posterior direction (sagittal plane), with a center-to-center distance of 20 mm to 30 mm.

[0076] Each channel group generates a high-frequency current, creating an interference low-frequency field in the corresponding muscle region. Specifically: Main channel group A (frequency...) ) and main channel group B (frequency) The current is superimposed deep within the belly of the genioglossus muscle (approximately 15 mm to 30 mm subcutaneously), generating a difference frequency. The interference low-frequency driving field. Auxiliary channel group C (frequency ) and main channel group B ( The same formation occurs in the region of the mylohyoid muscle. The interference field. Auxiliary channel group D (frequency) ) and auxiliary channel group C ( It also forms in the stylohyoid muscle region. The interference field. The instantaneous amplitude of the resultant electric field can be expressed by the following formula:

[0077] Nerve and muscle tissues have natural low-pass filtering characteristics for high-frequency carriers above 1kHz, and only respond to 10Hz envelope modulation signals. Therefore, an effective stimulation equivalent to 10Hz can be formed in the deep part of each target muscle, while the skin surface only receives high-frequency current and has no obvious stimulation.

[0078] Synchronized rhythmic contraction of three muscle groups achieves coordinated airway expansion: the genioglossus muscle contracts to pull the base of the tongue forward and downward, moving it forward 3mm–8mm and widening the glossopharyngeal space; the mylohyoid muscle contracts to lift the hyoid bone, synergistically expanding the oropharyngeal lateral wall; and the stylohyoid muscle contracts to pull the hyoid bone backward and upward, forming stable tension with the genioglossus muscle to maintain the tongue's position. This synergistic effect can further increase the narrowest cross-sectional area of ​​the airway by 20%–35% compared to single-site stimulation, achieving three-dimensional opening of the upper airway.

[0079] This embodiment provides an anti-snoring device based on closed-loop control of interference electrical stimulation, used to implement all five steps in the aforementioned method embodiment. See appendix. Figure 4 As shown, the device integrates the following four functional modules.

[0080] The acquisition and extraction module (401) is used to acquire sound signals and skin vibration signals, preprocess the sound signals and skin vibration signals, and determine the occurrence of snoring events based on the preprocessed signals; The stimulation circuit (402) is used to generate a first high-frequency stimulation wave and a second high-frequency stimulation wave with similar frequencies if a snoring event is detected; wherein the interference difference frequency between the first high-frequency stimulation wave and the second high-frequency stimulation wave is between 7 and 15 Hz. Electrode module (403) is used to apply a first high-frequency stimulation wave and a second high-frequency stimulation wave to the genioglossus muscle of the chin via a first set of stimulation electrodes and a second set of stimulation electrodes, respectively; under the interference of the first high-frequency stimulation pulse and the second high-frequency stimulation pulse, the target area motor neurons are activated, causing the genioglossus muscle to contract rhythmically to expand the airway cross-sectional area. The correction module (404) is used to continuously update the stimulation parameters based on the snoring sensor feedback after intervention and historical intervention effect data, and output the first high-frequency stimulation pulse and the second high-frequency stimulation pulse for individualized snoring cessation through a Bayesian optimization algorithm.

[0081] To improve user comfort, this embodiment provides an anti-snoring device based on closed-loop control of interference electrical stimulation, used to implement the two steps described in the extended description of the aforementioned method embodiment. The device integrates the following four functional modules.

[0082] The acquisition and extraction module is used to acquire sound signals and skin vibration signals, preprocess the sound signals and skin vibration signals, and determine the occurrence of snoring events based on the preprocessed signals. If a snoring event is detected, pulse wave signals and body movement signals are acquired. The adjacent heartbeat RR interval sequence is extracted from the pulse wave signal, and the time-domain and frequency-domain indices of heart rate variability are calculated. The mean and peak values ​​of body movement energy per unit time are extracted from the body movement signal to obtain a multidimensional sleep state representation sequence containing the time-domain and frequency-domain indices of heart rate variability and the mean and peak values ​​of body movement energy per unit time. The current sleep stage is determined based on the multidimensional sleep state representation sequence using a sleep staging estimation algorithm. The stimulation circuit is used to generate corresponding graded high-frequency stimulation wave parameter configurations based on the sleep stage, wherein the interference difference frequency of the graded high-frequency stimulation wave parameter configurations is between 7 and 15 Hz; and to generate a first high-frequency stimulation wave and a second high-frequency stimulation wave according to the graded high-frequency stimulation wave parameter configurations. The electrode module is used to apply a first high-frequency stimulation wave and a second high-frequency stimulation wave to the genioglossus muscle of the chin via a first set of stimulation electrodes and a second set of stimulation electrodes, respectively; under the interference of the first high-frequency stimulation pulse and the second high-frequency stimulation pulse, the target area motor neurons are activated, causing the genioglossus muscle to contract rhythmically to expand the airway cross-sectional area. The correction module is used to continuously update the stimulation parameters based on the snoring sensor feedback after intervention and historical intervention effect data, and output the first high-frequency stimulation pulse and the second high-frequency stimulation pulse for individualized snoring cessation through a Bayesian optimization algorithm.

[0083] The core innovation of this invention lies in applying the principle of interference electrical stimulation to the field of anti-snoring, achieving deep targeted stimulation of the genioglossus muscle. By applying a high-frequency current to the body surface and utilizing the low-pass filtering characteristics of tissue, an equivalent low-frequency stimulation of 10Hz is formed deep within the muscle. This effectively activates the genioglossus muscle 15mm–30mm subcutaneously without significant skin discomfort, solving the problem of balancing comfort and stimulation depth in traditional transcutaneous electrical stimulation.

[0084] Based on this, the present invention adopts a four-channel frequency stepped carrier design (2000Hz, 2010Hz, 2020Hz, 2030Hz), so that adjacent channels can form a 10Hz interference electric field, realizing the synchronous and coordinated work of the three muscles of the genioglossus, mylohyoidus, and stylohyoidus, forming a three-dimensional traction on the tongue and hyoid bone, which changes the limitation of the traditional solution that can only stimulate a single muscle.

[0085] Meanwhile, this invention combines sleep stage detection with graded stimulation, adjusting the stimulation intensity in segments according to light sleep, deep sleep, and REM sleep. While ensuring the anti-snoring effect, it minimizes sleep disturbance and makes up for the lack of sleep protection mechanism in existing electrical stimulation anti-snoring devices.

[0086] In addition, this invention innovatively introduces a Bayesian optimization algorithm, which can automatically optimize and adaptively adjust multi-dimensional parameters such as output current, stimulation duration, frequency difference, and channel ratio, so as to achieve personalized parameter matching for different users and overcome the shortcomings of traditional anti-snoring devices with fixed parameters and poor adaptability.

[0087] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural transformations made using the contents of the present invention's specification and drawings under the inventive concept of the present invention, or direct / indirect applications in other related technical fields, are included within the patent protection scope of the present invention.

[0088] A computer device according to embodiments of the present disclosure includes a memory and a processor. The memory is used to store non-transitory computer-readable instructions. Specifically, the memory may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may, for example, include random access memory (RAM) and / or cache memory. The non-volatile memory may, for example, include read-only memory (ROM), hard disk, flash memory, etc.

[0089] The processor may be a central processing unit (CPU) or other form of processing unit with data processing capabilities and / or instruction execution capabilities, and may control other components in the computer device to perform desired functions. In one embodiment of this disclosure, the processor is used to execute computer-readable instructions stored in the memory, causing the computer device to perform all or part of the steps of the methods described in the foregoing embodiments of this disclosure.

[0090] like Figure 5 This is a schematic diagram of a computer device provided for an embodiment of the present disclosure. It illustrates a structural schematic diagram suitable for implementing the computer device in the embodiments of the present disclosure. Figure 5 The computer device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein.

[0091] For a detailed description of this embodiment, please refer to the corresponding descriptions in the foregoing embodiments, which will not be repeated here.

Claims

1. A method of snore stoppage based on closed loop control of interferential electrical stimulation, characterized in that, Includes the following steps: Based on the acquisition of sound signals and skin vibration signals by sensors, the sound signals and skin vibration signals are preprocessed, and the occurrence of snoring events is determined based on the preprocessed signals; If the determination result indicates that the snoring event has occurred, a first high-frequency stimulation wave and a second high-frequency stimulation wave with similar frequencies are generated respectively; wherein the frequency range of the first high-frequency stimulation wave and the second high-frequency stimulation wave is 200Hz to 20kHz, and the interference difference frequency is between 7 and 15Hz. The first high-frequency stimulation wave and the second high-frequency stimulation wave are applied to the genioglossus muscle of the chin via the first group of stimulation electrodes and the second group of stimulation electrodes, respectively. The target area motor neurons are activated by the interference of the first high-frequency stimulation pulse and the second high-frequency stimulation pulse, causing the genioglossus muscle to contract rhythmically to expand the airway cross-sectional area. Based on the collected sound signals and skin vibration signals after intervention, as well as historical intervention effect data, the stimulation parameters are continuously updated through a Bayesian optimization algorithm to output personalized third and fourth high-frequency stimulation waves for snoring relief.

2. The method of claim 1, wherein, The preprocessing of the sound signal and skin vibration signal specifically includes: To remove low-frequency environmental noise and DC offset from the audio signal by passing it through a high-pass filter, a Hamming window is used to perform frame-by-frame processing on the filtered audio signal, and short-time energy and zero-crossing rate are calculated frame by frame. Perform a Fast Fourier Transform on each frame of audio signal to extract the energy percentage of the frequency band from 100Hz to 500Hz; Hilbert transform is performed on the vibration signal to extract the instantaneous envelope and estimate the power spectral density, and the average vibration power in the frequency band from 50Hz to 300Hz is extracted simultaneously. The short-time energy, zero-crossing rate, energy percentage, and average vibration power are concatenated into a feature vector, and a sliding feature vector sequence is formed with a fixed time step.

3. The method of claim 2, wherein, The first and second sets of stimulation electrodes are applied to the midline region of the chin below the mandible. The first set of stimulation electrodes consists of two electrodes symmetrically distributed along the coronal plane, while the second set of stimulation electrodes consists of two electrodes distributed along the sagittal plane.

4. The method of claim 1, wherein, The historical intervention effect data includes: The cumulative number of interventions and total duration of stimulation within the current sleep session are statistically analyzed, and a fatigue protection mechanism is used to limit continuous interventions. If the number of interventions interrupted by the fatigue protection mechanism exceeds a threshold value for more than two consecutive sleep cycles, automatically shorten the length of the single stimulation Increase the amplitude of the output current and shorten the observation window.

5. The method of claim 1, wherein, The process of continuously updating stimulation parameters using a Bayesian optimization algorithm and outputting individualized first and second high-frequency stimulation pulses for snoring prevention specifically includes: The variable parameters consisting of the output current amplitude of the first and second high-frequency stimulation pulses, the duration of a single intervention, and the frequency difference between the two channels are obtained. The optimal parameters of the variables are obtained by Gaussian process regression and expected improvement function.

6. The method of claim 5, wherein, The Gaussian process regression function is: where, and denote two arbitrary sample points in the parameter space, is the Euclidean distance between the two points; is the signal variance, is the length scale hyperparameter, the output of the kernel function denotes the covariance between the two points; The desired improvement in the acquisition function is: wherein, represents a sample point to be evaluated, is the value of the objective function at the sample point as predicted by the Gaussian process surrogate model, is the current best observed objective function value, represents taking the mathematical expectation over the random variables in the parentheses, represents that a positive improvement is only counted when the predicted value exceeds the current best value, otherwise it is zero.

7. A snore stopping method based on closed loop control of interferential electrical stimulation, characterized in that, Includes the following steps: Based on the acquisition of sound signals and skin vibration signals by sensors, the sound signals and skin vibration signals are preprocessed, and the occurrence of snoring events is determined based on the preprocessed signals; If the determination result indicates that the snoring event has occurred, pulse wave signal and body movement signal are collected. The adjacent heartbeat RR interval sequence is extracted from the pulse wave signal, and the time domain index and frequency domain index of heart rate variability are calculated. The mean and peak value of body movement energy per unit time are extracted from the body movement signal to obtain a multidimensional sleep state characterization sequence containing the time domain index and frequency domain index of heart rate variability, and the mean and peak value of body movement energy per unit time. The multidimensional sleep state representation sequence is used to determine the current sleep stage through a sleep stage estimation algorithm, and a corresponding graded high-frequency stimulation wave parameter configuration is generated based on the determined current sleep stage; the interference difference frequency of the graded high-frequency stimulation wave parameter configuration is between 7Hz and 15Hz. The first high-frequency stimulation wave and the second high-frequency stimulation wave are generated according to the configuration of the graded high-frequency stimulation wave parameters. The first high-frequency stimulation wave and the second high-frequency stimulation wave are applied to the genioglossus muscle of the chin through the first group of stimulation electrodes and the second group of stimulation electrodes, respectively. The target area motor neurons are activated by the interference of the first high-frequency stimulation pulse and the second high-frequency stimulation pulse, causing the genioglossus muscle to contract rhythmically to expand the airway cross-sectional area. Based on the collected sound signals and skin vibration signals after intervention, as well as historical intervention effect data, the stimulation parameters are continuously updated through a Bayesian optimization algorithm to output personalized third and fourth high-frequency stimulation waves for snoring relief.

8. The method of claim 7, wherein, The fifth and sixth high-frequency stimulation waves are generated according to the configuration of the graded high-frequency stimulation wave parameters. The fifth and sixth high-frequency stimulation waves are applied to the mylohyoid muscle via the third and fourth stimulation electrodes, respectively. The seventh and eighth high-frequency stimulation waves are generated according to the configuration of the graded high-frequency stimulation wave parameters. The seventh and eighth high-frequency stimulation waves are applied to the stylohyoid muscle via the fifth and sixth stimulation electrodes, respectively.

9. An anti-snoring device based on closed-loop control of interference electrical stimulation, characterized in that, The device includes: a data acquisition and extraction module, used to acquire sound signals and skin vibration signals, preprocess the sound signals and skin vibration signals, and determine the occurrence of snoring events based on the preprocessed signals; The stimulation circuit is used to generate a first high-frequency stimulation wave and a second high-frequency stimulation wave with similar frequencies if a snoring event is detected; wherein the interference difference frequency between the first high-frequency stimulation wave and the second high-frequency stimulation wave is between 7 Hz and 15 Hz. The electrode module is used to apply a first high-frequency stimulation wave and a second high-frequency stimulation wave to the genioglossus muscle of the chin via a first set of stimulation electrodes and a second set of stimulation electrodes, respectively; under the interference of the first high-frequency stimulation pulse and the second high-frequency stimulation pulse, the target area motor neurons are activated, causing the genioglossus muscle to contract rhythmically to expand the airway cross-sectional area. The correction module is used to continuously update the stimulation parameters based on the collected sound signals and skin vibration signals after intervention and historical intervention effect data, and output personalized anti-snoring third and fourth high-frequency stimulation waves through a Bayesian optimization algorithm.

10. An anti-snoring device based on closed-loop control of interference electrical stimulation, characterized in that, The device includes: The acquisition and extraction module is used to acquire sound signals and skin vibration signals, preprocess the sound signals and skin vibration signals, and determine the occurrence of snoring events based on the preprocessed signals. If a snoring event is detected, pulse wave signals and body movement signals are acquired. The adjacent heartbeat RR interval sequence is extracted from the pulse wave signal, and the time-domain and frequency-domain indices of heart rate variability are calculated. The mean and peak values ​​of body movement energy per unit time are extracted from the body movement signal to obtain a multidimensional sleep state representation sequence containing the time-domain and frequency-domain indices of heart rate variability and the mean and peak values ​​of body movement energy per unit time. The current sleep stage is determined based on the multidimensional sleep state representation sequence using a sleep staging estimation algorithm. The stimulation circuit is used to generate corresponding graded high-frequency stimulation wave parameter configurations based on the sleep stage, wherein the interference difference frequency of the graded high-frequency stimulation wave parameter configurations is between 7Hz and 15Hz; and to generate a first high-frequency stimulation wave and a second high-frequency stimulation wave according to the graded high-frequency stimulation wave parameter configurations. The electrode module is used to apply a first high-frequency stimulation wave and a second high-frequency stimulation wave to the genioglossus muscle of the chin via a first set of stimulation electrodes and a second set of stimulation electrodes, respectively; under the interference of the first high-frequency stimulation pulse and the second high-frequency stimulation pulse, the target area motor neurons are activated, causing the genioglossus muscle to contract rhythmically to expand the airway cross-sectional area. The correction module is used to continuously update the stimulation parameters based on the collected sound signals and skin vibration signals after intervention and historical intervention effect data, and output personalized anti-snoring third and fourth high-frequency stimulation waves through a Bayesian optimization algorithm.

Images