High-precision sleep electrocardio continuous monitoring system and method

Abstract

The invention relates to the technical field of medical monitoring, and provides a high-precision sleep electrocardio continuous monitoring system and method.The high-precision sleep electrocardio continuous monitoring system and method.A modular closed-loop framework is adopted, and synchronous acquisition of electrocardio, movement and respiration signals is achieved through a multi-modal sensing and high-precision synchronous acquisition module; the signal quality multi-dimensional traceability diagnosis module performs feature extraction and noise classification on the input signal; the dynamic traceability filtering processing module intelligently calls a corresponding filtering algorithm to perform signal purification according to the noise type; and the intelligent output and self-adaptive resource management module outputs high-quality electrocardiosignals and realizes dynamic optimization of system power consumption. By establishing a complete signal quality evaluation system and an intelligent processing mechanism, the whole process optimization of the sleep electrocardiosignals from collection to output is realized, and the accuracy of monitoring data and the cruising ability of the system are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical monitoring, in particular to a high-precision sleep electrocardiogram continuous monitoring system and method. BACKGROUND

[0002] With the rapid development of health monitoring technology, sleep electrocardiogram monitoring has become an important means of early screening and health management of cardiovascular diseases. However, the existing electrocardiogram monitoring technology faces many technical bottlenecks when applied to the sleep scene. First, the human body cannot actively cooperate during sleep, and the electrode contact state is easily affected by actions such as turning over, resulting in unstable signal quality. Second, the motion artifacts generated during sleep, baseline drift caused by respiration, and various noises such as environmental electromagnetic interference are intertwined with each other, and traditional filtering methods are difficult to effectively distinguish between noise and effective signal. Third, in order to realize long-term continuous monitoring, the system needs to balance the contradiction between processing performance and power consumption under limited battery capacity, and existing solutions often cannot be considered. In addition, the problem of time sequence asynchronization in the process of collecting multi-modal physiological signals also restricts the accuracy of signal fusion analysis. These factors together result in the existing sleep electrocardiogram monitoring system having defects such as poor signal quality, weak noise suppression ability, low power efficiency, and insufficient data analysis reliability.

[0003] Therefore, a high-precision sleep electrocardiogram continuous monitoring system and method is proposed, which adopts a strict modular design architecture and is composed of four core functional modules connected in sequence to form a complete processing link. The modules are connected through bidirectional communication to form a closed-loop control system, and the downstream modules adjust the parameters of the upstream modules in real time through the feedback channel. SUMMARY

[0004] In view of the deficiencies of the prior art, the present application provides a high-precision sleep electrocardiogram continuous monitoring system and method.

[0005] To achieve the above-mentioned solution, the present application provides the following technical scheme: a high-precision sleep electrocardiogram continuous monitoring system, comprising multi-modal sensing and high-precision synchronous acquisition module, signal quality multi-dimensional traceability diagnosis module, dynamic traceability filtering processing module and intelligent output and adaptive resource management module connected in sequence, the modules are connected through data bus and control bus to establish bidirectional communication, and the downstream modules adjust the parameters of the upstream modules in real time through the feedback channel to form a closed-loop control system.

[0006] The multi-modal sensing and high-precision synchronous acquisition module collects electrocardiogram, motion and respiration physiological signals and integrates multi-source data streams through a precise time synchronization mechanism.

[0007] The signal quality multi-dimensional traceability diagnosis module pre-processes, extracts features and classifies and identifies noise of the input signal and outputs signal quality index and noise type label.

[0008] The dynamic traceability filtering processing module calls corresponding filtering algorithms according to the noise classification result to perform targeted signal purification processing.

[0009] The intelligent output and adaptive resource management module outputs high-quality electrocardio signals and metadata and realizes dynamic optimization management of system power consumption and parameters.

[0010] Preferably, the multi-modal sensing and high-precision synchronous acquisition module includes a main electrocardio sensing unit, an auxiliary electrocardio sensing unit, a motion and physiological sensing unit, and a synchronous acquisition controller; the main electrocardio sensing unit is fixed to the second intercostal space of the right sternal margin and the fifth intercostal space of the midclavicular line according to the analog modified lead II standard layout using a flexible dry electrode material; the auxiliary electrocardio sensing unit is composed of a pair of independent flexible dry electrodes, with a spatial distance greater than or equal to 2 cm from the main electrode pair and a 45-90 degree included angle between the electrical vector directions; the motion and physiological sensing unit integrates a six-axis inertial measurement unit and a four-electrode bioimpedance sensor; the synchronous acquisition controller uses a high-performance low-power microcontroller, integrates a high-stability crystal oscillator with a precision of less than or equal to 1 microsecond, and synchronously acquires signals through parallel multi-channel ADC and adds time stamps.

[0011] Preferably, the signal quality multi-dimensional traceability diagnosis module includes a preprocessing and parallel feature extraction unit, a noise type classifier, and a comprehensive quality evaluation unit; the preprocessing and parallel feature extraction unit filters and extracts features from electrocardio, motion, and respiratory signals, and calculates time domain, frequency domain, and correlation features; the noise type classifier uses a lightweight machine learning model, inputs a 15-dimensional feature vector, and outputs five noise labels of ClassA high-quality signal, ClassB motion artifact, ClassC periodic artifact, ClassD respiratory baseline drift, and ClassE poor contact or signal drop; the comprehensive quality evaluation unit calculates a signal quality index SQI in the range of 0 to 1, and determines effectiveness according to classification confidence.

[0012] Preferably, the dynamic traceability filtering processing module includes an intelligent algorithm routing controller, a blind source separation engine, an adaptive resonance canceller, an impedance-based baseline corrector, and a data security and labeling unit; the intelligent algorithm routing controller dynamically schedules processing engines according to noise types and confidence; the blind source separation engine processes ClassB noise and performs fast independent component analysis; the adaptive resonance canceller processes ClassC noise and performs adaptive notch filtering; the impedance-based baseline corrector processes ClassD noise and performs a regression correction algorithm; the data security and labeling unit processes ClassE noise and labels invalid data segments.

[0013] Preferably, the intelligent output and adaptive resource management module comprises a purification signal and metadata output unit, a dynamic power management unit, and a self-aware feedback control unit; the purification signal and metadata output unit outputs high-fidelity electrocardiogram signals and structured metadata, the metadata including a timestamp, a signal quality index (SQI), a noise label, and a processing algorithm identifier; the dynamic power management unit activates a low-power mode when high-quality signals are continuously detected, and triggers an on-demand activation mode when noise features are detected; and the self-aware feedback control unit sends parameter adjustment instructions to the front-end acquisition module based on signal quality index history and noise pattern statistics.

[0014] Preferably, the low-power mode of the dynamic power management unit is realized by at least one of the following methods: reducing the sampling rate of the inertial measurement unit, turning off the auxiliary electrocardiogram acquisition channel, and intermittently disabling the blind source separation engine calculation unit.

[0015] The high-precision sleep electrocardiogram continuous monitoring method is applied to the high-precision sleep electrocardiogram continuous monitoring system, and comprises the following steps: S1. High-precision multi-modal signal synchronous acquisition: multi-modal sensor network is used to acquire electrocardiogram, motion, and respiration signals, a high-precision clock source is used to generate a unified sampling clock, all sensor channels are synchronously triggered, and time stamps are added to the data.

[0016] S2. Multi-dimensional signal quality traceability diagnosis: pre-processing, feature extraction, and noise classification are performed on the multi-modal signals, a signal quality index (SQI) and a classification confidence are calculated.

[0017] S3. Dynamic traceability filtering processing: filtering algorithms are scheduled according to noise types and confidence, and signal purification processing is performed.

[0018] S4. Intelligent output and adaptive resource management: purified electrocardiogram signals and metadata are output, and system power consumption and parameters are dynamically managed.

[0019] Preferably, the high-precision multi-modal signal synchronous acquisition step comprises: the main and auxiliary electrocardiogram electrodes are arranged in a spatially differentiated manner with a spacing of greater than or equal to 2 cm, and the electrical vector directions form an included angle of 45 to 90 degrees; and all signal channels are synchronously acquired at a sampling rate of 250 Hz by a synchronous acquisition controller.

[0020] Preferably, the multi-dimensional signal quality traceability diagnosis step comprises: the electrocardiogram signal is subjected to 0.5 to 40 Hz band-pass filtering, the motion signal is subjected to direct current removal processing, and the respiration signal is subjected to envelope detection; time domain, frequency domain, and correlation features are extracted to construct a 15-dimensional feature vector; a lightweight classification model is used to output five types of noise labels, and the classification is determined to be valid when the confidence is greater than or equal to 0.8.

[0021] Preferably, the dynamic traceability filtering processing step comprises: when the confidence is greater than or equal to 0.8, activating a corresponding processing engine according to the noise type; enabling a blind source separation engine for Class B noise; enabling an adaptive resonance canceller for Class C noise; enabling an impedance-based baseline corrector for Class D noise; and enabling a data security and marking unit for Class E noise.

[0022] Compared with the prior art, the present application provides a high-precision sleep electrocardiogram continuous monitoring system and method, which has the following beneficial effects: 1. In the present scheme, by adopting a main-aid dual-channel electrocardiogram sensing design, a homologous noise-different dual-channel input system with noise mode difference is constructed through strict geometric relationship and electrical vector differentiation layout. With a high-precision synchronous acquisition controller and a unified clock reference, strict synchronous acquisition of multi-modal signals is realized, fundamentally solving the sampling time delay and signal asynchronization problems existing in traditional monitoring systems.

[0023] 2. In the present scheme, through parallel feature extraction and lightweight machine learning model, accurate identification and classification of noise types are realized. Based on the dynamic traceability filtering processing mechanism of the diagnostic results, an accurate processing pipeline for different noise types is constructed, and through the intelligent routing mechanism, multiple special processing engines are coordinated to realize the process from signal processing to signal management, improving the accuracy and reliability of electrocardiogram signal analysis.

[0024] 3. In the present scheme, through the dynamic power management strategy and self-sensing feedback control mechanism, intelligent balance of system performance and energy consumption is realized. The system can intelligently identify specific problem patterns and automatically adjust working parameters based on signal quality history and noise mode statistics. This adaptive feedback regulation enables the system to intelligently respond to changes in the use environment and fluctuations in the user's physiological state, maintaining the optimal working state in the long-term continuous monitoring process. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 The figure is a schematic diagram of the overall architecture of the system of the present application; Figure 2 The figure is a schematic diagram of the hardware connection of the multi-modal sensing and high-precision synchronous acquisition module of the present application; Figure 3 The figure is a schematic diagram of the processing flow of the signal quality multi-dimensional traceability diagnosis module of the present application; Figure 4 The figure is a schematic diagram of the algorithm routing of the dynamic traceability filtering processing module of the present application; Figure 5 The figure is a schematic diagram of the output and power consumption management of the intelligent output and adaptive resource management module of the present application; Figure 6 The figure is a schematic diagram of the method flow of the present application. Detailed Implementation

[0026] 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. Example

[0027] Please see Figures 1-5 The high-precision sleep ECG continuous monitoring system is a dedicated ECG signal monitoring and processing system for sleep. It adopts a strict modular design architecture, with four core functional modules connected in sequence to form a complete processing link.

[0028] The multimodal sensing and high-precision synchronous acquisition module, serving as the system front-end, is responsible for acquiring various physiological signals such as ECG, motion, and respiration, and integrating multi-source data streams through a precise time synchronization mechanism. The multi-dimensional signal quality tracing and diagnostic module, connected downstream of the acquisition module, performs preprocessing, feature extraction, and noise classification of the input signals. The dynamic tracing and filtering module intelligently calls the appropriate filtering algorithm based on the diagnostic results to achieve targeted signal purification. The intelligent output and adaptive resource management module, serving as the system terminal, is responsible for outputting high-quality ECG signals and dynamically optimizing system power consumption and parameters.

[0029] Each module establishes a bidirectional communication connection via a data bus and a control bus, forming a complete closed-loop system from signal acquisition and processing to output. Downstream modules adjust the parameters of upstream modules in real time through feedback paths, ensuring that the system is always in optimal operating condition. This hierarchical processing architecture allows each module to focus on a specific function while maintaining the efficient and coordinated operation of the overall system.

[0030] The multimodal sensing and high-precision synchronous acquisition module consists of four physically integrated sensing sub-units and a central synchronous controller, forming a complete acquisition system.

[0031] The main ECG sensing unit uses a biocompatible flexible dry electrode material and is laid out in strict accordance with the electrical standards of the analog modified lead II. It is fixed at two key locations on the monitoring patch body: the second intercostal space at the right sternal border and the fifth intercostal space at the left midclavicular line. This unit is connected to the synchronous acquisition controller via a three-wire analog signal line, where two signal lines transmit differential ECG signals and a common ground line provides a reference potential, ensuring high-quality acquisition of the core ECG signal.

[0032] The auxiliary ECG sensing unit is composed of a pair of independent flexible dry electrodes, which maintain a strict geometric relationship with the main electrode pair on the monitoring patch layout. The spatial linear distance is ensured to be ≥2 cm, and the electrical vector direction forms an angle of 45-90 degrees with the main electrode pair. The independent dual-channel analog front end is connected to the synchronous acquisition controller. This differentiated design creates a second observation channel that is homologous to the main ECG signal but has significantly different noise patterns, providing key data support for subsequent blind source separation algorithms.

[0033] The motion and physiological sensing unit integrates two functionally complementary detection subunits. The inertial measurement unit uses a highly integrated six-axis motion sensor chip, which includes a three-axis accelerometer and a three-axis gyroscope. It communicates data with the synchronous acquisition controller through an I2C digital serial interface, accurately quantifying the motion state of the device. The bioimpedance sensor uses a precise four-electrode method design, with two excitation electrodes and two detection electrodes cleverly integrated on the surface of the ECG sensing electrodes. It realizes accurate extraction of respiratory signals through an alternating excitation signal and a synchronous demodulation circuit.

[0034] The synchronous acquisition controller uses a high-performance low-power microcontroller as the core processing unit, and integrates a high-stability crystal oscillator with a precision of ≤1 microsecond as the system clock reference. The controller synchronously acquires ECG and impedance signals through a parallel multi-channel analog-to-digital converter, and simultaneously reads IMU data in real time through a digital interface. All data channels share the same clock source, ensuring strict synchronization of different physical signals at the sampling time, completely eliminating the cross-sensor data fusion error caused by sampling time delay.

[0035] After the system is powered on, the synchronous acquisition controller performs initialization operations according to the preset process. First, configure all sensor interface and ADC channel parameters, send calibration signals to the ECG analog front end and bioimpedance analog front end for baseline calibration. Then start the high-precision timer to generate a 250Hz unified sampling clock, and at the rising edge of each sampling clock, the controller synchronously triggers all ADC channels for data conversion. After adding accurate time stamps to the converted multi-channel data, it is packaged into a standard data frame, and finally the complete data frame is transmitted to the downstream diagnostic module through the data bus. This provides a reliable timing basis for subsequent multi-source signal collaborative processing.

[0036] The signal quality multi-dimensional traceability diagnostic module includes three serial processing units, which perform step-by-step in-depth analysis on the input multi-modal signals, and construct a complete signal quality evaluation system.

[0037] After receiving the raw data stream from the acquisition module, the preprocessing and parallel feature extraction unit first performs targeted preprocessing on the three types of signals. For the ECG signal channel, a 0.5 Hz high-pass filter is applied to remove baseline drift, and a 40 Hz low-pass filter is applied to suppress high-frequency noise, while retaining the main energy frequency band of the QRS complex. For the IMU signal channel, a DC removal process is performed to eliminate static offset and highlight dynamic acceleration and angular velocity changes. For the respiratory impedance signal, an envelope detection algorithm is used to extract the respiratory fluctuation curve, remove high-frequency interference components, and retain effective respiratory information.

[0038] Based on preprocessing, this unit calculates three types of key feature parameters in parallel. Time-domain features include signal peak-to-peak range, variance, and zero-crossing rate, reflecting the overall fluctuation characteristics of the signal. Frequency-domain features include the energy proportion of the 0-5 Hz frequency band corresponding to the baseline and respiratory components, the 5-15 Hz frequency band corresponding to the QRS complex components, and the 15-40 Hz frequency band corresponding to the electromyographic noise components, revealing the frequency distribution characteristics of the signal. Correlation features include the maximum value of the cross-correlation function of the primary and secondary ECG signals, the coherence coefficient between ECG and IMU motion vectors, and the coherence coefficient between ECG and respiratory impedance signals, representing the degree of internal correlation between signals.

[0039] The noise type classifier uses a specially optimized lightweight machine learning model, implemented as a support vector machine or a multi-class decision tree architecture. The model is pruned and parameter-optimized to adapt to the computational resource limitations of the embedded platform, receives a 15-dimensional feature vector as input, containing all time-domain, frequency-domain, and correlation feature parameters, and outputs five explicit noise labels. The classification decision process sets a confidence threshold of 0.8, and results below this threshold are considered unreliable for classification, ensuring the accuracy of the classification results.

[0040] The comprehensive quality assessment unit receives the classification results and performs comprehensive calculations based on the deviation of each dimension of the feature vector from the ideal value, calculating a signal quality index SQI in the range of 0-1. Each data segment is labeled with a noise type and a classification confidence. When the confidence is ≥0.8, the classification is considered valid, otherwise it is marked as "to be confirmed" and awaits further processing. This refined quality assessment system provides a reliable decision basis for subsequent filtering algorithm selection.

[0041] The 5 categories of noise output by the classifier have clear and distinguishable feature patterns. Class A, a high-quality signal, has all feature parameters within the normal physiological range, frequency domain features show that the QRS complex energy ratio is >60%, and correlation features show that the primary and secondary ECGs are highly consistent. Class B, a motion artifact, has an ECG-IMU coherence coefficient >0.7, time domain features show non-physiological mutations, and the primary and secondary ECG signals show synchronous distortion characteristics. Class C, a periodic artifact, has isolated narrowband spectral peaks in the frequency domain features, with energy concentrated in non-physiological frequency bands and no harmonic relationship with the heart rhythm. Class D, a respiratory baseline drift, has an ECG-respiratory impedance signal coherence coefficient >0.6, abnormally high low-frequency energy ratio, and time domain signals showing slow periodic fluctuations. Class E, poor contact or signal loss, has time domain amplitude exceeding the ADC range or approximately zero, and abnormally low total energy in the frequency domain.

[0042] The dynamic traceability filtering processing module uses an intelligent routing mechanism to coordinate multiple dedicated processing engines, and constructs precise processing pipelines for different noise types. The intelligent algorithm routing controller, as the scheduling core of this module, reads the noise type label and confidence score output by the upstream diagnosis module in real time. According to the preset logic, the processing path is dynamically constructed: when the confidence is <0.8, the controller directly forwards the original signal to the output module and adds a "classification suspicious" label to avoid false processing; when the confidence is ≥0.8, the controller activates the corresponding dedicated processing engine according to the noise type, forming a precise mapping relationship of Class B activating blind source separation engine, Class C activating adaptive resonance canceller, Class D activating impedance-based baseline corrector, and Class E activating data security and labeling unit. The controller continuously monitors the execution state of each processing engine to ensure that the real-time requirement is met.

[0043] The blind source separation engine specifically handles Class B motion artifacts and executes a fast independent component analysis algorithm. First, input the primary and secondary ECG signals to form a two-dimensional observation vector, and perform centering and whitening preprocessing to eliminate correlation; then find the demixing matrix through the fixed-point iteration algorithm, maximize the non-Gaussianity of the output components to achieve signal separation; finally, based on the periodicity and morphological characteristics of R-wave rhythm, automatically identify the electrocardiogram source components, output pure electrocardiogram signals while suppressing motion interference components. This processing method effectively solves the serious impact of motion artifacts on electrocardiogram signal quality.

[0044] The adaptive resonance canceller is designed for Class C periodic artifacts, which performs an adaptive notch filtering algorithm. The interference frequency is accurately identified by IMU data spectral analysis or ECG signal autocorrelation analysis, and a second-order IIR notch filter is configured with the center frequency locked within the interference frequency ± 0.1 Hz. The adjustable bandwidth parameter is set to 0.5-2 Hz, which effectively filters out the interference while maximizing the preservation of QRS complex spectral components. The system tracks the change of interference frequency in real time and dynamically adjusts the filter parameters to ensure continuous suppression effect on periodic noise.

[0045] The impedance-based baseline corrector is specifically designed to handle Class D respiratory drift, which performs a regression correction algorithm. The simultaneously acquired respiratory impedance signal is used as a baseline reference to establish a linear regression model between the respiratory signal amplitude and the ECG baseline position. The baseline drift curve is accurately estimated according to the regression coefficient, and the estimated drift component is subtracted from the original ECG signal to output a baseline-stable electrocardiogram signal. This method fully utilizes the advantages of multi-modal signals and effectively corrects the respiratory baseline drift.

[0046] The data security and labeling unit is responsible for handling Class E signal quality issues, which performs an integrity check process. It detects whether the signal amplitude exceeds the reasonable physiological range, evaluates the signal-to-noise ratio and quality index of the signal segment. For non-recoverable signal degradation segments, it is marked as "no clinical diagnostic value" and inserts a detailed label frame containing the start time, end time, and failure reason into the data stream. This processing method ensures the safety and reliability of data usage, avoiding misleading of invalid data on subsequent analysis.

[0047] The intelligent output and adaptive resource management module ensures the system to provide high-quality data while optimizing energy efficiency through standardized output interfaces and intelligent management mechanisms. The purified signal and metadata output unit generates two types of strictly time-aligned output data. The high-fidelity electrocardiogram signal stream uses 16-bit linear encoding with a fixed sampling rate of 250 Hz to output signal data within the ±5mV physiological range, with a frequency bandwidth strictly controlled within the 0.5-40Hz range. The structured metadata stream contains complete information such as millisecond-level precision timestamps, signal quality index SQI values, noise type classification labels, used processing algorithm identifiers, and invalid data segment labels. The invalid data segment label details the start time, end time, and specific failure reason, providing comprehensive reference for subsequent data analysis.

[0048] The dynamic power manager implements two energy efficiency optimization strategies. The low-power mode is automatically activated when a Class A high-quality signal is detected for 3 consecutive seconds. This reduces power consumption by 30%–40% by reducing the IMU sampling rate from 250Hz to 125Hz, disabling the auxiliary ECG acquisition channel, and intermittently shutting down the blind source separation engine computing unit. The on-demand activation mode is triggered immediately when the diagnostic module detects Class B–Class D noise characteristics, completing the state switch within 10 milliseconds. It starts the corresponding sensor at full power and activates the relevant processing engine to a high-performance state, ensuring that real-time processing capabilities are not affected. This dynamic power management mechanism effectively balances system performance and battery life, making it particularly suitable for applications requiring long-term continuous monitoring.

[0049] The self-sensing feedback control unit continuously optimizes system performance through intelligent closed-loop adjustment. Based on continuously accumulated historical SQI data and noise pattern statistics, it identifies specific problem patterns such as signal saturation caused by sweating or amplitude fluctuations caused by poor contact. It sends precise parameter adjustment commands to the front-end acquisition module via a dedicated feedback control line. Specifically, when signal saturation is detected for 5 consecutive minutes and the amplitude reaches 90% of full scale, the command instructs the simulated front-end to appropriately reduce the gain by ≤3dB to prevent signal clipping; when a continuous increase in contact impedance is detected, the detection threshold and excitation current parameters are adjusted accordingly; and based on long-term noise type statistics, the classifier decision threshold is dynamically optimized to improve recognition accuracy. This self-sensing feedback mechanism enables the system to intelligently adapt to changes in the usage environment and fluctuations in the user's physiological state, maintaining optimal operating conditions during long-term continuous monitoring, ensuring continuous stability of data quality and efficient system operation.

[0050] Example 2: Please see Figure 6 A high-precision continuous sleep ECG monitoring system includes the following steps: Step 1: High-precision synchronous acquisition of multi-modal signals First, the system is initialized and configured, establishing a multimodal sensing network consisting of a primary ECG sensing electrode, an auxiliary ECG sensing electrode, an inertial measurement unit, and a bioimpedance sensor. The primary and auxiliary ECG electrodes employ a spatially differentiated layout design, ensuring that the distance between them is strictly controlled to be more than 2 cm, and that their electrical vector directions remain significantly different. This unique layout constructs a dual-channel input system with "same source, different noise" characteristics and different noise modes, laying the foundation for subsequent signal separation and processing.

[0051] After the hardware configuration is completed, the high-precision clock source generates a unified sampling clock signal of 250 Hz, which synchronously triggers all sensor channels to enter the data acquisition state. The specific acquisition process includes: acquiring the core ECG signal through the main ECG sensing channel, collecting the second ECG observation signal through the auxiliary ECG sensing channel, recording three-axis acceleration and angular velocity data using the inertial measurement unit, and monitoring the thoracic impedance change characteristics with the help of the bioimpedance sensor. All collected data are added with precise timestamps with a precision of 1 microsecond, ensuring the strict alignment of multi-source data streams in the time dimension.

[0052] The data integration stage packages the time-synchronized multi-modal signals into a unified data frame and transmits them to the signal processing unit through a high-speed data bus. This precise time synchronization mechanism effectively eliminates the sampling delay problem existing in traditional acquisition methods, providing a reliable timing basis for subsequent signal fusion analysis and is a key link to ensure the monitoring accuracy of the entire system.

[0053] Step two: multi-dimensional source diagnosis of signal quality After entering the signal quality evaluation stage, the system first performs layered preprocessing on the input multi-modal signals. For ECG signal characteristics, a 0.5-40 Hz bandpass filter combination is applied to remove baseline drift while suppressing high-frequency noise, and the main energy band of the QRS complex is completely retained. For motion signals, perform DC removal to eliminate static offset and highlight effective motion features. For respiratory impedance signals, use envelope detection algorithm to extract clear respiratory fluctuation curve and remove high-frequency interference components.

[0054] Based on preprocessing, the system calculates three types of key feature parameters in parallel, constructing a comprehensive feature vector containing 15 dimensions. Time domain feature analysis includes signal amplitude range, variance characteristics and zero crossing rate, reflecting the macro fluctuation characteristics of the signal; frequency domain features calculate the baseline respiratory component energy ratio of 0-5 Hz frequency band, QRS complex component energy ratio of 5-15 Hz frequency band and muscle noise component energy distribution of 15-40 Hz frequency band through fast Fourier transform, revealing the frequency spectrum characteristics of the signal; correlation features calculate the maximum value of the cross-correlation function of the main and auxiliary ECG signals, the coherence coefficient of ECG and motion vector, and the coherence coefficient of ECG and respiratory signal, to deeply analyze the internal correlation degree between signals.

[0055] After feature extraction, the system inputs the feature vector into the optimized lightweight classification model, and outputs the five-class noise recognition results. Class A represents high-quality signals, characterized by all parameters within the normal physiological range, and no additional processing is required. Class B identifies motion artifacts, characterized by high coherence between ECG and motion data and non-physiological distortion. Class C indicates periodic artifacts, which exhibit significant isolated narrowband peaks in the frequency domain. Class D corresponds to respiratory baseline drift, characterized by high coherence between ECG and respiratory signals and prominent low-frequency energy. Class E represents poor contact or signal loss, characterized by signal amplitude exceeding the limit or being close to zero. The system also calculates the signal quality index (SQI) in the range of 0 to 1. When the classification confidence is above 0.8, it is determined to be valid classification, otherwise it is marked as pending confirmation to ensure the reliability of the evaluation results.

[0056] Step three: dynamic traceability filtering processing Based on accurate noise diagnosis results, the system enters the dynamic filtering processing stage. The intelligent routing controller executes precise algorithm scheduling according to the noise type and confidence score. When the confidence is below 0.8, the system adopts a conservative strategy, directly outputs the original signal and marks the classification as doubtful, to avoid signal distortion caused by false processing. When the confidence meets the standard, the corresponding special processing engine is activated according to the specific noise type, forming a precise processing pipeline for "one type of noise, one set of solutions".

[0057] For Class B motion artifacts, the system starts the blind source separation engine. This engine inputs the main and auxiliary ECG signals and processes them through fast independent component analysis algorithm. First, the observation signal is centered and whitened for preprocessing to eliminate the correlation between signals. Then, the maximum Gaussian iteration operation is used to calculate the demixing matrix to separate the statistically independent signal components. Finally, based on the unique R-wave rhythm periodicity of ECG signals, the cardiac source component is automatically identified and selected, effectively suppressing motion interference and outputting pure ECG signals.

[0058] For Class C periodic artifacts, the adaptive resonance canceller starts working. This module accurately identifies the noise frequency through inertial measurement unit data or spectral analysis, configures an adaptive notch filter, locks the center frequency within ±0.1 Hz of the interference frequency, and sets the bandwidth between 0.5-2 Hz adjustable. The system tracks the noise frequency changes in real time and dynamically adjusts the filter parameters, effectively filtering out periodic interference while maximizing the preservation of useful spectral components of QRS complexes.

[0059] When dealing with Class D respiratory baseline drift, the system activates an impedance-based baseline corrector. This corrector takes the synchronously acquired respiratory impedance signal as a direct reference for baseline changes, establishes a mapping relationship between the respiratory signal amplitude and the ECG baseline position through a linear regression model, accurately estimates the baseline drift curve, and then subtracts the estimated drift component from the original ECG signal to achieve effective baseline correction and output stable ECG signals.

[0060] When encountering Class E poor contact or signal drop, the data safety and marking unit activates the protection mechanism. Instead of signal reconstruction or interpolation processing, this unit directly performs integrity assessment on the problematic data segment, marks the unrecoverable degraded signal as "no clinical diagnostic value", and inserts detailed marking frames containing the start time, end time, and specific failure reason into the data stream, ensuring the accuracy and reliability of subsequent data analysis.

[0061] Step four: intelligent output and adaptive resource management In the final output stage, the system generates two types of strictly time-aligned output data streams. The high-fidelity ECG signal stream outputs the purified ECG signal with 16-bit precision and a sampling rate of 250Hz, with the signal frequency band strictly controlled within the 0.5-40Hz physiological relevant range. The structured metadata stream provides complete auxiliary information, including millisecond-level precision timestamps, signal quality index SQI values, noise type classification labels, used processing algorithm identifiers, and detailed marking information for invalid data segments, providing comprehensive references for subsequent deep analysis and clinical diagnosis. The system also implements intelligent power management strategies, dynamically adjusting system energy consumption through two modes. The low-power mode automatically activates when Class A high-quality signals are detected for 3 seconds in a row, reducing the sampling rate of the inertial measurement unit and auxiliary ECG channels to 125Hz, or intermittently shutting down high-energy consumption computing units such as the blind source separation engine, achieving 30-40% power saving. The on-demand activation mode triggers immediately when the diagnostic module detects Class B to Class D noise characteristics, completing state switching within 10 milliseconds and starting the corresponding sensors and processing engines at full power, ensuring real-time processing performance is not affected.

[0062] A self-sensing feedback control mechanism based on long-term operational data continuously optimizes system performance. This mechanism intelligently identifies specific problems such as signal saturation caused by sweating and amplitude fluctuations caused by poor contact by analyzing accumulated signal quality indices and noise pattern statistics. When signal saturation reaches 90% of full scale, the system automatically instructs the analog front-end to appropriately reduce gain; when abnormal contact impedance is detected, the detection threshold and excitation current parameters are adjusted accordingly; and the decision threshold of the classifier is dynamically optimized based on long-term noise type distribution characteristics. This adaptive feedback adjustment enables the system to intelligently respond to changes in the usage environment and fluctuations in the user's physiological state, maintaining optimal operating conditions during long-term continuous monitoring. Finally, it should be noted that the above embodiments can be implemented entirely or partially through software, hardware, firmware, or any other combination. When implemented in software, the above embodiments can be implemented entirely or partially in the form of a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. However, any obvious changes or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A high-precision continuous sleep electrocardiogram monitoring system, characterized in that: It includes a multimodal sensing and high-precision synchronous acquisition module, a signal quality multi-dimensional source tracing and diagnosis module, a dynamic source tracing and filtering module, and an intelligent output and adaptive resource management module connected in sequence. Each module establishes a bidirectional communication connection through a data bus and a control bus. The downstream module adjusts the parameters of the upstream module in real time through a feedback path to form a closed-loop control system. The multimodal sensing and high-precision synchronous acquisition module acquires electrocardiogram, motion and respiratory physiological signals and integrates multi-source data streams through a precise time synchronization mechanism; The multi-dimensional signal quality source tracing and diagnosis module preprocesses the input signal, extracts features, classifies and identifies noise, and outputs a signal quality index and noise type label. The dynamic source tracing filtering module calls the corresponding filtering algorithm to perform targeted signal purification processing based on the noise classification results; The intelligent output and adaptive resource management module outputs high-quality ECG signals and metadata, and realizes dynamic optimization management of system power consumption and parameters.

2. The high-precision continuous sleep ECG monitoring system according to claim 1, characterized in that: The multimodal sensing and high-precision synchronous acquisition module includes a main ECG sensing unit, an auxiliary ECG sensing unit, a motion and physiological sensing unit, and a synchronous acquisition controller. The main ECG sensing unit uses flexible dry electrode material and is fixed to the second intercostal space on the right sternal border and the fifth intercostal space on the left midclavicular line according to the simulated modified lead II standard layout. The auxiliary ECG sensing unit consists of a pair of independent flexible dry electrodes, with a spatial distance of greater than or equal to 2 cm from the main electrode pair, and the electrical vector directions forming an angle of 45 to 90 degrees. The motion and physiological sensing unit integrates a six-axis inertial measurement unit and a four-electrode bioimpedance sensor. The synchronous acquisition controller uses a high-performance, low-power microcontroller and integrates a high-stability crystal oscillator with an accuracy of less than or equal to 1 microsecond, synchronously acquiring signals through parallel multi-channel ADCs and adding timestamps.

3. The high-precision continuous sleep ECG monitoring system according to claim 1, characterized in that: The multi-dimensional signal quality tracing and diagnosis module includes a preprocessing and parallel feature extraction unit, a noise type classifier, and a comprehensive quality assessment unit. The preprocessing and parallel feature extraction unit filters and extracts features from ECG, motion, and respiratory signals, calculating time-domain, frequency-domain, and correlation features. The noise type classifier uses a lightweight machine learning model, taking a 15-dimensional feature vector as input and outputting five noise labels: Class A (high-quality signal), Class B (motion artifacts), Class C (periodic artifacts), Class D (respiratory baseline drift), and Class E (poor contact or signal dropout). The comprehensive quality assessment unit calculates the Signal Quality Index (SQI) in the range of 0 to 1 and determines its validity based on classification confidence.

4. The high-precision continuous sleep ECG monitoring system according to claim 1, characterized in that: The dynamic source tracing filtering processing module includes an intelligent algorithm routing controller, a blind source separation engine, an adaptive resonant canceller, an impedance-based baseline corrector, and a data security and tagging unit; the intelligent algorithm routing controller dynamically schedules the processing engine according to the noise type and confidence level. The blind source separation engine processes Class B noise and performs fast independent component analysis; the adaptive resonant canceller processes Class C noise and performs adaptive notch filtering; the impedance-based baseline corrector processes Class D noise and performs regression correction algorithm; and the data security and marking unit processes Class E noise and marks invalid data segments.

5. The high-precision continuous sleep ECG monitoring system according to claim 1, characterized in that: The intelligent output and adaptive resource management module includes a purified signal and metadata output unit, a dynamic power consumption manager, and a self-sensing feedback control unit. The purified signal and metadata output unit outputs high-fidelity electrical signals and structured metadata, including timestamps, signal quality index (SQI), noise tags, and processing algorithm identifiers. The dynamic power consumption manager activates a low-power mode when continuously detecting high-quality signals and triggers an on-demand activation mode when noise characteristics are detected. The self-sensing feedback control unit sends parameter adjustment commands to the front-end acquisition module based on historical SQI data and noise pattern statistics.

6. The high-precision sleep ECG continuous monitoring system according to claim 5, characterized in that: The low-power mode of the dynamic power manager is achieved by reducing the sampling rate of the inertial measurement unit, turning off the auxiliary ECG acquisition channel, and intermittently disabling the blind source separation engine computing unit.

7. A high-precision continuous sleep electrocardiogram (ECG) monitoring method, wherein the method is applied to the high-precision continuous sleep ECG monitoring system according to any one of claims 1 to 6, characterized in that: The method includes the following steps: The S1 multimodal signal high-precision synchronous acquisition acquires ECG, motion, and respiratory signals through a multimodal sensor network, generates a unified sampling clock using a high-precision clock source, synchronously triggers all sensor channels, and adds timestamps to the data; S2 signal quality multi-dimensional source tracing diagnosis performs preprocessing, feature extraction, and noise classification on multimodal signals, and calculates the signal quality index (SQI) and classification confidence. S3 dynamic source tracing filtering process performs signal purification by scheduling the filtering algorithm according to the noise type and confidence level. The S4 intelligent output and adaptive resource management output purified ECG signals and metadata, and dynamically manages system power consumption and parameters.

8. The high-precision sleep electrocardiogram continuous monitoring method according to claim 7, characterized in that: The high-precision synchronous acquisition steps for multimodal signals include: the primary and auxiliary ECG electrodes are spatially differentiated with a spacing of 2 cm or more and the electrical vector directions form an angle of 45 to 90 degrees; and all signal channels are synchronously acquired at a sampling rate of 250 Hz through a synchronous acquisition controller.

9. The high-precision sleep electrocardiogram continuous monitoring method according to claim 7, characterized in that: The multi-dimensional signal quality tracing and diagnostic steps include: performing bandpass filtering on ECG signals from 0.5 to 40 Hz, removing DC from motion signals, and performing envelope detection on respiratory signals; extracting time-domain, frequency-domain, and correlation features to construct a 15-dimensional feature vector; using a lightweight classification model to output five types of noise labels, and determining the classification to be valid when the confidence level is greater than or equal to 0.

8.

10. The high-precision continuous sleep electrocardiogram monitoring method according to claim 7, characterized in that: The dynamic source tracing filtering process includes: when the confidence level is greater than or equal to 0.8, activating the corresponding processing engine according to the noise type; enabling the blind source separation engine for Class B noise; enabling the adaptive resonant canceller for Class C noise; enabling the impedance-based baseline corrector for Class D noise; and enabling the data security and tagging unit for Class E noise.

Images