Multimode fusion respiration monitoring method based on millimeter wave radar and Wi-Fi

Abstract

The invention discloses a breath detection method based on multimode fusion of millimeter waves and Wi-Fi signals. Firstly, echo signals of a target area are collected by using a frequency modulated continuous wave (FMCW) radar and wireless Wi-Fi equipment, and a multi-source signal fusion judgment module for human body target judgment and signal quality evaluation based on a radar echo signal heat map and a multi-source signal fusion method based on life signal quality screening correction and multivariate singular spectrum analysis (MSSA) are utilized to obtain a multi-source signal fusion result. The multimode fusion respiration sensing system is realized, and steady non-contact respiration monitoring can be carried out by jointly utilizing the multimode information of the millimeter wave radar and the Wi-Fi signal.

Description

Technical Field

[0001] This invention relates to the field of non-contact vital sign monitoring technology based on millimeter-wave radar and Wi-Fi multi-mode fusion, specifically to a respiratory monitoring method based on millimeter-wave radar and Wi-Fi multi-mode fusion. Background Technology

[0002] With the accelerating global aging process and the continuous growth in demand for population health management, respiratory monitoring, as a core technology for clinical diagnosis, home-based care, and daily health assessment, is becoming increasingly important. Reliable respiratory status monitoring not only provides key physiological indicators for the early screening and disease progression tracking of diseases such as sleep-disordered breathing, chronic obstructive pulmonary disease, and heart failure, but also generates human health assessment reports and provides early warnings of sudden respiratory abnormalities (such as sleep apnea and asthma) by continuously analyzing changes in respiratory rate, rhythm, and depth, thus gaining valuable time for timely intervention. In addition, non-contact respiratory monitoring technology avoids the discomfort and skin irritation caused by traditional contact sensors, improving user compliance with long-term use, while also taking into account the privacy protection needs in the home environment. It is suitable for scenarios such as newborn monitoring, elderly care, and isolation wards, and has significant clinical value and social benefits.

[0003] Current respiratory monitoring technology presents a landscape where traditional solutions have limitations while emerging technologies offer breakthroughs. Traditional contact-based solutions rely on breathing belts, nasal airflow sensors, spirometers, and polysomnography systems. While these offer high measurement accuracy under controlled conditions, they require direct contact with the human body, potentially causing skin discomfort, activity restrictions, and poor user compliance, and are particularly unsuitable for long-term continuous monitoring. Traditional non-contact solutions, such as optical cameras or infrared thermal imaging, can estimate respiration through changes in chest and abdominal movement or thermal radiation, but they are sensitive to lighting conditions, obstructions, and ambient temperature, and pose significant privacy risks. In recent years, millimeter-wave radar and Wi-Fi radio frequency sensing have become research hotspots due to their non-contact nature, ability to penetrate obstructions, independence from light, and privacy protection without image acquisition; related patents have validated their technological feasibility. Millimeter-wave radar, under ideal orientation (e.g., lying supine facing the radar), can achieve sub-millimeter-level respiratory displacement measurement, with significantly higher accuracy than other wireless solutions. Wi-Fi sensing, on the other hand, utilizes Channel State Information (CSI) to capture multipath changes caused by breathing, offering wide coverage, strong wall penetration, and good adaptability to changes in body posture. However, existing single-modal solutions still face key challenges: the effective sensing range of millimeter-wave radar is easily affected by natural sleep behaviors such as lying on one's side, turning over, or being obstructed by blankets, leading to signal attenuation or loss; while Wi-Fi sensing has better robustness, its respiratory estimation accuracy is lower than radar under ideal conditions and is susceptible to interference from non-respiratory movements in the environment. These limitations severely restrict their independent deployment performance in real-world, complex scenarios. Summary of the Invention

[0004] To address the instability of respiratory monitoring using single signals from millimeter-wave radar and Wi-Fi devices, this paper proposes a multi-mode fusion respiratory sensing system that leverages the multi-modal information from both millimeter-wave radar and Wi-Fi signals for robust non-contact respiratory monitoring. For the decision-making process of multi-modal signal fusion, this paper proposes a method based on human target identification using radar echo signal heatmaps and respiratory Rényi entropy signal quality assessment to determine whether fusion should proceed. For the problem of respiratory estimation through multi-source signal fusion, this paper proposes a method combining vital sign screening, dynamic time warping (DTW) correction, and multivariate singular spectrum analysis (MSSA) based multivariate signal fusion. The core of this method is as follows: firstly, based on human target identification and respiratory signal Rényi entropy, adaptively determine the data segments requiring multi-source signal fusion; for the data requiring fusion, further refine and enhance respiratory features through vital sign screening and correction; finally, fuse multivariate signals using multivariate singular spectrum analysis to stably estimate and monitor respiration in complex scenarios.

[0005] The technical solution of this invention is as follows: A multi-mode fusion respiratory monitoring method based on millimeter-wave radar and Wi-Fi includes: Step 1: Simultaneously acquire millimeter-wave radar echo data and CSI data from Wi-Fi devices. Perform FFT transformations on the radar echoes in the range, Doppler, and angle dimensions to generate range-Doppler and range-angle heatmaps, providing basic features for subsequent human target localization and respiratory signal extraction. Step 2: Apply Constant False Alarm Rate (CFAR) detection to locate potential human chest cavity targets on the heatmap. Extract the range cell with the highest number of detected targets from both the range-Doppler heatmap and the range-angle heatmap, denoted as _____. and This is used to determine the consistency of objectives in subsequent fusion decisions; Step 3: Extract the initial breathing signal from the radar echo. Perform phase unwrapping, detrending, phase differential, and outlier removal on the complex signal of the target range cell, and convert the signal to the velocity domain to enhance the periodic micro-motion characteristics caused by breathing and suppress slow body drift and abrupt noise; Step 4: Quantitatively evaluate the frequency domain quality of the radar breathing signal using Rényi entropy. Calculate the Rényi entropy of the breathing signal power spectral density in the local frequency band. Low entropy indicates concentrated energy and good periodicity, while high entropy indicates signal interference or low signal-to-noise ratio, providing a vital signal quality indicator for fusion decision-making. Step 5: Perform fusion decision. Determine whether fusion is necessary based on two criteria: first, the target distance. and The following criteria must be met: 1) Excessive deviation or azimuth angle exceeding the radar's high-gain region; 2) Rényi entropy of the respiratory signal exceeding a threshold. If either criterion is met, the radar signal quality is deemed insufficient, triggering multi-source fusion respiratory monitoring; otherwise, independent radar estimation is used directly. Step 6: Extract respiratory signals from Wi-Fi CSI data. Through subcarrier filtering, polynomial detrending, outlier filtering, Savitzky-Golay smoothing filtering, and discrete wavelet transform, respiratory waveforms reflecting chest wall movement are extracted within the respiratory frequency band, providing a second source signal for subsequent fusion. Step 7: Perform direction correction and time alignment on the radar and Wi-Fi breathing signals. Principal component analysis is used to unify the direction of each signal (based on the radar direction), and then dynamic time warping (DTW) is used to minimize the distance between waveforms, achieving precise alignment of multi-source signals in the time domain and eliminating hardware delay and direction ambiguity; Step 8: Multi-source signal fusion based on multivariate singular spectral analysis (MSSA). The aligned respiratory signals are used to construct a Hankel trajectory matrix and stacked vertically for singular value decomposition. The signal subspace is divided according to the cumulative energy contribution rate to reconstruct the respiratory signal. Fusion weights are dynamically allocated based on the signal-to-noise ratio of each source signal before and after reconstruction to obtain a weighted fused respiratory waveform. Step 9: Respiratory Monitoring and Respiratory Rate Estimation. The fused respiratory waveform is analyzed using short-time Fourier transform (SFT) for time-frequency analysis. The peak frequency within each time window is extracted as the instantaneous respiratory rate, and a continuous respiratory rate curve is output, enabling robust respiratory monitoring in complex scenarios.

[0006] Beneficial effects: 1. This invention continues the advantages of millimeter-wave radar and Wi-Fi non-contact detection, eliminating the need for physical contact with the monitored person and preventing privacy leaks from the technical source, thus fully taking into account both humanistic care and privacy needs in the health and wellness monitoring scenario. 2. This invention utilizes multi-mode fusion detection technology, combining the respective advantages of millimeter-wave radar and Wi-Fi in detection accuracy and robustness, to achieve stable respiratory monitoring in complex scenarios; 3. This invention, through the design of a fusion decision module and a multimodal fusion module, realizes an integrated system for the assessment and fusion of respiratory signals, providing valuable ideas and practical frameworks for the integrated application of multimodal perception in the field of health monitoring. Attached Figure Description

[0007] Figure 1 A schematic diagram of the method flow of the present invention; Figure 2 Examples of two-dimensional RDM and RAM diagrams for human target localization and CFAR detection and recognition; Figure 3A schematic diagram illustrating the assessment of respiratory signal vital sign quality based on Rényi entropy; Figure 4 1. Experimental scenario and experimental equipment diagram of the present invention; Figure 5 1. Experimental results of the method of the present invention (Figure); Figure 6 Figure 1 shows the results of the ablation experiment. Detailed Implementation

[0008] The accompanying drawings are for illustrative purposes only and should not be construed as limiting the scope of this patent. To better illustrate this embodiment, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual dimensions.

[0009] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0010] A multi-mode fusion respiratory monitoring method based on millimeter-wave radar and Wi-Fi, the flowchart of which is as follows: Figure 1 As shown, the method includes the following steps: Step 1: Simultaneously acquire millimeter-wave radar echo data and CSI data from Wi-Fi devices. Perform FFT transformations on the radar echoes in the range, Doppler, and angle dimensions to generate range-Doppler and range-angle heatmaps, providing basic features for subsequent human target localization and respiratory signal extraction.

[0011] Radar echoes will be used for fusion decision-making and the fusion module. The radar data and Wi-Fi data formats are as follows:

[0012] in These represent fast time, slow time, and angular dimension, respectively. ,in and It is the time interval between fast and slow time. This represents the wavelength of the linear frequency modulated pulse. Represents virtual antenna The relative distance introduced. It is the azimuth of the target.

[0013]

[0014] in, For time-varying phase shift, For static CSI components, Represents dynamic amplitude decay. Represents the length of each path. Indicates the number of dynamic paths. This represents the carrier frequency corresponding to the wavelength, which will vary for different subcarriers.

[0015] For millimeter-wave radar data, radar echo data processing is performed in 60-second windows. Range-dimensional FFT, Doppler-dimensional FFT, and angle-dimensional FFT are performed within each sliding window (1-second step) to obtain range-Doppler and range-angle heatmaps within the ROI region. The heatmaps and localization are shown below. Figure 2 As shown.

[0016] Step 2: Apply Constant False Alarm Rate (CFAR) detection to locate potential human chest cavity targets on the heatmap. Extract the range cell with the highest number of detected targets from both the range-Doppler heatmap and the range-angle heatmap, denoted as _____. and It is used to determine the consistency of objectives in subsequent fusion decisions.

[0017] Information within non-ROI regions of the distance-Doppler heatmap is filtered out. Constant false alarm rate (CFAR) detection is applied to potential human targets to obtain multiple potential human target locations. The distance cell corresponding to the location with the highest number of detected targets is selected as the possible distance of the human target. .distance This represents the potential target distance in the human chest cavity with a suitable velocity threshold, and will be used in the decision-making process for fusion detection.

[0018] The constant false alarm rate (CFAR) test was also applied to the distance-angle heatmap to obtain multiple target results. The distance cell corresponding to the location with the highest number of detected targets was selected as the possible distance to the human chest cavity. .

[0019] Step 3: Extract the initial breathing signal from the radar echo. Perform phase unwrapping, detrending, phase differential, and outlier removal on the complex signal of the target range cell, and convert the signal to the velocity domain to enhance the periodic micro-motion characteristics caused by breathing and suppress slow body drift and abrupt noise.

[0020] For any target distance obtained, a phase extraction and unwinding algorithm is applied to the complex signal within a short sliding window to initially recover the original phase signal. This phase signal contains components such as human respiration, heartbeat, other micro-movements, and interference. To eliminate extremely low-frequency interference, detrending is applied to the initial phase signal. Subsequently, phase difference and outlier removal processes are used to avoid abrupt changes in the phase waveform, while the signal is converted to the velocity domain to enhance respiratory characteristics.

[0021] Step 4: Quantitatively evaluate the frequency domain quality of the radar breathing signal using Rényi entropy. Calculate the Rényi entropy of the breathing signal power spectral density in local frequency bands. Low entropy indicates concentrated energy and good periodicity, while high entropy indicates signal interference or low signal-to-noise ratio, providing a vital signal quality indicator for fusion decision-making.

[0022] To quantify and measure the quality of respiratory signals in the frequency domain, Rényi entropy from information technology is introduced. In the frequency domain analysis of respiratory signals, the core idea is to use entropy values ​​to quantify the energy concentration of the power spectrum, thereby filtering out sharp spectral peaks generated by genuine periodic physiological activity and suppressing broad or random spectral peaks caused by noise and interference. Simultaneously, power spectral entropy also helps capture effective signals that are not otherwise considered good at characterizing vital signs. Based on a given respiratory frequency range, the aforementioned entropy value is further optimized to select a local Rényi entropy threshold for the normalized power spectrum. Normalization ensures that the selected Rényi entropy threshold can be robustly applied to different subjects. Local entropy avoids interference from multi-peak values ​​and can be robustly applied to situations where respiratory harmonics are present. Using the power spectral density as the probability density function for calculating Rayleigh entropy, the formula is transformed into:

[0023] in , , , This represents the power spectral density of the selected frequency band. It is a discrete probability distribution. Frequency bands representing local entropy It is the order of entropy, which controls the sensitivity to probabilistic events. Here, we choose... .

[0024] The periodic breathing activity of a stationary human body produces coherent phase modulation in radar echoes, concentrating energy near the breathing frequency and exhibiting low-entropy characteristics. In contrast, environmental interference and other aperiodic disturbances, lacking coherence, result in a dispersed spectral energy distribution, exhibiting high-entropy characteristics. Examples include... Figure 3 As shown. The calculated power spectral entropy values ​​for all respiratory signals. This will be applied to the subsequent unified fusion detection decision-making process.

[0025] Step 5: Perform fusion decision. Determine whether fusion is necessary based on two criteria: first, the target distance. and If either the deviation is too large or the azimuth angle exceeds the radar's high-gain region, or if the Rényi entropy of the respiratory signal exceeds a threshold, the radar signal quality is deemed insufficient, triggering multi-source fusion respiratory monitoring; otherwise, independent radar estimation is used directly.

[0026] The fusion decision criteria are primarily based on the two main parts mentioned above to determine the vital sign quality of the echo signal. The core idea of ​​the decision based on target localization and discrimination is that the chest cavity position of a real human target does not change significantly in a short period of time and is located within the radar's high-gain reflection zone. Therefore, the position of a range-Doppler target exhibiting respiratory characteristics should largely coincide with the position of a range-angle target based on respiratory rate, and the target's range and angle relative to the radar position should not be far from its high-gain reflection zone. The core idea of ​​the decision based on vital sign quality assessment is that high-quality respiratory signals should have good periodicity. Therefore, their power spectrum should be relatively concentrated, and the local entropy of the power spectral density should be low. Based on the above core ideas of fusion decision-making, the following two main fusion decision criteria were formulated.

[0027] 1. If If a false chest cavity target is detected, the radar echo signal-to-noise ratio for life signals is determined to be low; if However, its corresponding azimuth angle If the signal exceeds the high gain range of the radar, the signal-to-noise ratio of the radar echo signal is determined to be low.

[0028] 2. For the life signal extracted based on a determined target distance, the local entropy of its power spectral density is excessively high, i.e. If the signal quality of the echo is low, then the life signal quality of the echo is determined to be low.

[0029] If the acquired radar data meets one of the criteria, it indicates that the quality of the vital signs in the echo signal is not entirely reliable, and multi-source fusion respiration detection is required. If neither criterion is met, it indicates that the quality of the vital signs in the echo is high, and respiration estimation can be performed independently based on the radar echo. In the criteria, the range threshold and azimuth threshold depend on the specific radar model and parameters. In this paper, the azimuth threshold is set to 0.2m, and the azimuth angle threshold depends on the 6dB attenuation angle of the radar transmit-receive gain. The threshold for the local entropy of the power spectral density depends on the sampling rate and the defined range of the breathing frequency; the empirical threshold set in this paper is 3.6.

[0030] Step 6: Extract respiratory signals from Wi-Fi CSI data. Through subcarrier filtering, polynomial detrending, outlier filtering, Savitzky-Golay smoothing filtering, and discrete wavelet transform, respiratory waveforms reflecting chest wall movement are extracted within the respiratory frequency band, providing a second source signal for subsequent fusion.

[0031] For the raw CSI data, some negative impacts from communication technology are mitigated by removing the pilot subcarrier and performing data completion. Polynomial fitting is used to remove non-breathing trend terms from the signal, reducing the impact of DC bias and slowly varying offset on the CSI signal. Subsequently, outlier removal is used to filter out abrupt noise in the signal. For the breathing frequency band of interest, Savitzky-Golay smoothing filtering and discrete wavelet transform are used to smooth and extract the breathing signal of interest.

[0032] Step 7: Perform direction correction and time alignment on the radar and Wi-Fi breathing signals. Principal component analysis is used to unify the direction of each signal, and dynamic time warping (DTW) is used to minimize the distance between waveforms, thereby achieving accurate alignment of multi-source signals in the time domain.

[0033] Because the amplitude and phase characteristics of Wi-Fi signal breathing differ from those of radar, the direction of breathing fluctuations reflected by CSI signals does not necessarily match actual breathing. Therefore, the directions of multi-source signals need to be corrected before fusion. Furthermore, considering that hardware time alignment may not perfectly align multi-source signals, time alignment of signals from different sources is also necessary before fusion. To correct the different directions of Wi-Fi signals, we use principal component analysis to unify the breathing direction. For the denoised signals from different sources, we calculate the first principal component of the breathing waveform. The sign of the principal component indicates the breathing direction. Using the radar breathing direction as a reference, we add all directions to the original breathing signals from different sources, thus unifying the breathing direction. Subsequently, we use dynamic time warping to minimize the distance between the radar and Wi-Fi breathing signal waveforms, achieving alignment between different source signals.

[0034] Step 8: Multi-source signal fusion based on multivariate singular spectral analysis (MSSA). Construct Hankel trajectory matrices from the aligned respiratory signals and stack them vertically for singular value decomposition; divide the signal subspace according to the cumulative energy contribution rate to reconstruct the respiratory signals; dynamically allocate fusion weights based on the signal-to-noise ratio of each source signal before and after reconstruction to obtain a weighted fused respiratory waveform.

[0035] For the processed respiratory signals from different sources, multivariate singular spectrum analysis is applied to capture the same respiratory feature contained in the multi-source signals. The fused data matrix composed of radar respiratory signals and Wi-Fi respiratory signals can be represented in the following form:

[0036] in This represents the timing of the i-th breathing signal based on radar or Wi-Fi. It refers to the number of radar and Wi-Fi breathing signal sequences.

[0037] For each time series Construct its Hankel trajectory matrix:

[0038] in It is the length of the sequence. .

[0039] The trajectory matrices of the multimodal signals are then vertically stacked to comprehensively utilize the respiratory information shared by the multi-source signals.

[0040]

[0041] Multidimensional singular spectrum analysis is performed on the multimodal trajectory matrix, resulting in several component parts that make up the multi-source signal. The signal space and noise space are then divided based on the eigenvalue energy thresholds of these components. Based on the cumulative energy contribution rate, components with cumulative energy exceeding 85% of the total energy are considered as follows: The principal components form the signal subspace, while the remaining components form the noise subspace. Based on the signal subspace, the multimodal respiratory signal trajectory matrix is ​​reconstructed through an inverse decomposition process. Then, by anti-angle averaging, the separated respiratory signal sequence can be reconstructed.

[0042] Based on the reconstructed respiratory signal, the signal-to-noise ratio (SNR) of a given multi-source signal before reconstruction is defined as the ratio of the variance of the reconstructed respiratory signal to the variance of the corresponding noise signal. The quality of respiratory features in different respiratory signal sequences is determined based on the SNR, thereby dynamically allocating fusion weights. Each fusion weight is used to obtain a weighted fused respiratory signal. .

[0043] Step 9: Respiratory Monitoring and Respiratory Rate Estimation. The fused respiratory waveform is analyzed using short-time Fourier transform (SFT) for time-frequency analysis. The peak frequency within each time window is extracted as the instantaneous respiratory rate, and a continuous respiratory rate curve is output, enabling robust respiratory monitoring in complex scenarios.

[0044] After fusion, the respiratory waveform within the window can be obtained. Time-frequency domain analysis is performed using short-time Fourier transform, and the peak amplitude in each window is taken as the respiratory rate estimate for the corresponding time, thus obtaining the respiratory rate curve changing over time, thereby realizing respiratory monitoring.

[0045] The following experiment will verify this.

[0046] To verify the proposed multi-mode fusion respiratory monitoring method based on millimeter-wave radar and Wi-Fi, a field experiment was designed and analyzed. In this paper, a radar system was built using the Texas Instruments AWR1642BOOST radar sensor, and a Wi-Fi transceiver platform was built using the PicoScenes system for experimental verification. The radar, Wi-Fi, and respiratory truth acquisition equipment are as follows: Figure 4 As shown, the signal format adopts a linear frequency modulated continuous wave operating mode, and the radar adopts a 1-transmit 4-receive operating mode. The radar experimental parameters are shown in Table 1, and the Wi-Fi experimental parameters are shown in Table 2.

[0047] Table 1

[0048] Table 2

[0049] The experimental setup and apparatus are as follows: Figure 4 As shown in the figure. The experiment mainly tested actual indoor scenarios in a laboratory and a bedroom, with floor areas of approximately 6m×6m and 3.8m×2.2m, respectively. A 1.8m×0.9m bed was placed in the experimental area, with Wi-Fi transceivers on both sides of the bed. The direct distance between the Wi-Fi transceivers was 1.8m, and the distance from the head of the bed was approximately 0.3m. A radar system was located above the head of the bed, with a vertical distance of approximately 0.6m from the bed surface. Seven subjects simulated sleep in any position within the bed area. In the experiment, we set up various test scenarios considering many influencing factors such as lying flat, lying on the side, lying prone, being covered by bedding, and radar coverage, and used the respiratory rate estimation error of the empirical cumulative distribution function as the main evaluation index.

[0050] Table 3

[0051] To systematically verify the effectiveness and necessity of each key module in the proposed algorithm and to validate the system performance, we designed and conducted a series of ablation and comparative experiments. The ablation experiments used the complete system as the baseline model, subsequently removing or replacing core processing modules to construct multiple variant models. The comparative experiments compared several other methods, including baselines for radar only and Wi-Fi only. All experiments were conducted under a unified dataset and evaluation framework to quantify the contribution of each module to the overall performance and the performance of the proposed method. The comparison results are shown in Table 2 and... Figure 5 , Figure 6As shown in the figure, the proposed complete system achieves optimal performance, with a 90% absolute error of less than 0.38 bpm. In comparison, the performance of comparative methods M1, M2, and M3 is slightly worse than that of the proposed method, with 90% absolute errors of 0.86 bpm, 1.42 bpm, and 1.83 bpm, respectively. Ablation experiments clarified the roles of each module: removing the radar-based fusion decision module reduced the 90% absolute error index to 0.69 bpm, confirming the necessity of radar signal evaluation and fusion decision assessment before fusion; removing the Rényi entropy-based respiratory signal quality measurement and screening module reduced the 90% absolute error index to 2.4 bpm, indicating the presence of low-quality respiratory signals in the echo signals, which would lead to a significant increase in error if high-quality respiratory features were not evaluated and screened; after removing the principal component analysis and dynamic time warping-based respiratory signal alignment and correction module, the 90% absolute error index decreased to 0.44 bpm, indicating that the processing of inconsistent respiratory direction correction and time alignment has a positive effect on the result index; finally, after removing the MSSA-based multi-mode signal fusion module, the 90% absolute error index decreased to 1.33 bpm, highlighting the crucial importance of extracting and fusing common respiratory features from multiple sources for the accuracy and stability of the final respiratory signal and respiratory rate estimation. This patent rigorously verifies the effectiveness of the system and its module architecture through horizontal comparisons between different methods and vertical comparisons of ablation experiments. Experimental results confirm that the proposed multi-mode fusion respiratory monitoring system based on millimeter-wave radar and Wi-Fi has good detection accuracy and stability, and each module of the system plays a key role in the system's robustness.

[0052] In summary, the above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A multi-mode fusion respiratory monitoring method based on millimeter-wave radar and Wi-Fi, characterized in that, include: Step 1: Synchronously acquire millimeter-wave radar echo data and CSI data from Wi-Fi devices; FFT transformations were performed on the radar echo in the range, Doppler, and angle dimensions to generate range-Doppler heatmaps and range-angle heatmaps, providing basic features for subsequent human target localization and respiratory signal extraction. Step 2: Apply Constant False Alarm Rate (CFAR) detection to locate potential human chest cavity targets on the heatmap; extract the range cell with the highest number of detected targets from both the range-Doppler heatmap and the range-angle heatmap, denoted as . and This is used to determine the consistency of objectives in subsequent fusion decisions; Step 3: Extract the initial breathing signal from the radar echo; perform phase unwrapping, detrending, phase difference and outlier removal on the complex signal of the target range cell, and convert the signal to the velocity domain to enhance the periodic micro-motion characteristics caused by breathing and suppress slow body drift and abrupt noise. Step 4: Quantitatively evaluate the frequency domain quality of the radar breathing signal using Rényi entropy; calculate the Rényi entropy of the breathing signal power spectral density in the local frequency band. Low entropy indicates concentrated energy and good periodicity, while high entropy indicates that the signal is interfered with or has a low signal-to-noise ratio, providing a life signal quality index for fusion decision-making. Step 5: Perform fusion decision-making; determine whether fusion is needed based on two criteria: first, target distance. and If the deviation is too large or the azimuth angle exceeds the radar's high-gain region; or if the Rényi entropy of the respiratory signal exceeds the threshold; if either criterion is met, the radar signal quality is deemed insufficient, triggering multi-source fusion respiratory monitoring; otherwise, radar independent estimation is directly used. Step 6: Extract respiratory signals from Wi-Fi CSI data; through subcarrier filtering, polynomial detrending, outlier filtering, Savitzky-Golay smoothing filtering and discrete wavelet transform, extract respiratory waveforms reflecting chest wall movement within the respiratory frequency band to provide a second source signal for subsequent fusion. Step 7: Perform direction correction and time alignment on radar and Wi-Fi breathing signals; use principal component analysis to unify the direction of each signal (based on radar direction), and then use dynamic time warping (DTW) to minimize the distance between waveforms, so as to achieve accurate alignment of multi-source signals in the time domain and eliminate hardware delay and direction ambiguity. Step 8: Perform multi-source signal fusion based on multivariate singular spectrum analysis (MSSA); construct Hankel trajectory matrices from the aligned respiratory signals and stack them vertically for singular value decomposition; divide the signal subspace according to the energy accumulation contribution rate and reconstruct the respiratory signal; dynamically allocate fusion weights based on the signal-to-noise ratio of each source signal before and after reconstruction to obtain the weighted fused respiratory waveform; Step 9: Respiratory monitoring and respiratory rate estimation; Short-time Fourier transform is used to perform time-frequency analysis on the fused respiratory waveform. The peak frequency is extracted as the instantaneous respiratory rate in each time window, and a continuous respiratory rate curve is output to achieve robust respiratory monitoring in complex scenarios.

2. The multi-mode fusion respiratory monitoring method based on millimeter-wave radar and Wi-Fi as described in claim 1, characterized in that, In step three, in order to eliminate extremely low frequency interference, detrending is applied to the initial phase signal; then, phase difference and outlier removal processing are used to avoid abrupt changes in the phase waveform, while switching to the velocity domain to enhance respiratory characteristics.

3. The multi-mode fusion respiratory monitoring method based on millimeter-wave radar and Wi-Fi as described in claim 1, characterized in that, In step four, the power spectral density is used as the probability density function for calculating the Rayleigh entropy, and the formula is transformed as follows: ； in , , , This represents the power spectral density of the selected frequency band. It is a discrete probability distribution. Frequency bands representing local entropy It is the order of entropy, which controls the sensitivity to probabilistic events. .

4. The multi-mode fusion respiratory monitoring method based on millimeter-wave radar and Wi-Fi as described in claim 1, characterized in that, In step five, two criteria are used to determine whether fusion is necessary:

1. If If a false chest cavity target exists, the radar echo signal-to-noise ratio for life signals is determined to be low; if However, its corresponding azimuth angle If it exceeds the high gain region of the radar, the signal-to-noise ratio of the radar echo life signal is determined to be low.

2. For the life signals extracted based on a determined target distance, the local entropy of the power spectral density is too large, i.e. If so, the quality of the life signal in the echo is determined to be low; If the acquired radar data meets one of the criteria, it indicates that the quality of the vital signs in the echo signal is not entirely trustworthy and is deemed necessary for multi-source fusion respiratory detection; if none of the criteria are met, it indicates that the quality of the vital signs in the echo is high and respiratory estimation can be performed independently based on the radar echo.

5. The multi-mode fusion respiratory monitoring method based on millimeter-wave radar and Wi-Fi as described in claim 1, characterized in that, In step six, for the original CSI data, some of the negative impacts brought by communication technology are removed by removing the pilot subcarrier and data completion; polynomial fitting is used to remove non-breathing trend terms in the signal and reduce the impact of DC bias and gradual offset on the CSI signal. Outlier removal was used to filter out abrupt noise in the signal; for the respiratory band of interest, Savitzky-Golay smoothing filter and discrete wavelet transform were used to smooth and extract the respiratory signal of interest.

6. The multi-mode fusion respiratory monitoring method based on millimeter-wave radar and Wi-Fi as described in claim 1, characterized in that, In step seven, all directions are incorporated into the original breathing signals from different sources to unify the breathing direction; dynamic time warping is used to minimize the distance between the radar and Wi-Fi breathing signal waveforms, thereby achieving alignment between signals from different sources.

7. The multi-mode fusion respiratory monitoring method based on millimeter-wave radar and Wi-Fi as described in claim 1, characterized in that, In step eight, multidimensional singular spectrum analysis is performed on the multimodal trajectory matrix, decomposing it into several component parts that make up the multi-source signal. The signal space and noise space are then divided based on the eigenvalue energy thresholds of the decomposed components. Based on the energy accumulation contribution rate, components with accumulated energy exceeding 85% of the total energy are considered as follows: The principal components constitute the signal subspace, and the remaining components constitute the noise subspace. Based on the signal subspace, the multimodal respiratory signal trajectory matrix is ​​reconstructed through an inverse decomposition process; then, the separated respiratory signal sequence can be reconstructed by anti-angle averaging. Based on the reconstructed respiratory signal, the signal-to-noise ratio (SNR) of a multi-source signal before reconstruction is defined as the ratio of the variance of the reconstructed respiratory signal to the variance of the corresponding noise signal. Based on the SNR, the quality of respiratory features in different respiratory signal sequences is determined, thereby dynamically allocating fusion weights. Each fusion weight is used to obtain a weighted fused respiratory signal. .

Images