Respiration and heart rate signal separation method and device based on non-extraction wavelet transform

Abstract

The application discloses a method and device for separating respiratory and heart rate signals based on non-extraction wavelet transform, and the method comprises the following steps: performing distance processing on a millimeter wave radar echo signal to obtain a target distance gate complex phase time sequence signal; constructing a sample covariance matrix based on the target distance gate complex phase time sequence signal; extracting low-frequency approximation coefficients related to the respiratory signal and high-frequency detail coefficients related to the heart rate signal by using non-extraction wavelet transform; constructing a respiratory signal steering vector base matrix and a heart rate signal steering vector base matrix; constructing an estimation model based on the sample covariance matrix, the respiratory signal steering vector base matrix and the heart rate signal steering vector base matrix; obtaining corresponding spectral density functions by solving the estimation model; and searching in a frequency band based on the spectral density to output separated respiratory and heart rate spectrum graphs. The application can effectively separate the respiratory and heart rate signals under the background of the respiratory component and its harmonic interference, and improve the accuracy and stability of heart rate estimation.

Description

Technical Field

[0001] This invention belongs to the field of vital sign monitoring technology, and in particular relates to a method and device for separating respiratory and heart rate signals based on non-decimation wavelet transform. Background Technology

[0002] With the continuous development of millimeter-wave radar technology, its application in vital sign monitoring is gradually becoming more widespread. Compared to traditional contact monitoring equipment, millimeter-wave radar offers advantages such as non-contact operation, privacy protection, and all-weather monitoring, reducing measurement errors caused by factors such as discomfort during wear and poor contact. Millimeter-wave radar emits electromagnetic waves towards the target. The signal reaches the surface of the human chest cavity, generating a reflected echo, which the radar receives and processes. Because the chest cavity undergoes minute displacements with respiration and heartbeat, this displacement causes periodic changes in the echo phase. Analyzing these phase changes allows for the extraction of respiratory and heart rate signals. However, in millimeter-wave radar vital sign monitoring, the heartbeat signal is relatively weak and easily overwhelmed by the stronger respiratory signal and its harmonic interference, making it difficult to accurately locate the heart rate peak in the phase spectrum, thus affecting the accuracy of heart rate estimation. Therefore, effectively separating respiratory and heart rate signals is one of the key issues in millimeter-wave radar vital sign monitoring.

[0003] Current research on separating respiratory and heartbeat signals mainly includes three types of methods: First, separation methods based on Fourier Transform (FFT) combined with bandpass / notch filters are simple and easy to implement, but are limited by frequency resolution and window function effects. Furthermore, they are prone to spectral leakage and spurious peaks in the presence of respiratory harmonics, static clutter, and environmental disturbances, making it difficult to stably detect heart rate peaks. Second, multi-scale separation methods based on wavelet decomposition extract respiratory and heartbeat components by decomposing radar signals to different scales. However, due to the discreteness of scale and frequency band mapping and the fixed basis functions, in scenarios where the heartbeat signal amplitude is weak and adjacent to or partially overlaps with the spectrum of strong respiratory harmonics, it is often difficult to simultaneously achieve interference suppression and fine frequency resolution. Third, separation methods based on empirical mode decomposition (EMD) are data-driven and do not require preset basis functions, which can adapt to non-stationary signals to a certain extent. However, EMD lacks a unified and strict mathematical model constraint, which can easily lead to mode aliasing and endpoint effects. Especially when respiratory harmonics and heartbeat spectra are close to or overlap, the target component may be mixed into the same intrinsic mode function (IMF) or dispersed into multiple IMFs, resulting in frequency estimation bias and decreased stability.

[0004] In view of the above problems, there is an urgent need for a method to effectively suppress respiratory harmonic interference and alleviate spectral aliasing in the separation of respiratory and heart rate signals, thereby improving the accuracy and stability of heart rate estimation. Summary of the Invention The purpose of this invention is to provide a method and apparatus for separating respiratory and heart rate signals based on decimation-free wavelet transform. This invention achieves effective separation of respiratory and heart rate signals, improves the accuracy and stability of heart rate estimation, effectively suppresses respiratory harmonic interference and alleviates spectral aliasing, and also has strong anti-interference capabilities. To achieve the above objectives, this invention adopts the following technical solution: The present invention also provides a method for separating respiratory and heart rate signals based on decimation-free wavelet transform, the method comprising the following steps: Millimeter-wave radar emits electromagnetic waves toward the chest cavity of the target human body, acquires the millimeter-wave radar echo signal reflected after the electromagnetic waves reach the surface of the human chest cavity, performs range processing on the millimeter-wave radar echo signal, and obtains the target range gate complex phase timing signal. The sample covariance matrix R is constructed based on the target distance gate complex phase time series signal to characterize the statistical properties of the vital sign signal; Based on the target range gate complex phase time-series signal, a non-decimation wavelet transform is performed to extract low-frequency approximation coefficients related to the respiratory signal and high-frequency detail coefficients related to the heart rate signal. The respiratory signal steering vector basis matrix B is constructed using the low-frequency approximation coefficients related to the respiratory signal. r And constructing the heart rate signal steering vector basis matrix B using high-frequency detail coefficients related to the heart rate signal. c ; Based on the sample covariance matrix R and the basis matrix B of the respiratory signal steering vector r Heart rate signal steering vector basis matrix B c Minimum variance distortionless response spectrum estimation models for respiratory signals and heart rate signals are constructed respectively, and the corresponding spectral density functions are obtained by solving the minimum variance distortionless response spectrum estimation models for respiratory signals and heart rate signals respectively. Peak search is performed on the spectral density function within the preset respiratory and heart rate bands to separate the respiratory and heart rate frequencies, and the separated respiratory and heart rate spectra are generated. A further preferred embodiment of the above scheme, which constructs the sample covariance matrix R based on millimeter-wave radar echo signals, includes the following steps: The target range-gate complex phase time-series signal is processed by a sliding window with a window length L to obtain the sliding window data x(k) of the millimeter-wave radar echo signal, which is expressed by the following expression: , x(k) is the time-domain signal vector of the k-th window; Construct an L×L dimensional sample covariance matrix R based on the sliding window data x(k).

[0005] A further preferred embodiment of the above scheme involves extracting low-frequency approximation coefficients related to respiratory signals and high-frequency detail coefficients related to heart rate signals using non-decimation wavelet transform based on the target distance gate complex phase time-series signal. This specifically includes the following steps: The db4 wavelet basis is selected as the transform basis, and the decomposition level n and the original input signal cA0[n] are set based on the transform basis. x [n] represents the n-level decimation-free wavelet decomposition of the target range-gate complex phase time-series signal, where the db4 wavelet basis corresponds to a set of low-pass decomposition filters. h j and high-pass decomposition filter g j The low-pass filter h j A high-pass filter is used to extract the low-frequency approximate component of the target range-gate complex phase time-series signal. g j Used to extract high-frequency detail components of a signal; The initial signal cA0[n] of the target range gate complex phase timing signal is... x [n] passes through the first low-pass decomposition filter. h 1 and high-pass decomposition filter g 1. Perform convolution filtering to obtain the first layer of low-frequency approximation coefficients cA1 and the first layer of high-frequency detail coefficients cD1; continue passing them through a scale-expanded low-pass decomposition filter and a high-pass decomposition filter to obtain the second layer of low-frequency approximation coefficients cA2 and the second layer of high-frequency detail coefficients cD2; and so on. j Decomposition of the low-pass filter after scale expansion h j and high-pass decomposition filter g j For the lower frequency approximation coefficient cA of the previous layer j-1 Perform convolution to obtain the first... j Low-frequency approximation coefficient cA j [n] and the high-frequency detail coefficient cD of the j-th layer j [n], represented as: ; ; j Indicates the wavelet decomposition level. n This indicates the index of the sampling point for the current layer coefficient. m This represents the sampling point index of the lower-frequency approximation coefficient of the previous layer.

[0006] A further preferred embodiment of the above scheme utilizes low-frequency approximation coefficients related to respiratory signals to construct the respiratory signal steering vector basis matrix B. r And constructing the heart rate signal steering vector basis matrix B using high-frequency detail coefficients related to the heart rate signal. c Specifically, it includes the following steps: First, the low-frequency approximation coefficients cA obtained through j-layer decomposition are... j [n], High-frequency detail coefficients cD j [n] is subjected to a non-decimation wavelet transform to obtain the respiration-dominant time-domain reconstructed signal. x r (t) Heart rate-dominated time-domain reconstructed signal x c (t) , respectively represented as: x r (t) = IUDWT (cA5), x c (t) = IUDWT (cD3); IUDWT(·) represents the inverse decimation-free wavelet transform; Secondly, the reconstructed time-domain signal dominated by respiration x r (n) Heart rate-dominated time-domain reconstructed signal x c (n) Perform sliding window processing separately, setting the sliding window step size to S, the window length to W, and the reconstructed time-domain signal length to N; then the number of effective windows is: M = (NW) / S +1; Record No. i The sampling position at the end of each sliding window is k. i Where i = 1, 2, ..., M, for the respiratory-dominant time-domain reconstructed signal x r (n) , with k i To extract W consecutive sampling points from the end of the window, we construct the first... i The local steering vector of the first respiratory signal, the first i The local steering vectors of the heart rate signals are represented as follows: ; ; Finally, the M local steering vectors of the respiratory signals are arranged in chronological order to form the basis matrix B of the respiratory signal steering vectors.r The M local steering vectors of the heart rate signals are arranged in chronological order to form the basis matrix B of the heart rate signal steering vectors. c They respectively satisfy the following expressions: ; ; Among them, B r B c The matrix dimension is W×M.

[0007] The above scheme is further optimized, and the minimum variance distortionless response spectrum estimation models for respiratory signals and heart rate signals are constructed as follows: The minimum variance distortionless response spectrum estimation model for respiratory signals is as follows: ; Its constraints are: ; The minimum variance distortionless response spectrum estimation model for heart rate signals is as follows: ; Solving the minimum variance distortionless response spectrum estimation model of the respiratory signal yields the corresponding spectral density, which satisfies the following expression: ; ; Solve the minimum variance distortionless response spectrum estimation model of the heart rate signal to obtain the corresponding spectral density, which satisfies the following expression: ; ; in, e(f r ), e(f c ) Indicates the guide vector. f r Indicates the frequency of the respiratory signal. f c Indicates heart rate frequency. W Indicates the window length. f s Indicates the sampling frequency.

[0008] A further preferred embodiment of the above scheme involves performing a peak search on the spectral density function within a preset respiratory and heart rate frequency band to separate the respiratory and heart rate frequencies, including the following steps: Using the respiratory frequency band and heart rate frequency band as sub-bands respectively, the respiratory frequency and heart rate frequency within the frequency band of the spectral density are searched to find the respiratory frequency and heart rate frequency within each sub-band range; The maximum respiratory rate value is obtained by comparing the magnitudes of each respiratory rate and each heart rate within each sub-band range. f r and maximum heart rate f c Value, based on the maximum respiratory rate value f r ×60 is the final output respiratory rate, with the maximum heart rate as the base. f c ×60 is the final output heart rate frequency.

[0009] According to another aspect of the present invention, a respiratory and heart rate signal separation device is provided, the separation device being used to perform the respiratory and heart rate signal separation method based on decimation-free wavelet transform of the present invention, the separation device comprising: The radar signal transmitting unit is used to transmit millimeter-wave radar signals to the target detector; Radar signal receiving unit, used to receive reflected millimeter-wave radar echo signals; The signal processing module uses sliding window processing to obtain time-domain signal data from millimeter-wave radar echo signals, and calculates the covariance data of the time-domain signal data based on the time-domain signal data. The signal decomposition module is used to perform non-decimation wavelet multi-level decomposition on the target range gate complex phase time-series signal to obtain the low-frequency approximation coefficients and high-frequency detail coefficients of each level. The signal reconstruction module is used to perform wavelet transform reconstruction on the low-frequency approximation coefficients and high-frequency detail coefficients to obtain the time-domain reconstructed signal. The signal calculation module is used to calculate the spectral density of respiration and heart rate on the time-domain reconstructed signal, and search for the final respiratory and heart rate frequencies within the frequency band of the spectral density.

[0010] In another aspect, the present invention provides an electronic processing system, which may include a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of the present invention's method for separating respiratory and heart rate signals based on decimation-free wavelet transform.

[0011] In summary, due to the adoption of the above technical solutions, the present invention has the following significant effects: (1) The respiratory and heart rate signal separation method proposed in this invention utilizes the redundancy and translation invariance characteristics of UDWT to achieve multi-scale decomposition while maintaining the phase time domain continuity. The target sub-band coefficients obtained by reconstructing the decomposition are used to construct the MVDR steering vector basis, thereby improving the ability to suppress interference and noise at the covariance statistics level. This invention can effectively separate respiratory and heart rate signals under the background of strong respiratory components and their harmonic interference, and improve the accuracy and stability of heart rate estimation. (2) This invention proposes a minimum variance distortionless response algorithm based on the construction of a basis coordinate system using undecimated wavelet transform. It mainly utilizes the redundancy characteristics of undecimated wavelet transform (UDWT) to maintain the continuity of the phase in the time domain, directly transforming the wavelet coefficients into the guiding vector basis of minimum variance distortionless response (MVDR), and calculating the power spectral density to separate the respiratory and heart rate signals. This effectively suppresses respiratory harmonic interference and alleviates spectral aliasing, while also having strong anti-interference ability. Attached Figure Description

[0012] Figure 1 This is a flowchart illustrating a method for separating respiratory and heart rate signals based on non-decimation wavelet transform as described in this invention. Figure 2 This is a schematic diagram of a respiratory and heart rate signal separation device according to the present invention; Figure 3 This is a schematic diagram illustrating the mapping relationship between the UDWT decomposition levels and frequency band division of this invention; Figure 4 The waveform of the coefficients for UDWT multiscale decomposition; Figure 5 The spectrum of respiratory and heart rate is separated by traditional FFT and filter; Figure 6 This is a schematic diagram showing the results of separating respiration and heart rate according to the present invention. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of this invention clearer, preferred embodiments will be listed below with reference to the accompanying drawings to provide a clear and complete description of the invention. However, it should be noted that the examples in the specification are preferred examples, not all examples. Furthermore, many details listed in the specification are merely to provide the reader with a deep understanding of the key issues of this invention; the invention can be implemented even without these specific details.

[0014] Combination Figure 1The present invention provides a method for separating respiratory and heart rate signals based on non-decimation wavelet transform, the method comprising:

[0015] Step 1: The millimeter-wave radar emits electromagnetic waves toward the chest cavity of the target human body, and obtains the millimeter-wave radar echo signal reflected after the electromagnetic waves reach the surface of the chest cavity. The range direction of the millimeter-wave radar echo signal is processed to determine the range gate of the human target, and the complex echo sequence at the range gate of the target is extracted. The complex echo sequence is subjected to phase extraction and phase unwrapping to obtain the complex phase time sequence signal x[n] of the target range gate. Step 2: Construct a sample covariance matrix R based on the target range-gate complex phase time-series signal x[n] to characterize the statistical properties of the vital sign signal. The construction of the sample covariance matrix R based on the millimeter-wave radar echo signal x[n] includes the following steps: Using window length L to gate the complex phase timing signal of the target range x [n] Perform sliding window processing, taking the k-th sampling point as the end of the window, and extracting L consecutive sampling points to form the k-th window vector x(k), which is expressed by the following expression: , in, for 3D column vector, is the target range gate complex phase time sequence signal, L is the window length when constructing the covariance matrix, and T represents the transpose; Construct an L×L dimensional sample covariance matrix R based on multiple sliding window data x(k); ; in, K To determine the effective number of sliding windows, H The conjugate transpose is represented by R, which is an L×L dimensional sample covariance matrix used to characterize the statistical correlation of the target range gate complex phase time series signal within the local window. Step 3: Based on the target range gate complex phase time-series signal, perform decimation-free wavelet transform to extract low-frequency approximation coefficients related to the respiratory signal and high-frequency detail coefficients related to the heart rate signal. Use the low-frequency approximation coefficients related to the respiratory signal to construct the respiratory signal steering vector basis matrix B. r And constructing the heart rate signal steering vector basis matrix B using high-frequency detail coefficients related to the heart rate signal. c In this invention, based on the target range gate complex phase time-series signal, a non-decimation wavelet transform is performed to extract low-frequency approximation coefficients related to respiratory signals and high-frequency detail coefficients related to heart rate signals. Specifically, the steps include: Step 31: Select the db4 wavelet basis as the transform basis, and set the decomposition level n and the original input signal cA0[n] based on the transform basis.x [n] represents the n-level decimation-free wavelet decomposition of the target range-gate complex phase time-series signal, where the db4 wavelet basis corresponds to a set of low-pass decomposition filters. h j and high-pass decomposition filter g j The low-pass filter h j A high-pass filter is used to extract the low-frequency approximate component of the target range-gate complex phase time-series signal. g j Used to extract high-frequency detail components of a signal; Step 32, initialize the target range gate complex phase timing signal cA0[n] = x [n] passes through the first low-pass decomposition filter. h 1 and high-pass decomposition filter g 1. Perform convolution filtering to obtain the first layer of low-frequency approximation coefficients cA1 and the first layer of high-frequency detail coefficients cD1; continue passing them through a scale-expanded low-pass decomposition filter and a high-pass decomposition filter to obtain the second layer of low-frequency approximation coefficients cA2 and the second layer of high-frequency detail coefficients cD2; and so on. j Decomposition of the low-pass filter after scale expansion h j and high-pass decomposition filter g j For the previous level approximation coefficient cA j-1 Perform convolution to obtain the first... j Low-frequency approximation coefficient cA j [n] and the high-frequency detail coefficient cD of the j-th layer j [n], represented as: ; ; in ，j Indicates the wavelet decomposition level. n This indicates the index of the sampling point for the current layer coefficient. m Indicates the sampling point index of the lower-frequency approximation coefficient of the previous layer; Because a non-decimation wavelet transform is used, downsampling is not performed during each decomposition layer, therefore cA j [n] and cD j The length of [n] is consistent with the original input signal, the nth j The frequency band range of each subsequent layer of the low-frequency approximation coefficients is [0, f]. s / 2 j+1 ], No. j The bandwidth of the high-frequency detail coefficients of the layer is: [ f s / 2 j+1 ,f s / 2 j In this embodiment of the invention, the Daubechies wavelet basis db4 is used to implement a 5-level undecimated wavelet decomposition (UDWT decomposition). After the 5-level decomposition, the low-frequency approximation coefficients cA5 of the 5th level and the high-frequency detail coefficients cD1, cD2, cD3, cD4, and cD5 of the 1st to 5th levels are obtained, which satisfy the following: ; ; Finally, the low-frequency approximation coefficients cA5 of the respiratory-dominant layer and the high-frequency detail coefficients cD3 of the heart rate-dominant layer are retained, while other irrelevant layer coefficients are discarded to reduce subsequent computational complexity. Since this invention employs decimation-free wavelet transform, downsampling is not performed during each layer decomposition process; instead, multi-scale analysis is achieved by scaling the filters. Therefore, the wavelet coefficients of each layer maintain the same length as the original input signal, thus ensuring the temporal phase continuity of the respiratory signal correlation coefficient and the heart rate signal correlation coefficient. The respiratory signal steering vector basis matrix B is constructed using the low-frequency approximation coefficients related to the respiratory signal. r And constructing the heart rate signal steering vector basis matrix B using high-frequency detail coefficients related to the heart rate signal. c Specifically, it includes the following steps: First, the low-frequency approximation coefficients cA obtained through j-layer decomposition are... j [n], High-frequency detail coefficients cD j [n] is subjected to inverse decimation-free wavelet transform to obtain the reconstructed breathing-dominant time-domain signal. x r (t) Heart rate-dominated time-domain reconstructed signal x c (t) The extracted correlation coefficients of the respiratory signal and the heart rate signal are reconstructed in the time domain using single sub-bands. Specifically, all non-decimated wavelet decomposition coefficients except for the target sub-band are set to zero, retaining only the low-frequency approximation coefficients cA5 of the respiratory-dominant layer or the high-frequency detail coefficients cD3 of the heart rate-dominant layer. Then, the inverse non-decimated wavelet reconstruction process corresponding to the non-decimated wavelet decomposition is used to map the wavelet domain coefficients back to the time domain through a reconstruction filter bank, obtaining the respiratory-dominant time domain signal and the heart rate-dominant time domain signal, respectively, expressed as: x r (t) = IUDWT (cA5), x c (t) = IUDWT (cD3); IUDWT(·) represents the inverse decimation-free wavelet transform; Secondly, the reconstructed time-domain signal dominated by respiration x r (n) Heart rate-dominated time-domain reconstructed signal x c (n) Perform sliding window processing separately, setting the sliding window step size to S, the window length to W, and the reconstructed time-domain signal length to N; then the number of effective windows is: M = (NW) / S +1; Record No. i The sampling position at the end of each sliding window is k. i k i The corresponding sampling positions are denoted as i=1,2,...,M, for the respiratory-dominant time-domain reconstructed signal. x r (n) , with k i To extract W consecutive sampling points from the end of the window, we construct the first... i The local steering vector of the first respiratory signal, the first i The local steering vectors of the heart rate signals are represented as follows: ; ; Finally, the M local steering vectors of the respiratory signals are arranged in chronological order to form the basis matrix B of the respiratory signal steering vectors. r The M local steering vectors of the heart rate signals are arranged in chronological order to form the basis matrix B of the heart rate signal steering vectors. c In this process, a sliding window is applied to the reconstructed time-domain signal. The window length is W=258, and the step size is S=32 points. Respiratory signal steering vector sequences and heart rate signal steering vector sequences are constructed. Sliding windows are applied to both the respiratory-dominant and heart rate-dominant time-domain signals, resulting in a column vector of length W for each window. Each steering vector is a W×1 dimensional column vector, which serves as a local steering vector. All respiratory steering vectors are arranged in chronological order to form a respiratory basis matrix, and all heart rate steering vectors are arranged in chronological order to form a heart rate basis matrix. Therefore, if there are M effective windows, each basis matrix consists of M W×1 dimensional steering vectors, with a matrix dimension of W×M. The basis matrices satisfy the following expressions: ; ; Among them, B r B c The matrix dimension is W×M; Step 4, based on the sample covariance matrix R and the respiratory signal steering vector basis matrix B r Heart rate signal steering vector basis matrix B c Minimum variance distortionless response spectrum estimation models for respiratory and heart rate signals were constructed separately. The corresponding spectral density functions were obtained by solving these models. The goal was to minimize the total power of the output signal while ensuring the target signal passes through without distortion, with the constraint that the target signal components in the basis matrix remain unchanged. The constructed minimum variance distortionless response spectrum estimation models (MVDR optimization models) for respiratory and heart rate signals satisfy the following conditions: ; Its constraints are: ; The minimum variance spectral density of the respiratory signal solution satisfies: ; ; The minimum variance distortionless response spectrum estimation model (MVDR optimization model) associated with the heart rate signal is constructed and satisfies the following: ; ; ; in, e(f r ), e(f c ) Indicates the guide vector. f r Indicates the frequency of the respiratory signal. f c Indicates heart rate frequency. W Indicates the window length. f s This indicates the sampling frequency; the guide vector frequency is scanned in the range of 0-12.5 Hz, with the respiratory frequency band focusing on analysis of 0.1-0.5 Hz and the heart rate frequency band focusing on analysis of 0.8-2.0 Hz; Step 5: Within the preset respiratory and heart rate frequency bands, perform peak search on the spectral density function to separate the respiratory and heart rate frequencies, and generate the separated respiratory and heart rate spectra. Specifically, this includes the following steps: using the respiratory and heart rate frequency bands as sub-bands, search for the respiratory and heart rate frequencies within each sub-band range of the spectral density function; compare the magnitudes of each respiratory frequency and each heart rate frequency within each sub-band range to obtain the maximum respiratory frequency value. f rand maximum heart rate f c The value is output directly when the spectral peaks are clearly identifiable; when interference exists, the basis matrix parameters are automatically adjusted and recalculated; the maximum respiratory rate value is used as the basis matrix value. f r ×60 is the final output respiratory rate, with the maximum heart rate as the base. f c ×60 is used as the final output heart rate frequency; specifically, within the respiratory frequency range of 0.1–0.5 Hz, the MVDR spectral density function of the respiratory signal is... P r ( f Perform a search to determine the spectral density P. r ( f The frequency at which the maximum value is obtained is taken as the respiratory rate. f r Within the heart rate frequency band of 0.8–2.0 Hz, the MVDR spectral density function of the heart rate signal... P c ( f To perform a search, the spectral density will be determined. P c ( f The frequency at which the maximum value is obtained is taken as the heart rate frequency. f c The output of the spectrum search is the respiratory rate. f r and heart rate f c . In this invention, such as Figure 2 As shown, this invention provides a respiratory and heart rate signal separation device. The separation device is used to execute the respiratory and heart rate signal separation method based on decimation-free wavelet transform of this invention; the specific execution steps are not detailed here. The separation device includes: The radar signal transmitting unit is used to transmit millimeter-wave radar signals to the target detector; Radar signal receiving unit, used to receive reflected millimeter-wave radar echo signals; The signal processing module is used to perform sliding window processing on the millimeter-wave radar echo signal to obtain time-domain signal data, and to calculate the covariance data of the time-domain signal data based on the time-domain signal data. The signal processing module is used to perform range-direction processing on the millimeter-wave radar echo signal to obtain the target range-gate complex phase time-series signal, then perform sliding window processing on the target range-gate complex phase time-series signal to obtain time-domain signal data, and calculate the covariance data of the time-domain signal data based on the time-domain signal data. The signal decomposition module is used to perform non-decimation wavelet multi-level decomposition on the target range gate complex phase time-series signal to obtain the low-frequency approximation coefficients and high-frequency detail coefficients of each level. The signal reconstruction module is used to perform wavelet transform reconstruction on the low-frequency approximation coefficients and high-frequency detail coefficients to obtain the time-domain reconstructed signal. The signal calculation module is used to calculate the spectral density of respiration and heart rate on the time-domain reconstructed signal, and search for the final respiratory and heart rate frequencies within the frequency band of the spectral density.

[0016] Each module in the separation device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0017] In this invention, an electronic processing system according to an embodiment of the invention may include a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of the present invention's method for separating respiratory and heart rate signals based on decimation-free wavelet transform. The memory and the processor communicate with each other. The processor may be a central processing unit (CPU), a network processor (NP), etc.; the signal processor may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), or a field-programmable gate array (FPGA). The processor can call logic instructions in memory to execute the following computer program to perform the following methods: receiving reflected millimeter-wave radar echo signals; performing range-direction processing on the millimeter-wave radar echo signals to obtain target range-gate complex phase time-series signals; constructing a sample covariance matrix R based on the target range-gate complex phase time-series signals to characterize the statistical properties of the vital sign signals; performing decimation-free wavelet transform on the target range-gate complex phase time-series signals to extract low-frequency approximation coefficients related to respiratory signals and high-frequency detail coefficients related to heart rate signals; and constructing a respiratory signal steering vector basis matrix B using the low-frequency approximation coefficients related to respiratory signals. r And constructing the heart rate signal steering vector basis matrix B using high-frequency detail coefficients related to the heart rate signal. c Based on the sample covariance matrix R and the basis matrix B of the respiratory signal steering vector.r Heart rate signal steering vector basis matrix B c Minimum variance distortionless response spectrum estimation models (MVDR optimization models) for respiratory and heart rate signals are constructed separately. The corresponding spectral density functions are obtained by solving these models. Peak search is performed on the spectral density functions within preset respiratory and heart rate frequency bands to separate the respiratory and heart rate frequencies, generating the separated respiratory and heart rate spectra. The logical instructions in the memory can be implemented as software functional units and stored in a computer-readable storage medium. Each software functional unit includes several instructions to cause a computer device (such as a personal computer, server, or network device) to execute all or part of the steps of the respiratory and heart rate signal separation method described in this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks, and may also include non-volatile memory (NVDM). Volatile Memory (NVM) or Flash memory. The device embodiments described above are merely illustrative. 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, i.e., they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0018] In this invention, a 77 GHz millimeter-wave radar vital sign monitoring platform is specifically implemented. The radar front-end uses a TI AWR1843 chip with a working center frequency of 77 GHz, a bandwidth of 4 GHz, and a sampling rate of 10 MHz. After range processing, target range gate selection, and phase extraction, a target range-gated complex phase slow time series for vital sign analysis is obtained. In this embodiment, a mapping relationship between the UDWT decomposition hierarchy and frequency band division is established at the sampling rate of this slow time series; for example... Figure 3 As shown. From Figure 3 From this, we can obtain the low-frequency approximation coefficients cA5 for the fifth layer of the respiratory signal and the high-frequency detail coefficients cD3 for the third layer of the heart rate signal. Among these, Figure 4 The image shows the waveforms of the scale coefficients extracted from the original complex phase signal after 5-level UDWT decomposition. Figure 4 (a) is the target range gate complex phase time sequence signal, which contains mixed components such as breathing, heartbeat, noise and breathing harmonics; Figure 4 (b) is the low-frequency dominant respiratory component extracted by the low-frequency approximation coefficient cA5 of the 5th layer of UDWT, which mainly corresponds to the respiratory frequency band of 0.1–0.5Hz; Figure 4 (c) shows the high-frequency dominant heart rate component extracted using the UDWT layer 3 high-frequency detail coefficients cD3, primarily corresponding to the heart rate frequency band of 0.8–2.0 Hz. Figure 4 A comparison of the three figures (a, b, and c) shows that UDWT can separate the low-frequency respiratory component and the high-frequency weak heartbeat component in the original complex phase time series, providing input for the subsequent construction of the respiratory and heart rate steering vector basis matrix. Figure 4 It can be seen that, thanks to the non-extraction characteristic of UDWT, the extracted low-frequency approximation coefficients of the 5th layer and the high-frequency detail coefficients of the 3rd layer perfectly preserve the original phase alignment relationship, while successfully separating the low-frequency respiratory component from the weak high-frequency heartbeat component, effectively avoiding the mode aliasing problem that is easy to occur in traditional empirical mode decomposition. When the respiratory frequency range or heart rate frequency range changes, the number of layers of UDWT decomposition is adjusted accordingly, not limited to the fifth layer or the third layer.

[0019] In this embodiment of the invention, the spectra and accuracy of respiration and heart rate obtained by the traditional FFT algorithm and the UDWT-MVDR algorithm of the present invention are compared, specifically as follows: Figure 5 and Figure 6 As shown. Figure 5 This shows the results of spectral analysis of respiratory and heartbeat signals using the traditional FFT method. Among them, the upper... Figure 5 (a) is the FFT spectrum of the respiratory signal. Figure 5 (b) is the FFT spectrum of the heartbeat signal. Figure 5 It can be seen that the respiratory signal, due to its larger amplitude, can form a more obvious main peak in the spectrum; while the heartbeat signal has a weaker amplitude, with multiple interference peaks and spectral fluctuations in the heart rate frequency band, and the true heart rate peak is not prominent enough. This phenomenon indicates that, under the influence of respiratory harmonics, noise, and limited frequency resolution, it is difficult to achieve high-precision separation of respiratory and heart rate signals using the traditional FFT transform method. Figure 6 This is a comparison chart showing the results of separating and estimating the respiratory and heart rate signals using the UDWT-MVDR method of this invention, and also a chart showing the final separation experimental results of this invention; wherein, Figure 6 (a) is the MVDR spectrum of the respiratory signal. Figure 6 (b) is the MVDR spectrum of the heart rate signal. Figure 6(c) A comparison of the detected frequency and the true frequency; In the traditional FFT spectrum, the amplitude of chest cavity displacement caused by respiration is much greater than that of the heartbeat signal, resulting in strong respiratory high-order harmonics filling the spectrum. These harmonics leak and alias severely into the high frequencies, directly masking the weak true heartbeat frequency, making conventional methods prone to misjudgment and missed detection. In stark contrast, the UDWT-MVDR algorithm of this invention achieves a fundamental improvement in separation performance. It transforms the high-purity wavelet coefficients extracted by UDWT into steering vectors, ensuring that the target physiological characteristics pass through without distortion while adaptively forming deep nulls at interference frequencies, thereby maximally suppressing respiratory harmonics and background noise. Compared with the traditional Fourier transform method, the respiratory and heart rate signal separation method of this invention improves frequency resolution under limited observation time conditions, making the originally adjacent or partially aliased respiratory fundamental frequency and weak heart rate peaks present clearer and more concentrated spectral peaks in the spectrum. Further combined with Figure 6 The results of the "Frequency Detection Accuracy Comparison" show that the detected peak values ​​extracted using this method are highly consistent with the true frequencies. Specifically, the detection error for respiratory frequency is only 0.0002 Hz, and the detection error for heart rate is as low as 0.0013 Hz. Experimental verification demonstrates that in a 77 GHz millimeter-wave radar measured dataset, this invention can effectively separate respiration and heart rate while exhibiting strong anti-interference capabilities. This extremely small quantization error data illustrates that the UDWT-MVDR algorithm of this invention not only solves the problem of weak heart rate signals being submerged by harmonics, but also possesses extremely high accuracy and robustness in the separation and precise measurement of vital signs.

[0020] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of protection of the present invention.

Claims

1. A method for separating respiratory and heart rate signals based on decimation-free wavelet transform, characterized in that: The method for separating respiratory and heart rate signals includes the following steps: Millimeter-wave radar emits electromagnetic waves toward the chest cavity of the target human body, acquires the millimeter-wave radar echo signal reflected after the electromagnetic waves reach the surface of the human chest cavity, performs range processing on the millimeter-wave radar echo signal, and obtains the target range gate complex phase timing signal. The sample covariance matrix R is constructed based on the target distance gate complex phase time series signal to characterize the statistical properties of the vital sign signal; Based on the target range gate complex phase time-series signal, a non-decimation wavelet transform is performed to extract low-frequency approximation coefficients related to the respiratory signal and high-frequency detail coefficients related to the heart rate signal. The respiratory signal steering vector basis matrix B is constructed using the low-frequency approximation coefficients related to the respiratory signal. r And constructing the heart rate signal steering vector basis matrix B using high-frequency detail coefficients related to the heart rate signal. c ; Based on the sample covariance matrix R and the basis matrix B of the respiratory signal steering vector r Heart rate signal steering vector basis matrix B c Minimum variance distortionless response spectrum estimation models for respiratory signals and heart rate signals are constructed respectively, and the corresponding spectral density functions are obtained by solving the minimum variance distortionless response spectrum estimation models for respiratory signals and heart rate signals respectively. Within the preset respiratory and heart rate frequency bands, peak search is performed on the spectral density function to separate the respiratory and heart rate frequencies, and the separated respiratory and heart rate spectra are generated.

2. The method for separating respiratory and heart rate signals based on decimation-free wavelet transform according to claim 1, characterized in that: Constructing the sample covariance matrix R based on millimeter-wave radar echo signals includes the following steps: The target range-gate complex phase time-series signal is processed by a sliding window with a window length L to obtain the sliding window data x(k) of the millimeter-wave radar echo signal, which is expressed by the following expression: ， x(k) is the time-domain signal vector of the k-th window; Construct an L×L dimensional sample covariance matrix R based on the sliding window data x(k).

3. The method for separating respiratory and heart rate signals based on decimation-free wavelet transform according to claim 1, characterized in that: Based on the target distance gate complex phase time-series signal, a non-decimation wavelet transform is performed to extract low-frequency approximation coefficients related to respiratory signals and high-frequency detail coefficients related to heart rate signals. The specific steps include the following: The db4 wavelet basis is selected as the transform basis, and the decomposition level n and the original input signal cA0[n] are set based on the transform basis. x [n] represents the n-level decimation-free wavelet decomposition of the target range-gate complex phase time-series signal, where the db4 wavelet basis corresponds to a set of low-pass decomposition filters. h j and high-pass decomposition filter g j The low-pass filter h j A high-pass filter is used to extract the low-frequency approximate component of the target range-gate complex phase time-series signal. g j Used to extract high-frequency detail components of a signal; The initial signal cA0[n] of the target range gate complex phase timing signal is... x [n] passes through the first low-pass decomposition filter. h 1 and high-pass decomposition filter g 1. Perform convolution filtering to obtain the first layer of low-frequency approximation coefficients cA1 and the first layer of high-frequency detail coefficients cD1; continue passing them through a scale-expanded low-pass decomposition filter and a high-pass decomposition filter to obtain the second layer of low-frequency approximation coefficients cA2 and the second layer of high-frequency detail coefficients cD2; and so on. j Decomposition of the low-pass filter after scale expansion h j and high-pass decomposition filter g j For the lower frequency approximation coefficient cA of the previous layer j-1 Perform convolution to obtain the first... j Low-frequency approximation coefficient cA j [n] and the high-frequency detail coefficient cD of the j-th layer j [n], represented as: ； ； j Indicates the wavelet decomposition level. n This indicates the index of the sampling point for the current layer coefficient. m This represents the sampling point index of the lower-frequency approximation coefficient of the previous layer.

4. A method for separating respiratory and heart rate signals based on decimation-free wavelet transform according to claim 1 or 3, characterized in that: The respiratory signal steering vector basis matrix B is constructed using low-frequency approximation coefficients associated with the respiratory signal. r And constructing the heart rate signal steering vector basis matrix B using high-frequency detail coefficients related to the heart rate signal. c， Specifically, the steps include the following: First, the low-frequency approximation coefficients cA obtained through j-layer decomposition are... j [n], High-frequency detail coefficients cD j [n] is subjected to inverse decimation-free wavelet transform to obtain the reconstructed breathing-dominant time-domain signal. x r (t) Heart rate-dominated time-domain reconstructed signal x c (t) , respectively represented as: x r (t) = IUDWT (cA5), x c (t) = IUDWT (cD3) IUDWT(·) represents the inverse decimation-free wavelet transform; Secondly, the reconstructed time-domain signal dominated by respiration x r (n) Heart rate-dominated time-domain reconstructed signal x c (n) Perform sliding window processing separately, setting the sliding window step size to S, the window length to W, and the reconstructed time-domain signal length to N; then the number of effective windows is: M = (NW) / S +1; Record No. i The sampling position at the end of each sliding window is k. i Where i = 1, 2, ..., M, for the respiratory-dominant time-domain reconstructed signal x r (n) , with k i To extract W consecutive sampling points from the end of the window, we construct the first... i The local steering vector of the first respiratory signal, the first i The local steering vectors of the heart rate signals are represented as follows: ； ； Finally, the M local steering vectors of the respiratory signals are arranged in chronological order to form the basis matrix B of the respiratory signal steering vectors. r The M local steering vectors of the heart rate signals are arranged in chronological order to form the basis matrix B of the heart rate signal steering vectors. c They respectively satisfy the following expressions: ； ； Among them, B r B c The matrix dimension is W×M.

5. The method for separating respiratory and heart rate signals based on decimation-free wavelet transform according to claim 1, characterized in that: The minimum variance distortionless response spectrum estimation models for respiratory and heart rate signals are constructed as follows: The minimum variance distortionless response spectrum estimation model for respiratory signals is as follows: ； Its constraints are: ; The minimum variance distortionless response spectrum estimation model for heart rate signals is as follows: ； Solving the minimum variance distortionless response spectrum estimation model of the respiratory signal yields the corresponding spectral density, which satisfies the following expression: ； ； ； Solve the minimum variance distortionless response spectrum estimation model of the heart rate signal to obtain the corresponding spectral density, which satisfies the following expression: ； ； ； in, e(f r ), e(f c ) Indicates the guide vector. f r Indicates the frequency of the respiratory signal. f c Indicates heart rate frequency. W Indicates the window length. f s Indicates the sampling frequency.

6. The method for separating respiratory and heart rate signals based on decimation-free wavelet transform according to claim 1, characterized in that: The peak search of the spectral density function within the preset respiratory and heart rate bands to separate the respiratory and heart rate frequencies includes the following steps: Using the respiratory frequency band and heart rate frequency band as sub-bands respectively, the respiratory frequency and heart rate frequency within the frequency band of the spectral density are searched. The maximum respiratory rate value is obtained by comparing the magnitudes of each respiratory rate and each heart rate within each sub-band range. f r and maximum heart rate f c Value, based on the maximum respiratory rate value f r ×60 is the final output respiratory rate, with the maximum heart rate as the base. f c ×60 is the final output heart rate frequency.

7. A device for separating respiratory and heart rate signals, characterized in that: The separation device is used to perform the respiratory and heart rate signal separation method based on non-decimation wavelet transform as described in any one of claims 1 to 6, and the separation device comprises: The radar signal transmitting unit is used to transmit millimeter-wave radar signals to the target detector; Radar signal receiving unit, used to receive reflected millimeter-wave radar echo signals; The signal processing module is used to perform range-direction processing on the millimeter-wave radar echo signal to obtain the target range-gate complex phase time-series signal, then perform sliding window processing on the target range-gate complex phase time-series signal to obtain time-domain signal data, and calculate the covariance data of the time-domain signal data based on the time-domain signal data. The signal decomposition module is used to perform non-decimation wavelet multi-level decomposition on the target range gate complex phase time-series signal to obtain the low-frequency approximation coefficients and high-frequency detail coefficients of each level. The signal reconstruction module is used to perform wavelet transform reconstruction on the low-frequency approximation coefficients and high-frequency detail coefficients to obtain the time-domain reconstructed signal. The signal calculation module is used to calculate the spectral density of respiration and heart rate on the time-domain reconstructed signal, and search for the final respiratory and heart rate frequencies within the frequency band of the spectral density.

8. An electronic processing system may include a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the method for separating respiratory and heart rate signals based on decimation-free wavelet transform as described in any one of claims 1 to 5.

Images