Non-contact sleep vital signs detection method based on sector distributed radar
By distributing multiple independent radars on the semicircle for baseband signal processing and weighted fusion, the problems of contactless radar in sleep detection accuracy and signal-to-noise ratio are solved, and high accuracy and low-cost home detection are achieved. Popular to ordinary families.
Patent Information
- Application Number
- CN202210532471.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-05-12
- Publication Date
- 2024-12-20
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
When detecting human sleep vital signs, the measurement accuracy is limited, especially when sleeping posture changes, the signal-to-noise ratio is reduced. Traditional contact equipment is bulky, costly, poor user experience, and it is difficult to popularize it into the home environment.
Using a contactless sleep vital sign detection method based on a sector-shaped distributed radar, multiple independent single-send and single-receive radars are uniformly distributed on the semicircle, baseband signals are obtained and body movement recognition, imbalance calibration, demodulation, Respiratory rate and heart rate separation and weighted fusion improve detection accuracy.
It realizes the detection of breathing rate and heart rate during sleep with high accuracy under low cost and small equipment, avoiding the inconvenience of experience and insufficient accuracy of traditional contact equipment, and is suitable for popularization in home environments.
Smart Images
Figure CN114983382B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of distributed radar, and in particular to a non-contact sleep vital sign detection method based on sector-shaped distributed radar. Background Art
[0002] Traditionally, physiological parameters have been measured using traditional contact-based measurement equipment in hospitals. The data is used for clinical diagnosis and analysis to assess a person's health status. While these traditional measurement devices offer high accuracy, they are generally bulky, expensive, and require complex and cumbersome operation. With increasing health awareness, the demand for health monitoring is increasing, and a wide range of portable products for monitoring vital signs have emerged, such as smartwatches, smart bracelets, and wristband-type pulse and blood pressure monitors. These new portable devices are non-invasive and relatively safe, making them suitable for daily health monitoring at home. However, their contact-based nature has limited their widespread adoption: User experience is poor, particularly among the elderly, who find them inconvenient and reluctant to wear them daily. They are also difficult for users with burns or sensitive skin to use. Furthermore, the contact electrodes themselves can affect the user's physiological state, interfering with the measurement process and thus compromising the accuracy of vital sign parameters.
[0003] The use of non-contact methods to detect sleep vital signs has become a general trend. Non-contact radar can meet people's needs for home health monitoring and continuous testing, and has broad application prospects. Non-contact radar transmits electromagnetic waves to illuminate the target and receives echoes to obtain information such as the target's distance, direction, movement speed, and amplitude. The human body's breathing and heartbeat will cause the chest to rise and fall regularly, and these physiological characteristics will be modulated in the reflected electromagnetic waves. Non-contact medical detection radar transmits electromagnetic waves and performs down-conversion, filtering, and demodulation on the received echoes. It extracts and separates useful physiological information (primarily including breathing and heartbeat signals) from the echoes, thereby monitoring changes in human physiological sign parameters.
[0004] Currently, relatively accurate bioradars are generally expensive and large, typically found in specialized laboratories and unsuitable for home use. Conventional, low-cost radars are suitable for home health monitoring. However, most current research on the application of conventional, low-cost radars for detecting human vital signs relies on single or dual radars, resulting in limited measurement accuracy. Furthermore, changes in sleeping position during sleep can further reduce the signal-to-noise ratio (SNR) of radar measurements. Therefore, driven by the need to maintain a low cost and make them suitable for household use, improving the accuracy of non-contact radar vital sign detection has become a current research hotspot. Summary of the Invention
[0005] The purpose of the present invention is to solve the above-mentioned defects in the prior art and provide a non-contact sleep vital signs detection method based on sector-shaped distributed radar.
[0006] The purpose of the present invention can be achieved by taking the following technical solutions:
[0007] A non-contact sleep vital sign detection method based on sector-shaped distributed radar, the vital sign detection method comprising the following steps:
[0008] S1. Obtain the baseband signal output by the distributed radar, identify and eliminate the turning motion signal; wherein, the distributed radar has a total of N radars, where N ≥ 4 and N is an even number, fixed on a semicircle with a diameter of 1m, with the midpoint of the semicircle as the boundary, N / 2 radars on the left and N / 2 radars on the right, and the N / 2 radars on each side are evenly distributed on 1 / 4 of the circle, and the radars on the left and right sides are symmetrically arranged; each radar is a single-transmitter and single-receiver radar, and will not interfere with each other; each radar works independently and will simultaneously output I and Q baseband signals, and the N radars output a total of 2N baseband signals; the signal acquisition system collects the baseband signals output by the radar and wirelessly transmits them to the host computer via a WiFi module.
[0009] The baseband signal acquisition process is as follows:
[0010] When the radar system starts working, the radar transmitting antenna automatically emits electromagnetic waves T(t); when the transmitted electromagnetic wave T(t) hits the surface of the human chest, the breathing and heartbeat movement of the human chest will modulate the phase of T(t), and then generate an echo signal R(t), which is automatically received by the receiving antenna of the radar module;
[0011] When the echo signal R(t) is received by the radar receiving antenna, it first passes through a low-noise amplifier and then is split into two signals. These two signals are mixed with the radar local oscillator signal and the local oscillator signal with a 90° phase shift, respectively, to obtain two mixed signals. The two mixed signals are then low-pass filtered and then amplified to obtain the I and Q baseband signals. Each radar operates independently and outputs both I and Q baseband signals simultaneously. N radars output a total of 2N baseband signals.
[0012] The signal acquisition system collects the baseband signals output by the radar and transmits them wirelessly to the host computer via the WiFi module. The signal acquisition system has 2N acquisition channels and can simultaneously collect 2N baseband signals at a sampling frequency of 500Hz. After receiving the collected 2N baseband signals, the host computer performs subsequent digital signal processing.
[0013] The turning body motion signal is caused by turning body motion during sleep, which increases the amplitude and frequency of the baseband signal;
[0014] S2. Perform imbalance calibration and demodulation on the baseband signal output by the distributed radar to obtain N demodulated signals;
[0015] S3, separating the respiratory signal and the heartbeat signal from the N demodulated signals, and solving for N respiratory rates and N heart rates;
[0016] S4. Perform weighted fusion on the respiratory rate and heart rate detection results respectively to obtain a fused respiratory rate and a fused heart rate.
[0017] Furthermore, the process of step S1 is as follows:
[0018] S101, obtaining a 2N roadbed band signal output by a distributed radar, and performing windowing processing on the 2N roadbed band signal, where the added window is a rectangular window;
[0019] S102, preprocessing the obtained 2N baseband signals respectively and obtaining the product of the signal short-time zero-crossing rate and the short-time energy of the baseband signals; wherein the preprocessing process includes smoothing filtering and DC component removal; the signal short-time zero-crossing rate refers to the number of times a signal sampling point changes from a positive number to a negative number, or from a negative number to a positive number within 1 second; and the short-time energy refers to the energy of the signal within 1 second;
[0020] S103, perform body turning motion signal recognition, the recognition process is as follows: for the 2N roadbed band signal, calculate the product of the short-time zero-crossing rate and the short-time energy X i , where i is the serial number of the 2N roadbed signal, i = 1, 2, ..., 2N, and find the maximum value of the product max(X i ), a total of 2N maximum values are obtained; then the maximum value is continued to be obtained from the 2N maximum values to obtain maxFinal, and it is determined whether maxFinal exceeds a given threshold; if it exceeds, it is determined that a turning movement has occurred; otherwise, no turning movement has occurred;
[0021] Among them, the maxFinal value is used instead of using one of the max(X i ) value to identify the flipping action is that a single max(X i ) value may not be sensitive enough to the human body turning over movement, and may be easily affected by the human sleeping posture, resulting in incorrect recognition; while using the maxFinal value to identify the turning over movement can make full use of the radar output baseband signal information at multiple positions and improve the recognition accuracy.
[0022] S104: If a body turning movement occurs, directly slide the window to the next time period and do not execute subsequent steps S2 to S4; if no body turning movement occurs, continue to execute steps S2 to S4.
[0023] Among them, if a body rollover occurs, it means that the signal within the window time is greatly interfered with and the respiratory rate and heart rate cannot be accurately detected. Therefore, it is necessary to slide the window to the next time period to extract the respiratory rate and heart rate of the next time period.
[0024] Furthermore, the process of step S2 is as follows:
[0025] S201, performing Butterworth filtering on the I and Q baseband signals output by a single radar in the distributed radar to reduce noise;
[0026] Among them, Butterworth filtering has good flatness within the passband and is suitable for processing the baseband signal output by the radar;
[0027] S202, performing imbalance calibration on the I and Q baseband signals after noise reduction in step S201;
[0028] Under ideal conditions, the I and Q baseband signals are orthogonal. However, during radar measurement, stationary objects other than human targets can reflect the radar's transmitted signal. This, combined with hardware errors, can introduce DC offset, amplitude imbalance, and phase imbalance, collectively referred to as IQ imbalance. When IQ imbalance occurs, the shape of the curve in a coordinate graph, with the I baseband signal as the independent variable and the Q baseband signal as the dependent variable, can be viewed as part of an elliptical curve. Directly demodulating the unbalanced I and Q baseband signals will result in significant errors in the demodulated signals, necessitating imbalance calibration.
[0029] The imbalance calibration process is as follows:
[0030] First, perform elliptic curve fitting on the input I and Q baseband signals to calculate the amplitude imbalance factor A e , phase imbalance factor φ e 、I channel DC offset B I , Q-path DC offset B Q , and then use Schmidt orthogonalization to correct the baseband signal. Let the I and Q baseband signals before correction be After correction, they are I orth , Q orth , the correction process can be expressed as:
[0031]
[0032] S203: Perform arc tangent demodulation on the I and Q baseband signals after imbalance calibration. The process is as follows:
[0033] Assuming the demodulated signal is θ(t), the inverse tangent demodulation calculation formula is as follows:
[0034]
[0035] S204, phase-continuing the demodulated signal to obtain the phase-continuing demodulated signal; wherein, the reason for phase continuity is: the value range of the inverse tangent function in step S203 is However, the actual phase may exceed this value range, resulting in discontinuous phase of the demodulated signal; therefore, phase continuity is required to obtain a complete demodulated signal;
[0036] S205. Repeat steps S201 to S204 until the I and Q baseband signals of all radars in the distributed radar are demodulated to obtain N demodulated signals.
[0037] Furthermore, the process of step S3 is as follows:
[0038] S301, downsampling the obtained N demodulated signals so that the sampling frequency of the demodulated signals becomes a signal with a sampling frequency greater than 10 Hz and less than 20 Hz; the demodulated signal frequency after downsampling is greater than 10 Hz and less than 20 Hz, which can ensure that the required information of the respiratory signal and the heartbeat signal is intactly preserved and the complexity of subsequent calculations can be reduced;
[0039] S302. Perform wavelet decomposition and reconstruction on the N downsampled demodulated signals to obtain N respiratory signals and N heartbeat signals. The process is as follows: first, perform wavelet decomposition on the demodulated signals to obtain wavelet coefficients at each layer. Then, based on the frequency range of the respiratory and heartbeat frequencies, wavelet coefficients within appropriate frequency bands are selected to reconstruct the respiratory and heartbeat signals. Ultimately, the N demodulated signals can be reconstructed into N respiratory signals and N heartbeat signals. Wavelet decomposition and reconstruction fully utilize the multi-resolution characteristics of the wavelet transform to reconstruct respiratory and heartbeat signals with good waveform quality.
[0040] S303. Calculate N respiratory frequencies (hereinafter referred to as respiratory rates) from the reconstructed N respiratory signals using an autocorrelation function method. The process is as follows: Calculate the autocorrelation function of the respiratory signal, and find the time interval between the maximum peak and the second peak of the autocorrelation function. This time interval is the period of the respiratory signal. Take the inverse of the period to calculate the respiratory rate.
[0041] Among them, the amplitude of the respiratory signal is larger than that of the heartbeat signal, the respiratory rate detection is relatively easy, and the calculation process using the autocorrelation function method is relatively simple.
[0042] S304. Calculate N heartbeat frequencies (hereinafter referred to as heart rate) using discrete Fourier transform, spectrum interpolation, and spectrum peak finding on the reconstructed N heartbeat signals. The process is as follows: perform discrete Fourier transform on the heartbeat signal to obtain a spectrum of the heartbeat signal. Interpolate the spectrum of the heartbeat signal to obtain the frequency point corresponding to the spectrum peak, and calculate the heart rate.
[0043] Among them, the amplitude of the heartbeat signal is relatively weak compared to the respiratory signal. The discrete Fourier transform method can improve the detection accuracy, and further interpolation of the spectrum can improve the frequency resolution of the heartbeat signal spectrum.
[0044] Furthermore, the weighted fusion process of the respiratory rate detection results in step S4 is as follows:
[0045] For each of the N autocorrelation functions of the N respiratory signals obtained in step S303, the ratio between the second peak and the maximum peak is calculated, and this ratio is defined as ratio_b. Then, the N ratio_b values are compared with a given threshold, and the respiratory signal with a ratio_b value greater than the given threshold is selected. Assume that n respiratory signals are selected, where n <= N, and the maximum amplitude of the mth respiratory signal among the n respiratory signals is bA m , the weight coefficient is bW m , m=1,2,…,n, then the calculation formula of the weight coefficient of the mth respiratory signal is as follows:
[0046]
[0047] The respiratory rate fb corresponding to the above n respiratory signals m Perform weighted fusion to obtain the fused respiratory rate fb_fusion, which is calculated as follows:
[0048]
[0049] Among them, the purpose of selecting n channels from N respiratory signals is to avoid the respiratory rate detection results corresponding to the respiratory signals with poor signal quality from causing large errors in the subsequent fused respiratory rate, while the purpose of weighted fusion is to make full use of the respiratory rate detection results of each radar to improve the accuracy of respiratory rate detection.
[0050] Furthermore, the weighted fusion process of the heart rate detection results in step S4 is as follows:
[0051] For the N heartbeat signals obtained in step S302, the corresponding N autocorrelation functions are obtained, and the ratio between the second peak and the maximum peak of each of the N autocorrelation functions is determined. This ratio is defined as ratio_h. The N ratio_h values are then compared with a given threshold, and the heartbeat signal with a ratio_h value greater than the given threshold is selected. Assume that k heartbeat signals are selected, where k <= N. The maximum amplitude of the jth heartbeat signal among the k heartbeat signals is denoted as hA. j , the weight coefficient is hW j , j = 1, 2, ..., k, then the calculation formula of the weight coefficient of the j-th heartbeat signal is as follows:
[0052]
[0053] The heart rate fh corresponding to the above k-channel heartbeat signal j Perform weighted fusion to obtain the fused heart rate fh_fusion, which is calculated as follows:
[0054]
[0055] Among them, the purpose of selecting k heartbeat signals from N heartbeat signals is to avoid the heart rate detection results corresponding to heartbeat signals with poor signal quality from causing large errors in the subsequent fused heart rate, while the purpose of weighted fusion is to make full use of the heart rate detection results of each radar to improve the accuracy of heart rate detection.
[0056] The present invention has the following advantages and effects compared to the prior art:
[0057] 1) Based on the modulation effect of human breathing and heartbeat vibrations on radar electromagnetic waves, the present invention performs body movement recognition, baseband signal demodulation, wavelet decomposition and reconstruction, respiratory rate and heart rate detection and other processing on the radar output baseband signal, thereby extracting the respiratory rate and heart rate, realizing non-contact detection of respiratory rate and heart rate, and avoiding the problems of poor user experience and inconvenience brought by traditional contact detection equipment.
[0058] 2) The present invention uses a sector-shaped distributed radar to obtain the detection results of multiple radars on the human body's respiratory rate and heart rate during sleep. Through screening and weighted fusion, it avoids the problems of being easily interfered by clutter, easily affected by sleeping posture, and having large detection errors when using only a single radar, thereby improving the accuracy of non-contact radar respiratory rate and heart rate detection. Distributed radars are arranged in multiple locations to detect the human body from multiple angles. When a certain radar is greatly interfered by clutter, the signal selection process in this method can eliminate the detection results with greater interference, and then the detection results of other radars can be weighted fused to obtain the final result; when the human body's sleeping posture changes, the detection effect of the radar facing the back of the human body may be poor, while the detection effect of the radar facing the front of the human body remains at a good level. At this time, through the screening and weighted fusion methods in this method, a relatively accurate detection effect can still be obtained. Therefore, this method has good detection capabilities under various human sleeping postures.
[0059] 3) The radar used in the present invention is low-cost and small in size, and is suitable for popularization in home environments; the proposed method has strong real-time performance and high practicality. BRIEF DESCRIPTION OF THE DRAWINGS
[0060] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:
[0061] Figure 1 This is a flow chart of a method for detecting sleep vital signs based on a sector-shaped distributed radar disclosed in an embodiment of the present invention;
[0062] Figure 2 1 is a layout diagram of a sector-shaped distributed radar according to an embodiment of the present invention;
[0063] Figure 3 is a system framework diagram in an embodiment of the present invention;
[0064] Figure 4 This is a flow chart of the identification and elimination of sleep tossing motion signals according to an embodiment of the present invention;
[0065] Figure 5 is a flow chart of imbalance calibration and demodulation in an embodiment of the present invention;
[0066] Figure 6 is a flow chart of detecting respiratory rate and heart rate in an embodiment of the present invention;
[0067] Figure 7 12 is a waveform diagram of 12 baseband signals output by the distributed radar in an embodiment of the present invention;
[0068] Figure 81 is a graph showing the product of the short-time zero-crossing rate and the short-time energy of each of the 12 baseband signals when no body rolling movement occurs in an embodiment of the present invention;
[0069] Figure 9 1 is a graph showing the product of the short-time zero-crossing rate and the short-time energy of each of the 12 baseband signals when a body rollover occurs in an embodiment of the present invention;
[0070] Figure 10 4 is a waveform diagram of 6-channel demodulated signals in an embodiment of the present invention. DETAILED DESCRIPTION
[0071] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.
[0072] Example 1
[0073] This embodiment discloses a method for detecting sleep vital signs based on sector-shaped distributed radar. Figure 1 The specific steps are as follows:
[0074] S1. Obtain the baseband signal output by the distributed radar, identify and eliminate the turning motion signal;
[0075] In this embodiment, the distributed radar system comprises six radars, mounted on a semicircle with a diameter of 1 meter. The front ends of three radars on the left and three radars on the right are separated by the midpoint of the semicircle. The angles between the three radars on each side and the direction of chest breathing movement when a person lies supine are 20 degrees, 40 degrees, and 60 degrees, respectively. The radars on the left and right sides are arranged symmetrically. Each radar is a single-transmitter, single-receiver radar, preventing interference between them.
[0076] Figure 2 This is the layout diagram of the sector-shaped distributed radar.
[0077] The baseband signal acquisition process is as follows:
[0078] When the radar system begins operating, the radar's transmitting antenna automatically emits electromagnetic waves T(t). When these waves strike the surface of the human chest, the chest's breathing and heartbeat movements modulate the phase of T(t), generating an echo signal R(t), which is automatically received by the radar module's receiving antenna.
[0079] When the echo signal R(t) is received by the radar receiving antenna, it first passes through a low-noise amplifier and is then split into two signals. These two signals are mixed with the radar local oscillator signal and a 90° phase-shifted local oscillator signal, respectively, to produce two mixed signals. These two mixed signals are then low-pass filtered and amplified to produce the I and Q baseband signals. Each radar operates independently and simultaneously outputs both I and Q baseband signals. The six radars output a total of 12 baseband signals.
[0080] The signal acquisition system then collects the radar's baseband signals and transmits them wirelessly to a host computer via a WiFi module. The system has 12 acquisition channels, capable of simultaneously collecting 12 baseband signals at a 500Hz operating frequency. After receiving these 12 baseband signals, the host computer performs subsequent digital signal processing.
[0081] The turning over movement signal is caused by the turning over movement during sleep, which increases the amplitude and frequency of the baseband signal.
[0082] Figure 3 It is the system framework diagram. Figure 3 The layout of multiple radars should be fan-shaped, refer to Figure 2 The linear layout is drawn here to facilitate the explanation of the entire system.
[0083] Figure 4 Flowchart for identifying and eliminating body movement signals during sleep.
[0084] S101, obtaining 12 baseband signals output by the distributed radar and performing windowing processing on them;
[0085] In this embodiment, a 20-second rectangular window is used to intercept the signal, and a 20-second baseband signal is processed at a time;
[0086] S102. Preprocess the 12 obtained baseband signals respectively and obtain the product of the signal short-time zero-crossing rate and short-time energy of the baseband signals; wherein the preprocessing process includes smoothing filtering and DC component removal; the signal short-time zero-crossing rate refers to the number of times the signal sampling point changes from a positive number to a negative number, or from a negative number to a positive number within 1 second; and the short-time energy refers to the energy of the signal within 1 second.
[0087] S103, perform body turning motion recognition, the recognition process is as follows: calculate the product X of the short-time zero-crossing rate and the short-time energy of the 12 baseband signals respectively i , where i is the serial number of the 12-way baseband signal, and find the maximum value of the product max(X i), obtaining a total of 12 maximum values. The maximum of these 12 maximum values is then calculated to obtain maxFinal. It is then determined whether maxFinal exceeds a given threshold. If so, a rollover is determined to have occurred; otherwise, no rollover has occurred.
[0088] The reason for continuing to find the maximum value maxFinal for the 12 maximum values is that when the body turns over, the radars at different positions, max(X i ) differ greatly. For example, when a person rolls over from a supine position to the left, the body's movement displacement relative to the three radars on the left is smaller, resulting in less Doppler shift; while the body produces a "fast approach, then fast distance" movement relative to the three radars on the right, generating a violent baseband signal, that is, a signal with high amplitude and energy, significantly higher than the three radars on the left. Similarly, when a person rolls over from a supine position to the right, the three radars on the left produce signals with high amplitude and energy, significantly higher than the three radars on the right. Based on this characteristic, for a fixed threshold, the obtained maxFinal can better distinguish whether a rollover movement has occurred;
[0089] S104: If a body turning movement occurs, directly slide the window to the next time period and do not execute steps S2 to S4; if no body turning movement occurs, continue to execute the subsequent processing steps S2 to S4.
[0090] S2. Perform imbalance calibration and demodulation on the baseband signal output by the distributed radar to obtain 6 demodulated signals; Figure 5 The flowchart of imbalance calibration and demodulation of baseband signals.
[0091] S201. Use a Butterworth filter to perform low-pass filtering and noise reduction on the I and Q baseband signals output by a single radar in the distributed radar. The Butterworth filter has excellent flatness within the passband and is suitable for processing baseband signals output by the radar.
[0092] In this embodiment, the order of the Butterworth filter is set to 3, and the low-pass cutoff frequency is set to 10 Hz, which reduces noise without affecting the subsequent detection of respiratory rate and heart rate.
[0093] S202, performing imbalance calibration on the I and Q baseband signals after noise reduction in step S201;
[0094] Under ideal conditions, the I and Q baseband signals are orthogonal. However, during radar detection, in addition to human targets, other stationary objects can reflect the radar's transmitted signal. This, combined with errors in the hardware itself, can introduce DC bias, amplitude imbalance, and phase imbalance, collectively referred to as IQ imbalance. After IQ imbalance, the shape of the curve in the coordinate graph with the I baseband signal as the independent variable and the Q baseband signal as the dependent variable can be viewed as part of an elliptical curve. Directly demodulating the unbalanced I and Q baseband signals will result in large errors in the demodulated signals, necessitating imbalance calibration. The imbalance calibration process is as follows:
[0095] First, perform elliptic curve fitting on the input I and Q baseband signals to calculate the amplitude imbalance factor A e , phase imbalance factor φ e And I channel DC offset B I , Q-path DC offset B Q ; Then use Schmidt orthogonalization to correct the baseband signal. Let the I and Q baseband signals before correction be After correction, they are I orth , Q orth , the correction process can be expressed as:
[0096]
[0097] S203, performing inverse tangent demodulation on the I and Q baseband signals after imbalance calibration. The process is as follows: Assuming the demodulated signal is θ(t), the inverse tangent demodulation calculation formula is as follows:
[0098]
[0099] In this embodiment, the purpose of the arc tangent demodulation is to restore the real human chest motion signal, that is, the demodulated signal, to facilitate the subsequent extraction of the respiratory rate and heart rate.
[0100] S204, performing phase continuity on the demodulated signal to obtain a phase-continuous demodulated signal;
[0101] Phase continuity refers to the phase continuity of the phase discontinuity problem that may occur in step S203, and the restoration of the original phase information. Phase discontinuity is because the range of the inverse tangent function is However, the actual phase may exceed this value range, so phase continuity is required to obtain a complete demodulated signal;
[0102] S205. Repeat steps S201 to S204 until the I and Q baseband signals of all radars in the distributed radar are demodulated to obtain 6 demodulated signals.
[0103] S3, separating the respiratory signal and the heartbeat signal from the 6 demodulated signals, and solving for 6 respiratory rate and 6 heart rate values;
[0104] Figure 6 Flowchart for detecting respiratory rate and heart rate.
[0105] S301, downsampling the demodulated signal so that the sampling frequency of the demodulated signal becomes a signal greater than 10 Hz and less than 20 Hz, which can ensure that the required information of the respiratory and heartbeat signals can be intactly preserved and reduce the complexity of subsequent calculations;
[0106] In this embodiment, the sampling frequency after downsampling is 16 Hz.
[0107] S302. Wavelet decomposition and reconstruction are performed on the downsampled six demodulated signals to obtain six respiratory signals and six heartbeat signals. The process is as follows: first, wavelet decomposition is performed on the demodulated signals to obtain wavelet coefficients at each layer. Then, wavelet coefficients within the appropriate frequency band are selected based on the frequency range of the respiratory and heartbeat frequencies to reconstruct the respiratory and heartbeat signals. Ultimately, the six demodulated signals can be reconstructed into six respiratory signals and six heartbeat signals. Wavelet decomposition and reconstruction can fully utilize the multi-resolution characteristics of the wavelet transform to reconstruct the respiratory and heartbeat signals with good waveform quality.
[0108] S303. Calculate six respiratory frequencies (hereinafter referred to as respiratory rates) from the reconstructed six respiratory signals using an autocorrelation function method. The process is as follows: Calculate the autocorrelation function of the respiratory signal, and find the time interval between the maximum peak and the second peak of the autocorrelation function. This time interval is the period of the respiratory signal. Take the inverse of the period to calculate the respiratory rate.
[0109] Among them, the amplitude of the respiratory signal is larger than that of the heartbeat signal, the respiratory rate detection is relatively easy, and the calculation process using the autocorrelation function method is relatively simple.
[0110] S304. Calculate six heartbeat frequencies (hereinafter referred to as heart rates) from the reconstructed six-channel heartbeat signals using a discrete Fourier transform, spectrum interpolation, and spectrum peak finding method. The process is as follows: Perform a discrete Fourier transform on the heartbeat signal to obtain a spectrum of the heartbeat signal. Interpolate the spectrum of the heartbeat signal to obtain the frequency point corresponding to the spectrum peak, and then calculate the heart rate.
[0111] The heartbeat signal is weaker than the respiratory signal, so the discrete Fourier transform method can improve the detection accuracy, and further interpolation of the spectrum can improve the frequency resolution of the heartbeat signal spectrum. In this embodiment, cubic spline is selected for interpolation.
[0112] S4, respectively, weighted fusion of the respiratory rate and heart rate detection results. In step S4 of this embodiment, the weighted fusion process of the respiratory rate detection results is as follows:
[0113] For the six autocorrelation functions of the six respiratory signals obtained in step S303, the ratio between the second peak and the maximum peak is calculated, and the ratio is defined as ratio_b. Then, the six ratio_b values are compared with a given threshold, and the respiratory signal with a ratio_b value greater than the given threshold is selected. Assume that n respiratory signals are selected, where n <= 6, and the maximum amplitude of the mth respiratory signal among the n respiratory signals is recorded as bA m , the weight coefficient is bW m , m=1,2,…,n, then the calculation formula of the weight coefficient of the mth respiratory signal is as follows:
[0114]
[0115] The respiratory rate fb corresponding to the above n respiratory signals m Perform weighted fusion to obtain the fused respiratory rate fb_fusion, which is calculated as follows:
[0116]
[0117] Among them, the purpose of selecting n channels from the 6 respiratory signals is to avoid the respiratory rate detection results corresponding to the respiratory signals with poor signal quality from causing large errors in the subsequent fused respiratory rate, while the purpose of weighted fusion is to make full use of the respiratory rate detection results of each radar to improve the accuracy of respiratory rate detection.
[0118] In this embodiment, the weighted fusion process of the heart rate detection results in step S4 is as follows:
[0119] Calculate the corresponding six autocorrelation functions for the six heartbeat signals obtained in step S302, and calculate the ratio between the second peak and the maximum peak of each of the six autocorrelation functions. Define this ratio as ratio_h. Then compare the N ratio_h values with a given threshold and select the heartbeat signal with a ratio_h value greater than the given threshold. Suppose k heartbeat signals are selected, where k <= 6. Let the maximum amplitude of the jth heartbeat signal among the k heartbeat signals be hA. j , the weight coefficient is hW j , j = 1, 2, ..., k, then the calculation formula of the weight coefficient of the j-th heartbeat signal is as follows:
[0120]
[0121] The heart rate fh corresponding to the above k-channel heartbeat signal jPerform weighted fusion to obtain the fused heart rate fh_fusion, which is calculated as follows:
[0122]
[0123] Among them, the purpose of selecting k channels from the 6 heartbeat signals is to avoid the heart rate detection results corresponding to the heartbeat signals with poor signal quality from causing large errors in the subsequent fusion heart rate, while the purpose of weighted fusion is to make full use of the heart rate detection results of each radar to improve the accuracy of heart rate detection.
[0124] Example 2
[0125] Based on the non-contact sleep vital sign detection method based on sector-shaped distributed radar disclosed in Example 1, this example further provides an implementation process of the non-contact sleep vital sign detection method based on sector-shaped distributed radar. The specific steps are as follows:
[0126] S1. Based on Example 1, 12 baseband signals output by the distributed radar are obtained and windowed with a window length of 20s. The obtained signals are as follows: Figure 7 The product of the short-time zero-crossing rate and the short-time energy of the 12-channel baseband signals is obtained respectively. The result of the product is as follows Figure 8 shown.
[0127] In this embodiment, the judgment threshold of the turning over movement is set to 0.01.
[0128] Referring to the steps in Example 1, the maxFinal value is calculated to be 5×10 -4 , is less than the given threshold 0.01, then no turning over movement occurs, and the subsequent steps S2 to S4 can be continued.
[0129] If a body rollover occurs, the product of the short-time zero-crossing rate and the short-time energy of the baseband signal is as follows: Figure 9 As shown, the calculated maxFinal value is 0.045, which is obviously greater than the given threshold 0.01. In this case, the respiratory signal and the heartbeat signal cannot be accurately extracted. Therefore, the window is directly slid to the next time period, and the subsequent S2 to S4 processing steps are no longer executed.
[0130] S2. Based on Example 1, the obtained baseband signal is subjected to imbalance calibration, and then demodulated and phase-continuousized. The obtained 6-channel demodulated signals are as follows: Figure 10 shown.
[0131] Among them, the baseband signals output by radars 2 to 6 can be demodulated well, while the demodulation effect of radar 1 is not ideal. The reason may be that the radar is interfered by strong clutter, resulting in a poor signal-to-noise ratio of the baseband signal, which in turn affects the final demodulation result.
[0132] S3. Referring to the corresponding steps in Example 1, the respiratory signal and the heartbeat signal are separated from the six demodulated signals, and the respiratory rate and heart rate are calculated. The calculation results are shown in Table 1:
[0133] Table 1. Detection results of heart rate and respiratory rate in Example 2
[0134] Radar number Respiratory rate (times / minute) Heart rate (beats / minute) one 8.4 70.8 two 16.2 75.6 three 15.7 76.5 Four 15.8 75 five 15.7 81 six 16.3 69
[0135] S4. Refer to the corresponding steps in Example 1 and perform weighted fusion on the respiratory rate and heart rate detection results respectively.
[0136] Among them, the comparison between the fusion respiration rate and the reference value is shown in Table 2:
[0137] Table 2. Comparison table of fusion respiration rate and reference value in Example 2
[0138] Fusion respiratory rate (times / minute) Reference value (times / minute) Absolute error 15.8 16 0.2
[0139] The comparison between fused heart rate and reference value is shown in Table 3:
[0140] Table 3. Comparison table of fused heart rate and reference value in Example 2
[0141] Fusion heart rate (beats / min) Reference value (times / min) Absolute error 75.8 76 0.2
[0142] It can be seen that the errors between the fused respiratory rate and the fused heart rate and the reference values are relatively small, and the test results can meet the needs of health testing.
[0143] The comparison between the respiratory rate detected by each radar and the reference value is shown in Table 4:
[0144] Table 4. Comparison table of respiratory rates detected by each radar in Example 2 and reference values
[0145]
[0146] The comparison between the heart rate detection results of each radar and the reference value is shown in Table 5:
[0147] Table 5. Comparison table of heart rates detected by each radar separately and reference values in Example 2
[0148]
[0149] Compared to the reference values, the errors of individual radars varied considerably, with some performing better and others worse. The errors in the fused respiration rate and heart rate obtained using this method were both relatively small, and compared to single radars, they were able to reduce the detection errors of respiration rate and heart rate to a certain extent.
[0150] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be considered as equivalent replacement methods and are included in the scope of protection of the present invention.
Claims
1. A non-contact sleep vital sign detection method based on sector distributed radar, characterized in that: The vital signs detection method comprises the following steps: S1. Obtain the baseband signal output by the distributed radar, identify and eliminate the turning motion signal; wherein, the distributed radar has a total of N radars, wherein N≥4 and N is an even number, fixed on a semicircular ring with a diameter of 1m, with the midpoint of the semicircular ring as the boundary, N / 2 radars on the left, N / 2 radars on the right, the N / 2 radars on each side are evenly distributed on the 1 / 4 circle, and the layout of the radars on the left and right sides is symmetrical; each radar is a single-transmit and single-receive radar, and will not interfere with each other; each radar works independently, and will output I and Q baseband signals at the same time, and N radars output 2N baseband signals in total; the signal acquisition system collects the baseband signals output by the radar and transmits them wirelessly to the host computer through the WiFi module; wherein, the turning motion signal is caused by the turning motion during sleep; the process of step S1 is as follows: S101, obtaining a 2N roadbed band signal output by a distributed radar, and performing windowing processing on the 2N roadbed band signal, wherein the added window is a rectangular window; S102, preprocessing the obtained 2N baseband signals respectively and obtaining the product of the signal short-time zero-crossing rate and short-time energy of the baseband signals; wherein the preprocessing process includes smoothing filtering and removing direct current components; the signal short-time zero-crossing rate refers to the number of times the signal sampling point changes from a positive number to a negative number, or from a negative number to a positive number within 1 second; and the short-time energy refers to the energy of the signal within 1 second; S103, performing body turning motion signal recognition, the recognition process is as follows: for the 2N roadbed band signals, the product of the short-time zero-crossing rate and the short-time energy is calculated respectively. i , where i is the serial number of the 2N roadbed band signal, i=1,2,…,2N, and find the maximum value of the product max(X i ), a total of 2N maximum values are obtained; then the maximum value is continued to be obtained from the 2N maximum values to obtain maxFinal, and it is determined whether maxFinal exceeds a given threshold; if it exceeds, it is determined that a turning movement has occurred; otherwise, a turning movement has not occurred; S104, if a body turning movement occurs, directly slide the window to the next time period, and do not execute subsequent steps S2 to S4; if no body turning movement occurs, continue to execute steps S2 to S4; S2, performing imbalance calibration and demodulation on the baseband signal output by the distributed radar to obtain N demodulated signals; S3, separating the breathing signal and the heartbeat signal of the N demodulated signals, and solving for N breathing rates and N heart rates; S4. Perform weighted fusion on the respiratory rate and heart rate detection results respectively to obtain fused respiratory rate and fused heart rate.
2. The non-contact sleep vital sign detection method based on sector-shaped distributed radar according to claim 1 is characterized in that: The process of step S2 is as follows: S201, performing Butterworth filtering and noise reduction on the I and Q baseband signals output by a single radar in the distributed radar; S202, performing imbalance calibration on the I and Q baseband signals after noise reduction in step S201, the imbalance calibration process is as follows: First, perform elliptic curve fitting on the input I and Q baseband signals to obtain the amplitude imbalance factor A. e , Phase imbalance factor 、I-channel DC offset , Q-channel DC offset Then, the baseband signal is corrected by Schmidt orthogonalization. The I and Q baseband signals before correction are assumed to be , , after correction, they are , , the correction process can be expressed as: Formula (1) S203, performing arc tangent demodulation on the I and Q baseband signals after the imbalance calibration to obtain demodulated signals, the process is as follows: Assume the demodulated signal is , then the inverse tangent demodulation calculation formula is as follows: Formula (2) S204, performing phase continuity on the demodulated signal to obtain a phase-continuous demodulated signal; S205. Repeat the above steps S201 to S204 until the I and Q baseband signals output by all radars in the distributed radar are demodulated to obtain N demodulated signals.
3. The non-contact sleep vital sign detection method based on sector-shaped distributed radar according to claim 1, characterized in that: The process of step S3 is as follows: S301, down-sampling the obtained N demodulated signals respectively, so that the sampling frequency of the demodulated signals becomes a signal with a sampling frequency greater than 10 Hz and less than 20 Hz; S302, performing wavelet decomposition and reconstruction on the down-sampled N-channel demodulated signals respectively to obtain N-channel respiratory signals and N-channel heartbeat signals, the process is as follows: firstly performing wavelet decomposition on the demodulated signal to obtain wavelet coefficients of each layer, and then selecting wavelet coefficients in a suitable frequency band according to the frequency range of the respiratory frequency and the heartbeat frequency, reconstructing the respiratory signal and the heartbeat signal, and the N-channel demodulated signals can finally reconstruct the N-channel respiratory signals and the N-channel heartbeat signals; S303, using the autocorrelation method to calculate N respiratory frequencies (hereinafter referred to as respiratory rate) for the reconstructed N respiratory signals, the process is as follows: calculating the autocorrelation function of the respiratory signal, finding the time interval between the maximum peak point and the second peak point of the autocorrelation function, the time interval is the period of the respiratory signal, taking the inverse of the period to find the respiratory rate; S304. For the reconstructed N heartbeat signals, use discrete Fourier transform, spectrum interpolation, and spectrum peak finding methods to calculate N heartbeat frequencies, hereinafter referred to as heart rate. The process is as follows: perform discrete Fourier transform on the heartbeat signal to obtain the heartbeat signal spectrum, interpolate the heartbeat signal spectrum, obtain the frequency point corresponding to the spectrum peak, and calculate the heart rate.
4. The non-contact sleep vital sign detection method based on sector-shaped distributed radar according to claim 3 is characterized in that: The weighted fusion process of the respiratory rate detection results in step S4 is as follows: The ratio between the second peak and the maximum peak of the N autocorrelation functions of the N respiratory signals obtained in step S303 is respectively calculated, and the ratio is defined as ratio_b. Then, the N ratio_b values are compared with a given threshold, and a respiratory signal whose ratio_b value is greater than the given threshold is selected. Assuming that n respiratory signals are selected, where n<=N, the maximum amplitude of the mth respiratory signal among the n respiratory signals is recorded as , the weight coefficient is , m=1,2,…,n, then the calculation formula of the weight coefficient of the mth respiratory signal is as follows: Formula (3) The respiratory rate corresponding to the above n respiratory signals Perform weighted fusion to obtain the fused respiratory rate , the calculation formula is as follows: Formula (4).
5. The non-contact sleep vital sign detection method based on sector-shaped distributed radar according to claim 3 is characterized in that: The weighted fusion process of the heart rate detection results in step S4 is as follows: For the N heartbeat signals obtained in step S302, the corresponding N autocorrelation functions are obtained, and the ratio between the second peak and the maximum peak is respectively obtained for the N autocorrelation functions, and the ratio is defined as ratio_h. Then, the N ratio_h values are compared with a given threshold, and the heartbeat signal whose ratio_h value is greater than the given threshold is selected. Suppose k heartbeat signals are selected, where k<=N, and the maximum amplitude of the j-th heartbeat signal among the k heartbeat signals is recorded as , the weight coefficient is , j=1,2,…,k, then the calculation formula of the weight coefficient of the j-th heartbeat signal is as follows: Formula (5) The heart rate corresponding to the above k-channel heartbeat signal Perform weighted fusion to obtain the fused heart rate , the calculation formula is as follows: Official(6).