An electrocardiogram measurement method based on ultrasonic signals from smart speakers

Abstract

The present invention relates to the field of electrocardiogram measurement technology, and specifically refers to an electrocardiogram measurement method based on an intelligent speaker ultrasonic signal, comprising: receiving a plurality of mixed subcarrier signals of different frequencies; constructing and solving a spectrum function at each sampling time based on a received signal matrix, obtaining the true value of the propagation delay of each candidate object subcarrier signal on its corresponding target path at each sampling time, calculating the enhanced ultrasonic signal of each candidate object within a preset time period based on a carrier shaping technology, and determining the target user by using a spectrum amplitude ratio method and ultrasonic signal phase change; using an adaptive IIR comb notch filter, extracting the ultrasonic heartbeat signal of each target user within a preset time period and inputting it into a trained electrocardiogram signal prediction model, and outputting the predicted value of the electrocardiogram signal of each target user within the preset time period. The present invention has good monitoring effect on multi-user ultrasonic heartbeat signals, high accuracy, fast speed, and good real-time performance.

Description

Technical Field

[0001] The present invention relates to the technical field of electrocardiogram measurement, and in particular to an electrocardiogram measurement method based on intelligent speaker ultrasonic signals. Background Art

[0002] Among the existing medical testing equipment, the ECG signal is the external manifestation of the electrical activity of myocardial cells. Using an electrocardiogram diagnostic instrument to record the electrocardiogram corresponding to the patient's heart activity in each cardiac cycle from the body surface is one of the most commonly used technical means for clinical analysis of cardiovascular and cerebrovascular diseases. However, the existing electrocardiogram measurement method is inconvenient to operate and cannot be measured in real time.

[0003] In order to solve the above problems, acoustic vital signs monitoring technology has been proposed, which has attracted much attention in the field of mobile perception and has the characteristics of high detection accuracy and wide hardware application. However, since the sound signal strength will decay rapidly with the distance, the effective detection distance of the existing acoustic vital signs monitoring technology is only 1.2 meters, which limits the wide application of acoustic vital signs monitoring technology. Although the WiFi-based system can monitor the user's breathing within a large range, which solves the defect of the limited perception range of the existing acoustic vital signs monitoring technology, it is difficult to accurately detect the user's heartbeat. In order to accurately detect the user's heartbeat while expanding the perception range, a millimeter wave-based detection method has been proposed. Although this method can detect the heartbeat, a dedicated FMCW radar is used in the detection process, which is relatively expensive. In addition, the existing WiFi-based system and millimeter wave-based detection method are both contactless electrocardiogram signal measurement technologies, which can only detect the electrocardiogram signal of a single user at the same time. Therefore, the existing technology has the defects of limited perception range, high cost of use, and can only detect a single user when detecting the user's heartbeat. Summary of the invention

[0004] Therefore, the technical problem to be solved by the present invention is to overcome the problems in the prior art of limited sensing range, high use cost and only being able to detect a single user when detecting the user's heartbeat.

[0005] In order to solve the above technical problems, the present invention provides an electrocardiogram measurement method based on an intelligent speaker ultrasonic signal, comprising:

[0006] Speakers in smart speakers continuously generate Different frequency ultrasonic signals are modulated by orthogonal frequency division multiplexing. The subcarrier signals of different frequencies are then transmitted to multiple target users, and the microphone in the smart speaker receives mixed subcarrier signals of different frequencies; wherein, a horizontal line perpendicular to the connecting line is drawn through the midpoint of the connecting line between the loudspeaker and the microphone, and the angle between the connecting line between each target user and the midpoint of the connecting line and the horizontal line is within a preset angle range, and each target user does not block each other;

[0007] Get the mixed subcarrier signal vector received at each sampling time within the preset time period of the microphone , establish the signal receiving matrix , Indicates the total number of sampling moments, from 1 to The sampling moments constitute the preset period. Indicates The sampling time mixed subcarrier signal, characterizing the At the sampling moment, all candidate objects are transmitted on their corresponding target paths. The signal after the subcarrier signals are mixed; construct the At the sampling time, The first transmission on the target path corresponding to the candidate object Subcarrier signal , forming the Sampling time The subcarrier signal vector of the candidate ; Indicates the number of candidate objects;

[0008] Based on the received signal matrix, a signal covariance matrix is ​​constructed and eigen-decomposed. After obtaining the eigenvalues ​​of the signal covariance matrix and its corresponding eigenvectors, the signal subspace and the noise subspace are determined. Based on the orthogonality of the signal subspace and the noise subspace, the spectrum function at each sampling moment is constructed. After solving the peak value of the spectrum function at each sampling moment and obtaining the true value of the signal propagation delay on the target path of each candidate object at each sampling moment, the relative phase offsets of each subcarrier signal corresponding to each candidate object at each sampling moment and the first subcarrier signal are calculated, and a steering vector for each candidate object at each sampling moment is constructed. Based on the carrier shaping technology, the steering vector of each candidate object at each sampling moment, the mixed subcarrier signal vector at each sampling moment, and the noise value corresponding to each candidate object at each sampling moment are combined to form a steering vector for each candidate object at each sampling moment. The subcarrier signals are initially phase aligned to calculate the enhanced ultrasonic signal of each candidate object at each sampling moment, and then the enhanced ultrasonic signal of each candidate object in a preset time period is obtained;

[0009] Based on the enhanced ultrasonic signal of each candidate object within the preset time period, the spectrum concentration of each candidate object is calculated by using the spectrum amplitude ratio method; if the spectrum concentration of any candidate object is greater than the spectrum threshold, the current candidate object is a living being; the true value of the signal propagation delay on the target path of each living being at each sampling moment is used as the delay of the enhanced ultrasonic signal of each living being at each sampling moment, and the ultrasonic phase change of each living being at each sampling moment is calculated by combining the path expression and the phase expression; if the ultrasonic phase change of any living being at each sampling moment is greater than the phase change threshold, the current living being is the target user;

[0010] Using an adaptive IIR comb notch filter, the ultrasonic heartbeat signal of each target user within a preset time period is extracted;

[0011] The ultrasonic heartbeat signal of each target user within a preset time period is input into the trained electrocardiogram signal prediction model, and the predicted value of the electrocardiogram signal of each target user within the preset time period is output.

[0012] Preferably, the electrocardiogram signal prediction model is a neural network model for multi-level feature extraction; the neural network model for multi-level feature extraction is composed of an input layer, an output layer and a plurality of convolutional layers, including:

[0013] Input layer, inputting the ultrasonic heartbeat signal of the current target user within a preset time period;

[0014] The first convolutional layer consists of two 1x3 convolution operations plus two ReLU nonlinear activation functions; its input is the ultrasonic heartbeat signal of the current target user within the preset time period, and its output is the features corresponding to the ultrasonic heartbeat signal of the current target user within the preset time period. ;

[0015] The first downsampling layer consists of a 1x2 maximum pooling operation, which is used to downsample the feature values; its input is the feature , the output is The features obtained after downsampling through a 1x2 maximum pooling operation ;

[0016] The second convolutional layer consists of a 1xW convolution block, a 1x3 convolution operation, and a ReLU activation function; its input is the feature , whose output is the feature ;

[0017] The second downsampling layer consists of a 1x2 maximum pooling operation, which is used to downsample the feature values; its input is the feature , the output is The features obtained after downsampling through a 1x2 maximum pooling operation ;

[0018] The third convolutional layer consists of a 1xW convolution block, a 1x3 convolution operation, and a ReLU activation function; its input is the feature , whose output is the feature ;

[0019] The third downsampling layer consists of a 1x2 maximum pooling operation, which is used to downsample the feature values; its input is the feature , the output is The features obtained after downsampling through a 1x2 maximum pooling operation ;

[0020] The fourth convolutional layer consists of a 1xW convolution block, a 1x3 convolution, and a ReLU activation function; its input is the feature , whose output is the feature ;

[0021] The first upsampling layer consists of a 1x2 upsampling operation to upsample the features; its input is the feature , the output is The features obtained after a 1x2 upsampling operation ;

[0022] The fifth convolutional layer consists of two 1x3 convolution operations plus two ReLU nonlinear activation functions; its input is the feature , whose output is characterized ;

[0023] The second upsampling layer consists of a 1x2 upsampling operation to upsample the features; its input is the feature , the output is The features obtained after a 1x2 upsampling operation ;

[0024] The sixth convolutional layer consists of two 1x3 convolution operations plus two ReLU nonlinear activation functions; its input is the feature , whose output is the feature ;

[0025] The third upsampling layer consists of a 1x2 upsampling operation to upsample the features; its input is the feature , the output is The features obtained after a 1x2 upsampling operation ;

[0026] The seventh convolutional layer consists of two 1x3 convolution operations plus two ReLU nonlinear activation functions; its input is the feature , whose output is the feature ;

[0027] Output layer, based on features , through a 1x1 convolution, the output feature is generated, i.e., the electrocardiogram signal of the current target user within the preset time period;

[0028] Among them, the first convolution layer, the second convolution layer and the third convolution layer, as well as the first downsampling layer, the second downsampling layer and the third downsampling layer constitute the encoder; the fourth convolution layer is the bottom convolution layer, which serves as a bridge between the encoder and the decoder; the fifth convolution layer, the sixth convolution layer and the seventh convolution layer, as well as the first upsampling layer, the second upsampling layer and the third upsampling layer constitute the decoder;

[0029] The convolution block is composed of one-dimensional convolution, shortcut connection and depth-wise separable convolution. The eigenvalues ​​output by the downsampling layer are used as the input of the convolution block, and the data obtained by fusing the eigenvalues ​​processed by the one-dimensional convolution and depth-wise separable convolution with the eigenvalues ​​processed by the shortcut connection are used as the output of the convolution block.

[0030] Preferably, the loss function of the neural network model for multi-level feature extraction is a mean square error function , whose expression is:

[0031] ;

[0032] in, Indicates The true value of the ECG signal of samples; Indicates The predicted value of the ECG signal of samples; Indicates the total number of samples; Indicates the sample number.

[0033] Preferably, the method comprises constructing a signal covariance matrix based on a received signal matrix and performing eigendecomposition on the matrix, obtaining the eigenvalues ​​of the signal covariance matrix and the eigenvectors corresponding thereto, and then determining the signal subspace and the noise subspace; constructing a spectrum function at each sampling moment based on the orthogonality of the signal subspace and the noise subspace; solving the peak value of the spectrum function at each sampling moment to obtain the true value of the signal propagation delay on the target path of each candidate object at each sampling moment, including:

[0034] Based on signal receiving matrix , construct the signal covariance matrix , whose expression is:

[0035] ;

[0036] in, represents the transpose of the signal reception matrix;

[0037] The signal covariance matrix Perform characteristic decomposition and obtain the decomposition expression as follows:

[0038] ;

[0039] in, The eigenvector matrix representing the signal covariance matrix; The eigenvalue matrix representing the signal covariance matrix; represents the transpose of the eigenvector matrix of the signal covariance matrix;

[0040] Based on the number of candidate objects , we get the eigenvalues ​​of the signal covariance matrix Based on the speakers in the smart speaker, different frequency ultrasonic signals, then based on the sorting from large to small The first of the eigenvalues The eigenvectors corresponding to the eigenvalues ​​form the signal subspace; based on the remaining The eigenvectors corresponding to the eigenvalues ​​form the noise subspace;

[0041] Based on the orthogonality of the signal subspace and the noise subspace, the spectrum function at each sampling moment is constructed. The spectrum function of the sampling time The expression is:

[0042] ;

[0043] in, Indicates The turning vector of the candidate object at each sampling moment; Indicates The transpose of the steering vector of the candidate object at each sampling moment; Indicates The signal propagation delay on the target path of the candidate object at each sampling time; represents the noise subspace eigenvector matrix; represents the transpose of the noise subspace eigenvector matrix;

[0044] Through the The spectrum function of the sampling moment is analyzed to obtain the The peak value of the spectrum function corresponding to the sampling time, where each peak corresponds to a Based on the range of signal propagation delay on the target path of each candidate object at each sampling time, combined with the number of candidate objects, the first Sampling time The true value of the signal propagation delay on the target path of the candidate object , thus obtaining The turning vector of each candidate object at the sampling moment , and then get the The steering matrix composed of the steering vectors of each candidate object at the sampling time is: .

[0045] Preferably, based on the true value of the signal propagation delay on the target path of each candidate object at each sampling moment, the relative phase offsets of each subcarrier signal corresponding to each candidate object at each sampling moment and the first subcarrier signal are calculated respectively, and the steering vector of each candidate object at each sampling moment is formed, including:

[0046] No. Sampling time The candidate object corresponds to The relative phase offset between the subcarrier signal and the first subcarrier signal The expression is:

[0047] ;

[0048] No. Sampling time The turning vector of the candidate object The expression is:

[0049] ;

[0050] in, Indicates Sampling time The true value of the signal propagation delay on the target path of the candidate object; Indicates the total number of subcarrier signals; Represents the index of the subcarrier signal, ; Indicates the frequency spacing between adjacent subcarriers.

[0051] Preferably, the carrier shaping technology is based on the steering vector of each candidate object at each sampling moment, the mixed subcarrier signal vector at each sampling moment, and the noise value corresponding to each candidate object at each sampling moment. The subcarrier signals are initially phase aligned, and the enhanced ultrasonic signal of each candidate object at each sampling moment is calculated, and then the enhanced ultrasonic signal of each candidate object in the preset time period is obtained, including:

[0052] No. Sampling time Enhanced ultrasonic signal of candidate The expression is:

[0053] ;

[0054] in, Indicates Sampling time The turning vectors of candidate objects; express Moment The noise value corresponding to each candidate object;

[0055] Based on The enhanced ultrasonic signals of each candidate object at the sampling moment constitute The enhanced ultrasonic signal vector at time , whose expression is:

[0056] ;

[0057] in, ; Indicates The steering matrix composed of the steering vectors of each candidate object at the sampling time, ; Indicates The noise vector is composed of the noise values ​​corresponding to each candidate object at each sampling moment.

[0058] Preferably, the spectrum concentration of each candidate object is calculated by using the spectrum amplitude ratio method based on the enhanced ultrasonic signal of each candidate object within the preset time period; if the spectrum concentration of any candidate object is greater than the spectrum threshold, then the current candidate object is a living body, including:

[0059] Performing a fast Fourier transform on the enhanced ultrasonic signal of each candidate object within a preset time period, converting the enhanced ultrasonic signal of each candidate object from the time domain to the frequency domain, and obtaining the amplitude of the enhanced ultrasonic signal of each frequency of each candidate object;

[0060] Calculates the first The spectrum concentration of the candidate , whose expression is:

[0061] ;

[0062] in, Indicates The frequency of candidates is The amplitude of the enhanced ultrasonic signal; Indicates the preset frequency range. The sum of the amplitudes of the enhanced ultrasonic signals of each frequency of the candidate objects, and the preset frequency range is ; Indicates the overall frequency range The sum of the amplitudes of the enhanced ultrasonic signals of each frequency of the candidate objects, the overall frequency range is the frequency range corresponding to the preset time period, and its range is ;

[0063] If the spectrum concentration of any candidate object is greater than the spectrum threshold, then The candidate object is a living thing, otherwise it is a non-living thing.

[0064] Preferably, the real value of the signal propagation delay on the target path of each life form at each sampling moment is used as the delay of the enhanced ultrasonic signal of each life form at each sampling moment, and the ultrasonic phase change of each life form at each sampling moment is calculated in combination with the distance expression and the phase expression; if the ultrasonic phase change of any life form at each sampling moment is greater than the phase change threshold, then the current life form is the target user, including:

[0065] The distance expression is:

[0066] ;

[0067] in, Indicates Sampling time The time delay of the enhanced ultrasonic signal of a living organism; Indicates Sampling time The target path length corresponding to the enhanced ultrasonic signal of a living being; Indicates The fixed path length corresponding to the enhanced ultrasonic signal of a living being; Indicates Sampling time The length of the chest cavity fluctuation path corresponding to the enhanced ultrasonic signal of a living body; Indicates the speed of enhancing ultrasonic signal;

[0068] The phase expression is:

[0069] ;

[0070] in, Indicates Sampling time Ultrasonic phase changes of a living organism; Indicates Sampling time The wavelength of the enhanced ultrasonic signal of a living organism;

[0071] If the ultrasonic phase change of any living being at all sampling moments is less than the phase change threshold, the current living being is the target user, otherwise it is a non-target user.

[0072] Preferably, the extracting the ultrasonic heartbeat signal of each target user within a preset time period by using an adaptive IIR comb notch filter comprises:

[0073] By using Fourier transform, the time-frequency relationship of the enhanced ultrasonic signal of each target user in the preset time period is converted into a frequency-amplitude relationship to obtain the spectrum of each target user; based on the spectrum of each target user, the frequency of the prominent peak pointing to the fundamental frequency of the heartbeat motion of each user is found, which is recorded as , representing the The target user points to the frequency of the prominent peak of the heartbeat fundamental frequency; wherein the heartbeat fundamental frequency is within the scope of

[0074] Using adaptive IIR comb notch filter, combined with preset sampling frequency, The frequency of the prominent peak of the heartbeat fundamental frequency pointed by the target user is used to obtain the transfer function of the adaptive IIR comb notch filter, which is expressed as:

[0075] ;

[0076] in, represents a complex frequency variable; represents the feedback coefficient, ; represents the sampling frequency, represents the frequency of the signal to be eliminated, that is ;

[0077] Based on the transfer function, the first After the signal to be eliminated in the enhanced ultrasonic signal of the target user is eliminated, the first Sampling time The filtered ultrasonic signal of the target user ;

[0078] The difference between the enhanced ultrasonic signal of each target user at each sampling moment in the preset time period and its filtered enhanced ultrasonic signal is used as the ultrasonic heartbeat signal of each target user at each sampling moment in the preset time period, that is, Sampling time The difference between the enhanced ultrasonic signal of the target user and the filtered enhanced ultrasonic signal is taken as the Sampling time The ultrasonic heartbeat signal of the target user , whose expression is:

[0079] ;

[0080] in, Indicates the first Sampling time The enhanced ultrasonic signal of a target user.

[0081] Preferably, among the multiple propagation paths of the subcarrier signal of each candidate object, the path where the reflection intensity reaches the maximum value is used as the target path for transmission of the subcarrier signal of each candidate object.

[0082] The above technical solution of the present invention has the following beneficial effects compared with the prior art:

[0083] The electrocardiogram measurement method based on the ultrasonic signal of an intelligent speaker described in the present invention sets a static scene where the intelligent speaker and multiple target users are located, wherein the midpoint between the line connecting the speaker and the microphone in the intelligent speaker is used as the midpoint, and the straight line passing through the midpoint and perpendicular to the line connecting the speaker and the microphone is used as the horizontal line, the angle between the line connecting each target user and the midpoint and the horizontal line is within a preset angle threshold, and the target users do not block each other; in the above-mentioned multi-user static scene, the intelligent speaker is used to send multiple ultrasonic signals of different frequencies, and the microphone receives multiple subcarrier signals of different frequencies, and the cost of using the equipment is low, the operation is simple, and the practicability is strong; based on the carrier shaping technology, the phase difference between the subcarriers of different frequencies received by the microphone is compensated to realize constructive superposition and enhance the ultrasonic signal, and the enhanced ultrasonic signal is obtained, the ultrasonic signal is enhanced, the range of electrocardiogram signal monitoring is expanded, and the foundation is laid for improving the accuracy of electrocardiogram measurement, and each target user does not need to be measured at a fixed angle position. Measurement; Based on the mixed situation of ultrasonic signals in multi-user static scenarios, a new continuous-MUSIC method is used to obtain the true value of the propagation delay of the subcarrier signal received by the microphone on the target path of each candidate object at each moment based on the spectrum function, thereby achieving the task of separating the ultrasonic signal of each candidate object from the mixed ultrasonic signal; Based on the spectrum amplitude ratio method and the difference in the phase change of the ultrasonic signal caused by breathing, multiple target users are determined among all candidate objects, thereby improving the accuracy of target user judgment; An adaptive IIR comb notch filter is used to extract ultrasonic heartbeat signals, and the ultrasonic heartbeat signals of each target user are input into a multi-level feature extraction neural network model to obtain the electrocardiogram signals of each target user. The electrocardiogram signals of multiple users can be measured simultaneously, thereby shortening the electrocardiogram measurement time of multiple users; The convolution block in the multi-level feature extraction neural network model enables the encoder to capture the contextual information between the waveform characteristics, thereby improving the accuracy of electrocardiogram signal extraction. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] In order to make the content of the present invention more clearly understood, the present invention is further described in detail below according to specific embodiments of the present invention in conjunction with the accompanying drawings, wherein:

[0085] Figure 1 It is a flow chart of an electrocardiogram measurement method based on intelligent speaker ultrasonic signal provided by the present invention;

[0086] Figure 2 It is a schematic diagram of the locations of the speakers and microphones in the smart speaker and each target user;

[0087] Figure 3 This is a comparison chart of the effects of continuous-MUSIC and MUSIC; Figure 3 (a) shows the gain diagram of MUSIC algorithm effect; Figure 3 (b) in the figure shows the gain diagram of the continuous-MUSIC algorithm effect;

[0088] Figure 4 It is a graph of the continuous-MUSIC changes within 60 seconds;

[0089] Figure 5 This is a comparison chart between ultrasonic heartbeat signal and real signal;

[0090] Figure 6 It is a diagram of the ultrasonic electrocardiogram signal recovery network structure;

[0091] Figure 7 It is the structure diagram of the convolution block. DETAILED DESCRIPTION

[0092] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present invention and implement it, but the embodiments are not intended to limit the present invention.

[0093] Reference Figure 1 As shown, Figure 1 This is a flow chart of an electrocardiogram measurement method based on an intelligent speaker ultrasonic signal provided by the present invention; specifically comprising:

[0094] S1: Continuous generation of speakers in smart speakers Different frequency ultrasonic signals are modulated by orthogonal frequency division multiplexing. The subcarrier signals of different frequencies are then transmitted to multiple target users, and the microphone in the smart speaker receives mixed subcarrier signals of different frequencies; all candidate objects are all possible reflection objects in a static scene;

[0095] Wherein, a horizontal line perpendicular to the connecting line is drawn through the midpoint of the connecting line between the speaker and the microphone, and the angle between the connecting line between each target user and the midpoint of the connecting line and the horizontal line is within a preset angle range, and each target user does not block each other; specifically, Figure 2 As shown, Figure 2 The figure shows the positional relationship between the speaker, microphone and each target user in the smart speaker in a static scene.

[0096] S2: Since the propagation delay of the subcarrier signal on the target path of different candidate objects at different times is unknown, a corresponding algorithm is required for detection; in a stationary state, the main movement of the target user is the chest movement caused by breathing and heartbeat, which is a quasi-periodic behavior that can be sensed by the ultrasonic signal and produce corresponding phase changes; among the multiple propagation paths of the subcarrier signal for each candidate object, the path with the maximum reflection intensity is taken as the target path for the subcarrier signal transmission of each candidate object;

[0097] Get the mixed subcarrier signal vector received at each sampling time within the preset time period of the microphone , establish the signal receiving matrix , 1 to The sampling moments constitute the preset period , Indicates The sampling time A mixed subcarrier signal, , representing the At the sampling moment, all candidate objects are transmitted on their corresponding target paths. The signal after the subcarrier signals are mixed; construct the At the sampling time, The first transmission on the target path corresponding to the candidate object Subcarrier signal , forming the Sampling time The subcarrier signal vector of the candidate ; Indicates the number of candidate objects;

[0098] Based on the received signal matrix, a signal covariance matrix is ​​constructed and eigen-decomposed. After obtaining the eigenvalues ​​of the signal covariance matrix and its corresponding eigenvectors, the signal subspace and the noise subspace are determined. Based on the orthogonality of the signal subspace and the noise subspace, the spectrum function at each sampling moment is constructed. The peak value of the spectrum function at each sampling moment is solved to obtain the true value of the signal propagation delay on the target path of each candidate object at each sampling moment, including:

[0099] Based on signal receiving matrix , construct the signal covariance matrix , whose expression is:

[0100] ;

[0101] in, represents the transpose of the signal receiving matrix;

[0102] The signal covariance matrix Perform characteristic decomposition and obtain the decomposition expression as follows:

[0103] ;

[0104] in, The eigenvector matrix representing the signal covariance matrix; The eigenvalue matrix representing the signal covariance matrix; represents the transpose of the eigenvector matrix of the signal covariance matrix;

[0105] Based on the number of candidate objects , we get the eigenvalues ​​of the signal covariance matrix Based on the speakers in the smart speaker, different frequency ultrasonic signals, then based on the sorting from large to small The first of the eigenvalues The eigenvectors corresponding to the eigenvalues ​​form the signal subspace; based on the remaining The eigenvectors corresponding to the eigenvalues ​​form the noise subspace;

[0106] Based on the orthogonality of the signal subspace and the noise subspace, the spectrum function at each sampling moment is constructed. The spectrum function of the sampling time The expression is:

[0107] ;

[0108] in, Indicates The turning vector of the candidate object at each sampling moment; Indicates The transpose of the steering vector of the candidate object at each sampling moment; Indicates The signal propagation delay on the target path of the candidate object at each sampling time; represents the noise subspace eigenvector matrix; represents the transpose of the noise subspace eigenvector matrix;

[0109] Through the The spectrum function of the sampling moment is analyzed to obtain the The peak value of the spectrum function corresponding to the sampling time, where each peak corresponds to a Based on the range of signal propagation delay on the target path of each candidate object at each sampling time, combined with the number of candidate objects, the first Sampling time The true value of the signal propagation delay on the target path of the candidate object , thus obtaining The turning vector of each candidate object at the sampling moment , and then get the The steering matrix composed of the steering vectors of each candidate object at the sampling time is: .

[0110] S3: According to the principle of OFDM, the frequencies of different subcarrier signals are different, and their corresponding wavelengths are also different; according to the formula , the initial phase of any signal The size of the corresponding propagation path length and wavelength The propagation path is the path of the subcarrier signal emitted from the loudspeaker, reflected by the candidate object, and then returned to the microphone; therefore, when the same microphone obtains different subcarriers, there are differences between the initial phases corresponding to each subcarrier. Therefore, it is necessary to align the initial phases of the subcarrier signal of each candidate object at each sampling moment, so as to obtain the enhanced ultrasonic signal of each candidate object at each sampling moment;

[0111] Based on the true value of the signal propagation delay on the target path of each candidate object at each sampling moment, the relative phase offsets of each subcarrier signal corresponding to each candidate object at each sampling moment and the first subcarrier signal are calculated respectively, and the steering vector of each candidate object at each sampling moment is formed, that is, the first Sampling time The candidate object corresponds to The relative phase offset between the subcarrier signal and the first subcarrier signal The expression is:

[0112] ;

[0113] No. Sampling time The turning vector of the candidate object The expression is:

[0114] ;

[0115] in, Indicates Sampling time The true value of the signal propagation delay on the target path of the candidate object; Indicates the total number of subcarrier signals; Represents the index of the subcarrier signal, ; Indicates the frequency spacing between adjacent subcarriers;

[0116] Based on the carrier shaping technology, the steering vector of each candidate object at each sampling moment, the mixed subcarrier signal vector at each sampling moment, and the noise value corresponding to each candidate object at each sampling moment are combined to calculate the noise value of each candidate object at each sampling moment. The subcarrier signals are initially phase aligned to calculate the enhanced ultrasonic signal of each candidate object at each sampling moment, i.e. Sampling time Enhanced ultrasonic signal of candidate The expression is:

[0117] ;

[0118] in, Indicates Sampling time The turning vectors of candidate objects; express Moment The noise value corresponding to each candidate object;

[0119] In summary, the enhanced ultrasonic signal of each candidate object within the preset time period can be obtained, that is, based on the The enhanced ultrasonic signals of each candidate object at the sampling moment constitute The enhanced ultrasonic signal vector at time , whose expression is:

[0120] ;

[0121] in, ; Indicates The steering matrix composed of the steering vectors of each candidate object at the sampling time, ; Indicates A noise vector composed of the noise values ​​corresponding to each candidate object at each sampling moment;

[0122] In a real scene of a specific embodiment of the present invention, the continuous MUSIC algorithm obtains the user's position effect as follows: Figure 3 As shown; Figure 3 (a) Using the standard MUSIC method, we can see a single strong peak representing a strong Line of Sight (LOS) path, indicating that other reflection path signals are covered by the strong LOS. The reflection path refers to the path that the signal takes from the transmitting source (speaker), passes through one or more reflection objects, and finally returns to the receiving device (microphone), that is, the propagation path of the signal. Figure 3In (b), the signal reflected by the chest heave at a propagation path length of ten meters can be well detected and is less affected by LOS. To further observe the performance of this method, we collect signals within 60 seconds and apply the continuous-MUSIC method. The results are shown in Figure 4 As shown, the ultrasonic phase changes caused by the user's breathing are observed, manifested as bright lines on the relevant path, proving that weak target reflections with subtle movements can be accurately detected.

[0123] S4: Based on the enhanced ultrasonic signal of each candidate object within a preset period, the spectrum amplitude ratio method is used to distinguish the living and non-living objects among all candidate objects; the phase change of the ultrasonic signal of each living object is used to distinguish the target user and the non-target user among all living objects, including:

[0124] Performing a fast Fourier transform on the enhanced ultrasonic signal of each candidate object within a preset time period, converting the enhanced ultrasonic signal of each candidate object from the time domain to the frequency domain, that is, converting the time-frequency relationship of the enhanced ultrasonic signal of each candidate object into a frequency-amplitude relationship, and obtaining the amplitude of the enhanced ultrasonic signal of each frequency of each candidate object;

[0125] Calculates the first The spectrum concentration of the candidate , whose expression is:

[0126] ;

[0127] in, Indicates The frequency of candidates is The amplitude of the enhanced ultrasonic signal; Indicates the frequency range of the preset The sum of the amplitudes of the enhanced ultrasonic signals of each frequency of the candidate objects, and the preset frequency range is ; Indicates the overall frequency range The sum of the amplitudes of the enhanced ultrasonic signals of each frequency of the candidate objects, the overall frequency range is the frequency range corresponding to the preset time period, and its range is ;

[0128] If the spectrum concentration of any candidate object is greater than the spectrum threshold, then The candidate object is a living thing, otherwise it is a non-living thing;

[0129] The true value of the signal propagation delay on the target path of each living body at each sampling moment is used as the delay of the enhanced ultrasonic signal of each living body at each sampling moment;

[0130] The distance expression is:

[0131] ;

[0132] in, Indicates Sampling time The time delay of the enhanced ultrasonic signal of a living organism; Indicates Sampling time The target path length corresponding to the enhanced ultrasonic signal of a living being; Indicates The fixed path length corresponding to the enhanced ultrasonic signal of a living being; Indicates Sampling time The length of the chest cavity fluctuation path corresponding to the enhanced ultrasonic signal of a living body; Indicates the speed of enhancing ultrasonic signal;

[0133] The phase expression is:

[0134] ;

[0135] in, Indicates Sampling time Ultrasonic phase changes of a living organism; Indicates Sampling time The wavelength of the enhanced ultrasonic signal of a living organism;

[0136] If the ultrasonic phase change of any life form at all sampling moments is less than the phase change threshold, the current life form is the target user, otherwise it is a non-target user;

[0137] In summary, the user's breathing behavior is a quasi-periodic motion, and its frequency is generally in the range of 0.1 ~ 0.5 Hz, which can be used as a unique feature to distinguish users from non-living objects. It is set to 0.75; at the same time, in order to prevent the existence of objects with similar frequency patterns in the environment, the user cannot be well distinguished from non-living objects based solely on the spectrum; since natural breathing causes millimeter-level (i.e., 5-12 mm) chest displacement, and most objects have larger scale displacements, the phase change range of the ultrasonic signal caused by chest movement can be used to identify the target user; therefore, the phase change range is set to 15 radians to filter out most of the interference objects with similar frequency patterns in the indoor environment.

[0138] S5: Using an adaptive IIR comb notch filter, extracting the ultrasonic heartbeat signal of each target user within a preset period of time, including:

[0139] By using Fourier transform, the time-frequency relationship of the enhanced ultrasonic signal of each target user in the preset time period is converted into a frequency-amplitude relationship to obtain the spectrum of each target user; based on the spectrum of each target user, the frequency of the prominent peak pointing to the fundamental frequency of the heartbeat motion of each user is found, which is recorded as , representing the The target user points to the frequency of the prominent peak of the heartbeat fundamental frequency; wherein the heartbeat fundamental frequency is within the scope of

[0140] Using adaptive IIR comb notch filter, combined with preset sampling frequency, The frequency of the prominent peak of the heartbeat fundamental frequency pointed by the target user is used to obtain the transfer function of the adaptive IIR comb notch filter, which is expressed as:

[0141] ;

[0142] in, represents a complex frequency variable; represents the feedback coefficient, ; represents the sampling frequency, represents the frequency of the signal to be eliminated, that is ;

[0143] Based on the transfer function, the first After the signal to be eliminated in the enhanced ultrasonic signal of the target user is eliminated, the first Sampling time The filtered ultrasonic signal of the target user ;

[0144] The difference between the enhanced ultrasonic signal of each target user at each sampling moment in the preset period and its filtered enhanced ultrasonic signal is used as the ultrasonic heartbeat signal of each target user at each sampling moment in the preset period, that is, Sampling time The difference between the enhanced ultrasonic signal of the target user and the filtered enhanced ultrasonic signal is taken as the Sampling time The ultrasonic heartbeat signal of the target user , whose expression is:

[0145] ;

[0146] in, Indicates the first Sampling time Enhanced ultrasonic signal of a target user;

[0147] Since the user is in a static state, CFR measurement mainly consists of two parts: breathing and heartbeat; however, the chest movement caused by breathing is greater than the movement caused by heartbeat, causing the heartbeat signal to be submerged in the breathing signal. The key point of the problem is to recover the heartbeat signal from the mixed ultrasonic signal; select the ultrasonic signal every ten seconds, and apply the above process, we can get the ultrasonic heartbeat signal; among them, the extracted heartbeat signal is as follows Figure 5 As shown, the peak value of the extracted heartbeat signal corresponds one-to-one with the real electrocardiogram in terms of timestamp, proving the feasibility of the method.

[0148] S6: inputting the ultrasonic heartbeat signal of each target user within the preset time period into the trained electrocardiogram signal prediction model, and outputting the electrocardiogram signal prediction value of each target user within the preset time period;

[0149] The electrocardiogram signal prediction model is a neural network model with multi-level feature extraction;

[0150] Reference Figure 7 As shown, Figure 7 The network structure of the neural network model for multi-level feature extraction; the neural network model for multi-level feature extraction is composed of an input layer, an output layer and multiple convolutional layers. In the encoding part, three convolutional modules and maximum pooling are used for compression; in the decoding part, convolution and upsampling are used to decode the compressed features, and finally a signal of the same length as the input signal is output, including:

[0151] Input layer, inputting the ultrasonic heartbeat signal of the current target user within a preset time period;

[0152] The first convolutional layer consists of two 1x3 convolution operations plus two ReLU nonlinear activation functions; its input is the ultrasonic heartbeat signal of the current target user within the preset time period, and its output is the features corresponding to the ultrasonic heartbeat signal of the current target user within the preset time period. ;

[0153] The first downsampling layer consists of a 1x2 maximum pooling operation, which is used to downsample the feature values; its input is the feature , the output is The features obtained after downsampling through a 1x2 maximum pooling operation ;

[0154] The second convolutional layer consists of a 1xW convolution block, a 1x3 convolution operation, and a ReLU activation function; its input is the feature , whose output is the feature ;

[0155] The second downsampling layer consists of a 1x2 maximum pooling operation, which is used to downsample the feature values; its input is the feature , the output is The features obtained after downsampling through a 1x2 maximum pooling operation ;

[0156] The third convolutional layer consists of a 1xW convolution block, a 1x3 convolution operation, and a ReLU activation function; its input is the feature , whose output is the feature ;

[0157] The third downsampling layer consists of a 1x2 maximum pooling operation, which is used to downsample the eigenvalues; its input is the eigenvalues , the output is The features obtained after downsampling through a 1x2 maximum pooling operation ;

[0158] The fourth convolutional layer consists of a 1xW convolution block, a 1x3 convolution and a ReLU activation function; its input is the feature , whose output is the feature ;

[0159] The first upsampling layer consists of a 1x2 upsampling operation to upsample the features; its input is the feature , the output is The features obtained after a 1x2 upsampling operation ;

[0160] The fifth convolutional layer consists of two 1x3 convolution operations plus two ReLU nonlinear activation functions; its input is the feature , whose output is the feature ;

[0161] The second upsampling layer consists of a 1x2 upsampling operation to upsample the features; its input is the feature , the output is The features obtained after a 1x2 upsampling operation ;

[0162] The sixth convolutional layer consists of two 1x3 convolution operations plus two ReLU nonlinear activation functions; its input is the feature , whose output is the feature ;

[0163] The third upsampling layer consists of a 1x2 upsampling operation to upsample the features; its input is the feature , the output is The features obtained after a 1x2 upsampling operation ;

[0164] The seventh convolutional layer consists of two 1x3 convolution operations plus two ReLU nonlinear activation functions; its input is the feature , whose output is the feature ;

[0165] Output layer, based on features , through a 1x1 convolution, generate and output the electrocardiogram signal of the current target user within the preset time period;

[0166] Among them, the first convolution layer, the second convolution layer and the third convolution layer, as well as the first downsampling layer, the second downsampling layer and the third downsampling layer constitute the encoder; the fourth convolution layer is the bottom convolution layer, which serves as a bridge between the encoder and the decoder; the fifth convolution layer, the sixth convolution layer and the seventh convolution layer, as well as the first upsampling layer, the second upsampling layer and the third upsampling layer constitute the decoder; the decoder part is symmetrical with the encoder and consists of multiple 1x2 upsampling layers and convolution layers; after upsampling, the input of the vth layer is obtained , through the skip connection, the feature map of the corresponding level of the encoder Splicing together helps preserve the details and edge information of the image; the connected feature map is then subjected to two 1x3 convolutions and two ReLU activation functions;

[0167] The convolution block consists of one-dimensional convolution, shortcut connection, and depth-separable convolution, such as Figure 6 As shown, it is used to learn feature information at different levels; the feature values ​​output by the downsampling layer are used as the input of the convolution block, and the data obtained by fusing the feature values ​​processed by the one-dimensional convolution and the depth-separable convolution with the feature values ​​processed by the shortcut link is used as the output of the convolution block;

[0168] Due to the sampling rates of ultrasonic signals and real electrocardiogram signals, the number of sampling points of the two signals in the same time is also different; for simplicity, the present invention chooses to use a fixed-length electrocardiogram signal to extract the electrocardiogram signal; the mean square error function (MSE) is used as the loss function of the neural network model for multi-level feature extraction, and its expression is:

[0169] ;

[0170] in, Indicates The true value of the ECG signal of samples; Indicates The predicted value of the ECG signal of samples; Indicates the total number of samples; Indicates the sample number.

[0171] go through Figure 3 (b) Figure 4 and Figure 5 The experimental results show that for medium and long distances (within 10 meters) and multi-user scenarios in two-dimensional space, the present invention has a better monitoring effect on ultrasonic heartbeat signals, with higher accuracy, faster speed, better real-time performance, and better performance; and the use of ultrasonic heartbeat signals to restore electrocardiogram signals has been proven to have good feasibility through experimental simulation.

[0172] The present invention discloses an electrocardiogram measurement method based on ultrasonic signals of smart speakers, that is, using ultrasonic waves to monitor chest fluctuation signals caused by heartbeats and separating the heartbeat signals of specific users to restore electrocardiograms. The method has the characteristics of low cost, high precision and long distance. The present invention selects Orthogonal Frequency Division Multiplexing (OFDM) as the basic scheme, and uses a microphone to receive ultrasonic signals in OFDM form. Then, the carrier shaping technology is used to constructively superimpose and enhance ultrasonic signals by compensating the phase difference between subcarriers of different frequencies received by the microphone, that is, phase compensation is performed on different subcarriers received by the same microphone to enhance the signal. Then, for the mixed situation of ultrasonic signals in multi-user scenarios, a new continuous-MUSIC (Multiple The ultrasonic signal classification method is used to separate multi-user ultrasonic signals according to the time delay and frequency phase of the ultrasonic wave, that is, to separate the signal of each candidate object from the mixed signal; then, the method based on the spectrum amplitude ratio and the difference in the phase change of the ultrasonic signal caused by breathing is used to distinguish users from non-living objects; an adaptive IIR comb notch filter is used to extract the ultrasonic heartbeat signal of the target user's chest rise and fall caused by the heartbeat, and the ultrasonic heartbeat signal of the target user is used as input. The real numerical value collected on the electrocardiogram device is used as a label, and a multi-level feature extraction neural network model is designed and trained to predict the electrocardiogram signal of the target user; thereby solving the problems of the existing electrocardiogram measurement method in detecting the user's heartbeat, such as limited perception range, high cost, inconvenient operation and inability to measure in real time.

[0173] Obviously, the above embodiments are merely examples for the purpose of clear explanation and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived therefrom are still within the scope of protection of the present invention.

Claims

1. An electrocardiogram measurement method based on intelligent speaker ultrasound, characterized in that: include: Speakers in smart speakers continuously generate Different frequency ultrasonic signals are modulated by orthogonal frequency division multiplexing. The subcarrier signals of different frequencies are then transmitted to multiple target users, and the microphone in the smart speaker receives mixed subcarrier signals of different frequencies; wherein, a horizontal line perpendicular to the connecting line is drawn through the midpoint of the connecting line between the loudspeaker and the microphone, and the angle between the connecting line between each target user and the midpoint of the connecting line and the horizontal line is within a preset angle range, and each target user does not block each other; Get the mixed subcarrier signal vector received at each sampling time within the preset time period of the microphone , establish the signal receiving matrix , Indicates the total number of sampling moments, from 1 to The sampling moments constitute the preset period. Indicates The sampling time mixed subcarrier signal, characterizing the At the sampling moment, all candidate objects are transmitted on their corresponding target paths. The signal after the subcarrier signals are mixed; construct the At the sampling time, The first transmission on the target path corresponding to the candidate object Subcarrier signal , forming the Sampling time The subcarrier signal vector of the candidate ; Indicates the number of candidate objects; Based on the received signal matrix, a signal covariance matrix is ​​constructed and eigen-decomposed. After obtaining the eigenvalues ​​of the signal covariance matrix and its corresponding eigenvectors, the signal subspace and the noise subspace are determined. Based on the orthogonality of the signal subspace and the noise subspace, the spectrum function at each sampling moment is constructed. After solving the peak value of the spectrum function at each sampling moment and obtaining the true value of the signal propagation delay on the target path of each candidate object at each sampling moment, the relative phase offsets of each subcarrier signal corresponding to each candidate object at each sampling moment and the first subcarrier signal are calculated, and a steering vector for each candidate object at each sampling moment is constructed. Based on the carrier shaping technology, the steering vector of each candidate object at each sampling moment, the mixed subcarrier signal vector at each sampling moment, and the noise value corresponding to each candidate object at each sampling moment are combined to form a steering vector for each candidate object at each sampling moment. The subcarrier signals are initially phase aligned to calculate the enhanced ultrasonic signal of each candidate object at each sampling moment, and then the enhanced ultrasonic signal of each candidate object in a preset time period is obtained; Based on the enhanced ultrasonic signal of each candidate object within the preset time period, the spectrum concentration of each candidate object is calculated by using the spectrum amplitude ratio method; if the spectrum concentration of any candidate object is greater than the spectrum threshold, the current candidate object is a living being; the true value of the signal propagation delay on the target path of each living being at each sampling moment is used as the delay of the enhanced ultrasonic signal of each living being at each sampling moment, and the ultrasonic phase change of each living being at each sampling moment is calculated by combining the path expression and the phase expression; if the ultrasonic phase change of any living being at each sampling moment is greater than the phase change threshold, the current living being is the target user; Using an adaptive IIR comb notch filter, the ultrasonic heartbeat signal of each target user within a preset time period is extracted; The ultrasonic heartbeat signal of each target user within the preset time period is input into the trained electrocardiogram signal prediction model, and the predicted value of the electrocardiogram signal of each target user within the preset time period is output.

2. The electrocardiogram measurement method based on intelligent speaker ultrasound according to claim 1 is characterized in that: The electrocardiogram signal prediction model is a neural network model for multi-level feature extraction; the neural network model for multi-level feature extraction is composed of an input layer, an output layer and multiple convolutional layers, including: Input layer, inputting the ultrasonic heartbeat signal of the current target user within a preset time period; The first convolutional layer consists of two 1x3 convolution operations plus two ReLU nonlinear activation functions; its input is the ultrasonic heartbeat signal of the current target user within the preset time period, and its output is the features corresponding to the ultrasonic heartbeat signal of the current target user within the preset time period. ; The first downsampling layer consists of a 1x2 maximum pooling operation, which is used to downsample the feature values; its input is the feature , the output is The features obtained after downsampling through a 1x2 maximum pooling operation ; The second convolutional layer consists of a 1xW convolution block, a 1x3 convolution operation, and a ReLU activation function; its input is the feature , whose output is the feature ; The second downsampling layer consists of a 1x2 maximum pooling operation, which is used to downsample the feature values; its input is the feature , the output is The features obtained after downsampling through a 1x2 maximum pooling operation ; The third convolutional layer consists of a 1xW convolution block, a 1x3 convolution operation, and a ReLU activation function; its input is the feature , whose output is the feature ; The third downsampling layer consists of a 1x2 maximum pooling operation, which is used to downsample the feature values; its input is the feature , the output is The features obtained after downsampling through a 1x2 maximum pooling operation ; The fourth convolutional layer consists of a 1xW convolution block, a 1x3 convolution, and a ReLU activation function; its input is the feature , whose output is the feature ; The first upsampling layer consists of a 1x2 upsampling operation to upsample the features; its input is the feature , the output is The features obtained after a 1x2 upsampling operation ; The fifth convolutional layer consists of two 1x3 convolution operations plus two ReLU nonlinear activation functions; its input is the feature , whose output is the feature ; The second upsampling layer consists of a 1x2 upsampling operation to upsample the features; its input is the feature , the output is The features obtained after a 1x2 upsampling operation ; The sixth convolutional layer consists of two 1x3 convolution operations plus two ReLU nonlinear activation functions; its input is the feature , whose output is the feature ; The third upsampling layer consists of a 1x2 upsampling operation to upsample the features; its input is the feature , the output is The features obtained after a 1x2 upsampling operation ; The seventh convolutional layer consists of two 1x3 convolution operations plus two ReLU nonlinear activation functions; its input is the feature , whose output is the feature ; Output layer, based on features , through a 1x1 convolution, the output feature is generated, i.e., the electrocardiogram signal of the current target user within the preset time period; Among them, the first convolution layer, the second convolution layer and the third convolution layer, as well as the first downsampling layer, the second downsampling layer and the third downsampling layer constitute the encoder; the fourth convolution layer is the bottom convolution layer, which serves as a bridge between the encoder and the decoder; the fifth convolution layer, the sixth convolution layer and the seventh convolution layer, as well as the first upsampling layer, the second upsampling layer and the third upsampling layer constitute the decoder; The convolution block is composed of one-dimensional convolution, shortcut connection and depth-wise separable convolution. The eigenvalues ​​output by the downsampling layer are used as the input of the convolution block, and the data obtained by fusing the eigenvalues ​​processed by the one-dimensional convolution and depth-wise separable convolution with the eigenvalues ​​processed by the shortcut connection are used as the output of the convolution block.

3. The electrocardiogram measurement method based on intelligent speaker ultrasound according to claim 2 is characterized in that: The loss function of the neural network model for multi-level feature extraction is the mean square error function , whose expression is: ； in, Indicates The true value of the ECG signal of samples; Indicates The predicted value of the ECG signal of samples; Indicates the total number of samples; Indicates the sample number.

4. The electrocardiogram measurement method based on intelligent speaker ultrasound according to claim 1 is characterized in that: The method comprises: constructing a signal covariance matrix based on a received signal matrix and performing eigendecomposition on the matrix to obtain the eigenvalues ​​of the signal covariance matrix and the corresponding eigenvectors thereof, and then determining the signal subspace and the noise subspace; constructing a spectrum function at each sampling moment based on the orthogonality of the signal subspace and the noise subspace; solving the peak value of the spectrum function at each sampling moment to obtain the true value of the signal propagation delay on the target path of each candidate object at each sampling moment. Based on signal receiving matrix , construct the signal covariance matrix , whose expression is: ； in, represents the transpose of the signal reception matrix; The signal covariance matrix Perform characteristic decomposition and obtain the decomposition expression as follows: ； in, The eigenvector matrix representing the signal covariance matrix; The eigenvalue matrix representing the signal covariance matrix; represents the transpose of the eigenvector matrix of the signal covariance matrix; Based on the number of candidate objects , we get the eigenvalues ​​of the signal covariance matrix Based on the speakers in the smart speaker, different frequency ultrasonic signals, then based on the sorting from large to small The first of the eigenvalues The eigenvectors corresponding to the eigenvalues ​​form the signal subspace; based on the remaining The eigenvectors corresponding to the eigenvalues ​​form the noise subspace; Based on the orthogonality of the signal subspace and the noise subspace, the spectrum function at each sampling moment is constructed. The spectrum function of the sampling time The expression is: ； in, Indicates The turning vector of the candidate object at each sampling moment; Indicates The transpose of the steering vector of the candidate object at each sampling moment; Indicates The signal propagation delay on the target path of the candidate object at each sampling time; represents the noise subspace eigenvector matrix; represents the transpose of the noise subspace eigenvector matrix; Through the The spectrum function of the sampling moment is analyzed to obtain the The peak value of the spectrum function corresponding to the sampling time, where each peak corresponds to a Based on the range of signal propagation delay on the target path of each candidate object at each sampling time, combined with the number of candidate objects, the first Sampling time The true value of the signal propagation delay on the target path of the candidate object , thus obtaining The turning vector of each candidate object at the sampling moment , and then get the The steering matrix composed of the steering vectors of each candidate object at the sampling time is: .

5. The electrocardiogram measurement method based on intelligent speaker ultrasound according to claim 1 is characterized in that: Based on the true value of the signal propagation delay on the target path of each candidate object at each sampling moment, the relative phase offsets of each subcarrier signal corresponding to each candidate object at each sampling moment and the first subcarrier signal are calculated respectively, and the steering vector of each candidate object at each sampling moment is formed, including: No. Sampling time The candidate object corresponds to The relative phase offset between the subcarrier signal and the first subcarrier signal The expression is: ； No. Sampling time The turning vector of the candidate object The expression is: ； in, Indicates Sampling time The true value of the signal propagation delay on the target path of the candidate object; Indicates the total number of subcarrier signals; Represents the index of the subcarrier signal, ; Indicates the frequency spacing between adjacent subcarriers.

6. The electrocardiogram measurement method based on intelligent speaker ultrasound according to claim 1 is characterized in that: The carrier shaping technology is based on the steering vector of each candidate object at each sampling moment, the mixed subcarrier signal vector at each sampling moment, and the noise value corresponding to each candidate object at each sampling moment. The subcarrier signals are initially phase aligned, and the enhanced ultrasonic signal of each candidate object at each sampling moment is calculated, and then the enhanced ultrasonic signal of each candidate object in the preset time period is obtained, including: No. Sampling time Enhanced ultrasonic signal of the candidate The expression is: ； in, Indicates Sampling time The turning vectors of candidate objects; express Moment The noise value corresponding to each candidate object; Based on The enhanced ultrasonic signals of each candidate object at the sampling moment constitute The enhanced ultrasonic signal vector at time , whose expression is: ； in, ; Indicates The steering matrix composed of the steering vectors of each candidate object at the sampling time, ; Indicates The noise vector is composed of the noise values ​​corresponding to each candidate object at each sampling moment.

7. The electrocardiogram measurement method based on intelligent speaker ultrasound according to claim 1 is characterized in that: The spectrum concentration of each candidate object is calculated by using the spectrum amplitude ratio method based on the enhanced ultrasonic signal of each candidate object within the preset time period; If the spectrum concentration of any candidate object is greater than the spectrum threshold, the current candidate object is a living being including: Performing a fast Fourier transform on the enhanced ultrasonic signal of each candidate object within a preset time period, converting the enhanced ultrasonic signal of each candidate object from the time domain to the frequency domain, and obtaining the amplitude of the enhanced ultrasonic signal of each frequency of each candidate object; Calculates the first The spectrum concentration of the candidate , whose expression is: ； in, Indicates The frequency of candidates is The amplitude of the enhanced ultrasonic signal; Indicates the preset frequency range. The sum of the amplitudes of the enhanced ultrasonic signals of each frequency of the candidate objects, and the preset frequency range is ; Indicates the overall frequency range The sum of the amplitudes of the enhanced ultrasonic signals of each frequency of the candidate objects, the overall frequency range is the frequency range corresponding to the preset time period, and its range is ; If the spectrum concentration of any candidate object is greater than the spectrum threshold, then The candidate object is a living thing, otherwise it is a non-living thing.

8. The electrocardiogram measurement method based on intelligent speaker ultrasound according to claim 1 is characterized in that: The actual value of the signal propagation delay on the target path of each life form at each sampling moment is used as the delay of the enhanced ultrasonic signal of each life form at each sampling moment, and the ultrasonic phase change of each life form at each sampling moment is calculated by combining the distance expression and the phase expression; if the ultrasonic phase change of any life form at each sampling moment is greater than the phase change threshold, the current life form is the target user, including: The distance expression is: ； in, Indicates Sampling time The time delay of the enhanced ultrasonic signal of a living organism; Indicates Sampling time The target path length corresponding to the enhanced ultrasonic signal of a living being; Indicates The fixed path length corresponding to the enhanced ultrasonic signal of a living being; Indicates Sampling time The length of the chest cavity fluctuation path corresponding to the enhanced ultrasonic signal of a living body; Indicates the speed of enhancing ultrasonic signals; The phase expression is: ； in, Indicates Sampling time Ultrasonic phase changes of a living organism; Indicates Sampling time The wavelength of the enhanced ultrasonic signal of a living organism; If the ultrasonic phase change of any living being at all sampling moments is less than the phase change threshold, the current living being is the target user, otherwise it is a non-target user.

9. The electrocardiogram measurement method based on intelligent speaker ultrasound according to claim 1, characterized in that: The method of extracting the ultrasonic heartbeat signal of each target user within a preset time period by using an adaptive IIR comb notch filter comprises: By using Fourier transform, the time-frequency relationship of the enhanced ultrasonic signal of each target user in the preset time period is converted into a frequency-amplitude relationship to obtain the spectrum of each target user; based on the spectrum of each target user, the frequency of the prominent peak pointing to the fundamental frequency of the heartbeat motion of each user is found, which is recorded as , representing the The target user points to the frequency of the prominent peak of the heartbeat fundamental frequency; wherein the heartbeat fundamental frequency is within the scope of Using adaptive IIR comb notch filter, combined with preset sampling frequency, The frequency of the prominent peak of the heartbeat fundamental frequency pointed by the target user is used to obtain the transfer function of the adaptive IIR comb notch filter, which is expressed as: ； in, represents a complex frequency variable; represents the feedback coefficient, ; represents the sampling frequency, represents the frequency of the signal to be eliminated, that is ; Based on the transfer function, the first After the signal to be eliminated in the enhanced ultrasonic signal of the target user is eliminated, the first Sampling time The filtered ultrasonic signal of the target user ; The difference between the enhanced ultrasonic signal of each target user at each sampling moment in the preset period and its filtered enhanced ultrasonic signal is used as the ultrasonic heartbeat signal of each target user at each sampling moment in the preset period, that is, Sampling time The difference between the enhanced ultrasonic signal of the target user and the filtered enhanced ultrasonic signal is taken as the Sampling time The ultrasonic heartbeat signal of the target user , whose expression is: ； in, Indicates the first Sampling time The enhanced ultrasonic signal of a target user.

10. The electrocardiogram measurement method based on intelligent speaker ultrasound according to claim 1, characterized in that: Among the multiple propagation paths of the subcarrier signal of each candidate object, the path where the reflection intensity reaches the maximum value is used as the target path for the transmission of the subcarrier signal of each candidate object.

Images