Method and system for synchronously measuring multiple electrophysiological signals based on adaptive algorithm

Abstract

The invention relates to the technical field of signal processing, and provides a multi-electrophysiological signal synchronous measurement method and system based on an adaptive algorithm, and the method comprises the steps: obtaining digital signals of a plurality of electrophysiological signals collected by a multi-channel sensor; constructing a signal synchronous calibration model based on an adaptive algorithm, and inputting the digital signals of the various electrophysiological signals into the signal synchronous calibration model; a signal alignment module is adopted to carry out time sequence alignment on the multiple digital signals, and preliminary synchronization signals corresponding to the electrophysiological signals are obtained; a noise suppression module is adopted to carry out noise suppression on all the preliminary synchronizing signals, and noise-reduced signals corresponding to all the electrophysiological signals are obtained; and performing time-frequency domain error correction on each denoised signal by adopting an error correction module to obtain a high-precision synchronizing signal corresponding to each electrophysiological signal. According to the embodiment of the invention, through the cooperative processing model constructed by the adaptive algorithm, the accuracy and consistency of signal synchronization are improved, and the reliability and applicability of the electrophysiological signal data are enhanced.

Description

Technical Field

[0001] This disclosure relates to the field of signal processing technology, and more specifically, to a method and system for synchronous measurement of multiple electrophysiological signals based on an adaptive algorithm. Background Technology

[0002] In the field of biomedical signal measurement, the simultaneous acquisition of multiple electrophysiological signals, such as electrocardiograms (ECG), electroencephalograms (EEG), and electromyograms (EMG), is crucial for revealing the coupling mechanisms of the nervous, muscular, and cardiac systems. Currently, related technologies can achieve simultaneous acquisition of multi-channel signals, for example, through multimodal hardware acquisition systems, high-density electrode arrays, and dedicated equipment for specific environments (such as MRI compatibility).

[0003] However, although these systems can acquire multiple types of signals, their synchronization mechanisms mostly rely on hardware clocks or fixed algorithms. When dealing with multi-device collaboration, channel drift, and complex environmental noise, their synchronization accuracy and adaptability are often insufficient, making it difficult to meet the stringent requirements of high-precision medical research for signal timing consistency. Summary of the Invention

[0004] This disclosure provides at least one method and system for synchronizing multiple electrophysiological signals based on an adaptive algorithm. The collaborative processing model constructed through the adaptive algorithm improves the accuracy and consistency of signal synchronization, reduces timing errors and noise effects in multi-channel acquisition, and enhances the reliability and applicability of electrophysiological signal data.

[0005] This disclosure provides a method for synchronous measurement of multiple electrophysiological signals based on an adaptive algorithm, including: Acquire digital signals of various electrophysiological signals collected by multi-channel sensors; A signal synchronization calibration model is constructed based on an adaptive algorithm, and the digital signals of the various electrophysiological signals are input into the signal synchronization calibration model; wherein, the signal synchronization calibration model includes a signal alignment module, a noise suppression module, and an error correction module; The signal alignment module is used to perform time-series alignment of the digital signals of the various electrophysiological signals to obtain preliminary synchronization signals corresponding to each electrophysiological signal; The noise suppression module is used to suppress noise in each of the preliminary synchronization signals to obtain the noise-reduced signals corresponding to each electrophysiological signal. The error correction module is used to perform time-frequency domain error correction on each of the noise-reduced signals to obtain a high-precision synchronization signal corresponding to each electrophysiological signal.

[0006] This disclosure provides a system for synchronously measuring multiple electrophysiological signals based on an adaptive algorithm, including: A multi-channel signal acquisition module is configured to acquire multiple electrophysiological signals through a multi-channel sensor and output digital signals of the multiple electrophysiological signals; wherein each channel sensor is used to acquire a digital signal of one electrophysiological signal. The signal synchronization calibration processor has an internal signal synchronization calibration model built based on an adaptive algorithm. The signal synchronization calibration processor is communicatively connected to the multi-channel signal acquisition module and is used to receive digital signals of the various electrophysiological signals and input the digital signals into the signal synchronization calibration model for processing. The signal synchronization calibration model includes: The signal alignment unit is configured to perform time-series alignment on the digital signals of multiple input electrophysiological signals and output a preliminary synchronization signal corresponding to each electrophysiological signal. The noise suppression unit is communicatively connected to the signal alignment module and is configured to receive each of the preliminary synchronization signals and perform noise suppression processing, and output a noise-reduced signal corresponding to each electrophysiological signal. The error correction unit is communicatively connected to the noise suppression module and is configured to receive each of the noise-reduced signals and perform time-frequency domain error correction processing, and finally output a high-precision synchronization signal corresponding to each electrophysiological signal.

[0007] This disclosure provides a device for synchronizing the measurement of multiple electrophysiological signals based on an adaptive algorithm, including: The signal acquisition module is used to acquire digital signals of various electrophysiological signals collected by a multi-channel sensor. The model processing module is used to construct a signal synchronization calibration model based on an adaptive algorithm and input the digital signals of the various electrophysiological signals into the signal synchronization calibration model; wherein, the signal synchronization calibration model includes a signal alignment module, a noise suppression module, and an error correction module; Specifically, the model processing module is used for: The signal alignment module is used to perform time-series alignment of the digital signals of the various electrophysiological signals to obtain preliminary synchronization signals corresponding to each electrophysiological signal; The noise suppression module is used to suppress noise in each of the preliminary synchronization signals to obtain the noise-reduced signals corresponding to each electrophysiological signal. The error correction module is used to perform time-frequency domain error correction on each of the noise-reduced signals to obtain a high-precision synchronization signal corresponding to each electrophysiological signal.

[0008] This disclosure provides a computer device, including a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the computer device is running, the processor communicates with the memory via the bus. When the machine-readable instructions are executed by the processor, they perform a method for synchronous measurement of multiple electrophysiological signals based on an adaptive algorithm as described in any of the above possible embodiments.

[0009] This disclosure provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements a method for simultaneous measurement of multiple electrophysiological signals based on an adaptive algorithm as described in any of the possible embodiments above.

[0010] The method and system for synchronous measurement of multiple electrophysiological signals based on adaptive algorithms provided in this disclosure utilize a synchronization calibration model constructed based on an adaptive algorithm signal alignment module, a noise suppression module, and an error correction module to sequentially perform time-series alignment, noise suppression, and time-frequency domain error correction on multi-channel electrophysiological digital signals. Thus, the collaborative processing model constructed by the adaptive algorithm in this disclosure improves the accuracy and consistency of signal synchronization, reduces the impact of timing errors and noise in multi-channel acquisition, and enhances the reliability and applicability of electrophysiological signal data.

[0011] To make the above-mentioned 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

[0012] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the accompanying drawings referenced in the embodiments will be briefly described below. These drawings are incorporated in and constitute a part of this specification. They illustrate embodiments conforming to this disclosure and, together with the specification, serve to explain the technical solutions of this disclosure. It should be understood that the following drawings only show some embodiments of this disclosure and should not be considered as limiting the scope. Those skilled in the art can obtain other related drawings based on these drawings without creative effort.

[0013] Figure 1 A flowchart of a method for simultaneous measurement of multiple electrophysiological signals based on an adaptive algorithm, provided in an embodiment of this disclosure, is shown. Figure 2 A flowchart of a signal synchronization calibration model construction method provided by an embodiment of this disclosure is shown; Figure 3 A flowchart illustrating a timing alignment method for signals provided in an embodiment of this disclosure is shown; Figure 4 A flowchart of a noise suppression method for a signal provided by an embodiment of this disclosure is shown; Figure 5 A flowchart of a time-frequency domain error correction determination method for a signal provided by an embodiment of this disclosure is shown; Figure 6 A flowchart of a data parsing and interactive control method provided in an embodiment of this disclosure is shown; Figure 7 A schematic diagram of the structure of a multi-electrophysiological signal synchronous measurement system based on an adaptive algorithm provided in an embodiment of this disclosure is shown. Figure 8 A schematic diagram of the structure of a multi-electrophysiological signal synchronization measurement device based on an adaptive algorithm provided in an embodiment of this disclosure is shown. Figure 9 A schematic diagram of the structure of a computer device provided in an embodiment of this disclosure is shown. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. The components of the embodiments of this disclosure described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this disclosure provided in the accompanying drawings is not intended to limit the scope of the claimed disclosure, but merely represents selected embodiments of this disclosure. All other embodiments obtained by those skilled in the art based on the embodiments of this disclosure without inventive effort are within the scope of protection of this disclosure.

[0015] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0016] In this document, the term "and / or" merely describes a relationship, indicating that three relationships can exist. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. Furthermore, the term "at least one" in this document means any combination of at least two of any one or more elements. For example, including at least one of A, B, and C can mean including any one or more elements selected from the set consisting of A, B, and C.

[0017] In the field of biomedical signal processing and measurement technology, the simultaneous acquisition and analysis of multi-channel, multi-modal electrophysiological signals (such as electrocardiogram (ECG), electroencephalogram (EEG), and electromyography (EMG)) has become a key technology for neuroscience research, clinical diagnosis, and health monitoring. With the development of sensing technology and computing power, researchers are now able to simultaneously acquire multiple types of signals from different physiological sources, thereby enabling comprehensive observation and analysis of complex physiological processes.

[0018] However, due to factors such as diverse signal sources, heterogeneous acquisition equipment, and complex environmental noise, achieving high-precision, real-time synchronous measurement of multiple electrophysiological signals remains a challenging technical problem. Especially in application scenarios requiring simultaneous analysis of multiple system interactions such as heart-brain coupling and neuromuscular coordination, the temporal alignment accuracy between signals directly affects the effectiveness of subsequent feature extraction and pattern recognition.

[0019] Research has revealed several technological solutions currently available for the simultaneous acquisition and processing of multi-channel electrophysiological signals. For example, multi-channel / multi-modal acquisition systems (such as MADQ) can simultaneously acquire multiple types of physiological signals; multi-modal fusion analysis techniques attempt to integrate EEG, EMG, ECG, and other signals at the algorithmic level; high-density non-invasive recording methods further improve the spatial resolution of signal acquisition; novel sensing technologies such as microneedle dry electrodes have also made progress in improving signal quality and wearability; and, in addition, noise suppression has been specifically studied for systems compatible with special environments (such as MRI).

[0020] Nevertheless, although the relevant technologies can acquire various types of signals, their synchronization mechanisms mostly rely on hardware clocks or fixed algorithms. When dealing with multi-device collaboration, channel drift, and complex environmental noise, their synchronization accuracy and adaptability are often insufficient, making it difficult to meet the stringent requirements of high-precision medical research for signal timing consistency.

[0021] Based on the above research, this disclosure provides a method and system for synchronous measurement of multiple electrophysiological signals based on an adaptive algorithm. The synchronous calibration model, constructed by a signal alignment module, a noise suppression module, and an error correction module based on the adaptive algorithm, sequentially performs time alignment, noise suppression, and time-frequency domain error correction processing on the multi-channel electrophysiological digital signals.

[0022] In this embodiment of the disclosure, the collaborative processing model constructed by the adaptive algorithm improves the accuracy and consistency of signal synchronization, reduces timing errors and noise effects in multi-channel acquisition, and enhances the reliability and applicability of electrophysiological signal data.

[0023] To facilitate understanding of this embodiment, the execution subject of the method for simultaneous measurement of multiple electrophysiological signals based on adaptive algorithms provided in this disclosure will first be described in detail. The execution subject of the method for simultaneous measurement of multiple electrophysiological signals based on adaptive algorithms provided in this disclosure is a computer device. This computer device can be a terminal device or a server. The terminal device can also be a mobile device, a user terminal, a terminal, a handheld device, a computing device, an in-vehicle device, a wearable device, etc. The server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud storage, big data, and artificial intelligence platforms. Optionally, this method can also be applied to an implementation environment composed of computer devices and servers.

[0024] The following detailed description, with reference to the accompanying drawings, illustrates the methods for synchronous measurement of various electrophysiological signals based on adaptive algorithms provided in the embodiments of this application. See also... Figure 1 The diagram shown is a flowchart of a method for synchronously measuring multiple electrophysiological signals based on an adaptive algorithm, according to an embodiment of this disclosure. The method includes the following steps S101 to S105: S101 acquires digital signals of various electrophysiological signals collected by a multi-channel sensor.

[0025] Understandably, a multi-channel sensor is a sensor device capable of simultaneously acquiring multiple different types of signals. It possesses multiple independent signal acquisition channels, each designed to acquire specific electrophysiological signals. Here, the sensor can include EEG electrodes, ECG leads, EMG sensors, etc., to simultaneously acquire analog electrical signals generated by different physiological activities of an organism. Electrophysiological signals are electrical signals generated by an organism during its life activities. Common electrophysiological signals include ECG signals, EEG signals, and EMG signals. For example, in medical monitoring scenarios, using a multi-channel sensor can simultaneously acquire ECG signals generated by the heartbeat, ECG signals generated by brain neural activity, and EMG signals generated by muscle contraction. These acquired analog signals can be digitally sampled via an analog-to-digital converter, converted into discrete time-series data that can be processed by a computer, thereby obtaining the digital signal stream corresponding to each electrophysiological signal.

[0026] In some possible embodiments, when acquiring multi-channel signals, the sampling frequency of each channel can be set to [value]. The sample value of each signal is ,in Indicates the channel number. Representing a point in time, each signal can be converted into a discrete signal after acquisition using the sampling theorem: .

[0027] In some other embodiments, after step S101, the process may further include: performing self-calibration and channel health monitoring on each channel sensor to obtain the gain parameter, phase offset parameter, and health status index of each channel, and adaptively weighting the subsequent synchronization process based on the parameters and indexes.

[0028] Specifically, regarding the first The electrophysiological signals from each channel can be denoted as the acquired discrete time series. By inputting a preset calibration signal to the multi-channel sensor Alternatively, a calibration sequence can be constructed using physiological event characteristics (such as the R wave in ECG), and a linear calibration model can be established in the time domain as follows: ; in, For channel gain, For channel time delay, This represents the noise term. Parameter estimates can be obtained through least squares or maximum likelihood estimation. ; After calibration, the original channel signals are normalized and aligned preprocessed: ; In addition, a health scoring function is constructed for each channel to monitor its health status. This score takes into account the signal-to-noise ratio. Electrode contact resistance Packet loss rate Indicators such as: ; in These are the weighting coefficients. This represents the normalization operator.

[0029] In subsequent signal alignment and noise suppression modules, it can be based on Dynamically adjust the participation weight of each channel. : ; in This represents the temperature coefficient. When the health score of a certain channel drops significantly, the weight of that channel in the synchronization strategy optimization process is automatically reduced to avoid the low-quality channel adversely affecting the overall synchronization accuracy.

[0030] In some possible embodiments, considering the energy-constrained characteristics of wearable devices, the method for synchronizing multiple electrophysiological signals based on adaptive algorithms further includes adaptively configuring the sampling frequency and data upload frequency of each channel under synchronization accuracy constraints, so as to reduce energy consumption while ensuring preset synchronization performance.

[0031] Specifically, record the first The sampling frequency of each channel is The energy consumption per unit time can be approximated as: ; in, For the number of channels, This represents the power consumption coefficient that increases linearly with the sampling frequency. This represents static power consumption. Synchronization error can be modeled using an estimate from the output of the deep Q-network in the signal alignment module. Let the overall synchronization error metric be... ,in The adaptive sampling strategy can then be determined through the following constrained optimization problem: ; in, To preset the upper limit of synchronization error, This refers to the sampling frequency boundary under hardware and signal bandwidth constraints.

[0032] In practical implementation, the above constrained optimization problem can be transformed into an unconstrained objective function with a penalty term: ; in, This is the penalty coefficient. It is updated online through a method based on gradient descent or reinforcement learning policy search. This enables adaptive adjustment of the sampling frequency. The synchronization calibration model forms a closed loop between the acquisition and processing ends: when a large synchronization error margin is detected, the sampling rate of some channels can be appropriately reduced; when the error approaches the threshold... At the same time, the sampling rate of key channels is automatically increased to ensure synchronization performance.

[0033] In some possible embodiments, when constructing the reinforcement learning state space and optimization objective of the signal alignment module, the physiological synergy between different electrophysiological signal types is explicitly modeled as coherence features, thereby driving the adaptive update of the synchronization strategy. Specifically, for various electrophysiological signals such as EEG, ECG, and EMG, the coherence between different signal pairs in the frequency domain is estimated based on multi-channel cross-spectral density, and a multi-dimensional coherence feature vector is constructed as part of the reinforcement learning state input. This allows the synchronization strategy to focus not only on the magnitude of the timing error but also on the physiological consistency of different signals in key frequency bands after synchronization.

[0034] For the With the Electrophysiological signals of each channel First, calculate its discrete Fourier transform: ; Obtain the power spectral density , cross-spectral density : ; Further calculations at frequency Coherence indices at the location: ; frequency band of interest We obtain the normalized cross-signal coherence features by weighted integration of the coherence within the signal: .

[0035] For all channel pairs By performing permutations and combinations, construct a multidimensional coherence vector: .

[0036] In the signal alignment module of this invention, the aforementioned coherence characteristics are combined with the multi-channel timing error vector. Historical synchronous action coding Concatenate the vectors to form the reinforcement learning state vector: ; The immediate reward function in reinforcement learning not only penalizes temporal errors but also penalizes a decrease in consistency across signal types coherence, and is constructed as follows: ; in, The weighting parameter is used here. This invention guides the deep Q-network to maximize frequency domain consistency across signal types while reducing temporal differences when searching for synchronization action sequences, thus making the final synchronization result more physiologically reasonable.

[0037] S102, construct a signal synchronization calibration model based on an adaptive algorithm, and input the digital signals of the various electrophysiological signals into the signal synchronization calibration model.

[0038] Here, the adaptive algorithm is an algorithm that can automatically adjust its own parameters and structure according to the characteristics and changes of the input data. It can be flexibly optimized and adjusted according to different electrophysiological signal characteristics to achieve better signal processing results.

[0039] In this disclosure, a signal synchronization calibration model is constructed based on an adaptive algorithm, and the digital signals of various electrophysiological signals obtained in step S101 are used as inputs into this model. The signal synchronization calibration model constructed here is an integrated software processing framework, whose core comprises three functionally coordinated modules: a signal alignment module, a noise suppression module, and an error correction module. The signal alignment module performs time alignment processing on various electrophysiological signals acquired at different times, ensuring they have a consistent starting point and time interval on the time axis; the noise suppression module reduces various noise interferences present in the signal, improving signal purity; and the error correction module corrects errors that may occur during signal acquisition and transmission, improving signal accuracy.

[0040] Specifically, refer to Figure 2 As shown, the signal synchronization calibration model can be constructed through the following steps S201~S204: S201, The signal alignment module is constructed based on a deep reinforcement learning algorithm, and the signal alignment module aims to minimize the timing error between multi-channel signals.

[0041] Here, the signal alignment module is constructed using a deep reinforcement learning algorithm. Deep reinforcement learning is a machine learning paradigm where the agent learns the optimal decision-making strategy by interacting with the environment and based on reward signals. During construction, the synchronization problem of multi-channel signals can be modeled as a sequential decision-making process: the agent observes the current temporal offset between signals and then outputs an adjustment action (such as shifting the time of a specific channel signal forward or backward); the environment then provides a reward signal, which is typically designed to be negatively correlated with the synchronization error—the smaller the error, the higher the reward. Through continuous iterative training, the agent can learn an optimal alignment strategy that minimizes the temporal error between multi-channel signals. For example, when acquiring multi-channel signals such as electrocardiograms (ECGs) and electroencephalograms (EEGs), due to differences in the response time of the acquisition devices and signal transmission delays, signals from different channels may have temporal deviations. The signal alignment module constructed using the deep reinforcement learning algorithm can automatically analyze signal characteristics, adjust the time base of each channel signal, and achieve alignment of multi-channel signals on the time axis.

[0042] S202, The noise suppression module is constructed based on generative adversarial networks and adaptive filtering algorithms, with the goal of dynamic noise modeling and adaptive noise suppression.

[0043] As is understandable, a Generative Adversarial Network (GAN) consists of a generator and a discriminator. The generator is responsible for generating noise samples that resemble real noise, while the discriminator distinguishes between the generated noise samples and real noise samples. Through adversarial training between the two, the generator can continuously optimize the generated noise samples, making them closer to the distribution of real noise, thereby achieving accurate modeling of dynamic noise. Adaptive filtering algorithms automatically adjust the filter parameters based on the statistical characteristics of the input signal and noise to effectively suppress noise.

[0044] Specifically, in the noise suppression module, a generator can learn to separate or synthesize representative noise samples from real signals, and a discriminator can distinguish noise in the real signals from noise generated by the generator. This allows the noise model generated by the generative adversarial network to provide more accurate prior noise information for the adaptive filtering algorithm. The adaptive filtering algorithm can then adjust its filter coefficients in real time and automatically based on the noise characteristic estimates provided by the generative adversarial network, thereby accurately subtracting the estimated noise components from the input signal and achieving adaptive noise suppression. For example, when acquiring EEG signals, they are subject to interference from various dynamic noises such as electromyographic noise and electrooculographic noise. The noise suppression module constructed using the generative adversarial network and adaptive filtering algorithm can dynamically adjust the noise model and filtering parameters in real time according to changes in the EEG signal and noise, effectively removing noise from the signal and improving the signal-to-noise ratio of the EEG signal.

[0045] S203, The error correction module is constructed based on joint time-frequency domain analysis and adaptive error correction algorithm, and the error correction module aims at time-frequency domain error correction.

[0046] Here, joint time-frequency domain analysis refers to a tool that analyzes signals simultaneously from both time and frequency dimensions. It combines the advantages of both time-domain and frequency-domain analysis, enabling the simultaneous acquisition of signal information in both time and frequency dimensions (such as timing drift and phase error characteristics). By performing time-frequency domain transformations on the signal, such as short-time Fourier transforms and wavelet transforms, the signal can be decomposed into sub-signals with different time and frequency components, thus providing a more comprehensive understanding of the signal's characteristics and error distribution. Adaptive error correction algorithms automatically adjust correction parameters based on the results of time-frequency domain analysis to correct time-frequency domain errors in the signal. For example, a Kalman filter can be used as an adaptive estimation algorithm to optimally estimate and predict the signal's state (including time deviation); or a gradient descent-based adaptive algorithm can be used to iteratively adjust the correction parameters.

[0047] Specifically, in the error correction module, joint time-frequency domain analysis provides error information for the adaptive error correction algorithm, which can then target errors in different time and frequency components. For example, in electroencephalography (EEG) signals, errors in the time-frequency domain occur due to factors such as electromagnetic interference from the surrounding environment and interference between different neurons within the brain during signal propagation. The error correction module, constructed using joint time-frequency domain analysis and the adaptive error correction algorithm, can accurately identify and locate these errors and, through adaptive adjustment of correction parameters, effectively correct time-frequency domain errors in EEG signals, thereby improving signal accuracy and reliability.

[0048] S204, integrate the signal alignment module, the noise suppression module and the error correction module to construct the signal synchronization calibration model.

[0049] Understandably, after completing the construction of the three modules (signal alignment module, noise suppression module, and error correction module), they can be integrated to form a complete signal synchronization calibration model. Here, the signal processed by the signal alignment module serves as the input to the noise suppression module, and the signal processed by the noise suppression module serves as the input to the error correction module. The modules cooperate with each other to complete the task of synchronously calibrating the signal.

[0050] S103, the signal alignment module is used to perform time-series alignment of the digital signals of the various electrophysiological signals to obtain preliminary synchronization signals corresponding to each electrophysiological signal.

[0051] It is understandable that signals from different sensors or channels may have initial clock asynchrony, slight differences in sampling rate, or transmission delay, causing their correspondence on the same time axis to shift. Therefore, after the digital signals of multiple electrophysiological signals are input into the signal synchronization calibration model, they first enter the signal alignment module. The signal alignment module is used to perform time alignment of the digital signals of each electrophysiological signal to compensate for these shifts.

[0052] For example, refer to Figure 3 As shown, the specific process of timing alignment of digital signals of various electrophysiological signals using a signal alignment module may include the following steps S301~S308: S301, for each type of electrophysiological signal, determine the timing mapping relationship between the digital signal and the digital signals corresponding to other electrophysiological signals; and based on the timing mapping relationship between the digital signal and the digital signals corresponding to other electrophysiological signals, determine the timing error between the digital signal and the digital signals corresponding to other electrophysiological signals.

[0053] Specifically, the temporal mapping relationship describes the correspondence between different signals at corresponding points on the time axis. This relationship can be established by calculating the cross-correlation function or by matching based on feature points (such as the R-wave peak of an electrocardiogram signal). Based on this temporal mapping relationship, the temporal error of the digital signal relative to each other signal can be quantified. The temporal error is the deviation between different signals in time, usually manifested as a time offset, such as a feature point in an electrocardiogram signal being preceded or delayed by a certain number of milliseconds compared to the corresponding feature point in an electroencephalogram signal.

[0054] S302, input the timing error of the digital signal corresponding to various electrophysiological signals to the signal alignment module.

[0055] Here, the timing mapping relationship of all electrophysiological signals corresponding to digital signals and the calculated timing error are used as initial state information and input to the signal alignment module.

[0056] S303, based on the timing error, the deep Q network in the signal alignment module outputs timing correction actions for each electrophysiological signal; wherein, the deep Q network optimizes the synchronization strategy through a reward function, and the reward function is constructed based on the objective of minimizing the timing error between multi-channel signals.

[0057] Understandably, a Deep Q-Network (DQN) is a reinforcement learning model that combines deep neural networks with Q-Learning. It outputs corresponding actions (temporal correction actions) based on the input state (in this case, information such as timing errors). It can evaluate the long-term benefit of taking an action in a specific state by learning an action value function. Internally, the DQN optimizes the synchronization strategy through a reward function constructed based on the objective of minimizing the timing error between multi-channel signals. The reward value can be designed as the negative of the root mean square of the timing error, thus incentivizing the network to select actions that continuously reduce the error. During model operation, when the timing correction action output by the DQN reduces the timing error between multi-channel signals, a reward is given; otherwise, a penalty is imposed. Through continuous learning and optimization, the DQN can output more effective timing correction actions.

[0058] Specifically, when determining the timing correction action for each electrophysiological signal based on a deep Q-network in deep reinforcement learning, the timing error of each channel signal and its changing trend can be encoded into a digital state vector (e.g., containing the current delay in milliseconds for each channel such as ECG, EEG, and EMG). Next, this state vector is input into the trained deep Q-network. The network's output layer calculates a Q-value for each preset discrete adjustment action (e.g., "shifting the EEG signal forward by 0.1ms"), representing the long-term expected benefit of performing this action. Then, based on the output Q-value, a greedy strategy can be used to directly select the action with the highest Q-value, or an ε-greedy strategy can be used for exploratory selection, ultimately outputting a specific timing correction action instruction for each electrophysiological signal channel.

[0059] Here, the Q-value function in a deep Q-value network can be expressed as: ; in, Current state and actions Q value under, For instant rewards, As a discount factor, The maximum Q value for the next state; The reward function in a deep Q-value network can be expressed as: ; ; in, Indicates the first Signal from each channel, This is the signal after timing error synchronization correction; It is the reward value in the reinforcement learning process; the smaller the error, the higher the reward value.

[0060] S304, for each type of electrophysiological signal, based on the timing correction action of the electrophysiological signal by the signal alignment module, the digital signal of the electrophysiological signal is time-shifted and adjusted to obtain the time-aligned signal corresponding to the electrophysiological signal.

[0061] Furthermore, after obtaining the timing correction action for each electrophysiological signal, the signal alignment module will adjust the specific time offset of the corresponding digital signal according to the timing correction action output by the deep Q network for each electrophysiological signal. This action may involve instructing digital signal processing operations such as interpolation, translation, or resampling of the signal sequence to compensate for its time deviation, making the signal more accurately aligned with other signals on the time axis, thereby obtaining a set of preliminarily adjusted time-aligned signals.

[0062] S305, the time-aligned signals of the various electrophysiological signals are taken as input, and the multi-channel temporal correlation model in the signal alignment module is used to process the time-aligned signals of the various electrophysiological signals to model the temporal dependency between the multi-channel signals and generate the prediction signal corresponding to each electrophysiological signal.

[0063] Specifically, to further improve synchronization accuracy, the inherent coupling or dependency relationships between multi-channel signals can be considered. A multi-channel temporal correlation model can be pre-built and trained in the signal alignment module. The time-aligned signal obtained in the previous step is used as input and fed into the multi-channel temporal correlation model in the signal alignment module for processing. Here, the multi-channel temporal correlation model is used to model the temporal dependency relationships between multi-channel signals, that is, the mutual influence and correlation of different electrophysiological signals in time. For example, the beating of the heart may affect the electrical activity of the brain. This influence has a certain regularity in time. The multi-channel temporal correlation model can capture this regularity and generate a predicted signal corresponding to each electrophysiological signal. The predicted signal is the signal form of the electrophysiological signal under ideal conditions inferred from the temporal dependency relationships between multi-channel signals.

[0064] In some possible embodiments, the aforementioned multi-channel temporal correlation model can be constructed based on a bidirectional long short-term memory (LSTM) network. A LSTM network is a recurrent neural network designed with two independent hidden layers (forward and backward) to simultaneously capture the forward and backward dependencies of sequence data in the time dimension. During construction, the pre-adjusted time-aligned signals of each channel can be organized by time steps and used as the input sequence for the network. Through learning, the network can model complex cross-channel temporal dependencies between signals from multiple channels; for example, the appearance of a specific waveform feature in one channel may indicate that a specific change is about to occur in another channel. The model is typically trained using a large amount of historically acquired, precisely synchronized multi-channel electrophysiological signal data as supervision samples, with the training objective being to minimize the difference between the network's predicted signal and the actual synchronized signal. Mean squared error is often used as the loss function. The trained model is integrated into a signal alignment module, and its high-precision predicted signal is used to generate feedback signals to drive the iterative optimization of the deep Q-network strategy, thus forming a closed-loop adaptive system from global temporal correlation modeling to local action correction.

[0065] S306, for each electrophysiological signal, calculate the feature difference between the predicted signal corresponding to the electrophysiological signal and the time-aligned signal, and generate a feedback signal corresponding to the electrophysiological signal.

[0066] Here, to evaluate the effect of the initial adjustment and form an optimized closed loop, for each channel, the characteristic difference between its predicted signal and the time-aligned signal can be calculated. This characteristic difference can include differences in signal amplitude, frequency, phase, etc. This difference value can be constructed as a feedback signal, characterizing the residual error between the current time-aligned signal and the ideal predicted signal on that channel using the current synchronization strategy. The feedback signal can be expressed as: .

[0067] S307, the feedback signal corresponding to each electrophysiological signal is fed back to the deep Q network; the deep Q network is used to adaptively iteratively optimize the output timing correction action based on the feedback signal to obtain the updated timing correction action corresponding to each electrophysiological signal.

[0068] Specifically, after obtaining the feedback signals corresponding to each electrophysiological signal, the feedback signals generated by all channels can be aggregated and fed back to the deep Q-network. The deep Q-network can use these feedback signals to determine whether the previously output timing correction actions are effective. If the feedback signals show significant differences in characteristics, it indicates that the timing correction actions are ineffective, and the network will adjust its own parameters, thus achieving adaptive iterative optimization of its internal policy network. This process can typically be achieved by updating the network weights through the backpropagation algorithm, thereby obtaining a set of updated, theoretically better timing correction actions.

[0069] S308, based on the updated timing correction action corresponding to each electrophysiological signal, return to step S304 until the feature difference generated in step S306 meets the preset synchronization condition, and determine the time alignment signal corresponding to each electrophysiological signal obtained in the last update as the preliminary synchronization signal corresponding to the electrophysiological signal.

[0070] Here, based on this set of updated timing correction actions, the process returns to step S304 to begin a new round of signal adjustment, prediction, evaluation, and optimization. This loop continues until the feature differences generated for all channels in step S306 decrease to meet a preset synchronization condition (set according to actual needs, for example, the maximum timing error between all channels is less than 0.1 milliseconds). When the condition is met, the loop terminates, and the time-aligned signal obtained after the last iteration update is determined as the preliminary synchronization signal corresponding to each electrophysiological signal.

[0071] In some possible embodiments, since the raw electrophysiological digital signals typically contain significant noise and artifacts, such as power line interference, baseline drift, and electromyography artifacts, the signal synchronization calibration model may also include a signal preprocessing module to improve the accuracy and stability of subsequent synchronization processing. This module can apply intelligent denoising algorithms based on convolutional neural networks and adaptive filtering techniques to perform real-time noise modeling, artifact suppression, and signal enhancement on the raw input signal, effectively separating and retaining useful physiological electrical activity components from the raw signal. Then, the high-quality preprocessed signal is input to the signal alignment module, providing a clearer and more stable data foundation for timing alignment operations, completing a high-quality conversion from the raw signal to the initial synchronization signal.

[0072] S104, the noise suppression module is used to suppress noise in each of the preliminary synchronization signals to obtain the noise-reduced signal corresponding to each electrophysiological signal.

[0073] Understandably, electrophysiological signal acquisition is subject to various noise interferences, such as power frequency interference, electrode contact noise, electromyography artifacts, or environmental electromagnetic noise. Noise can mask the true characteristics of the signal, affecting signal quality and the accuracy of analysis results. The noise suppression module's task is to identify and attenuate these interference components. This module employs various algorithms and techniques, such as filtering algorithms and wavelet transforms, to process the initial synchronization signal and remove noise components. For example, for power frequency interference noise in electrocardiogram signals, the noise suppression module can design appropriate filters to remove it, resulting in a cleaner, denoised signal and improved signal-to-noise ratio.

[0074] Specifically, in order to achieve dynamic noise modeling and adaptive noise suppression, when processing each preliminary synchronization signal using the noise suppression module, refer to... Figure 4 As shown, for each type of electrophysiological signal corresponding to the preliminary synchronization signal, the following noise suppression steps S401~S404 are performed in the noise suppression module: S401, the preliminary synchronization signal is input into the generative adversarial network in the noise suppression module. Through the adversarial training process between the generator and the discriminator in the generative adversarial network, the dynamic noise features of the signal are learned and extracted from the preliminary synchronization signal. Based on the extraction results, a dynamic noise model for the preliminary synchronization signal is established. An adaptive filter for the dynamic noise model is constructed.

[0075] Specifically, a Generative Adversarial Network (GAN) consists of a generator and a discriminator. During adversarial training, the generator aims to learn and generate data that simulates the noise characteristics of real signals, while the discriminator aims to accurately distinguish whether the input data is real noise from the real signal or simulated noise generated by the generator. Through adversarial training, the network can automatically learn and extract complex, time-varying dynamic noise features from the initial synchronization signal, such as interference patterns in specific frequency bands or motion-related artifacts. Dynamic noise features refer to the variation patterns of noise at different times and frequencies. Based on the extraction results, a dynamic noise model for the initial synchronization signal can be established, which describes the noise generation mechanism and variation patterns. Simultaneously, an adaptive filter is constructed for this dynamic noise model. This adaptive filter can automatically adjust its parameters according to the characteristics of the input signal to better suppress noise.

[0076] Here, the loss function of the generative adversarial network can be expressed as: ; in: It is the output of the discriminator, representing the signal. Is it real noise? These are noise samples generated by the generator; It is the input noise of the generator.

[0077] Here, the update rule for the adaptive filter is as follows: ; in, It is the output of the filter; It is the filter's error signal; It is the step size factor, used to control the adjustment speed of the filter.

[0078] S402, the preliminary synchronization signal is decomposed into time-frequency domain components at multiple scales; and the adaptive filter is used to filter the time-frequency domain components at the multiple scales respectively to suppress the components related to the dynamic noise model, so as to obtain the filtered time-frequency domain components of the preliminary synchronization signal; and, based on the filtered time-frequency domain components of the preliminary synchronization signal, a preliminary noise reduction signal corresponding to the preliminary synchronization signal is determined.

[0079] Here, time-frequency domain analysis allows for the simultaneous observation of signal characteristics in both time and frequency. Decomposition enables detailed analysis of the signal at different scales and frequencies. Specifically, time-frequency analysis tools, such as wavelet transform, can be used to decompose the initial input synchronization signal into time-frequency domain components at multiple scales. Wavelet transform expands the signal to a series of different resolutions (scales), with each scale reflecting the signal's characteristics over time within a specific frequency band. For example, one scale might correspond to the high-frequency components of the QRS complex in an ECG signal, while another scale might correspond to the low-frequency components of baseline drift. Adaptive filters are then used to filter the time-frequency domain components at multiple scales, suppressing components that match the characteristics of the dynamic noise model. For example, if the dynamic noise model indicates strong noise within a certain frequency range, the adaptive filter will attenuate the components within that frequency range. Finally, all the filtered time-frequency domain components are recombined (wavelet reconstruction) to obtain a preliminary denoised signal that shows partial noise suppression in the time domain.

[0080] Here, the formula for wavelet transform can be expressed as: ; in, It is a signal wavelet transform, It is the mother wavelet. and These are the scale and translation parameters, respectively.

[0081] S403, perform feature extraction and signal enhancement processing on the preliminary noise reduction signal to obtain an enhanced signal; and calculate the signal difference between the enhanced signal and the preliminary noise reduction signal to obtain the noise suppression error corresponding to the preliminary noise reduction signal.

[0082] Understandably, feature extraction aims to automatically extract representative quantitative indicators from signal waveforms that can distinguish different states or components, such as the mean amplitude and peak interval in the time domain, or the dominant frequency and power spectral density in the frequency domain. Signal enhancement processing refers to improving the clarity and usability of a signal through specific algorithms. For example, it may use deep autoencoder networks to learn a compact representation of the signal to recover details obscured by noise, or use image restoration techniques based on convolutional neural networks to deblur the time-frequency plot of the signal. When calculating the signal difference between the enhanced signal and the initial denoised signal, this difference mainly includes residual noise components that are further suppressed during the enhancement process, as well as signal distortion that may be introduced due to processing. This difference is quantified into a scalar indicator, namely the noise suppression error, to measure the effectiveness of the current round of noise suppression.

[0083] S404, based on the noise suppression error, update the dynamic noise model in step S401 and the adaptive filter parameters in step S403; and based on the updated model and parameters, repeat steps S401 to S403 until the noise suppression error meets the preset noise conditions, and determine the enhanced signal obtained from the last update as the denoised signal corresponding to the electrophysiological signal.

[0084] Specifically, if the noise suppression error is large, it indicates that the current model and parameters are not effective at suppressing noise and need to be adjusted. Backpropagation is performed to update the dynamic noise model parameters in step S401 (e.g., adjusting the weights of the generative adversarial network) and the coefficients of the adaptive filter, so that they can better adapt to the noise characteristics in the signal. Then, based on the updated model and parameters, steps S401 to S403 are repeated to continuously optimize the noise suppression effect until the noise suppression error meets a preset noise condition. The preset noise condition is an error threshold set according to actual needs, for example, the error is lower than a certain threshold or the signal-to-noise ratio improvement reaches the expected target. When the condition is met, the loop terminates, and the enhanced signal obtained after the last iteration is identified as the denoised signal corresponding to the electrophysiological signal.

[0085] For example, when processing electroencephalogram (EEG) signals, the signals are affected by factors such as electrooculography (EOG) artifacts caused by eye movements and blinking, and electromyography (EMG) artifacts caused by involuntary tension in the head muscles during acquisition. These factors can lead to non-stationary, transient amplitude abrupt changes and frequency contamination in the time-frequency domain. These artifacts manifest as abnormally high-energy components appearing within specific time periods in the time-frequency domain, with frequency distributions that overlap with but differ from actual EEG activity. An error correction module constructed using joint time-frequency domain analysis (such as wavelet transform) can accurately identify and locate the time-frequency regions (i.e., the joint interval of time and frequency) corresponding to these artifacts in the time-frequency plane. Subsequently, an integrated adaptive error correction algorithm (e.g., combining Kalman filter state tracking with neural network-based nonlinear correction) can adaptively adjust the correction intensity and strategy for the identified time-frequency region based on the time-frequency characteristics of the error, thereby selectively suppressing or correcting distortions caused by artifacts while preserving the true components of EEG activity to the greatest extent possible. This allows for effective correction of time-frequency domain errors in EEG signals, improving the accuracy and reliability of subsequent EEG feature extraction and analysis.

[0086] In some other embodiments, when using a noise suppression module to suppress noise in electrocardiogram (ECG) signals, typical baseline drift (a type of low-frequency noise) and power line interference (a type of narrowband high-frequency noise) occupy different regions in the time-frequency domain. Joint time-frequency domain analysis can clearly separate these two types of interference from key features of the ECG, such as the QRS complex and ST segment. Based on this information, the error correction module can then apply a specialized baseline correction algorithm to the low-frequency region and perform adaptive notch filtering on specific power line frequency points, thereby achieving accurate and adaptive error correction of the ECG signal according to its signal components.

[0087] S105, the error correction module is used to perform time-frequency domain error correction on each of the noise-reduced signals to obtain a high-precision synchronization signal corresponding to each electrophysiological signal.

[0088] Understandably, in addition to noise interference, various errors may exist during signal acquisition and transmission, such as frequency and phase errors. These errors can affect the accuracy and synchronization of the signal. The error correction module analyzes and processes the denoised signal from both the time and frequency domains, correcting errors by adjusting parameters such as frequency and phase. For example, in EEG signal analysis, because EEG signals have a wide frequency range, different frequency components may be affected by different errors. The error correction module precisely corrects different frequency components based on the spectral characteristics of the EEG signal, thereby obtaining a high-precision synchronization signal.

[0089] Specifically, to ultimately achieve high-precision synchronization, when processing each noise-reduced signal using the aforementioned error correction module, refer to... Figure 5 As shown, for each type of electrophysiological signal and its corresponding denoised signal, the following time-frequency domain error correction steps S501~S504 are performed in the error correction module: S501, perform time-frequency domain decomposition on the denoised signal to obtain time-frequency coefficients corresponding to the denoised signal; and determine the frequency information and phase information of the denoised signal in the time-frequency domain based on the time-frequency coefficients.

[0090] Here, time-frequency domain decomposition is a method for analyzing signals in both time and frequency dimensions. It can convert a one-dimensional time signal into a two-dimensional time-frequency representation, simultaneously displaying the frequency components of the signal at different times. Commonly used time-frequency domain decomposition methods include wavelet transform and short-time Fourier transform. Taking wavelet transform as an example, by selecting appropriate wavelet basis functions, the signal can be decomposed into wavelet coefficients at different scales. These coefficients constitute the time-frequency coefficients corresponding to the denoised signal. The time-frequency coefficients are a set of values, each associated with a specific time point and frequency band, representing the energy intensity of the original signal at that time and frequency component. For example, performing a 5-level discrete wavelet decomposition on an EEG signal yields 5 sets of detail coefficients at different frequency resolutions and a set of approximation coefficients. These coefficients together constitute the time-frequency representation of the signal. Based on these time-frequency coefficients, the frequency information (such as the dominant frequency distribution and instantaneous frequency) and phase information (such as the relative time position of the signal waveform at a specific frequency) of the signal in the time-frequency domain can be further analyzed. Frequency information reflects the distribution of different frequency components in a signal. For example, in electroencephalogram (EEG) signals, different frequency bands of brain waves correspond to different physiological states. Phase information describes the phase state of a signal at different times and can be used to analyze the phase relationship and synchronicity of the signal.

[0091] S502, based on the frequency information and phase information, adaptive error correction is performed on the noise-reduced signal to obtain a preliminary corrected signal.

[0092] Understandably, after obtaining the frequency signal and phase information, this information can be used as a guide to adaptively correct residual systematic errors in the denoised signal. The purpose of adaptive error correction is to automatically adjust the correction strategy according to the actual situation of the signal in order to more effectively eliminate errors.

[0093] Specifically, adaptive error correction comprises two parallel sub-processes: First, dynamic correction of linear error components is achieved using a Kalman filter. The Kalman filter is a recursive linear minimum variance estimation method that predicts and updates the "true state" of a signal (such as its ideal temporal position) through a recursive model containing state and measurement equations, enabling real-time estimation and correction of linear errors in the system. In this scenario, the temporal error of the signal is treated as the state of a dynamic system, optimally estimating and correcting linear drift by continuously fusing new observations (i.e., the actual measured signal time points). For example, in electrocardiogram (ECG) signal processing, slight changes in human physiological state or sensor instability may introduce linear errors. The Kalman filter can dynamically adjust its filtering parameters based on the statistical characteristics of the ECG signal to correct these linear errors. Secondly, deep neural networks can be used to learn and correct nonlinear error components. A deep neural network is a machine learning model with multiple layers of nonlinear transformations, such as a network containing multiple fully connected layers. It can learn the complex nonlinear mapping relationship between error-laden signal features and error-free signal features through a large amount of training data, thereby correcting nonlinear distortions that Kalman filters cannot model, such as irregular phase distortions caused by complex physiological interactions. For example, in electromyography (EMG) signal processing, the nonlinear characteristics of EMG signals are quite significant. Deep neural networks can accurately correct nonlinear errors by learning the differences between normal EMG signals and EMG signals containing errors.

[0094] S503, based on the joint optimization objective function, the time-domain and frequency-domain characteristics of the preliminary correction signal are jointly optimized.

[0095] Here, the joint optimization objective function is designed to comprehensively consider the signal's performance in both the time and frequency domains. It is used to collaboratively optimize the time-domain characteristics (such as waveform alignment on the time axis) and frequency-domain characteristics (such as spectral purity and consistency) of the initial correction signal. This function can include timing error terms and frequency offset error terms. The timing error term primarily measures the signal's deviation on the time axis, such as signal delay or advance, and is used to penalize time deviations between signals. The frequency offset error term focuses on the difference between the signal frequency and the actual frequency, and is used to penalize unwanted drift of the signal's frequency components. In communication signal processing, frequency offset leads to signal distortion and increased bit error rate. By comprehensively optimizing these two error terms, the initial correction signal can achieve better performance in both the time and frequency domains. Gradient descent can be used in the optimization process to iteratively adjust the signal representation (such as fine-tuning coefficients in the time and frequency domains) to minimize the value of this objective function that integrates timing and frequency constraints, thereby obtaining a signal that is closer to the ideal synchronization state in both time and frequency dimensions.

[0096] Specifically, the goal of joint optimization is to minimize the time synchronization error and frequency drift error of the signal, which can be expressed as: ; in, It's a time error. It's a frequency shift. and It is the weighting coefficient.

[0097] S504, the co-optimized signal is input to the signal enhancement network for signal detail recovery and signal noise suppression processing to obtain a high-precision synchronization signal corresponding to the electrophysiological signal.

[0098] Specifically, signal enhancement networks typically employ deep learning architectures, such as convolutional neural networks and recurrent neural networks. They can learn latent features and patterns in signals, enabling deeper detail recovery (e.g., sharpening blurred waveform inflection points) and residual noise suppression (e.g., filtering out subtle background noise that may still exist after optimization). In terms of signal detail recovery, it can recover weak features lost during preprocessing, such as recovering low-amplitude EEG components in EEG signals. Regarding signal noise suppression, it can further suppress potentially residual noise, improving the signal-to-noise ratio. After processing by the signal enhancement network, a high-precision synchronization signal corresponding to the electrophysiological signal can be obtained. These signals have higher accuracy and reliability, providing more effective support for subsequent physiological state analysis and disease diagnosis.

[0099] In some possible embodiments, to further improve the reliability of multi-channel electrophysiological signal synchronization, after obtaining the high-precision synchronization signal corresponding to each electrophysiological signal, the uncertainty of each channel synchronization signal can be estimated, and the multi-channel fusion weight can be adaptively adjusted based on the uncertainty.

[0100] Specifically, the high-precision synchronization signal output by the error correction module can be denoted as... The predicted signal generated by the multi-channel time-series correlation model or signal enhancement network is denoted as... For each channel, its residual sequence can be calculated: ; Path uncertainty estimation based on the statistical properties of residuals ,For example: ; in To estimate the window length, uncertainty is considered. Mapped to channel fusion weights: ; This yields a weighted fusion signal based on uncertainty estimation: ; When the residual variance of a certain channel increases significantly, the corresponding uncertainty is... The boost causes the weight of that channel to decrease automatically, thereby reducing the impact of abnormal channels on the overall analysis results.

[0101] In some embodiments, the aforementioned uncertainty estimation can also be fed back to the deep Q-network and noise suppression module as an auxiliary reward indicator and noise modeling constraint, so that the synchronous calibration model can consider both "time error" and "prediction stability" during optimization.

[0102] In some other embodiments, a three-track collaborative adaptive error correction architecture can be proposed, which deeply integrates a reinforcement learning-based temporal alignment track, a generative adversarial network-based dynamic noise suppression track, and a time-frequency domain error correction track based on linear-nonlinear joint modeling. This simultaneously optimizes synchronization accuracy at the time, frequency, and statistical structure levels. Compared with traditional error correction techniques that only use a single filter or a single deep network, the three-track collaborative design can form a complete closed-loop process from coarse alignment and noise modeling to fine-grained error correction. (1) Timing alignment track: The time offset correction of each channel signal is output by the deep reinforcement learning module. A preliminary alignment signal is obtained. ; (2) Dynamic noise suppression trajectory: Conditional generative adversarial network (cGAN) is used to model multi-channel noise and generate noise estimates. The noise-suppressed signal is obtained by combining adaptive filtering in the time-frequency domain. ; (3) Time-frequency domain error correction track: A linear Kalman filter and a nonlinear bidirectional residual network (Bi-ResNet) are used in parallel to jointly estimate and compensate for the residual error.

[0103] For orbit (2), the present invention constructs a conditional GAN, generator. Received noisy reference channel and wavelet feature concatenation vector of physiological signal Output noise estimation : ; Discriminator A two-stream architecture is used, with the time-domain branch input being... The frequency domain branch input is The consistency between generated noise and real noise is determined by sharing high-level features. The GAN loss function is: ; Simultaneously, a reconstruction error constraint is introduced: ; The comprehensive loss function for dynamic noise suppression tracks is: ; For orbit (3), the linear state estimate of the Kalman filter output in this invention is assumed to be: The nonlinear correction value output by Bi-ResNet is The two signals are then weighted and fused in the frequency domain to obtain the final corrected signal. ; in, The frequency-dependent adaptive fusion weights can be dynamically adjusted based on the Kalman filter estimation error covariance and the residual network prediction uncertainty, and are finally obtained through inverse transformation: .

[0104] Here, the overall objective function can be expressed as: ; in, Characterizing multi-channel timing errors, To characterize frequency domain residual distortion, this invention jointly trains a three-track module under a unified optimization framework, enabling the three tracks to converge collaboratively.

[0105] In some possible embodiments, since obtaining a high-precision synchronization signal is not the ultimate goal, but rather serves advanced data analysis and application decision-making, an integrated data parsing and interactive control process may be included after signal synchronization is completed to achieve real-time monitoring and intelligent assessment of physiological states. (See also...) Figure 6 As shown, after obtaining the high-precision synchronization signals corresponding to each electrophysiological signal, the following steps S601~S605 may also be included: S601, extract the signal feature information of the high-precision synchronization signal corresponding to each electrophysiological signal.

[0106] Here, signal feature information includes time-domain features, frequency-domain features, and time-frequency-domain features. Time-domain features refer to statistics directly derived from the time-series waveform of the signal, such as the RR interval of an electrocardiogram (ECG) signal, or the average amplitude or zero-crossing rate of an electroencephalogram (EEG) signal. Frequency-domain features are descriptive parameters obtained after converting the signal to the frequency domain using techniques such as Fourier transform, such as the dominant frequency of the signal, the peak value of the power spectrum, or the energy proportion of a specific frequency band (e.g., the 8-13Hz alpha wave in an EEG). Time-frequency-domain features combine information from both the time and frequency domains, reflecting the changes in the signal in both time and frequency. Methods such as short-time Fourier transform and wavelet transform can be used to extract time-frequency-domain features. For example, in electromyography (EMG) signal analysis, time-frequency-domain features can better demonstrate the frequency changes at different times during muscle contraction.

[0107] S602 inputs the signal characteristic information of each high-precision synchronization signal into the preset signal analysis model and outputs the signal classification information corresponding to each electrophysiological signal.

[0108] Understandably, a pre-built signal analysis model is a pre-constructed model used to analyze and classify signal features. It can be trained using methods such as machine learning and deep learning. This model can learn and analyze the input signal feature information, thereby outputting corresponding signal classification information. The signal classification information includes signal classification results and / or physiological state assessment results. Signal classification results categorize electrophysiological signals according to certain rules or standards, such as classifying electrocardiogram (ECG) signals into different types like normal ECG and arrhythmia; and classifying electroencephalogram (EEG) signals into different states like wakefulness and sleep. Physiological state assessment results evaluate an individual's physiological state based on the characteristics of the electrophysiological signals, such as assessing an individual's fatigue level and attention state by analyzing EEG signals, or assessing the functional state of the heart by analyzing ECG signals.

[0109] In some possible implementations, for multi-channel signals where there is a strong spatial dependency between them (e.g., the spatiotemporal relationship between EEG signals from different brain regions), a Graph Convolutional Network (GCN) can be introduced. This treats the signals as a graph structure, where nodes represent different signal channels and edges represent the relationships between channels. GCNs can utilize the spatial and temporal relationships between signals for pattern recognition. Here, the graph convolution operation formula for GCN can be expressed as: ; in, For the first The node feature matrix of the layer; For the first The weight matrix of the layer, Let be the normalized adjacency matrix of the graph; It is an activation function.

[0110] S603, the high-precision synchronization signal, signal classification information, and module parameters in the signal synchronization calibration model corresponding to each electrophysiological signal are visualized through the user interface.

[0111] Specifically, the user interface is the window through which users interact with the system. It presents complex data and information to users (operators or clinicians) in intuitive graphical and chart formats. Within this interface, high-precision synchronization signals can be displayed as waveforms, allowing users to visually observe the signal's shape and changes. Signal classification information can be displayed as text or labels, facilitating quick understanding of the signal's classification. The parameters of each module in the signal synchronization calibration model can be presented in tabular or numerical form, enabling users to clearly understand the model's specific configuration. For example, in a medical monitoring system, doctors can view the patient's ECG waveform, ECG classification results (such as whether it's a normal heart rhythm), and parameter settings for filtering, correction, and other operations in the signal synchronization calibration model through the user interface.

[0112] In some possible implementations, when the user interface displays information such as signal synchronization errors and noise interference, these deviations can be automatically detected and fed back to the backend algorithm. Through this feedback mechanism, the backend adjusts the synchronization algorithm, noise suppression parameters, and correction coefficients based on the real-time signal status to ensure signal synchronization accuracy and quality. For example, if the synchronization error is too large, the parameters of the deep learning synchronization algorithm can be automatically adjusted, and signal correction can be performed again; if the noise suppression effect is insufficient, the complexity of noise modeling can be increased or the noise suppression model can be retrained. Here, the feedback mechanism formula can be expressed as: ; in, It is the first Feedback correction amount for each signal and It provides real-time feedback of synchronization error and signal-to-noise ratio data, and this correction amount is used to automatically adjust the synchronization algorithm and noise suppression strategy.

[0113] Here, the real-time feedback mechanism can include not only synchronization error and noise suppression adjustments, but also dynamic adjustment of algorithm parameters. For example, when poor signal quality is detected, the learning rate of the deep learning model can be dynamically adjusted, or the model's hyperparameters can be adjusted based on user input, thereby optimizing the system's synchronization and noise processing performance. The real-time parameter adjustment formula can be expressed as: ; in, These are the updated algorithm parameters. These are the current algorithm parameters. It is a loss function. It is the learning rate.

[0114] In some possible embodiments, a set of adjustable parameters is provided on the interface. This includes cross-signal coherence weights. Three-track loss function weights Users can adjust the above parameters according to their actual tasks (e.g., focusing more on R-wave positioning accuracy or EEG band consistency) to form a user-desired parameter vector. This invention maintains a set of basic parameters internally. With adjustment increment The parameters are updated using the following formula: ; in, This is the step size factor. The updated parameters are used to construct the new reward function. and loss function The system can quickly reconfigure the synchronization strategy without retraining the global model, thus achieving online adaptation.

[0115] S604, receive adjustment instructions input by the user through the user interface, and adjust the parameters or strategies of at least one module in the signal synchronization calibration model based on the adjustment instructions.

[0116] Here, during use, users may request adjustments to the signal synchronization calibration model based on actual needs or observations of the signal processing effect. These adjustments can be specific parameter modifications, such as increasing the filter cutoff frequency; or strategy selection, such as temporarily switching to a different synchronization algorithm in the signal alignment module. For example, if a user finds that the processed signal still has some noise, they can input instructions to adjust the filter module parameters through the user interface to increase the filter strength; or if a user believes that the current calibration strategy is ineffective in handling certain special cases, they can input instructions to adjust the calibration strategy.

[0117] S605 performs real-time synchronous processing of subsequently acquired electrophysiological signals based on the adjusted model.

[0118] Furthermore, after the parameters or strategies of the model are adjusted, the internal parameters or processing strategies of one or more modules (such as noise suppression module and error correction module) in the signal synchronization calibration model are dynamically adjusted online. Then, this manually optimized updated model is used to synchronize and process the newly acquired electrophysiological signal streams in real time.

[0119] This ensures that the signal processing can be dynamically optimized based on actual conditions, transforming it from a static, open-loop processing pipeline into a dynamic closed-loop system that integrates expert experience and adapts to individual differences or special scenarios. This continuously guarantees and optimizes the effectiveness and reliability of the signal processing results. For example, in a long-term physiological monitoring project, as monitoring time progresses, an individual's physiological state may change, and the acquisition environment may also differ. By continuously adjusting the model based on user feedback, it can be ensured that subsequently acquired electrophysiological signals are accurately and effectively processed synchronously.

[0120] In some possible embodiments, to improve the generalization ability of the synchronous measurement method under different individuals, physiological state changes, and long-term monitoring scenarios, the signal synchronization calibration model also supports an individualized online transfer learning mechanism.

[0121] Specifically, the general model parameters obtained from offline training for multiple subjects can be denoted as... This includes network weights in the signal alignment module, noise suppression module, and error correction module. For newly entered target individuals, lightweight fine-tuning is performed using short-term labeled data (such as ECG / EEG fragments with precise event markers) to obtain individualized parameters. .

[0122] Transfer learning can be modeled as the following regularization optimization problem: ; in, For the synchronization loss function targeting the individual, These are regularization weights used to limit the deviation between individualized parameters and general parameters. The above optimizations can be performed online in mini-batch iterations on edge devices or servers.

[0123] During online operation, local parameter updates can be dynamically triggered based on real-time synchronization errors and user feedback, with fine-tuning only applied to network layers highly correlated with the individual, thereby controlling computational overhead while ensuring synchronization accuracy. For long-term monitoring scenarios, parameters that have stabilized and converged can be frozen periodically, with updates only applied to sub-modules related to changes in the current operating conditions (such as the dynamic noise model in the noise suppression module).

[0124] In some embodiments, multiple electrophysiological signal synchronous measurement methods can be deployed in a cloud-edge collaborative architecture: edge devices are responsible for multi-channel signal acquisition and lightweight preprocessing, while remote servers are responsible for complex model inference and historical data modeling, in order to support high-precision synchronization across devices and scenarios.

[0125] Specifically, let the function executed on the edge device side be... This includes analog-to-digital conversion, channel self-calibration, and coarse synchronization for the original signal. The feature representation is obtained after processing. .Will The data is uploaded to the server via a secure channel, where the server performs a full synchronous calibration of the model. Output high-precision synchronization signal And physiological status assessment results:

[0126] in, The model parameters maintained on the server side can be updated periodically based on historical data from multiple users and synchronized to each edge device through model distribution to achieve consistency of algorithm versions.

[0127] For heterogeneous data acquisition devices (with different sampling rates and different time bases), a unified time reference can be maintained on the server side. And introduce device-level clock skew estimation parameters. , will the Timestamp uploaded by each device Mapped to a unified timeline:

[0128] It can be estimated by aligning periodic time synchronization signals or cross-device characteristic events, and participate in optimization during synchronization calibration.

[0129] The method and system for synchronous measurement of multiple electrophysiological signals based on adaptive algorithms provided in this disclosure improve the accuracy and consistency of signal synchronization through the collaborative processing model constructed by the adaptive algorithm, reduce the timing errors and noise effects in multi-channel acquisition, and enhance the reliability and applicability of electrophysiological signal data.

[0130] Those skilled in the art will understand that, in the above-described method of the specific implementation, the order in which each step is written does not imply a strict execution order and does not constitute any limitation on the implementation process. The specific execution order of each step should be determined by its function and possible internal logic.

[0131] Based on the same inventive concept, this disclosure also provides a system for synchronous measurement of multiple electrophysiological signals based on an adaptive algorithm, corresponding to the method for synchronous measurement of multiple electrophysiological signals based on an adaptive algorithm. Since the principle of the system in this disclosure is similar to the method for synchronous measurement of multiple electrophysiological signals based on an adaptive algorithm described above, the implementation of the system can refer to the implementation of the method, and the repeated parts will not be described again.

[0132] Reference Figure 7The diagram shown is a schematic of a system for synchronously measuring multiple electrophysiological signals based on an adaptive algorithm, according to an embodiment of this disclosure. The system includes: A multi-channel signal acquisition module is configured to acquire multiple electrophysiological signals through a multi-channel sensor and output digital signals of the multiple electrophysiological signals; wherein each channel sensor is used to acquire a digital signal of one electrophysiological signal. The signal synchronization calibration processor has an internal signal synchronization calibration model built based on an adaptive algorithm. The signal synchronization calibration processor is communicatively connected to the multi-channel signal acquisition module and is used to receive digital signals of the various electrophysiological signals and input the digital signals into the signal synchronization calibration model for processing. The signal synchronization calibration model includes: The signal alignment unit is configured to perform time-series alignment on the digital signals of multiple input electrophysiological signals and output a preliminary synchronization signal corresponding to each electrophysiological signal. The noise suppression unit is communicatively connected to the signal alignment module and is configured to receive each of the preliminary synchronization signals and perform noise suppression processing, and output a noise-reduced signal corresponding to each electrophysiological signal. The error correction unit is communicatively connected to the noise suppression module and is configured to receive each of the noise-reduced signals and perform time-frequency domain error correction processing, and finally output a high-precision synchronization signal corresponding to each electrophysiological signal.

[0133] In one possible embodiment, the signal synchronization calibration processor is specifically used for: The signal alignment module is constructed based on a deep reinforcement learning algorithm, and the signal alignment module aims to minimize the timing error between multi-channel signals. The noise suppression module is constructed based on generative adversarial networks and adaptive filtering algorithms, with the goal of dynamic noise modeling and adaptive noise suppression. The error correction module is constructed based on joint time-frequency domain analysis and an adaptive error correction algorithm, with the error correction module aiming at time-frequency domain error correction. The signal synchronization calibration model is constructed by integrating the signal alignment module, the noise suppression module, and the error correction module.

[0134] In one possible embodiment, the signal alignment unit is specifically used for: Step 1: For each type of electrophysiological signal, determine the timing mapping relationship between the digital signal and the digital signals corresponding to other electrophysiological signals; and based on the timing mapping relationship between the digital signal and the digital signals corresponding to other electrophysiological signals, determine the timing error between the digital signal and the digital signals corresponding to other electrophysiological signals. Step 2: Input the timing errors of the digital signals corresponding to various electrophysiological signals into the signal alignment module; Step 3: Based on the timing error, the deep Q-network in the signal alignment module outputs timing correction actions for each electrophysiological signal; wherein, the deep Q-network optimizes the synchronization strategy through a reward function, which is constructed based on the objective of minimizing the timing error between multi-channel signals; Step 4: For each type of electrophysiological signal, based on the timing correction action of the electrophysiological signal by the signal alignment module, the digital signal of the electrophysiological signal is adjusted by time offset to obtain the time-aligned signal corresponding to the electrophysiological signal. Step 5: Using the time-aligned signals of the various electrophysiological signals as input, the multi-channel temporal correlation model in the signal alignment module is used to process the time-aligned signals of the various electrophysiological signals in order to model the temporal dependency between the multi-channel signals and generate the prediction signal corresponding to each electrophysiological signal. Step 6: For each electrophysiological signal, calculate the feature difference between the predicted signal and the time-aligned signal corresponding to the electrophysiological signal, and generate a feedback signal corresponding to the electrophysiological signal; Step 7: Feedback signals corresponding to each electrophysiological signal are fed back to the deep Q network; the deep Q network is used to adaptively iteratively optimize the output timing correction action based on the feedback signals to obtain the updated timing correction action corresponding to each electrophysiological signal. Step 8: Based on the updated timing correction action corresponding to each electrophysiological signal, return to step 4 until the feature difference generated in step 6 meets the preset synchronization condition. Then, determine the time alignment signal corresponding to each electrophysiological signal obtained from the last update as the preliminary synchronization signal corresponding to the electrophysiological signal.

[0135] In one possible embodiment, the noise suppression unit is specifically used for: For each type of electrophysiological signal corresponding to the preliminary synchronization signal, the following noise suppression steps are performed in the noise suppression module: Step A: Input the preliminary synchronization signal into the generative adversarial network in the noise suppression module. Through the adversarial training process between the generator and the discriminator in the generative adversarial network, learn and extract the dynamic noise features of the signal from the preliminary synchronization signal, and establish a dynamic noise model for the preliminary synchronization signal based on the extraction results; and construct an adaptive filter for the dynamic noise model. Step B: Decompose the preliminary synchronization signal into time-frequency domain components at multiple scales; and use the adaptive filter to filter the time-frequency domain components at the multiple scales respectively to suppress components related to the dynamic noise model, thereby obtaining the filtered time-frequency domain components of the preliminary synchronization signal; and, based on the filtered time-frequency domain components of the preliminary synchronization signal, determine the preliminary noise reduction signal corresponding to the preliminary synchronization signal. Step C: Perform feature extraction and signal enhancement processing on the preliminary denoising signal to obtain an enhanced signal; and calculate the signal difference between the enhanced signal and the preliminary denoising signal to obtain the noise suppression error corresponding to the preliminary denoising signal. Step D: Update the dynamic noise model in step A and the adaptive filter parameters in step C based on the noise suppression error; and repeat steps A to C based on the updated model and parameters until the noise suppression error meets the preset noise conditions, and determine the enhanced signal obtained from the last update as the denoised signal corresponding to the electrophysiological signal.

[0136] In one possible embodiment, the error correction unit is specifically used for: For each type of electrophysiological signal and its corresponding denoised signal, the following time-frequency domain error correction steps are performed in the error correction module: The denoised signal is decomposed in the time-frequency domain to obtain time-frequency coefficients corresponding to the denoised signal; and based on the time-frequency coefficients, the frequency information and phase information of the denoised signal in the time-frequency domain are determined. Based on the frequency and phase information, adaptive error correction is performed on the denoised signal to obtain a preliminary corrected signal; wherein, the adaptive error correction includes using a Kalman filter to dynamically correct the linear error component, and using a deep neural network to learn and correct the nonlinear error component. Based on a joint optimization objective function, the time-domain and frequency-domain characteristics of the preliminary correction signal are jointly optimized; wherein, the joint optimization objective function includes a timing error term and a frequency offset error term; The co-optimized signal is input into a signal enhancement network for signal detail recovery and signal noise suppression to obtain a high-precision synchronization signal corresponding to the electrophysiological signal.

[0137] In one possible embodiment, the signal synchronization calibration processor is further configured to: The signal feature information of the high-precision synchronization signal corresponding to each electrophysiological signal is extracted respectively; wherein, the signal feature information includes time domain features, frequency domain features, and time-frequency domain features; The signal characteristic information of each high-precision synchronization signal is input into a preset signal analysis model, and the signal classification information corresponding to each electrophysiological signal is output; wherein, the signal classification information includes signal classification results and / or physiological state assessment results; The high-precision synchronization signal, signal classification information, and parameters of each module in the signal synchronization calibration model corresponding to each electrophysiological signal are visualized through the user interface. Receive adjustment instructions input by the user through the user interface, and adjust the parameters or strategies of at least one module in the signal synchronization calibration model based on the adjustment instructions; The adjusted model is used to process the subsequently acquired electrophysiological signals in real time.

[0138] Based on the same inventive concept, this disclosure also provides a device for synchronous measurement of multiple electrophysiological signals based on an adaptive algorithm, which corresponds to the method for synchronous measurement of multiple electrophysiological signals based on an adaptive algorithm. Since the principle of the device in this disclosure is similar to the method for synchronous measurement of multiple electrophysiological signals based on an adaptive algorithm described above, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be described again.

[0139] Reference Figure 8 The diagram shown is a schematic of a multi-electrophysiological signal synchronous measurement device 800 based on an adaptive algorithm provided in an embodiment of this disclosure. The device includes: The signal acquisition module 801 is used to acquire digital signals of various electrophysiological signals collected by a multi-channel sensor; The model processing module 802 is used to construct a signal synchronization calibration model based on an adaptive algorithm, and input the digital signals of the various electrophysiological signals into the signal synchronization calibration model; wherein, the signal synchronization calibration model includes a signal alignment module, a noise suppression module, and an error correction module; Specifically, the model processing module 802 is used for: The signal alignment module is used to perform time-series alignment of the digital signals of the various electrophysiological signals to obtain preliminary synchronization signals corresponding to each electrophysiological signal; The noise suppression module is used to suppress noise in each of the preliminary synchronization signals to obtain the noise-reduced signals corresponding to each electrophysiological signal. The error correction module is used to perform time-frequency domain error correction on each of the noise-reduced signals to obtain a high-precision synchronization signal corresponding to each electrophysiological signal.

[0140] In one possible embodiment, the model processing module 802 is specifically used for: The signal alignment module is constructed based on a deep reinforcement learning algorithm, and the signal alignment module aims to minimize the timing error between multi-channel signals. The noise suppression module is constructed based on generative adversarial networks and adaptive filtering algorithms, with the goal of dynamic noise modeling and adaptive noise suppression. The error correction module is constructed based on joint time-frequency domain analysis and an adaptive error correction algorithm, with the error correction module aiming at time-frequency domain error correction. The signal synchronization calibration model is constructed by integrating the signal alignment module, the noise suppression module, and the error correction module.

[0141] In one possible embodiment, the model processing module 802 is specifically used for: Step 1: For each type of electrophysiological signal, determine the timing mapping relationship between the digital signal and the digital signals corresponding to other electrophysiological signals; and based on the timing mapping relationship between the digital signal and the digital signals corresponding to other electrophysiological signals, determine the timing error between the digital signal and the digital signals corresponding to other electrophysiological signals. Step 2: Input the timing errors of the digital signals corresponding to various electrophysiological signals into the signal alignment module; Step 3: Based on the timing error, the deep Q-network in the signal alignment module outputs timing correction actions for each electrophysiological signal; wherein, the deep Q-network optimizes the synchronization strategy through a reward function, which is constructed based on the objective of minimizing the timing error between multi-channel signals; Step 4: For each type of electrophysiological signal, based on the timing correction action of the electrophysiological signal by the signal alignment module, the digital signal of the electrophysiological signal is adjusted by time offset to obtain the time-aligned signal corresponding to the electrophysiological signal. Step 5: Using the time-aligned signals of the various electrophysiological signals as input, the multi-channel temporal correlation model in the signal alignment module is used to process the time-aligned signals of the various electrophysiological signals in order to model the temporal dependency between the multi-channel signals and generate the prediction signal corresponding to each electrophysiological signal. Step 6: For each electrophysiological signal, calculate the feature difference between the predicted signal and the time-aligned signal corresponding to the electrophysiological signal, and generate a feedback signal corresponding to the electrophysiological signal; Step 7: Feedback signals corresponding to each electrophysiological signal are fed back to the deep Q network; the deep Q network is used to adaptively iteratively optimize the output timing correction action based on the feedback signals to obtain the updated timing correction action corresponding to each electrophysiological signal. Step 8: Based on the updated timing correction action corresponding to each electrophysiological signal, return to step 4 until the feature difference generated in step 6 meets the preset synchronization condition. Then, determine the time alignment signal corresponding to each electrophysiological signal obtained from the last update as the preliminary synchronization signal corresponding to the electrophysiological signal.

[0142] In one possible embodiment, the model processing module 802 is specifically used for: For each type of electrophysiological signal corresponding to the preliminary synchronization signal, the following noise suppression steps are performed in the noise suppression module: Step A: Input the preliminary synchronization signal into the generative adversarial network in the noise suppression module. Through the adversarial training process between the generator and the discriminator in the generative adversarial network, learn and extract the dynamic noise features of the signal from the preliminary synchronization signal, and establish a dynamic noise model for the preliminary synchronization signal based on the extraction results; and construct an adaptive filter for the dynamic noise model. Step B: Decompose the preliminary synchronization signal into time-frequency domain components at multiple scales; and use the adaptive filter to filter the time-frequency domain components at the multiple scales respectively to suppress components related to the dynamic noise model, thereby obtaining the filtered time-frequency domain components of the preliminary synchronization signal; and, based on the filtered time-frequency domain components of the preliminary synchronization signal, determine the preliminary noise reduction signal corresponding to the preliminary synchronization signal. Step C: Perform feature extraction and signal enhancement processing on the preliminary denoising signal to obtain an enhanced signal; and calculate the signal difference between the enhanced signal and the preliminary denoising signal to obtain the noise suppression error corresponding to the preliminary denoising signal. Step D: Update the dynamic noise model in step A and the adaptive filter parameters in step C based on the noise suppression error; and repeat steps A to C based on the updated model and parameters until the noise suppression error meets the preset noise conditions, and determine the enhanced signal obtained from the last update as the denoised signal corresponding to the electrophysiological signal.

[0143] In one possible embodiment, the model processing module 802 is specifically used for: For each type of electrophysiological signal and its corresponding denoised signal, the following time-frequency domain error correction steps are performed in the error correction module: The denoised signal is decomposed in the time-frequency domain to obtain time-frequency coefficients corresponding to the denoised signal; and based on the time-frequency coefficients, the frequency information and phase information of the denoised signal in the time-frequency domain are determined. Based on the frequency and phase information, adaptive error correction is performed on the denoised signal to obtain a preliminary corrected signal; wherein, the adaptive error correction includes using a Kalman filter to dynamically correct the linear error component, and using a deep neural network to learn and correct the nonlinear error component. Based on a joint optimization objective function, the time-domain and frequency-domain characteristics of the preliminary correction signal are jointly optimized; wherein, the joint optimization objective function includes a timing error term and a frequency offset error term; The co-optimized signal is input into a signal enhancement network for signal detail recovery and signal noise suppression to obtain a high-precision synchronization signal corresponding to the electrophysiological signal.

[0144] In one possible embodiment, the model processing module 802 is further configured to: The signal feature information of the high-precision synchronization signal corresponding to each electrophysiological signal is extracted respectively; wherein, the signal feature information includes time domain features, frequency domain features, and time-frequency domain features; The signal characteristic information of each high-precision synchronization signal is input into a preset signal analysis model, and the signal classification information corresponding to each electrophysiological signal is output; wherein, the signal classification information includes signal classification results and / or physiological state assessment results; The high-precision synchronization signal, signal classification information, and parameters of each module in the signal synchronization calibration model corresponding to each electrophysiological signal are visualized through the user interface. Receive adjustment instructions input by the user through the user interface, and adjust the parameters or strategies of at least one module in the signal synchronization calibration model based on the adjustment instructions; The adjusted model is used to process the subsequently acquired electrophysiological signals in real time.

[0145] Based on the same technical concept, this disclosure also provides a computer device. (See also...) Figure 9 The diagram shown is a structural schematic of a computer device 900 provided in an embodiment of this disclosure, including a processor 901, a memory 902, and a bus 903. The memory 902 stores execution instructions and includes a main memory 9021 and an external memory 9022. The main memory 9021, also called internal memory, is used to temporarily store computational data in the processor 901, as well as data exchanged with external memory 9022 such as a hard disk. The processor 901 exchanges data with the external memory 9022 through the main memory 9021.

[0146] In this embodiment, the memory 902 is specifically used to store application code that executes the solution of this application, and its execution is controlled by the processor 901. That is, when the computer device 900 is running, the processor 901 communicates with the memory 902 through the bus 903, causing the processor 901 to execute the application code stored in the memory 902, thereby executing the method described in any of the foregoing embodiments.

[0147] The memory 902 may be, but is not limited to, random access memory, read-only memory, programmable read-only memory, erasable read-only memory, electrically erasable read-only memory, etc.

[0148] Processor 901 may be an integrated circuit chip with signal processing capabilities. The aforementioned processor can be a general-purpose processor, including a central processing unit (CPU), network processor, etc.; it can also be a digital signal processor, application-specific integrated circuit (ASIC), field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this invention. The general-purpose processor can be a microprocessor or any conventional processor.

[0149] It is understood that the structures illustrated in the embodiments of this application do not constitute a specific limitation on the computer device 900. In other embodiments of this application, the computer device 900 may include more or fewer components than illustrated, or combine some components, or split some components, or have different component arrangements. The illustrated components may be implemented in hardware, software, or a combination of software and hardware.

[0150] This disclosure also provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program performs the steps of the method for synchronous measurement of multiple electrophysiological signals based on an adaptive algorithm as described in the above-described method embodiments. The storage medium can be volatile or non-volatile computer-readable storage.

[0151] This disclosure also provides a computer program product carrying program code. The program code includes instructions that can be used to execute the steps of the method for synchronous measurement of multiple electrophysiological signals based on an adaptive algorithm as described in the above method embodiments. For details, please refer to the above method embodiments, which will not be repeated here.

[0152] The aforementioned computer program product can be implemented through hardware, software, or a combination thereof. In one optional embodiment, the computer program product is specifically embodied in a computer storage medium; in another optional embodiment, the computer program product is specifically embodied in a software product, such as a software development kit (SDK).

[0153] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and devices described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. In the several embodiments provided in this disclosure, it should be understood that the disclosed systems and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division; in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Another point is that the displayed or discussed mutual coupling or direct coupling or communication connection may be through some communication interfaces; the indirect coupling or communication connection of devices or units may be electrical, mechanical, or other forms.

[0154] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0155] In addition, the functional units in the various embodiments of this disclosure can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0156] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this disclosure, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this disclosure. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory, random access memory, magnetic disks, or optical disks.

[0157] Finally, it should be noted that the above-described embodiments are merely specific implementations of this disclosure, used to illustrate the technical solutions of this disclosure, and not to limit it. The protection scope of this disclosure is not limited thereto. Although this disclosure has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the scope of the technology disclosed in this disclosure. Such modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this disclosure, and should all be covered within the protection scope of this disclosure. Therefore, the protection scope of this disclosure should be determined by the protection scope of the claims.

Claims

1. A method for synchronous measurement of multiple electrophysiological signals based on an adaptive algorithm, characterized in that, include: Acquire digital signals of various electrophysiological signals collected by multi-channel sensors; A signal synchronization calibration model is constructed based on an adaptive algorithm, and the digital signals of the various electrophysiological signals are input into the signal synchronization calibration model; wherein, the signal synchronization calibration model includes a signal alignment module, a noise suppression module, and an error correction module; The signal alignment module is used to perform time-series alignment of the digital signals of the various electrophysiological signals to obtain preliminary synchronization signals corresponding to each electrophysiological signal; The noise suppression module is used to suppress noise in each of the preliminary synchronization signals to obtain the noise-reduced signals corresponding to each electrophysiological signal. The error correction module is used to perform time-frequency domain error correction on each of the noise-reduced signals to obtain a high-precision synchronization signal corresponding to each electrophysiological signal.

2. The method according to claim 1, characterized in that, The signal synchronization calibration model based on the adaptive algorithm includes: The signal alignment module is constructed based on a deep reinforcement learning algorithm, and the signal alignment module aims to minimize the timing error between multi-channel signals. The noise suppression module is constructed based on generative adversarial networks and adaptive filtering algorithms, with the goal of dynamic noise modeling and adaptive noise suppression. The error correction module is constructed based on joint time-frequency domain analysis and an adaptive error correction algorithm, with the error correction module aiming at time-frequency domain error correction. The signal synchronization calibration model is constructed by integrating the signal alignment module, the noise suppression module, and the error correction module.

3. The method according to claim 2, characterized in that, The step of using the signal alignment module to perform time-series alignment of the digital signals of the various electrophysiological signals includes: Step 1: For each type of electrophysiological signal, determine the timing mapping relationship between the digital signal and the digital signals corresponding to other electrophysiological signals; and based on the timing mapping relationship between the digital signal and the digital signals corresponding to other electrophysiological signals, determine the timing error between the digital signal and the digital signals corresponding to other electrophysiological signals. Step 2: Input the timing errors of the digital signals corresponding to various electrophysiological signals into the signal alignment module; Step 3: Based on the timing error, the deep Q-network in the signal alignment module outputs timing correction actions for each electrophysiological signal; wherein, the deep Q-network optimizes the synchronization strategy through a reward function, which is constructed based on the objective of minimizing the timing error between multi-channel signals; Step 4: For each type of electrophysiological signal, based on the timing correction action of the electrophysiological signal by the signal alignment module, the digital signal of the electrophysiological signal is adjusted by time offset to obtain the time-aligned signal corresponding to the electrophysiological signal. Step 5: Using the time-aligned signals of the various electrophysiological signals as input, the multi-channel temporal correlation model in the signal alignment module is used to process the time-aligned signals of the various electrophysiological signals in order to model the temporal dependency between the multi-channel signals and generate the prediction signal corresponding to each electrophysiological signal. Step 6: For each electrophysiological signal, calculate the feature difference between the predicted signal and the time-aligned signal corresponding to the electrophysiological signal, and generate a feedback signal corresponding to the electrophysiological signal; Step 7: Feedback signals corresponding to each electrophysiological signal are fed back to the deep Q network; the deep Q network is used to adaptively iteratively optimize the output timing correction action based on the feedback signals to obtain the updated timing correction action corresponding to each electrophysiological signal. Step 8: Based on the updated timing correction action corresponding to each electrophysiological signal, return to step 4 until the feature difference generated in step 6 meets the preset synchronization condition. Then, determine the time alignment signal corresponding to each electrophysiological signal obtained from the last update as the preliminary synchronization signal corresponding to the electrophysiological signal.

4. The method according to claim 2, characterized in that, The noise suppression module is used to suppress noise in each of the preliminary synchronization signals, including: For each type of electrophysiological signal corresponding to the preliminary synchronization signal, the following noise suppression steps are performed in the noise suppression module: Step A: Input the preliminary synchronization signal into the generative adversarial network in the noise suppression module. Through the adversarial training process between the generator and the discriminator in the generative adversarial network, learn and extract the dynamic noise features of the signal from the preliminary synchronization signal, and establish a dynamic noise model for the preliminary synchronization signal based on the extraction results; and construct an adaptive filter for the dynamic noise model. Step B: Decompose the preliminary synchronization signal into time-frequency domain components at multiple scales; and use the adaptive filter to filter the time-frequency domain components at the multiple scales respectively to suppress components related to the dynamic noise model, thereby obtaining the filtered time-frequency domain components of the preliminary synchronization signal; and, based on the filtered time-frequency domain components of the preliminary synchronization signal, determine the preliminary noise reduction signal corresponding to the preliminary synchronization signal. Step C: Perform feature extraction and signal enhancement processing on the preliminary denoising signal to obtain an enhanced signal; and calculate the signal difference between the enhanced signal and the preliminary denoising signal to obtain the noise suppression error corresponding to the preliminary denoising signal. Step D: Update the dynamic noise model in step A and the adaptive filter parameters in step C based on the noise suppression error; and repeat steps A to C based on the updated model and parameters until the noise suppression error meets the preset noise conditions, and determine the enhanced signal obtained from the last update as the denoised signal corresponding to the electrophysiological signal.

5. The method according to claim 2, characterized in that, The step of using the error correction module to perform time-frequency domain error correction on each of the noise-reduced signals includes: For each type of electrophysiological signal and its corresponding denoised signal, the following time-frequency domain error correction steps are performed in the error correction module: The denoised signal is decomposed in the time-frequency domain to obtain time-frequency coefficients corresponding to the denoised signal; and based on the time-frequency coefficients, the frequency information and phase information of the denoised signal in the time-frequency domain are determined. Based on the frequency and phase information, adaptive error correction is performed on the denoised signal to obtain a preliminary corrected signal; wherein, the adaptive error correction includes using a Kalman filter to dynamically correct the linear error component, and using a deep neural network to learn and correct the nonlinear error component. Based on a joint optimization objective function, the time-domain and frequency-domain characteristics of the preliminary correction signal are jointly optimized; wherein, the joint optimization objective function includes a timing error term and a frequency offset error term; The co-optimized signal is input into a signal enhancement network for signal detail recovery and signal noise suppression to obtain a high-precision synchronization signal corresponding to the electrophysiological signal.

6. The method according to claim 1, characterized in that, After obtaining the high-precision synchronization signal, the process also includes: The signal feature information of the high-precision synchronization signal corresponding to each electrophysiological signal is extracted respectively; wherein, the signal feature information includes time domain features, frequency domain features, and time-frequency domain features; The signal characteristic information of each high-precision synchronization signal is input into a preset signal analysis model, and the signal classification information corresponding to each electrophysiological signal is output; wherein, the signal classification information includes signal classification results and / or physiological state assessment results; The high-precision synchronization signal, signal classification information, and parameters of each module in the signal synchronization calibration model corresponding to each electrophysiological signal are visualized through the user interface. Receive adjustment instructions input by the user through the user interface, and adjust the parameters or strategies of at least one module in the signal synchronization calibration model based on the adjustment instructions; The adjusted model is used to process the subsequently acquired electrophysiological signals in real time.

7. A system for synchronously measuring multiple electrophysiological signals based on an adaptive algorithm, characterized in that, include: A multi-channel signal acquisition module is configured to acquire multiple electrophysiological signals through a multi-channel sensor and output digital signals of the multiple electrophysiological signals; wherein each channel sensor is used to acquire a digital signal of one electrophysiological signal. The signal synchronization calibration processor has an internal signal synchronization calibration model built based on an adaptive algorithm. The signal synchronization calibration processor is communicatively connected to the multi-channel signal acquisition module and is used to receive digital signals of the various electrophysiological signals and input the digital signals into the signal synchronization calibration model for processing. The signal synchronization calibration model includes: The signal alignment unit is configured to perform time-series alignment on the digital signals of multiple input electrophysiological signals and output a preliminary synchronization signal corresponding to each electrophysiological signal. The noise suppression unit is communicatively connected to the signal alignment module and is configured to receive each of the preliminary synchronization signals and perform noise suppression processing, and output a noise-reduced signal corresponding to each electrophysiological signal. The error correction unit is communicatively connected to the noise suppression module and is configured to receive each of the noise-reduced signals and perform time-frequency domain error correction processing, and finally output a high-precision synchronization signal corresponding to each electrophysiological signal.

8. A device for synchronously measuring multiple electrophysiological signals based on an adaptive algorithm, characterized in that, include: The signal acquisition module is used to acquire digital signals of various electrophysiological signals collected by a multi-channel sensor. The model processing module is used to construct a signal synchronization calibration model based on an adaptive algorithm and input the digital signals of the various electrophysiological signals into the signal synchronization calibration model; wherein, the signal synchronization calibration model includes a signal alignment module, a noise suppression module, and an error correction module; Specifically, the model processing module is used for: The signal alignment module is used to perform time-series alignment of the digital signals of the various electrophysiological signals to obtain preliminary synchronization signals corresponding to each electrophysiological signal; The noise suppression module is used to suppress noise in each of the preliminary synchronization signals to obtain the noise-reduced signals corresponding to each electrophysiological signal. The error correction module is used to perform time-frequency domain error correction on each of the noise-reduced signals to obtain a high-precision synchronization signal corresponding to each electrophysiological signal.

9. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 6.

10. A computer device, comprising a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 6.

Images