A miniaturized card-type human vital signs monitoring device

Abstract

The present application relates to the field of intelligent health monitoring technology, and in particular to a miniaturized card-type human vital signs monitoring device, comprising a millimeter-wave radar module, a flexible substrate, an ADC conversion unit, and a signal processing unit; wherein the millimeter-wave radar module comprises an antenna array unit and an antenna substrate; the antenna array unit is etched on the antenna substrate using folded antenna technology; the flexible substrate and the antenna substrate are bonded; the ADC conversion unit is connected to the millimeter-wave radar module; and the signal processing unit is connected to the ADC conversion unit, for extracting human vital signs signals and calculating medical parameters. The present application adopts flexible substrate and folded antenna array technology, which greatly reduces the core module area of ​​the monitoring device, making it the size of a credit card, and can adapt to a variety of application scenarios. It can be made into the shape of a mobile phone case and coupled with a mobile phone to achieve a closed-loop monitoring system. It is simple to use and easy to carry, does not need to be worn on the surface of the human body, and can achieve non-contact medical-grade health monitoring.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent health monitoring, and in particular to a miniaturized card-type human vital signs monitoring device. Background Art

[0002] In modern life, young people stay up late, internet cafe users overuse the internet, the elderly need to protect their health, and drivers need to ensure safety. These scenarios all create an urgent need for continuous monitoring of human vital signs. Extensive data shows that continuous monitoring of vital signs can effectively reduce the risk of heart disease. Currently, health monitoring has three core requirements: continuous monitoring, non-invasive monitoring, and intelligent early warning.

[0003] However, existing monitoring technologies have shortcomings. Traditional medical equipment such as ECG (Electrocardiogram) monitors are limited by electrode contact, making it difficult to achieve long-term continuous monitoring. They are also complex to operate and the data is isolated. Millimeter wave monitoring devices are mostly fixed and large in size, making it difficult to meet portability requirements. Summary of the Invention

[0004] This application provides a miniaturized card-type human vital signs monitoring device, which adopts a miniature card-type design, equipped with a flexible substrate and intelligent interconnection technology to achieve portable and continuous intelligent health monitoring, meeting the needs of modern society for health monitoring.

[0005] This is achieved specifically through the following technical solutions:

[0006] The present application provides a miniaturized card-type human vital signs monitoring device, including a millimeter-wave radar module, a flexible substrate, an ADC conversion unit and a signal processing unit; wherein the millimeter-wave radar module includes an antenna array unit and an antenna substrate; the antenna array unit is etched on the antenna substrate using folded antenna technology; the antenna array unit is used to send a linear frequency modulated continuous wave signal and receive a heartbeat echo signal based on the linear frequency modulated continuous wave signal; the flexible substrate and the antenna substrate are bonded; the ADC conversion unit is connected to the millimeter-wave radar module and includes a mixer and an analog-to-digital converter; the mixer is used to mix the linear frequency modulated continuous wave signal and the heartbeat echo signal to generate a heartbeat intermediate frequency signal; the analog-to-digital converter is used to convert the heartbeat intermediate frequency signal into a heartbeat digital signal; the signal processing unit is connected to the ADC conversion unit, used to extract human vital signs signals from the heartbeat digital signal, and calculate medical parameters based on the human vital signs signals.

[0007] Preferably, the antenna array unit adopts a microstrip patch antenna, and the microstrip patch antenna adopts a serpentine layout that conforms to a sine wave curve.

[0008] Preferably, the flexible substrate and the antenna base are bonded by oxygen plasma activation and a silane coupling agent.

[0009] Preferably, the antenna array unit is a 4x4 antenna unit, which is embedded between flexible substrate layers through a Z-shaped folding layout, with a vertical stacking spacing of 0.3mm and an equivalent aperture of 4cm². Each antenna unit is connected by a flexible strip line with a bending radius greater than 0.1mm, and the bending loss of the flexible strip line is less than 0.2dB.

[0010] Preferably, the signal processing unit extracts human vital signs signals from the heartbeat digital signal, including: the signal processing unit generates a distance dimension signal and a Doppler dimension signal through the millimeter wave radar module, performs range-Doppler analysis on the heartbeat digital signal, and separates the target echo reflection signal, and the target echo reflection signal is used to characterize human vital signs.

[0011] Preferably, the signal processing unit processes the distance dimension signal within a single acquisition cycle. Perform fast Fourier transform analysis on the heartbeat digital signal to calculate the target distance between the monitoring device and the human body; and calculate the time window through the Doppler signal. The fast Fourier transform analysis of multiple heartbeat digital signals continuously collected in the system is used to calculate the physiological movement speed of the human body. The number of FFT points of the fast Fourier transform analysis is ; The signal processing unit uses the target distance and the human body physiological movement speed to separate and obtain the target echo reflection signal.

[0012] Preferably, the signal processing unit is further used to output the signal difference between two heartbeat digital signals collected at adjacent moments through the differential operation of the MTI filter, suppress the static component in the target echo reflection signal, and obtain the target echo dynamic reflection signal after eliminating static clutter interference.

[0013] Preferably, the signal processing unit is also used to filter the target echo dynamic reflection signal through a bandpass filter to obtain a target echo heart rate reflection signal; and, using a fast independent component analysis algorithm, separate the target echo heart rate reflection signal to obtain a target echo heartbeat reflection signal, a target echo respiratory reflection signal and a target echo body motion reflection signal.

[0014] Preferably, the signal processing unit is further configured to:

[0015] A deep learning model integrating a one-dimensional convolutional neural network and a Transformer is constructed, and the deep learning model is used to extract heartbeat waveform features from the target echo heartbeat reflection signal, respiratory waveform features from the target echo respiratory reflection signal, and body motion waveform features from the target echo body motion reflection signal; the heartbeat waveform features, respiratory waveform features, body motion waveform features, and heartbeat echo signal are used to train a generative adversarial network; wherein the discriminator of the adversarial network outputs a probability value of the authenticity of the heartbeat echo signal; the human vital sign signal is input into the deep learning model, and the heart rate, respiratory rate, and heart rate variability are output.

[0016] Preferably, the monitoring device further comprises a communication module, and the communication module is used to send the medical parameters to a mobile terminal.

[0017] Preferably, the monitoring device is in the shape of a mobile phone case as a whole, and the mobile terminal includes a smart phone; wherein the monitoring device in the shape of a mobile phone case is coupled to the smart phone.

[0018] As can be seen from the above, compared with the prior art, this application has the following beneficial technical effects:

[0019] This application uses flexible substrates and folding antenna array technology to significantly reduce the core module area of ​​the monitoring device, reducing it to the size of a credit card, making it easy to use and carry. Furthermore, by using high-frequency millimeter wave technology to form an antenna array, it can adapt to a variety of application scenarios. It does not need to be worn on the human body, enabling non-contact medical-grade health monitoring and seamless integration into daily life for all-weather health management. For example, the micro-card-shaped monitoring device of this application can be embedded in the hinge of a laptop computer, approximately 0.6 meters away from the chest, and can monitor human vital signs through millimeter waves through clothing. At the same time, it is not limited by the contact electrodes of traditional ECG electrocardiograms, enabling long-term continuous monitoring and simple operation.

[0020] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present application, nor is it intended to limit the scope of the present application. Other features of the present application will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0022] Figure 1A schematic structural diagram of a miniaturized card-type human vital signs monitoring device provided in an embodiment of the present application;

[0023] Figure 2 A schematic diagram of a flow chart of a miniaturized card-type human vital signs monitoring method provided in an embodiment of the present application;

[0024] Figure 3 A flowchart of a method for detecting and identifying device behavior status provided in an embodiment of the present application. DETAILED DESCRIPTION

[0025] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0026] See also Figure 1 , Figure 1 This is a schematic diagram of the structure of a miniaturized card-type human vital signs monitoring device provided in an embodiment of the present application, such as Figure 1 As shown, the monitoring device includes: a millimeter-wave radar module 100, a flexible substrate 200, an ADC conversion unit 300, and a signal processing unit 400. The millimeter-wave radar module 100 includes an antenna array unit and an antenna substrate; the antenna array unit is etched on the antenna substrate using folded antenna technology; the antenna array unit is used to transmit a linear frequency modulated continuous wave signal and receive a heartbeat echo signal based on the linear frequency modulated continuous wave signal; the flexible substrate 200 is bonded to the antenna substrate; the ADC conversion unit 300 is connected to the millimeter-wave radar module 100 and includes a mixer and an analog-to-digital converter; the mixer is used to mix the linear frequency modulated continuous wave signal and the heartbeat echo signal to generate a heartbeat intermediate frequency signal; the analog-to-digital converter is used to convert the intermediate frequency signal into a heartbeat digital signal; the signal processing unit 400 is connected to the ADC conversion unit and is used to extract human vital sign signals from the heartbeat digital signal and calculate medical parameters based on the human vital sign signals.

[0027] Specifically, the millimeter-wave radar module 100 uses an integrated millimeter-wave radar SOC (System-on-Chip for Millimeter-Wave Radar) in the 77 GHz frequency band. The chip supports FMCW (Frequency-Modulated Continuous Wave) frequency modulated continuous wave mode, with a bandwidth of ≥4 GHz and a distance resolution of ≤1 mm.

[0028] The antenna array unit is designed as a 4x4 antenna array, using two transmitters and four receivers to form a virtual 4x4 MIMO array, which helps improve spatial resolution. A microstrip patch antenna can be used, etched onto a polyimide substrate. The polyimide substrate serves as the antenna base, with a thickness of less than 0.5 mm and a dielectric constant of ε = 3.5.

[0029] The flexible substrate 200 is made of PDMS (Polydimethylsiloxane), a biocompatible flexible material with a thickness of 0.5 mm and a stretchability of >50%. The PDMS flexible substrate utilizes a multi-layered heterogeneous composite structure, balancing flexibility, high-frequency signal transmission, and mechanical strength. The 0.5 mm thick PDMS flexible substrate has a dielectric constant ε ≈ 2.7 and is formed on a glass carrier via a spin coating process and peeled off after curing. A 10 μm thick polyimide film is laminated on the PDMS surface, serving as the substrate for the 4x4 antenna array to reduce dielectric loss.

[0030] In some feasible embodiments, PDMS is bonded to the polyimide film through oxygen plasma activation and silane coupling agent, with an adhesion force >2N / cm² to avoid interlayer delamination.

[0031] In some feasible embodiments, a copper mesh (mesh density 200 mesh) is embedded in the PDMS bottom layer to shield external electromagnetic interference while maintaining the bendability of the substrate.

[0032] In some feasible implementations, the monitoring device uses a flexible solid-state battery with a capacity of 50 mAh, and the millimeter-wave radar SOC chip has an operating power consumption of <100 mW and a standby power consumption of <1 mW.

[0033] In some feasible implementations, the millimeter-wave radar SOC chip can also be integrated with a Bluetooth 5.0 module, which can transmit data to mobile terminals such as mobile phones / tablets in real time, allowing users to view the vital signs of the monitored subjects in real time.

[0034] In some feasible implementations, the 4x4 antenna array is embedded between PDMS layers in a Z-shaped folded layout using folded antenna technology, with a vertical stacking spacing of 0.3mm and an equivalent aperture compressed to 4cm². Embedding a serpentine wire layout between PDMS layers can prevent circuit breakage during bending. Specifically, the microstrip patch antenna uses a sinusoidal curve (wavelength λ = 0.5mm, amplitude A = 0.2mm), and the stretchability is increased to more than 50%. Inkjet printed silver nanowires (line width 20μm, thickness 5μm), with a resistivity of <5×10⁻ 8Ω·m. Laser-drilled holes (30μm in diameter) are filled with conductive silver paste to achieve vertical interconnection between PDMS layers, with an impedance matching error of <5%. The microstrip patch antenna operates at 77GHz and measures 1.5mm×1.5mm. Each antenna element is connected by a flexible stripline with a bending radius greater than 0.1mm and a bending loss of less than 0.2dB.

[0035] In some feasible embodiments, the monitoring device further includes a communication module 500 for transmitting the medical parameters to a mobile terminal. It is understood that the medical parameters can also be transmitted to the cloud, enabling automatic synchronization of monitoring data with a medical cloud platform, providing periodic health trend reports and professional advice, and acting as a cloud-based AI doctor.

[0036] In some feasible embodiments, the monitoring device is in the shape of a mobile phone case as a whole, and the mobile terminal includes a smart phone; wherein the monitoring device in the shape of a mobile phone case is coupled with the smart phone to realize a closed monitoring loop. That is, the miniaturized card-type human vital signs monitoring device described in this application is manufactured into the shape of a mobile phone case and coupled with a smart phone. The high-frequency millimeter wave of the device can penetrate the mobile phone case and clothing, extract the heart rate or respiratory rate, and realize non-contact continuous monitoring of human vital signs. For example, when a user plays games / browses short videos for a long time, it is easy to cause hidden health risks such as a surge in cardiac load, respiratory alkalosis, or missed detection of sudden death precursors. Through the method described in the embodiment of the present application, non-contact continuous monitoring can be realized while the user is using a smart phone. If the user's vital signs indicators are monitored to exceed the preset risk threshold, an early warning prompt will be issued through the smart phone, which is conducive to reducing the hidden health risks caused by long-term gaming or short video browsing.

[0037] The monitoring device described in this application utilizes a 0.5mm-thick PDMS flexible substrate and foldable antenna array technology, reducing the core module area to the size of a credit card. It is simple to use and portable. The antenna array, formed using high-frequency millimeter-wave technology, is adaptable to a variety of application scenarios. It does not require wearable clothing, enabling contactless, medical-grade health monitoring and seamless integration into daily life for around-the-clock health management. Furthermore, it is free from the contact electrodes of traditional ECG monitors, enabling long-term continuous monitoring with simple operation.

[0038] In summary, the monitoring device described in this application can be applied in a variety of application scenarios, such as:

[0039] The decline in cardiopulmonary function caused by prolonged sitting, hidden heart rate abnormalities caused by work stress, and the inability of traditional equipment to continuously monitor in office settings. The card-type monitoring device described in this application is embedded in the hinge of a laptop computer (0.6m from the chest), using millimeter waves to penetrate clothing to monitor key indicators.

[0040] Baby monitors require a wristband, which can cause skin allergies; camera monitoring poses a privacy risk. The card-type monitoring device described in this application can be embedded in the crib rail (at a distance of 0.8-1.2m) to monitor breathing rhythm using millimeter waves that penetrate the quilt.

[0041] Heart rate monitors restrict the chest and affect athletic performance, and smartwatches are affected by sweat and have reduced detection accuracy. Therefore, magnetic wear can be used to attach the card-type monitoring device described in this application to the outside of sportswear to dynamically track vital signs while running or swimming.

[0042] Patients with atrial fibrillation require continuous monitoring for abnormal heartbeats, while those with COPD need to track changes in breathing patterns. Therefore, the card-type monitoring device described in this application can be embedded in a smart medicine box (0.5 meters from the chest) to simultaneously monitor vital sign fluctuations during medication use. This automatically generates a vital sign change curve for remote review by doctors. In the event of cardiac arrest, community emergency response can be automatically called.

[0043] Fatigue-related accidents among long-distance drivers account for 37%, and existing DMS systems have a high false alarm rate (30% of missed detections occur due to facial occlusion). The card-type monitoring device described in this application can be embedded in the steering wheel (0.3m from the driver's chest), allowing millimeter waves to penetrate clothing for continuous monitoring.

[0044] Long-term gaming / short video browsing can easily lead to hidden health risks:

[0045] ① Increased cardiac load: Competitive games can cause abnormal heart rhythm (>120 BPM for 10 minutes).

[0046] ② Respiratory alkalosis: Tense hyperventilation (respiratory rate > 25 / minute).

[0047] ③Missed detection of signs of sudden death: Sudden ventricular arrhythmia when using a mobile phone alone at night.

[0048] A micro millimeter wave monitoring card can be integrated into a mobile phone case. The millimeter waves penetrate the phone case and clothing to extract heart rate / respiration rate.

[0049] The above are typical application scenarios. Users can extend them to more scenarios as needed. The application scenarios of this invention are extremely wide.

[0050] Here's how it works:

[0051] See also Figure 2 , Figure 2 A flowchart diagram of a miniaturized card-type human vital signs monitoring method provided in an embodiment of the present application is shown in FIG. Figure 2 As shown, including the following:

[0052] 201. The millimeter wave radar module generates a linear frequency modulated continuous wave signal.

[0053] The millimeter-wave radar SOC chip integrated in the millimeter-wave radar module 100 generates a linear frequency-modulated continuous wave (FMCW) signal. The frequency starts at 77 GHz and increases linearly to 81 GHz over 1 ms at a frequency modulation rate of 2 × 10^12 Hz / s. This generates a single chirp signal (i.e., a chirp signal) with a duration of Tc = 1 ms. A GaAs (Gallium Arsenide) power amplifier amplifies the chirp signal to 12 dBm, driving a 4 × 4 MIMO antenna array to transmit the millimeter-wave signal, also known as the linear frequency-modulated continuous wave (FMCW). This millimeter-wave signal penetrates clothing (attenuating approximately 3 dB) and the surface of the skin (penetration depth 8 cm), reaching the heart and chest surface. Heartbeats cause periodic chest displacement (amplitude 0.1-0.5 mm), resulting in a phase shift in the reflected wave (every 0.1 mm of displacement corresponds to a 0.12° phase difference).

[0054] Based on the modeling of the above transmission signal, the mathematical expression of the FMCW signal can be:

[0055] ;

[0056] in,

[0057] is the FMCW signal, expressed as a time domain signal;

[0058] A is the amplitude of the FMCW signal;

[0059] is the starting frequency of the FMCW signal, i.e. 77 GHz mentioned above.

[0060] S is the frequency modulation slope, which is 2×10^12 Hz / s mentioned above;

[0061] t is the time variable (in seconds).

[0062] 202. The millimeter wave radar module receives a heartbeat echo signal based on the linear frequency modulation continuous wave signal.

[0063] The four receiving antennas capture the reflected signals based on the linear frequency modulated continuous wave signal and amplify the weak echoes through a low noise amplifier (LNA, gain 25dB).

[0064] 203. The mixer of the ADC conversion unit mixes the heartbeat echo signal with the transmitted linear frequency modulation continuous wave signal to generate a heartbeat intermediate frequency signal.

[0065] Heartbeat echo signal delay , then the mathematical expression of the mixed heartbeat intermediate frequency signal can be:

[0066] ;

[0067] After ignoring the high-order terms, the frequency of the heartbeat intermediate frequency signal is for:

[0068] ;

[0069] in,

[0070] S is the frequency modulation slope;

[0071] It is the delay time of the heartbeat echo signal;

[0072] R is the target distance, wherein the target distance represents the physical distance between the monitoring device and the human body (heart or chest cavity);

[0073] c is the speed of light.

[0074] in, represents the linear phase term related to the target distance R and velocity v, which determines the frequency of the heartbeat intermediate frequency signal; It represents the fixed phase offset caused by the starting frequency of the FMCW signal and the delay of the heartbeat echo signal; represents a high-order phase term (quadratic term), which can be ignored at short distances (where τ is minimal).

[0075] The frequency difference of the heartbeat intermediate frequency signal is proportional to the target distance, as shown in the following formula:

[0076] ;

[0077] Wherein, Δf is the frequency difference of the heartbeat intermediate frequency signal.

[0078] 204. The analog-to-digital converter of the ADC conversion unit is used to convert the heartbeat intermediate frequency signal into a heartbeat digital signal.

[0079] The heartbeat intermediate frequency signal is sampled at 4 GSPS by a 14-bit ADC and converted to a digital signal. A digital down-conversion (DDC) brings the signal to baseband, where it is decimated and filtered, reducing the data rate to 400 kSPS.

[0080] 205. The signal processing unit performs range-Doppler analysis on the digital heartbeat signal.

[0081] The millimeter-wave radar SOC chip generates distance and Doppler signals to separate the reflected signals from the heart area. This includes:

[0082] (1) The signal processing unit processes the distance dimension signal within a single acquisition cycle. The heartbeat digital signal is subjected to fast Fourier transform analysis to calculate the target distance between the monitoring device and the human body.

[0083] Perform a 1024-point FFT (Fast Fourier Transform) analysis on each chirp signal to calculate the target distance. The calculation formula is as follows:

[0084] ;

[0085] in,

[0086] bin peak is the bin number where the peak value in the FFT spectrum is located;

[0087] fs is the sampling rate (4 GSPS, i.e. 4×10 9 Samples / s)

[0088] S is the frequency modulation slope (2×10 12 Hz / s)

[0089] is the number of first FFT points.

[0090] Each chirp undergoes a 1024-point FFT analysis, meaning that the 1024 time-domain signal points collected within a single chirp cycle are analyzed using a Fast Fourier Transform. By calculating the target distance, a range profile is generated. A range threshold (e.g., 0.3-1.2 m) is set to isolate reflected signals from the heart or chest area and filter out non-human targets.

[0091] (2) The Doppler signal is used to calculate the time window The fast Fourier transform analysis of multiple heartbeat digital signals continuously collected in the system is used to calculate the physiological movement speed of the human body. The number of FFT points of the fast Fourier transform analysis is .

[0092] For example, for 128 consecutive Chirp signals (time window ) to perform 256-point FFT analysis to calculate the human body's physiological motion speed (for example, including heart rate: the periodic displacement speed of the chest cavity caused by heart contraction / relaxation, and respiratory rate: the chest cavity rise and fall speed caused by lung expansion and contraction). The calculation formula is as follows:

[0093] ;

[0094] in,

[0095] ;

[0096] in,

[0097] λ is the millimeter wave wavelength (3.9 mm);

[0098] fd is the Doppler frequency shift;

[0099] PRF is the pulse repetition frequency, which is the number of pulses emitted per second; ;

[0100] bin peak is the bin number where the peak value in the FFT spectrum is located;

[0101] is the number of points in the second FFT.

[0102] By performing 256-point FFT analysis on 128 consecutive chirps, velocity information is extracted (resolution 0.05 m / s) and information other than human physiological movement speed is filtered out.

[0103] 206. The signal processing unit uses the target distance and the physiological movement speed of the human body to separate and obtain the target echo reflection signal, and the target echo reflection signal is used to represent the vital signs of the human body.

[0104] 207. The signal processing unit outputs the signal difference between the two heartbeat digital signals collected at adjacent moments through the differential operation of the MTI filter, suppresses the static component in the target echo reflection signal, and obtains the target echo dynamic reflection signal after eliminating static clutter interference.

[0105] The MTI (Moving Target Indication) filter is used to eliminate interference from static objects in the environment, such as static background reflections from walls and furniture. Its mathematical expression can be expressed through the transfer function:

[0106] ;

[0107] in,

[0108] H(z) is the frequency domain characteristic of the filter, which suppresses the static component by the difference between the current signal and the previous sampling point signal;

[0109] Indicates a unit delay operation (i.e., the signal of the previous sampling point).

[0110] Differential operation:

[0111] ;

[0112] in,

[0113] x[n] is the input signal at the current moment (nth sampling point);

[0114] x[n−1] is the input signal at the previous moment (n−1th sampling point);

[0115] y[n] is the output signal after filtering, which represents the signal difference between two adjacent sampling points.

[0116] Frequency Response:

[0117] ;

[0118] in,

[0119] f is the frequency of the input signal (unit: Hz).

[0120] T is the pulse repetition period (unit: seconds), T=1 / PRF.

[0121] The cutoff frequency is 0.1Hz, which suppresses the static component of f<0.1Hz, eliminates static background reflections such as walls and furniture, retains the dynamic heartbeat / respiration signals, and obtains the dynamic reflection signal of the target echo.

[0122] 208. The signal processing unit filters the target echo dynamic reflection signal through a bandpass filter to obtain a target echo heart rate reflection signal;

[0123] For example, by setting the parameters of the band-pass filter, the frequency band of 0.5 to 5 Hz is retained, corresponding to the heart rate signal of 30 to 300 BPM.

[0124] 209. The signal processing unit uses a fast independent component analysis algorithm to separate the target echo heart rate reflection signal to obtain a target echo heartbeat reflection signal, a target echo breathing reflection signal and a target echo body motion reflection signal.

[0125] (1) Preprocessing the target echo heart rate reflection signal:

[0126] Centralize the target echo heart rate reflection signal:

[0127] ;

[0128] in,

[0129] is the original observation signal matrix with dimension N×T, where N is the number of sensor channels and T is the number of time samples;

[0130] is the mean vector (dimension N×1) of each channel signal, which is used to eliminate the DC component in the signal.

[0131] Perform whitening processing on the target echo heart rate reflection signal:

[0132] ;

[0133] Where Wwhiten is the whitening matrix, which is obtained by eigendecomposition of the covariance matrix; Wwhiten=Λ−1 / 2·UT; U and Λ are The eigenvector matrix and eigenvalue diagonal matrix of . Z is the signal matrix after whitening processing, which satisfies Z·ZT=I and eliminates the second-order correlation between channels.

[0134] (2) Iterative optimization:

[0135] The objective function is to maximize the non-Gaussianity (approximate negative entropy) function:

[0136] ;

[0137] in,

[0138] ω is the separation weight vector, dimension m×1;

[0139] G(u) is a nonlinear function (logcosh(u)), G(u)=logcosh(u);

[0140] v is a standard Gaussian variable (mean 0, variance 1) used to contrast non-Gaussianity.

[0141] E[⋅] is the mathematical expectation (statistical mean).

[0142] Weight update:

[0143] ;

[0144] g(ωTZ)=tanh(ωTZ): derivative of the nonlinear function.

[0145] g′(ωTZ)=1−tanh2(ωTZ): second-order derivative (implicit in the formula).

[0146] Normalization: ω=w+ / ||w+||, ensures that the weight vector is normalized.

[0147] Until convergence.

[0148] (3) Separation signal:

[0149] The column vectors of the unmixing matrix W are independent components, and the signal S=W·Z.

[0150] W is an m×m matrix, each column of which corresponds to a separation weight vector ω.

[0151] S is the separated independent signal component matrix with dimension m×n.

[0152] Output the target echo heartbeat reflection signal, target echo breathing reflection signal and target echo body motion reflection signal components.

[0153] 210. The signal processing unit calculates medical parameters.

[0154] The signal processing unit constructs a deep learning model that is a fusion of a one-dimensional convolutional neural network and a Transformer, and uses the deep learning model to extract heartbeat waveform features from the target echo heartbeat reflection signal, extract respiratory waveform features from the target echo respiratory reflection signal, and extract body motion waveform features from the target echo body motion reflection signal; uses the heartbeat waveform features, respiratory waveform features, body motion waveform features and heartbeat echo signals to train a generative adversarial network; wherein the discriminator of the adversarial network outputs a probability value of the authenticity of the heartbeat echo signal; inputs the human vital sign signal into the deep learning model, and outputs heart rate, respiratory rate and heart rate variability.

[0155] (1) 1D-CNN (One-Dimensional Convolutional Neural Network) and Transformer fusion model:

[0156] Input layer: The input size is 512×4 (512 time steps, 4 antenna channels).

[0157] 1D-CNN convolutional layer: 64 filters, kernel size 5, stride 1, activation function ReLU is:

[0158] ;

[0159] in,

[0160] i is the output time step index (range: 0 ≤ i<508);

[0161] j is the filter index (output channel, range: 0 ≤ j < 64);

[0162] k is the temporal position of the convolution kernel (0 ≤ k<5);

[0163] d is the input channel index (range: 0 ≤ d<4);

[0164] is the weight of the convolution kernel at time position k and input channel d;

[0165] is the value of the input data at time step (i+k) and channel (j−d);

[0166] b is the bias term.

[0167] Output layer: The output size is 508×64.

[0168] Transformer Encoder:

[0169] Multi-head self-attention (4 heads):

[0170] ;

[0171] Here, Q (Query), K (Key), and V (Value) represent the three matrices obtained by linearly transforming the input signal, representing the query, key, and value, respectively. The input feature dimension is 64 (corresponding to the 64-dimensional Transformer encoder input size of 508×64). The dimension of each attention head is dk (here dk = 64 / 4 = 16), indicating that the 64-dimensional input is split into four independent self-attention heads (each with 16 dimensions). dk is a scaling factor used to prevent excessive dot product results from causing vanishing gradients.

[0172] The feed-forward network (FFN) is:

[0173] ;

[0174] W1 and W2 represent learnable weight matrices; W1 maps the input from 64 dimensions to a higher dimension, and after ReLU activation, it is mapped back to 64 dimensions through W2. b1 and b2 represent bias terms.

[0175] The output size of the feed-forward network (FFN) is 508×64.

[0176] Global average pooling: pooling along the time dimension 508×64→64

[0177] Fully connected layer: outputs three medical parameters: heart rate (HR), respiratory rate (BR), and heart rate variability (HRV).

[0178] Training configuration:

[0179] Loss function: mean squared error (MSE) and cross entropy loss of peak heart rate;

[0180] Optimizer: Adam (learning rate 3×10−4).

[0181] (2) Generative Adversarial Network (GAN):

[0182] Network structure:

[0183] Generator input: mmWave signal x∈R512×4;

[0184] Contains 5 layers of 1D transposed convolution, upsampling to ECG waveform y∈R512;

[0185] The last layer activation function: Tanh (output normalized to [-1, 1]);

[0186] Discriminator:

[0187] Input: real ECG or generated signal y;

[0188] Structure: 4 layers of 1D convolution (kernel size 5, stride 2) + fully connected layer;

[0189] Output: probability value p∈[0,1];

[0190] The probability value p∈[0,1] represents the discriminator’s judgment on the authenticity of the input signal. If p = 1, the discriminator believes that the input signal is a real medical-grade physiological signal; if p = 0, the discriminator believes that the input signal is a simulated signal forged by the generator.

[0191] Loss function:

[0192] Wasserstein GAN loss (stable training): ;

[0193] Where E represents the mathematical expectation, which means averaging the losses of batch samples; D represents the discriminator network, which outputs a score for the authenticity of the input signal (a larger value indicates a more realistic one); G(x) represents the generator network, which converts the input millimeter-wave signal x into a simulated ECG waveform; x represents the input of the generator, with a dimension of x∈R512×4, representing the signal of 512 time steps and 4 antenna channels obtained from the millimeter-wave radar.

[0194] ;

[0195] Where E[D(y)] represents the expected value of the discriminator's score on the true ECG signal y; y represents the true ECG waveform with dimension y∈R512; E[D(G(x))] represents the expected value of the discriminator's score on the generated signal G(x); λ represents the gradient penalty coefficient, which controls the weight of the gradient penalty term; the gradient penalty represents the regularization term that constrains the discriminator's gradient to prevent training collapse.

[0196] In addition, the signal processing unit of the monitoring device described in this application is also used to dynamically adjust the filter parameters of the 1D-CNN convolutional layer using the Recursive Least Squares (RLS) algorithm to achieve adaptive filtering to accommodate different users' body fat percentages and posture changes. When the monitoring device described in this application is used for the first time, a 30-second benchmark test is performed to establish a user's heartbeat waveform template. In subsequent monitoring, dynamic time warping (DTW) matching is used to achieve personalized user modeling.

[0197] The specific implementation is as follows:

[0198] (1) Recursive least squares (RLS) filtering:

[0199] First initialize the filter:

[0200] ;

[0201] ;

[0202] in,

[0203] represents the initial value of the filter weight;

[0204] represents the initial value of the inverse covariance matrix;

[0205] δ is the regularization parameter; I is the identity matrix.

[0206] Then iterate and update:

[0207] The following validation error is used to measure the deviation between the prediction result of the current weight estimate ω(n−1) for the input signal vector x(n) and the true value d(n):

[0208] ;

[0209] in,

[0210] x(n) represents the input signal vector at the nth moment;

[0211] d(n) represents the expected output at the nth moment.

[0212] The gain vector for weight updates is dynamically adjusted by the following formula:

[0213] ;

[0214] in, It represents the product of the inverse covariance matrix P(n-1) and the input x(n), reflecting the spatial distribution of the input signal; Contains the forgetting factor λ and the projection of the input signal, which is used to normalize the gain.

[0215] Finally, the weights are updated as follows:

[0216] ;

[0217] in,

[0218] Represents the updated filter weights.

[0219] The weights are adjusted by Kalman gain and error to make the prediction closer to the true value.

[0220] The inverse covariance matrix is ​​updated as follows:

[0221] ;

[0222] in,

[0223] The forgetting factor λ=0.99, λ-1 is used to exponentially decay the influence of old data;

[0224] represents the updated inverse covariance matrix.

[0225] The outer product term k(n)x(n)ᵀP(n−1) can reduce the uncertainty of the estimate.

[0226] Dynamically adjust the filter parameters to adapt to the differences in signal attenuation caused by different users' body fat percentages and posture changes, completing adaptive filtering.

[0227] (2) Dynamic Time Warping (DTW)

[0228] When a user uses the device for the first time, a 30-second benchmark test is performed to establish a personalized heartbeat waveform template, i.e., a template signal, as shown below: t 1 ,t 2 ,...,t M};in, t i Indicates the template signal i The amplitude value of each sampling point (such as heartbeat waveform voltage); M is the total number of sampling points of the template signal (i.e., the time window length).

[0229] The currently detected heartbeat waveform fragment, that is, the real-time signal is: S={ s 1 ,s 2 ,...,s N } ;in, sj Indicates real-time signal j The amplitude value of each sampling point (such as heartbeat waveform voltage); N is the total number of sampling points of the real-time signal (i.e., the length of the time window).

[0230] The cost matrix D is constructed based on the template signal and the real-time signal as follows:

[0231] ;

[0232] in,

[0233] ;

[0234] It can be seen that the cost matrix D is the local distance matrix between each point of the template signal and the real-time signal. Indicates the template signal i Point and real-time signal j The absolute difference of the points reflects the local similarity.

[0235] Then the optimal path is solved by dynamic planning through recursive formula, as shown below:

[0236] ;

[0237] Dacc(i,j) represents the minimum cumulative cost from the starting point (1,1) to the current point (i,j);

[0238] Dacc(i−1,j) represents the cumulative cost of moving from the left to the current point;

[0239] Dacc(i,j−1) represents the cumulative cost of moving from the top to the current point;

[0240] Dacc(i−1,j−1) represents the cumulative cost of moving from the diagonal to the current point.

[0241] Path constraints: monotonicity, continuity, and boundary alignment.

[0242] Finally, the similarity score is calculated by the minimum cumulative cost, as shown in the following formula:

[0243] ;

[0244] Among them, score represents the total difference after the template signal and the real-time signal are aligned. The smaller the value, the higher the similarity.

[0245] Finally, the threshold θ is used for judgment. If score < θ, it is determined to be a normal heartbeat waveform.

[0246] In some feasible implementations, in order to achieve all-weather endurance, the monitoring device described in this application also includes an energy consumption management module for adaptively optimizing device power consumption through mode division and state migration.

[0247] In the embodiment of the present application, the intelligent hierarchical energy consumption management of the device is mainly explained using a three-level perception mode.

[0248] Level 1, Continuous Monitoring Mode (100mW): Full-featured medical-grade monitoring suitable for emergency response or abnormality tracking. Triggers include active activation or automatic activation when HRV ≥ 50ms or apnea ≥ 10 seconds is detected.

[0249] Level 2, Interval Monitoring Mode (25mW): For daily health monitoring, the 6-axis sensor automatically switches to active mode when the user detects inactivity. When the radar sampling rate (initial setting: 80Hz) drops to 10Hz, the redundant RF channels are disabled (retaining the 2T4R array).

[0250] Level 3, trigger monitoring mode (1mW): When the device is idle, it only maintains basic signal detection, the millimeter wave radar is in deep sleep, and the 6-axis motion sensor and MCU perform low-power monitoring.

[0251] The built-in 6-axis motion sensor can automatically identify the device status (handheld / desktop / wearable) to switch to the optimal working mode.

[0252] In some feasible embodiments, to achieve multi-scenario adaptive monitoring, the monitoring device described herein also includes a behavior state recognition module for detecting and identifying the device's behavior state. A multimodal perception network is implemented by integrating a six-axis motion sensor (a three-axis accelerometer and a three-axis gyroscope). The hardware configuration specifically utilizes an inertial measurement unit (MEMS IMU), which has a dynamically adjustable range of ±8g / ±16g acceleration and ±250dps / ±2000dps gyroscope. The behavior state recognition module connects to the signal processing unit's main microcontroller unit (MCU) via an SPI bus (Serial Peripheral Interface), supporting a 100Hz sampling rate and DMA (Direct Memory Access) transmission to ensure timing synchronization error of <1ms. The acceleration signal undergoes an IIR low-pass filter (cutoff frequency 5Hz), and the gyroscope data is applied with a sliding average window (window length 10 points) to achieve dynamic noise reduction. When the device is first started, a 9-axis calibration (combined with the geomagnetic sensor) is performed to eliminate installation posture deviation.

[0253] See also Figure 3 , Figure 3 A flow chart of the device behavior status detection and identification method provided in the embodiment of the present application is shown as follows: Figure 3 As shown, the following is a detailed description of the steps for the behavior state recognition module to detect and identify the device behavior state:

[0254] 301. Calculate the variance of the device acceleration modulus.

[0255] The calculation formula is as follows:

[0256] ;

[0257] in,

[0258] is the device acceleration modulus variance;

[0259] N is the number of sampling points in the time window;

[0260] is the composite modulus of the three-axis acceleration at the i-th sampling point;

[0261] is the mean value of the acceleration modulus within the time window.

[0262] 302.Judgment Is it less than a preset variance threshold (which can be set to 0.05g² in the embodiment of the present application); if so, go to step 303; otherwise, go to step 305.

[0263] 303. Calculate the pitch angle of the device.

[0264] The pitch angle of the device is calculated in real time using the quaternion method. The calculation formula is as follows:

[0265] ;

[0266] in,

[0267] is the pitch angle of the device;

[0268] is the x-axis component of the three-axis accelerometer;

[0269] is the y-axis component of the three-axis accelerometer;

[0270] is the z-axis component of the three-axis accelerometer.

[0271] 304. Determine whether the pitch angle of the device is equal to 0. If so, identify the device behavior state as desktop state. Otherwise, go to step 305.

[0272] 305. Perform FFT analysis on the acceleration Z-axis signal to extract the energy proportion (i.e., periodicity index) of the preset target frequency band (which can be set to 1-4 Hz in the embodiment of the present application).

[0273] 306. Determine whether the energy ratio is less than a preset ratio threshold (which can be set to 0.7 in the embodiment of the present application). If so, identify the device behavior state as a wearable state; otherwise, identify the device behavior state as a handheld state.

[0274] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of the systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or portion of code, which contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions marked in the boxes can also occur in an order different from that marked in the accompanying drawings. For example, two boxes shown in succession can actually be executed substantially in parallel, or they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, as well as combinations of boxes in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified functions or operations, or can be implemented using a combination of dedicated hardware and computer instructions.

[0275] The above description is merely a preferred embodiment of the present application and an illustration of the technical principles employed. Those skilled in the art should understand that the scope of the invention herein is not limited to technical solutions formed by a specific combination of the aforementioned technical features, but also encompasses other technical solutions formed by any combination of the aforementioned technical features or their equivalents without departing from the inventive concept. For example, a technical solution formed by replacing the aforementioned features with (but not limited to) technical features with similar functions disclosed in this invention.

Claims

1. A miniaturized card-type human vital signs monitoring device, characterized in that: The monitoring device includes a millimeter wave radar module, a flexible substrate, an ADC conversion unit and a signal processing unit; wherein, The millimeter wave radar module includes an antenna array unit and an antenna substrate; the antenna array unit is etched on the antenna substrate using folded antenna technology; the antenna array unit is used to send a linear frequency modulation continuous wave signal and receive a heartbeat echo signal based on the linear frequency modulation continuous wave signal; The flexible substrate is bonded to the antenna base; The ADC conversion unit is connected to the millimeter wave radar module and includes a mixer and an analog-to-digital converter; the mixer is used to mix the linear frequency modulation continuous wave signal and the heartbeat echo signal to generate a heartbeat intermediate frequency signal; the analog-to-digital converter is used to convert the heartbeat intermediate frequency signal into a heartbeat digital signal; The signal processing unit is connected to the ADC conversion unit and is used to extract human vital sign signals from the heartbeat digital signal and calculate medical parameters based on the human vital sign signals; The signal processing unit extracts a human vital sign signal from the heartbeat digital signal, comprising: the signal processing unit performs range-Doppler analysis on the heartbeat digital signal through the range dimension signal and the Doppler dimension signal generated by the millimeter wave radar module, and separates the target echo reflection signal to obtain the target echo reflection signal, wherein the target echo reflection signal is used to represent the human vital sign; The signal processing unit processes the distance dimension signal within a single acquisition cycle. Perform fast Fourier transform analysis on the heartbeat digital signal to calculate the target distance between the monitoring device and the human body; and calculate the time window through the Doppler signal. The fast Fourier transform analysis of multiple heartbeat digital signals continuously collected in the system is used to calculate the physiological movement speed of the human body. The number of FFT points of the fast Fourier transform analysis is The signal processing unit uses the target distance and the human body physiological movement speed to separate the target echo reflection signal; The signal processing unit is further configured to output a signal difference between two heartbeat digital signals collected at adjacent moments through a differential operation of an MTI filter, thereby suppressing a static component in the target echo reflection signal and obtaining a target echo dynamic reflection signal after eliminating static clutter interference; The signal processing unit is also used to filter the target echo dynamic reflection signal through a bandpass filter to obtain a target echo heart rate reflection signal; and use a fast independent component analysis algorithm to separate the target echo heart rate reflection signal to obtain a target echo heartbeat reflection signal, a target echo respiratory reflection signal and a target echo body motion reflection signal.

2. A miniaturized card-type human vital signs monitoring device as claimed in claim 1, characterized in that: The antenna array unit adopts a microstrip patch antenna, and the microstrip patch antenna adopts a serpentine layout that conforms to a sine wave curve.

3. The miniaturized card-type human vital signs monitoring device according to claim 1, characterized in that: The flexible substrate and the antenna base are bonded by oxygen plasma activation and a silane coupling agent.

4. The miniaturized card-type human vital signs monitoring device according to claim 1, characterized in that: The signal processing unit is further configured to: Constructing a deep learning model that integrates a one-dimensional convolutional neural network and a Transformer, and using the deep learning model to extract heartbeat waveform features from the target echo heartbeat reflection signal, respiratory waveform features from the target echo respiratory reflection signal, and body motion waveform features from the target echo body motion reflection signal; Using the heartbeat waveform features, respiratory waveform features, body motion waveform features and heartbeat echo signals, a generative adversarial network is trained; wherein a discriminator of the adversarial network outputs a probability value of authenticity of the heartbeat echo signal; The human body vital sign signal is input into the deep learning model, and the heart rate, respiratory rate and heart rate variability are output.

5. A miniaturized card-type human vital signs monitoring device according to any one of claims 1 to 4, characterized in that: The system also includes a communication module, which is used to send the medical parameters to a mobile terminal.

6. The miniaturized card-type human vital signs monitoring device according to claim 5, characterized in that: The monitoring device is in the shape of a mobile phone case as a whole, and the mobile terminal includes a smart phone; wherein the monitoring device in the shape of a mobile phone case is coupled to the smart phone.

Images