Robust respiration sensing method and system based on multi-dimensional wi-fi signals

Abstract

The application belongs to the technical field of wireless signal sensing, and discloses a robust respiration sensing method and system based on multi-dimensional Wi-Fi signals. The sensing method comprises the following steps: collecting channel state information data under the respiration state of personnel through a Wi-Fi transceiving device; preprocessing the channel state information data, calculating CSI entropy, and calculating the maximum CSI entropy; projecting and filtering the CSI entropy signal, screening out subcarrier respiration signals sensitive to respiration and weakly affected by interference; analyzing the time-frequency domain characteristics of the subcarrier respiration signals, reconstructing complete and pure subcarrier respiration signals, estimating the respiration frequency based on peak value detection, and taking the average value as the final result. The application realizes fine-grained user behavior sensing, realizes systematic enhancement of the respiration signal, significantly optimizes the user experience, and improves the robustness of the existing sensing system.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of wireless signal sensing, and particularly relates to a robust respiration sensing method and system based on multi-dimensional Wi-Fi signals. BACKGROUND

[0002] Wireless sensing technology is a technology for sensing the surrounding environment or target behavior activity by using the ubiquitous radio frequency signals in the environment. The research and development of wireless sensing technology have created wireless possibilities for indoor positioning, smart home, human-computer interaction, gesture recognition, health monitoring, etc. Wireless sensing technology is mainly divided into radar-based, radio frequency identification (RFID) based and Wi-Fi signal based sensing technology according to different implementation means. Among them, Wi-Fi-based sensing technology has become a promising implementation means in wireless sensing technology due to its wide deployment, strong ease of use, good universality and low cost.

[0003] Early Wi-Fi signal-based respiration sensing systems attempt to observe the changes in received signal strength (RSS) for sensing. However, due to the coarse granularity of RSS itself, the performance of these systems is not outstanding, and RSS is easily overwhelmed by environmental noise, which cannot be used in dynamic environments. With the development of Wi-Fi devices, it is much easier to obtain fine-grained CSI signals. At present, there are a large number of studies based on the Fresnel region model, which have established a relatively complete respiration sensing theory in an ideal single-person scene, and have achieved high-precision monitoring in a pure respiration environment. However, these methods all assume that there is only a static or single respiration target in the environment, and do not consider the dynamic interference commonly seen in actual scenes, especially for respiration and other small-scale motion sensing, which is more seriously affected by interference.

[0004] To deal with the problem of external interference, some researchers attempt to suppress interference through adaptive beamforming technology. However, it relies on precise beam control and special hardware, is sensitive to position errors, and has limited deployment flexibility. Another approach focuses on the receiving end and uses signal post-processing to eliminate interference.

[0005] Chinese patent application ZL2024111766817 discloses a method and system for enhancing respiration signals in an interference scenario. This method uses human position information to reduce interference and develops a short-time window shifting method to directly remove walking interference in the environment.

[0006] However, the prior art still has obvious deficiencies: on the one hand, commercial Wi-Fi devices usually have multiple antennas and adopt orthogonal frequency division multiplexing (OFDM) technology, that is, signals are transmitted simultaneously on subcarriers at different frequencies through multiple antennas, while the existing work often only optimizes from the perspective of antennas or subcarriers, lacks a systematic signal screening framework, and fails to fully exert the spatial diversity and frequency diversity characteristics of Wi-Fi devices themselves. On the other hand, the existing technology only focuses on external walking interference and ignores self-interference caused by the detection object's own actions. In actual complex environments, the above two kinds of interference often occur coupled, which brings more serious impact to the respiratory monitoring, and there is still no work that can effectively solve this problem. SUMMARY

[0007] In view of the problem of how to stably monitor respiration in a complex interference environment in the prior art and improve the robustness of the existing Wi-Fi sensing system, the present application provides a robust respiratory sensing method and system based on multi-dimensional Wi-Fi signals to improve the robustness of the existing Wi-Fi sensing system.

[0008] In order to achieve the above-mentioned purpose, the present application is realized by the following technical scheme:

[0009] The present application is a robust respiratory sensing method and system based on multi-dimensional Wi-Fi signals, which is specifically for complex motion interference scenarios and includes the following steps:

[0010] Step 1, erecting a Wi-Fi transceiver device in an interference environment, and collecting channel state information (CSI) data of personnel in a respiratory state through the Wi-Fi transceiver device;

[0011] Step 2, preprocessing channel state information (CSI) data received by at least two receiving antennas, dynamically selecting antennas to calculate CSI entropy according to signal volatility characteristics, and calculating the maximum CSI entropy;

[0012] Step 3, projecting and filtering the CSI entropy signal, and applying double indicators to select different frequency subcarriers in the CSI entropy preprocessed in step 2 in the link layer, screening out subcarrier respiratory signals sensitive to respiration and less affected by interference, and realizing indirect enhancement of respiratory signals;

[0013] Step 4, for complex interference in the monitoring environment, using discrete wavelet decomposition (DWT) to analyze the frequency domain characteristics of the subcarrier respiratory signal screened out in step 3, separating out regular low-frequency subcarrier respiratory signals in walking interference, using a sliding window to analyze the low-frequency subcarrier respiratory signals in segments based on respiratory signal-to-noise ratio and similarity, and removing severely distorted signals caused by self-interference according to the change of time-frequency characteristics, and splicing the remaining pure respiratory segments to reconstruct the complete subcarrier respiratory signal;

[0014] Step 5, the sub-carrier breath signal screened out in step 3 and de-noised and reconstructed in step 4, its breath frequency is estimated based on peak detection, and their average value is taken as the final result.

[0015] The further improvement of the present application is that in step 2, the pre-processing of channel state information (CSI) data includes interpolation, CSI entropy calculation and dynamic antenna selection, specifically including the following steps:

[0016] Step 2.1, interpolating the channel state information (CSI) data to a fixed length, the received carrier frequency is The time is The channel state information (CSI) data is expressed as:

[0017]

[0018] Wherein, represents the static component, represents the dynamic component The complex attenuation of The phase shift of the dynamic component The path length of the dynamic component The dynamic component The first dynamic component, The second dynamic component, The first dynamic component, The second dynamic component, The first dynamic component The first dynamic component The second dynamic component The second dynamic component The signal wavelength, The first dynamic component, The second dynamic component, The complex attenuation of the first dynamic component The complex attenuation of the second dynamic component

[0019] Step 2.2, calculate the CSI entropy to get the CSI phase difference signal, and the calculated CSI entropy of the antennas And Is expressed as:

[0020]

[0021] Wherein, The CSI data received by the antenna The carrier frequency is The time is The CSI data, The antenna​ The received carrier frequency is Time is CSI data, Represents CSI entropy. Indicates random phase shift. Indicates antenna static components, Indicates antenna static components, Indicates antenna The dynamic components, Indicates antenna The dynamic components, Represents dynamic components Complex decay, Represents dynamic components phase shift, Represents dynamic components Path length, Represents dynamic components Complex decay, Represents dynamic components phase shift, Represents dynamic components Path length, Indicates the number index of dynamic components;

[0022] Step 2.3: For each subcarrier, select the one with the largest CSI amplitude variance from at least two receiving antennas as the numerator, and select the one with the smallest CSI amplitude variance from multiple receiving antennas as the denominator. By optimizing the antenna pair selection, effectively suppress the fluctuation of the denominator signal and maximize the CSI entropy, as expressed in:

[0023]

[0024] in, The carrier frequency received by the antenna with the largest CSI amplitude variance is: Time is CSI data, The carrier frequency received by the antenna with the smallest CSI amplitude variance is: Time is CSI data, This means maximizing CSI entropy. This represents the static component of the antenna with the largest CSI amplitude variance. This represents the static component of the antenna with the smallest CSI amplitude variance. The dynamic component representing the antenna with the largest CSI amplitude variance. This represents the dynamic component of the antenna with the smallest CSI amplitude variance. a complex attenuation of a dynamic component of an antenna with the largest variance of the amplitude of the CSI a phase shift of a dynamic component of an antenna with the largest variance of the amplitude of the CSI a path length of a dynamic component of an antenna with the largest variance of the amplitude of the CSI a complex attenuation of a dynamic component of an antenna with the smallest variance of the amplitude of the CSI a phase shift of a dynamic component of an antenna with the smallest variance of the amplitude of the CSI a path length of a dynamic component of an antenna with the smallest variance of the amplitude of the CSI a complex attenuation of a dynamic component of an antenna with the smallest variance of the amplitude of the CSI a phase shift of a dynamic component of an antenna with the smallest variance of the amplitude of the CSI a path length of a dynamic component of an antenna with the smallest variance of the amplitude of the CSI a number index of a dynamic component. Further improvement of the present application is that in step 3, a double index is applied to select the subcarriers of different frequencies in the preprocessed CSI entropy, including complex plane projection, basic filtering and subcarrier selection, specifically including the following steps: Step 3.1, for each subcarrier, the projection of the real part and the imaginary part of the channel state information (CSI) data with the largest variance on the rotating coordinate axis is taken as the final respiratory feature sequence, and the projection on the axis

[0025] is expressed as:

[0026]

[0027]

[0028] wherein, and respectively represent the real part and the imaginary part of the channel state information (CSI) data, is a projection angle, and the projection is the projection angle with the largest variance is the best projection angle, and the projection corresponding to the best projection angle is the final respiratory feature sequence, is a transpose;

[0029] Step 3.2, after fully combining the amplitude and phase information of the CSI, a Hampel filter and a Savitzky-Golay filter are used to preliminarily denoise the respiratory feature sequence of each subcarrier, and the high-frequency components are smoothed while the key features of the respiratory signal are maintained;

[0030] ​​​​Step 3.3, for each sub-carrier of the preliminary denoising processing, calculate the final respiratory feature sequence of the respiratory signal-to-noise ratio (BNR) value and the high-frequency suppression respiratory ratio (HSBR) value, and select at least three sub-carrier respiratory signals with the highest respiratory signal-to-noise ratio value and high-frequency suppression respiratory ratio (HSBR) value :

[0031]

[0032]

[0033] wherein, represents the highest peak energy in the respiratory frequency spectrum range, is the index number of the highest peak energy in the respiratory frequency spectrum range, , represents the typical human respiratory frequency range, represents the peak energy, represents the total energy of the signal in the typical human respiratory frequency range, represents the total energy of the signal in the high-frequency interference frequency range, represents the total energy of the signal, represents the high-frequency interference frequency range, is the length of the CSI spectrum, is the frequency index of each frequency point in the spectrum, is the respiratory signal-to-noise ratio value of the respiratory feature sequence, is the high-frequency suppression respiratory ratio value of the respiratory feature sequence;

[0034] Further improvements of the present application are that in step 4, the interference elimination includes walking interference removal based on discrete wavelet decomposition, self-interference segment removal based on time-frequency features, and respiratory signal reconstruction, specifically including the following steps:

[0035] Step 4.1, each selected sub-carrier respiratory signal will go through the same denoising process including steps 4.1-4.3, and for the interference caused by the walking of other people, the discrete wavelet decomposition (DWT) is used to separate the low-frequency sub-carrier respiratory signal and suppress the high-frequency walking noise;

[0036] Step 4.2, for the signal distortion caused by self-motion, a sliding window is used to analyze the low-frequency sub-carrier respiratory signal separated in step 4.1 segment by segment, detect and remove the severely distorted segments;

[0037] Step 4.3, after removing the severely distorted segments, the pure respiratory segments without interference are left, in order to better detect the respiration, all the pure respiratory segments are spliced into a complete respiratory signal, which is convenient for subsequent extraction of respiratory rate.

[0038] The further improvement of the present application is that step 4.2 specifically comprises the following steps:

[0039] Step 4.2.1, based on the fast Fourier transform (FFT), the spectrum of each small signal intercepted by each sliding window during movement is obtained, the respiratory signal-to-noise ratio (BNR) value of the spectrum is calculated, and the change of the spectrum with the sliding window is recorded, and the calculation method is the same as step 3.3;

[0040] Step 4.2.2, the carrier respiratory signal corresponding to the sliding window with the maximum respiratory signal-to-noise ratio (BNR) value is taken as a reference, and the similarity (MED) between the carrier respiratory signal and each small signal segment intercepted by the sliding window during movement is calculated, the time domain feature is extracted, and the change of the time domain feature with the sliding window is also recorded;

[0041] Step 4.2.3, the distortion degree of the signal is calculated by comprehensively considering the respiratory signal-to-noise ratio (BNR) value and the similarity (MED), and the window signal higher than the distortion threshold is a serious distortion segment caused by self-interference, which is directly removed to obtain a pure respiratory signal.

[0042] The robust respiratory sensing method of the present application is realized through a robust respiratory sensing system, and the robust respiratory sensing system comprises:

[0043] A data preprocessing module: pre-processes the channel state information (CSI) data received by at least two receiving antennas, and dynamically selects antennas to calculate CSI entropy according to signal volatility characteristics;

[0044] A respiratory waveform selection module: after complex plane projection combined with amplitude and phase information, the channel state information (CSI) data is preliminarily denoised, and respiratory signal-to-noise ratio (BNR) and high-frequency suppression respiratory ratio (HSBR) are applied in the link layer to select sub-carrier respiratory signals sensitive to respiration and weakly affected by interference, so as to realize indirect enhancement of respiratory signals;

[0045] A complex interference removal module: adopts multi-scale wavelet decomposition to separate regular low-frequency sub-carrier respiratory signals from walking interference, and uses a sliding window to analyze and remove serious distortion signals caused by self-interference in each segment of the low-frequency sub-carrier respiratory signals, and reconstructs a pure respiratory signal;

[0046] A respiratory frequency extraction module: estimates the respiratory frequency of each filtered and interference-removed sub-carrier respiratory signal based on peak value detection, and takes the average value of the respiratory frequency as the final result.

[0047] The beneficial effects of the present application are:

[0048] The application mines the environmental features contained in the Wi-Fi channel state information (CSI), realizes fine-grained user behavior perception, breaks through the limitations of traditional technologies, and achieves high-precision, non-intrusive perception target without the user wearing any device, thereby significantly optimizing the user experience.

[0049] The application calculates the CSI entropy based on the fluctuation and dynamically selects the receiving antenna, and jointly screens the sub-carrier signals in stages based on BNR and HSBR, thereby realizing the systematic enhancement of the respiratory signal. The cooperative processing strategy changes the scheme of optimizing the signal from the antenna or sub-carrier angle in the traditional system, constructs a systematic signal screening framework, and fully utilizes the spatial diversity and frequency diversity characteristics of the Wi-Fi device itself.

[0050] The application removes the influence of complex interference by combining multi-scale wavelet decomposition and the time-frequency characteristics of the signal, and reconstructs the pure respiratory signal for frequency detection. The above denoising method significantly improves the robustness of the existing perception system, overcomes the problems of interference and low detection accuracy caused by external and self-motion, and lays a good foundation for the landing application of related technologies in actual life. BRIEF DESCRIPTION OF DRAWINGS

[0051] Figure 1 It is a typical interference scene diagram in the embodiment of the application.

[0052] Figure 2 It is a principle diagram of cutting the Fresnel zone of the respiratory signal, external walking and self-trunk motion in the interference scene in the embodiment of the application.

[0053] Figure 3 It is a framework diagram of the robust respiratory detection method of the application.

[0054] Figure 4 It is a processing effect diagram of the robust respiratory detection method of the application.

[0055] Figure 5 It is a data processing flowchart of the robust respiratory detection system of the application.

[0056] Figure 6 It is a framework diagram of the robust respiratory detection system of the application.

[0057] Figure 7 It is a CDF comparison diagram of the robust respiratory detection method of the application and the existing method.

[0058] Figure 8 It is a detection rate comparison diagram of the robust respiratory detection method of the application and the existing method.

[0059] Figure 9Error comparison chart of the robust respiration detection method of the present application and existing methods at different sensing distances.

[0060] Figure 10 Influence comparison chart of different walking interferences when walking interference alone occurs in the embodiment of the present application.

[0061] Figure 11 Influence comparison chart of different walking interferences when two kinds of interferences occur simultaneously in the embodiment of the present application.

[0062] Figure 12 Influence comparison chart when self-interference alone occurs in the embodiment of the present application.

[0063] Figure 13 Influence comparison chart of self-interference on system performance under mixed interference in the embodiment of the present application.

[0064] Figure 14 Influence comparison chart of different testers in the embodiment of the present application.

[0065] Figure 15 Comparison chart of different monitoring environments in the embodiment of the present application. DETAILED DESCRIPTION

[0066] Embodiments of the present application will be described below with reference to drawings. Many practical details will be described in the following description in order to provide a thorough understanding of the present application. However, it will be apparent to those skilled in the art that these practical details are not intended to limit the present application. That is, in some embodiments of the present application, these practical details are not necessary. In addition, for the sake of simplicity, some conventional structures and components will be shown in the drawings in a simplified schematic manner.

[0067] As shown in FIG. 1, in actual applications such as hospital wards or home monitoring environments, the frequent movement of medical staff during routine ward rounds will produce continuous walking interference, and the necessary actions of patients during examination, such as the side movement during back auscultation and the body position adjustment during pressure ulcer examination, will cause sudden body movement. More complex is the phenomenon of coupling of the two kinds of interference, which significantly aggravates the degree of signal distortion, bringing great challenges to continuous monitoring. Therefore, the present application is committed to solving this problem. Figure 1 As shown in FIG. 2, the present application provides a robust respiration sensing system, and the robust respiration sensing method is implemented through the robust respiration sensing system. The robust respiration sensing system comprises:

[0068] As shown in FIG. 2, the present application provides a robust respiration sensing system, and the robust respiration sensing method is implemented through the robust respiration sensing system. The robust respiration sensing system comprises: Figures 5-6 A data preprocessing module: pre-processes channel state information (CSI) data received by at least two receiving antennas, and dynamically selects antennas to calculate CSI entropy according to signal volatility characteristics;

[0069] A data preprocessing module: pre-processes channel state information (CSI) data received by at least two receiving antennas, and dynamically selects antennas to calculate CSI entropy according to signal volatility characteristics;

[0070] Respiratory waveform selection module: After combining the amplitude and phase information through complex plane projection, the channel state information (CSI) data is preliminarily denoised, and the respiratory signal-to-noise ratio (BNR) and the high-frequency suppression respiratory ratio (HSBR) are applied in the link layer to select the respiratory waveform in stages, so as to screen out the sub-carrier respiratory signal sensitive to the respiratory signal and weakly affected by the interference, and realize the indirect enhancement of the respiratory signal.

[0071] Complex interference removal module: For complex interference in the monitoring environment, the present application uses multi-scale wavelet decomposition to separate the regular low-frequency sub-carrier respiratory signal in the walking interference, and at the same time, uses a sliding window to analyze and remove the serious distortion signal caused by self-interference in each segment of the low-frequency sub-carrier respiratory signal, and reconstructs the pure respiratory signal.

[0072] Respiratory frequency extraction module: For each screened and interference-removed reconstructed pure respiratory signal, the respiratory frequency is estimated based on peak value detection, and the average value of the respiratory frequency is taken as the final result.

[0073] As shown in Figures 2-4 The present application discloses a robust respiratory sensing method and system based on multi-dimensional Wi-Fi signal, which is a robust respiratory detection method using a fine-grained signal screening mechanism to reduce the influence of interference. First, the present application collects the CSI data of the detected person in a complex dynamic environment, dynamically selects the antenna and combines two indicators to select two levels of sub-carriers, amplifies the respiratory characteristics and reduces the influence of interference in the physical-link layer. In addition, for complex interference in the monitoring environment, the present application combines the time-frequency characteristics of the signal to propose a complex interference removal method. This method uses multi-scale wavelet decomposition to analyze the frequency domain characteristics of the received signal, separates the regular low-frequency respiratory component in the walking interference. At the same time, based on the time-frequency characteristics of the window signal, the serious distortion signal caused by self-interference is detected and removed, and the characteristics of the respiratory signal are completely preserved. Finally, the respiratory rate of the detected person is calculated according to the peak value detection.

[0074] Specifically, the present application is a robust respiratory sensing method based on multi-dimensional Wi-Fi signal, which includes the following steps:

[0075] Step 1, erect a Wi-Fi transceiver device in a typical interference environment, and collect channel state information (CSI) data of personnel in a respiratory state through the Wi-Fi transceiver device.

[0076] In this embodiment, two desktop computers equipped with Intel 5300 network cards are used as transceiver devices. The Wi-Fi signal is transmitted on channel 64 with a center frequency of 5.32 GHz and a sampling rate of 100 Hz. Both the transmitter and receiver are equipped with three omnidirectional antennas, but this embodiment only uses one antenna from the transmitter. The antenna height is 1.5 m, and the distance between the transmitter and receiver devices is 1.5 m. During the experiment, sample data is generated every 30 seconds, ensuring both a certain data length and real-time performance.

[0077] Typical interference environments include private offices and open cluster office spaces, with other people present besides the person being monitored. The constant movement of these other people creates continuous walking interference, while the person being monitored periodically performs different physical movements such as bending over or standing up, creating short-term strong self-interference.

[0078] Step 2: Preprocess the Channel State Information (CSI) data received by at least two receiving antennas, dynamically select antennas based on signal fluctuation characteristics, calculate CSI entropy, and calculate the maximum CSI entropy. This includes the following steps:

[0079] Step 2.1: First, in wireless sensing systems, due to factors such as channel fading and multipath effects, packet loss is inevitable during transmission, resulting in the actual number of received data points often being less than the theoretically expected value. Therefore, linear interpolation is used in the data preprocessing stage to complete the data sequence to 3000 sampling points, maintaining the temporal continuity of the original signal. The received carrier frequency is... Time is The Channel State Information (CSI) data is represented as follows:

[0080]

[0081] in, Represents static components. Represents dynamic components Complex attenuation, Represents dynamic components phase shift, Represents dynamic components Path length, Indicates the first A dynamic component Indicates the first in the environment A dynamic path, It is additive white Gaussian noise. Represents the imaginary unit. It is the signal wavelength. This represents the first dynamic component. represents the 2nd dynamic component, represents the 1st dynamic component complex attenuation of, represents the 2nd dynamic component complex attenuation of;

[0082] Step 2.2, due to the hardware defects of commercial Wi-Fi devices and the time delay of transceivers, each received CSI packet carries random phase noise, causing random phase offset of the received signal. This application introduces CSI entropy to solve this problem. At the same time, the CSI phase difference signal with higher perception accuracy is obtained to calculate the CSI entropy of the antenna and The CSI entropy calculated is expressed as:

[0083]

[0084] wherein, represents the CSI data received by the antenna at the carrier frequency at time , represents the CSI data received by the antenna at the carrier frequency at time , represents the CSI entropy, represents the random phase offset, represents the static component of the antenna , represents the static component of the antenna , represents the dynamic component of the antenna , represents the dynamic component of the antenna , represents the complex attenuation of the dynamic component , represents the phase shift of the dynamic component , represents the path length of the dynamic component , represents the complex attenuation of the dynamic component , represents the phase shift of the dynamic component , represents the path length of the dynamic component , represents the number index of the dynamic component.

[0085] Step 2.3, for each subcarrier, select the antenna with the largest variance of the CSI amplitude as the numerator and the antenna with the smallest variance of the CSI amplitude as the denominator, optimize the selection of the antenna pair to effectively suppress the fluctuation of the denominator signal, maximize the dynamic range of the CSI entropy signal, and enhance the respiratory representation, at this time, the CSI entropy signal is expressed as:

[0086]

[0087] wherein, represents the carrier frequency received by the antenna with the largest variance of the CSI amplitude, represents the carrier frequency received by the antenna with the smallest variance of the CSI amplitude, represents the CSI data at time represents the CSI data at time represents the maximum CSI entropy, represents the static component of the antenna with the largest variance of the CSI amplitude, represents the static component of the antenna with the smallest variance of the CSI amplitude, represents the dynamic component of the antenna with the largest variance of the CSI amplitude, represents the dynamic component of the antenna with the smallest variance of the CSI amplitude, represents the complex attenuation of the dynamic component of the antenna with the largest variance of the CSI amplitude, represents the phase shift of the dynamic component of the antenna with the largest variance of the CSI amplitude, represents the path length of the dynamic component of the antenna with the largest variance of the CSI amplitude, represents the complex attenuation of the dynamic component of the antenna with the smallest variance of the CSI amplitude, represents the phase shift of the dynamic component of the antenna with the smallest variance of the CSI amplitude, represents the path length of the dynamic component of the antenna with the smallest variance of the CSI amplitude, represents the number index of the dynamic component. Step 3, project and filter the CSI entropy signal, and select the subcarrier respiratory signal sensitive to respiration and weakly affected by interference in the link layer according to the double indicators applied to the different frequency subcarriers in the preprocessed CSI entropy in step 2, to realize indirect enhancement of the respiratory signal. Specifically, the following steps are included:

[0088] Step 3, project and filter the CSI entropy signal, and select the subcarrier respiratory signal sensitive to respiration and weakly affected by interference in the link layer according to the double indicators applied to the different frequency subcarriers in the preprocessed CSI entropy in step 2, to realize indirect enhancement of the respiratory signal. Specifically, the following steps are included:

[0089] ​​​​​​​Step 3.1. The selection of respiratory waveform from the perspective of amplitude or phase difference alone is prone to "blind spot" problem, which cannot fully consider the sensing performance of part of subcarriers. In order to fully combine the amplitude and phase difference information of CSI and eliminate the "blind spot" problem, the projection with the largest variance on the rotating coordinate axis of the real part and the imaginary part of the CSI signal is taken as the final respiratory feature sequence, which is equivalent to combining amplitude and phase, thereby avoiding the selection of subcarriers using amplitude or phase difference alone, providing more and better options for selecting respiratory mode signals, and the projection at the axis is represented as:

[0090]

[0091] wherein, and respectively represent the real part and the imaginary part of the channel state information (CSI) data, is the projection angle, and the projection is the projection angle with the largest variance is the best projection angle, and the projection corresponding to the best projection angle is the final respiratory feature sequence, is the transpose;

[0092] Step 3.2. In order to reduce the background noise in the respiratory signal and fully combine the amplitude and phase information of CSI, the present application adopts Hampel filter and Savitzky-Golay filter to preliminarily denoise the respiratory feature sequence of each subcarrier. Hampel filter removes outliers that are significantly different from adjacent CSI measurements based on the median statistical characteristics of the sliding window, while the most significant feature of Savitzky-Golay filter is that it can maintain the shape and width of the signal unchanged while filtering out noise, effectively fitting the low-frequency respiratory component in the signal and smoothing out the high-frequency component. This combination filtering strategy fully takes advantage of the advantages of the two filters, ensuring the robustness of the signal and maintaining the key features of the respiratory signal. Considering the problem of computational complexity, the above filtering is performed after the projection in the complex plane, rather than directly in the data preprocessing part as in most works. The received CSI original signal is a complex sequence and the projected signal is a real sequence, which has lower computational complexity and more obvious filtering effect.

[0093] ​Step 3.3, Different subcarriers have different center frequencies, so the sensing performance of breath and interference is different. In the complex interference scene, the high quality of the breath signal must have obvious periodicity and lower noise interference at the same time, and the two features can effectively represent the ability to sense the breath in a short time. Therefore, the present application combines BNR and HSBR to carry out two-stage screening to select the best subcarrier signal. The BNR (Breath Signal-to-Noise Ratio) index represents the percentage of the energy of the breath signal in the total energy in the spectrum diagram, reflecting the quality of the waveform. In the embodiment, the BNR is calculated through the following standardization process: first, the fast Fourier transform (FFT) is performed on the CSI signal of each subcarrier to obtain the frequency spectrum; then, in the typical human breath frequency range, i.e. 10-37 bpm, corresponding to 0.167-0.617 Hz, the spectral component with the maximum energy is identified; finally, the BNR value is determined by calculating the ratio of the peak energy to the total energy of the signal.

[0094]

[0095] wherein, represents the highest peak energy in the breath spectrum range, is the index number of the highest peak energy in the breath spectrum range, , represents the peak energy, represents the total energy of the signal, is the length of the CSI spectrum, indicates the frequency index of each frequency point in the spectrum, is the breath signal-to-noise ratio value of the breath feature sequence, is the high-frequency suppression breath ratio value of the breath feature sequence; specifically: according to the breath signal-to-noise ratio (BNR), the first screening is carried out, and at least 5 subcarrier breath feature sequences with the highest breath signal-to-noise ratio value are obtained.

[0096] However, in the scene of complex interference, only guaranteeing strong respiratory energy is not enough to ensure the accuracy of respiratory detection results, especially in the presence of self-interference. Considering that the energy of complex interference is concentrated in the high frequency band, that is, the signal change speed caused by walking and body shaking is much higher than that of breathing, this application proposes a novel index, High-suppressed Breath Ratio (HSBR). The HSBR index represents the ratio of respiratory signal energy to high-frequency interference signal energy, which is used to quantify the degree of high-frequency noise interference on the perception signal. After obtaining the frequency spectrum of CSI, according to the clinical respiratory monitoring standard, the normal respiratory frequency range of [10 bpm, 30 bpm] is converted to the frequency spectrum interval of [0.1667 Hz, 0.5 Hz], and [0.5 Hz, 10 Hz] is determined as the typical high-frequency noise distribution band. The HSBR of the subcarrier is obtained by calculating the energy ratio of the two characteristic frequency bands:

[0097]

[0098] wherein, represents the typical human respiratory frequency range, represents the frequency range of high-frequency interference, represents the total energy of the signal in the typical human respiratory frequency range, represents the total energy of the signal in the frequency range of high-frequency interference.

[0099] In this embodiment, at least 5 subcarrier signals with good respiratory characteristics are first screened out according to BNR, and the second screening is aimed at high-frequency noise components, quantifying the influence of complex interference on breathing, and screening out at least 3 subcarriers with the highest HSBR Respiratory detection is performed. After two-stage screening of subcarriers, this application fully utilizes the frequency diversity to obtain a signal to be detected with obvious respiratory characteristics and weak interference, reducing the difficulty of subsequent interference removal.

[0100] Step 4, for complex interference in the monitoring environment, discrete wavelet decomposition (DWT) is used to analyze the frequency domain characteristics of the subcarrier respiratory signals screened out in step 3, to separate the regular low-frequency subcarrier respiratory signals from the walking interference. Based on the respiratory signal-to-noise ratio and similarity, a sliding window is used to analyze the low-frequency subcarrier respiratory signals segment by segment, and according to the change of time-frequency characteristics, the severely distorted signals caused by self-interference are removed, and the remaining pure respiratory segments are spliced to reconstruct the complete subcarrier respiratory signal. Specifically, the following steps are included:

[0101] Step 4.1, each screened subcarrier respiratory signal All of them will go through the same de-noising process, i.e. step 4.1-step 4.3. For the interference caused by the walking of other people, discrete wavelet decomposition (DWT) is used to separate the low-frequency sub-carrier respiratory signal and suppress the high-frequency walking noise.

[0102] Due to the rapid variability and irregularity of the walking signal, the walking signal and the respiratory signal differ in frequency, and the discrete wavelet change (DWT) can separate the low-frequency part of the CSI signal, i.e. the respiratory signal component. DWT decomposes the signal into different scale approximation and detail coefficients through a series of wavelet filtering and downsampling operations, realizing multi-scale analysis of the signal. The wavelet base selected in this embodiment is the fourth wavelet of Daubechies series, usually abbreviated as "db4", which performs well on signals showing sharp transitions and is usually used to process signals with prominent features, balancing the needs of time and frequency localization. The selection of the decomposition level depends on the sampling rate of the signal and the frequency range of interest. In this embodiment, the signal sampling rate is 100 Hz, and the target is to cover the respiratory frequency range of about 0.1667 Hz to 0.5 Hz. Considering that each level of decomposition roughly halves the frequency range, a 6-level decomposition allows analysis of signals below 100 / 2^7 = 0.78 Hz, including the typical frequency range of the respiratory signal.

[0103] Step 4.2, during long-term respiratory monitoring, it is impossible for the monitored person to remain absolutely still. Random motion such as body sway can seriously affect the accuracy of respiratory detection. Therefore, this application provides an interference removal method to locate and filter out the severely distorted respiratory signal segment caused by self-motion. Based on the dual characteristics of the signal in the frequency domain and the time domain, this application can distinguish between normal respiratory sign segments and severely distorted segments on a shorter time scale.

[0104] In this embodiment, sliding window segment analysis is used to detect and remove severely distorted segments caused by self-motion. It is worth noting that this embodiment uses a sliding window with a window length of 600 sampling points to traverse the entire time window based on the longest interval of normal breathing (6s). Subsequently, based on the FFT, the frequency spectrum of each window signal is obtained, its BNR value is calculated and its change with window sliding is recorded, after all, the frequency domain peak caused by the periodicity of breathing is still one of the most obvious judgment features.

[0105] However, there are also false positives in the judgment of distortion segments only according to the periodicity of the signal. In step 4.1, the wavelet decomposition is performed to remove the walking interference, so that the distortion segment caused by self-interference only leaves the low-frequency respiratory component with severe distortion. Although it loses the correct periodicity, it may retain a slow fluctuation that is not periodic with respiration. When this slow fluctuation covers the sliding window, it will cause an abnormal increase in BNR, leading to a false positive. Therefore, the present application combines the time-domain characteristics of the signal for joint detection. Specifically, the signal in the window with the maximum BNR (the most obvious and standard part of respiration in the entire signal) is taken as a reference to calculate the minimum Euclidean distance (MED) with other window signals, so as to measure the similarity between two sequences with the same length. The similar part to the reference signal in all sliding windows is a pure respiratory segment, and the part with low similarity is naturally a distortion segment. The MED calculation method is as follows:

[0106]

[0107] In this embodiment, the change of MED with the movement of the window is also recorded. In order to combine the time-frequency domain characteristics, the BNR and the MED are weighted and summed in a ratio of 6:4 to obtain the distortion degree of the sliding window. The higher the distortion degree, the more distortion signals the sliding window contains. Screening with a dynamic threshold (0.9 times the maximum value of the distortion degree) can effectively detect sliding windows containing more distortion signals, and record the starting positions of these windows. By superimposing the window length on the starting position, the actual distortion signal segment can be mapped and removed.

[0108] Step 4.3, after screening and removing the severe distortion segment, if the distortion segment is in the middle of the time-domain signal, the original signal will be cut into several small segments, and it is obviously not suitable to perform respiration detection on these small segments. Therefore, the present application uses the maximum point splicing method to search for the nearest maximum point on both sides of the breakpoint position, combines the height adjustment to realize the smooth splicing of the entire respiratory signal, and splicing at the maximum point also avoids the trouble of phase correction and the change of respiratory interval caused by simple splicing.

[0109] Step 5, for the sub-carrier respiratory signal screened in step 3 and reconstructed after the interference removal and reconstruction in step 4, the respiratory frequency is estimated based on peak detection, and the average value thereof is taken as the final result. Specifically, after the foregoing signal screening and processing, the present application removes the interference caused by external walking and self-interference caused by self-motion, and obtains a pure respiratory signal close to the theoretical condition. The present application uses peak detection to obtain the respiratory rate, and the average value P of at least three processed sub-carrier peak-to-peak intervals is taken to obtain the period of the respiratory signal, and the estimated respiratory frequency can be calculated as 60 / P bpm. Finally, the average respiratory frequency of these sub-carriers is taken as the respiratory estimation in this time period.

[0110] To comprehensively evaluate the performance, we conducted experiments in both independent office and open cluster office space, and collected 540 groups of data, each group of data collected for 30s, a total of 162000s data. In order to intuitively show the performance of the system, we compared it with four advanced Wi-Fi-based breath sensing methods. We selected two widely recognized systems (PhaseBeat and FarSense) to verify the accuracy of respiration and two systems with certain anti-interference performance (TB and RespEnh) to verify the anti-interference advantage of the present application. The results are shown in Figure 8 The integrated accuracy of respiration rate estimation in two scenarios is 93.3%, which is significantly better than the other four algorithms. Compared with existing work, FarSense lacks corresponding interference consideration and is only suitable for pure breathing scenarios; the short-time window shifting method of RespEnh is based on the premise of complete respiration signal; TB only considers the frequency domain features of the signal, and the detection accuracy of self-interference segment needs to be further improved. In summary, the present application is the first breath sensing system that can handle both environmental and self-interference. Figure 7 The error distribution of the present application and the baseline algorithm is given. The steep rise of the present application curve at very low error values indicates that most of the estimates have achieved abnormally high accuracy. The rapid convergence of its CDF to 100% also confirms that the present application can effectively constrain the maximum error, again proving the excellent performance of the present system.

[0111] To more comprehensively evaluate the performance of the present application, we measured its sensing effect under various different factors and verified its effectiveness in different experimenters and test environments. For each different value of the factor, we collected 20 groups of data samples, each lasting for 30s, and used box plots to display the results after statistics.

[0112] Influence of sensing distance:

[0113] The distance between the detection target and the transceiver device is an important indicator affecting system performance. To explore the influence of different breathing distances on experimental results, we gradually increased the distance between the detection target and the transceiver device while keeping the same interference conditions and counted the experimental results. Figure 9 The influence of breathing distance on breath sensing accuracy is shown. As the distance increases, the respiration signal decays continuously, so the interference easily overwhelms the useful respiration signal, which in turn leads to an increase in respiration detection error.

[0114] Influence of different walking interference:

[0115] As an anti-interference respiration detection system, different interference conditions are the focus of our attention. In order to explore the influence of walking interference on the system performance, the present application carries out experiments in two cases of walking interference occurring alone and walking interference and self-interference occurring simultaneously. The distance between the interferer and the detected person is changed to change the influence of walking interference, while the same self-interference condition is ensured.

[0116] When walking interference occurs alone, as shown in FIG. 6, the performance of PhaseBeat is significantly poorer than that of other methods. The maximum error can even reach more than 6 bpm in the case that the distance between the detected person and the walking interference source is 1 m. Other methods are relatively stable, indicating that the anti-interference ability of respiration extracted from phase difference is still insufficient. Figure 10

[0117] Figure 11 The error of the system when two kinds of interference occur simultaneously is shown. By comparing the results under the same walking interference distance, it can be seen that the MAE of all methods increases by 50% compared with the MAE when only walking interference occurs, revealing the huge impact of self-motion on respiration detection. In addition, the farther the walking interference distance, the smaller the detection error, because the farther the distance between the interferer and the detected person, the smaller the proportion of walking interference in the respiration signal, and the closer the obtained waveform to the pure respiration scene. The performance gap between the present application and other baseline methods is further widened when mixed interference occurs, indicating the superior robustness of the system proposed by the present application.

[0118] Influence of different self-interference:

[0119] In order to explore the influence of self-interference on the system performance, the present application controls the amplitude of body shaking and changes the duration, and experiments are carried out in two cases of self-interference occurring alone and mixed with two kinds of interference.

[0120] Figure 12 The results when self-interference occurs alone are given. Compared with walking interference, the interference caused by non-respiratory motion of the body is a greater challenge. When 16s of non-respiratory motion occurs during CSI acquisition, the error of PhaseBeat is close to 9 bpm, and the maximum error of FarSense also reaches 6.36 bpm, which completely fails to function. It can be seen that the traditional Wi-Fi sensing system still needs to be continuously developed to function in actual environment. Although TB takes certain measures to reduce the influence of self-interference, the present application finds that threshold-based methods are prone to errors and omissions in actual testing, and fail to fully utilize the characteristics of the signal.

[0121] Figure 13 ​The impact of self-interference on system performance under mixed interference is demonstrated. The results highlight the robustness of the present application, which maintains high accuracy with minimal error degradation as the self-interference duration increases from 1 s to 16 s. In sharp contrast, all baseline methods, including RespEnh, exhibit severe performance degradation even under short self-motion and rapidly increase in error as the self-interference duration increases. This highlights the critical issue that existing methods designed for isolated interference types fail under coupled interference conditions.

[0122] Impact of different monitors:

[0123] The breathing rates and amplitudes of different individuals vary greatly. To explore the applicability of the present application to different testers, experiments were conducted on five different volunteers under the same interference conditions, with a walking interference distance of 3 m and a self-interference duration of 4 s. The results are shown in FIG. 6. Figure 14 The results show that although there are inherent physiological differences between individuals, the system of the present application still maintains high accuracy. The median absolute error of the five subjects remains below 1.0 bpm, and the interquartile range is narrow, indicating stable performance. Specifically, the mean errors of Person 1 ~ 5 are 0.70, 0.88, 0.98, 0.64, and 0.78 bpm, respectively. Although some of the data detection errors of Person 3 are larger, the present application overall exhibits strong generalization ability and independence from specific user characteristics, which is crucial for practical deployment in different populations.

[0124] Impact of different monitoring environments:

[0125] To explore the impact of the environment on system performance, experiments were conducted in an independent office and an open cluster office space. Figure 15 The breathing rate estimation errors of the present application in different scenarios are shown, and the interference conditions of the experiment are the same as above, with a 3 m walking interference and a 4 s self-interference. The spatial scale of the independent office is smaller, and the density of furniture and other items is greater than that of the open office, exacerbating the multipath effect in the indoor environment, so the error of breathing detection is greater, but careful comparison shows that the estimation errors of the two are not much different, with an average estimation error difference of only 0.1498 bpm, verifying the robustness of the present application to different scenarios.

[0126] The above merely illustrates the embodiments of the present application and is not intended to limit the present application. The present application can have various modifications and changes for those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the scope of the claims of the present application.

Claims

1. A robust respiration sensing method based on multi-dimensional Wi-Fi signals, for complex motion interference scenarios, characterized in that: The robust respiration sensing method comprises the following steps: Step 1, erecting a Wi-Fi transceiver device in an interference environment, and collecting channel state information data of personnel in a breathing state through the Wi-Fi transceiver device; Step 2, preprocessing channel state information data received by at least two receiving antennas, dynamically selecting antennas according to signal volatility characteristics, calculating CSI entropy, and maximizing CSI entropy; Step 3, projecting and filtering the CSI entropy signal, and applying double indicators to select subcarriers of different frequencies in the preprocessed CSI entropy in step 2, and screening out subcarrier breathing signals sensitive to breathing and weakly affected by interference; Step 4, using discrete wavelet decomposition to analyze the frequency domain characteristics of the subcarrier breathing signals screened out in step 3, separating the regular low-frequency subcarrier breathing signals from the walking interference, using a sliding window to analyze the low-frequency subcarrier breathing signals in segments based on the breathing signal-to-noise ratio and similarity, and removing the severely distorted signals caused by self-interference according to the change of time-frequency characteristics, and splicing the remaining pure breathing segments to reconstruct complete and pure subcarrier breathing signals; Step 5, for the subcarrier breathing signals screened out in step 3 and reconstructed after the interference removal and reconstruction in step 4, estimating the breathing frequency based on peak value detection, and taking their average value as the final result, wherein: In step 2, preprocessing the channel state information data includes interpolation point filling, CSI entropy calculation and dynamic antenna selection, which specifically includes the following steps: Step 2.1, Interpolating channel state information data to fixed length, received carrier frequency is Time is Channel state information data is expressed as: in, Represents static components. Represents dynamic components Complex attenuation, Represents dynamic components phase shift, Represents dynamic components Path length, Indicates the first One dynamic component, Indicates the first in the environment A dynamic path, It is additive white Gaussian noise. Represents the imaginary unit. It is the signal wavelength. This represents the first dynamic component. This represents the second dynamic component. Indicates the first dynamic component Complex decay, Indicates the second dynamic component Complex decay; Step 2.2, compute CSI entropy to get the CSI phase difference signal, in antenna and antenna The computed CSI entropy is represented as: wherein denotes an antenna received carrier frequency is time is CSI data, denotes an antenna received carrier frequency is time is CSI data, denotes a CSI entropy, denotes a random phase offset, denotes a static component of an antenna , denotes a static component of an antenna , denotes a dynamic component of an antenna , denotes a dynamic component of an antenna , denotes a complex attenuation of a dynamic component , denotes a phase shift of a dynamic component , denotes a path length of a dynamic component , denotes a complex attenuation of a dynamic component , denotes a phase shift of a dynamic component , denotes a path length of a dynamic component , denotes a number index of a dynamic component; Step 2.3, for each subcarrier, selecting the one with the maximum CSI amplitude variance from at least two receiving antennas as the numerator, and selecting the one with the minimum CSI amplitude variance as the denominator, and the maximum CSI entropy is expressed as: in, The carrier frequency received by the antenna with the largest CSI amplitude variance is: Time is CSI data, The carrier frequency received by the antenna with the smallest CSI amplitude variance is: Time is CSI data, This represents maximizing CSI entropy. This represents the static component of the antenna with the largest CSI amplitude variance. This represents the static component of the antenna with the smallest CSI amplitude variance. This represents the dynamic component of the antenna with the largest CSI amplitude variance. This represents the dynamic component of the antenna with the smallest CSI amplitude variance. The dynamic component of the antenna with the largest CSI amplitude variance. Complex decay, The dynamic component of the antenna with the largest CSI amplitude variance. phase shift, The dynamic component of the antenna with the largest CSI amplitude variance. Path length, The dynamic component of the antenna with the smallest CSI amplitude variance represents the antenna. Complex decay, The dynamic component of the antenna with the smallest CSI amplitude variance represents the antenna component. phase shift, The dynamic component of the antenna with the smallest CSI amplitude variance represents the antenna component. Path length; In step 3, double indicators are applied to select subcarriers of different frequencies in the preprocessed CSI entropy, including complex plane projection, basic filtering and subcarrier selection, which specifically includes the following steps: Step 3.1, For each subcarrier, the projection of the real and imaginary parts of the channel state information data onto the axis of maximum variance in the rotated coordinate system is taken as the final respiratory feature sequence, the projection onto the axis at Step 3.1 is denoted as: wherein, and respectively represent the real part and the imaginary part of the channel state information data, is a projection angle, the projection is the projection angle with the largest variance is the optimal projection angle, and the projection is the final respiratory feature sequence, is a transpose; Step 3.2, using Hampel filter and Savitzky-Golay filter to preliminarily denoise the breathing feature sequence of each subcarrier; Step 3.3, for each sub-carrier of the preliminary de-noising processed final breath feature sequence, calculate the breath signal-to-noise ratio value and the high frequency suppressed breath ratio value of the final breath feature sequence, and screen out at least three sub-carrier breath signals with the highest breath signal-to-noise ratio value and the highest high frequency suppressed breath ratio value : wherein, represents the highest peak energy in the respiratory frequency spectrum range, is an index number of the highest peak energy in the respiratory frequency spectrum range, , represents the typical human respiratory frequency range, represents the peak energy, represents the total energy of the signal, represents the total energy of the signal in the typical human respiratory frequency range, represents the total energy of the signal in the frequency range of high frequency interference, represents the frequency range of high frequency interference, is the length of the CSI spectrum, is the frequency index of each frequency point in the spectrum, is the respiratory signal-to-noise ratio value of the respiratory feature sequence, is the high frequency suppression respiratory ratio value of the respiratory feature sequence.

2. The method of claim 1, wherein: In step 4, interference elimination includes walking interference removal based on discrete wavelet decomposition, self-interference segment removal based on time-frequency characteristics, and breathing signal reconstruction, which specifically includes the following steps: Step 4.1, each selected sub-carrier respiration signal The low-frequency sub-carrier respiration signal is separated by using the discrete wavelet decomposition, and the high-frequency walking noise is suppressed. Step 4.2, using a sliding window to analyze the low-frequency subcarrier breathing signals separated in step 4.1 in segments, detecting and removing severely distorted segments; Step 4.3, after removing the severely distorted segments, leaving the pure breathing segments without interference, and splicing all the pure breathing segments into a complete breathing signal.

3. The method of claim 2, wherein: Step 4.2 specifically includes the following steps: Step 4.2.1, based on fast Fourier transform (FFT), obtaining the frequency spectrum of each small signal segment intercepted in the moving process of each sliding window, calculating the breathing signal-to-noise ratio value of the frequency spectrum and recording the change of the frequency spectrum with the sliding window; Step 4.2.2, taking the carrier breathing signal corresponding to the sliding window with the maximum breathing signal-to-noise ratio value as a reference, calculating the similarity with each small signal segment intercepted in the moving process of the sliding window, extracting the time domain characteristics, and also recording the change of the time domain characteristics with the sliding window; Step 4.2.

3. Calculate the distortion degree of the signal by combining the respiratory signal-to-noise ratio and the similarity, and remove the window signal with a distortion degree higher than a distortion threshold to obtain a pure respiratory signal.

4. The method of claim 1, wherein: The robust respiratory sensing method is implemented by a robust respiratory sensing system, and the robust respiratory sensing system comprises: A data preprocessing module: pre-processes channel state information data received by at least two receiving antennas, and dynamically selects antennas to calculate CSI entropy according to signal volatility characteristics; A respiratory waveform selection module: performs preliminary denoising processing on the channel state information data by complex plane projection combined with amplitude and phase information, and performs phased respiratory waveform selection by using a respiratory signal-to-noise ratio and a high-frequency suppression respiratory ratio to screen out sub-carrier respiratory signals sensitive to respiration and weakly affected by interference, thereby indirectly enhancing the respiratory signal; A complex interference removal module: separates regular low-frequency sub-carrier respiratory signals from walking interference by using multi-scale wavelet decomposition, and removes self-interference-caused severely distorted signals by using a sliding window to analyze and remove the severely distorted signals in segments, thereby reconstructing a pure respiratory signal; A respiratory frequency extraction module: estimates the respiratory frequency of each screened and interference-removed sub-carrier respiratory signal based on peak value detection, and takes the average value of the respiratory frequencies as the final result.

Images