Non-invasive multi-target blood pressure monitoring using passive beamforming mm-wave FMCW radar

Abstract

Provided herein are a system and method for non-contact, non-invasive, and continuous monitoring of blood pressure in at least one subject or target. The system and method utilize beamforming radar, for example a mm-wave frequency modulated continuous wave (mm-wave FMCW) radar, with beam switching and beam scanning abilities to focus or scan a beam sequentially on multiple areas of a subject's body, to capture signals reflected therefrom and to process the captured signals via at least one machine learning algorithm or via analog neural network as predicted blood pressure.

Description

[0001] NON-INVASIVE MULTI-TARGET BLOOD PRESSURE MONITORING USING PASSIVE BEAMFORMING MM-WAVE FMCW RADAR

[0002] Cross-Reference to Related Applications

[0003] This international patent application claims benefit of priority under 35 U.S.C. §119(e) of provisional patent application U.S. Serial No. 63 / 634,106, filed April 15, 2024, the entirety of which is hereby incorporated in its entirety.

[0004] Federal Funding Legend

[0005] This invention was made with government support under Grant Number 2336852 awarded by the National Science Foundation. The government has certain rights in the invention.

[0006] BACKGROUND OF THE INVENTION

[0007] Field of the Invention

[0008] The present invention relates generally to the fields of medicine and monitoring systems. More specifically, the present invention relates to non-contact blood pressure monitoring device via a beamforming radar.

[0009] Description of the Related Art

[0010] Developing continuous vital signs monitoring is essential for elderly people and young adults to maintain a healthy life (1-4). Invasive blood pressure monitoring is the direct method of arterial blood pressure monitoring where a catheter is placed in the patient and a cannula needle is inserted through an artery. This method is used for patients during surgery in hospitals (5, 6). Although this method is accurate, it affects the patient during daily monitoring and increases the risk of injury (7). Non-invasive blood pressure methods include cuff-based and cuff-less techniques (8). Cuff-less techniques consist of mobile applications and wearable health devices (WHD) (9-12) using wearable smartwatches, bracelets, etc. Also, sensors such as inertial sensors including accelerometers, magnetometers, and gyroscopes measure body movements (10). However, these devices cannot provide accurate results and cause discomfort especially during physical activity. Cuff-based sphygmomanometer devices are used by doctors in hospitals to analyze the blood pressure of a patient while placing the stethoscope on the patient’s chest (12). Recently, an oscillometric cuff-based device called OMRON 10 series is used by many individuals to monitor their blood pressure at home continuously and comfortably, and completes the measurements 45 seconds (13).

[0011] To overcome the various challenges faced in contact vital signs measurements, radar-based measurements are utilized. Yang Zhen et al. (14, 15) proposed a novel system for vital signs detection using IR-UWB radar. Owing to high-range resolutions, a denoising algorithm improves the signal-to-noise (SNR) ratio to detect the respiration rates and heart rates of humans and is useful for other applications, such as people counting systems (16), human gesture recognition (17), location tracking of objects (18). However, it has a high cost and is much more sensitive to heart rate detection.

[0012] CW Doppler radars can transmit and receive signals continuously, but are ineffective in detecting multiple targets (19). However, FMCW radars can detect various targets in comparison with CW radars (20, 21 ). Texas Instruments utilizes an FMCW mm-wave radar operating at 77 GHz that can detect vital signs such as heart rate and breathing rate using the feature extraction algorithms and XGBoost classifier (22). Furthermore, a recent paper (23) uses a Tl’s IWR1443 mm-wave radar to detect the vital signs of indoor walking persons over three route maps and a complete ensemble empirical mode decomposition with an adaptive noise method to detect vital signs for multiple targets.

[0013] Current studies utilize deep learning models compared with other machine learning models which is preferred for constructing many datasets and for tuning the hyperparameters. Models such as Long short-term memory (LSTM) neural networks, convolutional neural networks (CNN), and deep neural networks (DNN) have been used for both systolic blood pressure (SBP), and diastolic blood pressure (DBP) prediction (25, 26) where the signals are pre-processed using Pulse transit time (PTT) based features and ECG statistical features (27) in this study for blood pressure estimation.

[0014] Thus, there is a need in the art for a system that greatly enhances the efficiency and accuracy of blood pressure monitoring in a healthcare setting, providing real-time, non-contact monitoring for multiple individuals. Specifically, there is a need for a system which integrates a Rotman lens's multibeam and wide-angle scanning capabilities with the precise wave control and directionality of the holographic leaky wave meta surfaces that monitors blood pressure non-invasively both to detect multiple individuals and to focus on specific targets on an individual. The present invention fulfills this long-standing need in the art.

[0015] SUMMARY OF THE INVENTION

[0016] The present invention is directed to a system for continuously monitoring blood pressure in at least one subject. The system has a passive beamforming radar with a beam switching mechanism, and a beam scanning mechanism. A receiver is configured to receive signals reflected from the at least one subject. The system has means for processing signals continuously that are received from the passive beamforming radar and are processed to measure blood pressure in the at least one subject. The system has a display on which to continuously show blood pressure values.

[0017] The present invention is further directed to a method for non-invasive monitoring of blood pressure for at least one subject in real time. In this method, at least one beam from the passive beamforming radar comprising the system described herein is focused continuously at a distance from and sequentially onto a plurality of areas on the at least one subject and signals reflected from each of the plurality of areas on the at least one subject are captured and the signals are passed through the receiver. The signals received from the receiver are inputted into the at least one machine learning algorithm or into an analog neural network / machine learning methods. Measured values for blood pressure for the at least one subject processed from the inputted signals are continuously outputted. The present invention is directed to a related method further comprising displaying the measured values for blood pressure.

[0018] The present invention is directed further to a system for continuously monitoring blood pressure in multiple targets. The system has a mm-wave frequency modulated continuous wave (mm-wave FMCW) radar with a Rotman lens at a front end configured for beam steering, a quasi-holographic leaky-wave antenna (quasi-HLWA) disposed thereon configured for beam scanning and a 25° linear tapered substrate configured to improve power transfer between the Rotman lens and the quasi-HLWA. A six-port receiver is configured for phase detection of signals. An analog neural network is configured to process continuously signals received as input from the mm- wave FMCW radar after phase detection, where the signals are processed as output as blood pressure values in each subject in the plurality. The system has a display on which to continuously show the outputted blood pressure values of the multiple targets.

[0019] The present invention is directed further still to a non-invasive method for tracking blood pressures of multiple targets in real time. In the method a plurality of beams from the mm-wave frequency modulated continuous wave (mm-wave FMCW) radar comprising the system described herein are continuously scanned at a distance from and onto a plurality of areas on the multiple targets. The signals reflected from each of the plurality of areas on each of the multiple targets are captured and the signals are inputted into an analog neural network. The inputted signals are processed as measured values for blood pressures and the measured values for blood pressures are outputted continuously. The measured values for blood pressures for the multiple targets are displayed as they are outputted.

[0020] The present invention is directed further still to a non-invasive method for detecting changes in cognitive and behavioral states in a subject. In the method a plurality of beams from the mm-wave frequency modulated continuous wave (mm- wave FMCW) radar comprising the system as described herein are scanned continuously at a distance from and onto a plurality of areas on the subject. The signals reflected from each of the plurality of areas on the subject are captured and inputted into an analog neural network; The inputted signals are processed as blood pressure values and correlated with physiological changes associated with the cognitive and behavioral states in the subject. The present invention is directed to a related method further comprising measuring heart rate or respiratory rate or a combination thereof; and integrating values for the heart rate or values for the respiratory rate with the blood pressure values prior to the correlating step.

[0021] Other and further aspects, features, benefits, and advantages of the present invention will be apparent from the following description of the presently preferred embodiments of the invention given for the purpose of disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] So that the matter in which the above-recited features, advantages and objects of the invention, as well as others that will become clear, are attained and can be understood in detail, more particular descriptions of the invention briefly summarized above may be had by reference to certain embodiments thereof that are illustrated in the appended drawings. These drawings form a part of the specification. It is to be noted, however, that the appended drawings illustrate preferred embodiments of the invention and therefore are not to be considered limiting in their scope.

[0023] FIG. 1 is an illustration showing two radars directed at the carotid artery and at the heart and the resulting phase changes.

[0024] FIG. 2 is the flow chart for obtaining the pulse waveform where a range-FFT is applied for a single target range bin and extracting the phases to obtain the pulse waveform for chest and neck displacements.

[0025] FIG. 3 is the flow chart for blood pressure prediction.

[0026] FIGS. 4A-4B shows the waveforms for the filtered chest waveform (FIG. 4A) and the filtered neck waveform (FIG. 4B).

[0027] FIGS. 5A-5B shows the power spectral densities (PSD) of the chest waveform (FIG. 5A) and the neck waveform (FIG. 5B).

[0028] FIGS. 6A-6B shows the autocorrelations of the function of the chest waveform (FIG. 6A) and the function of the neck waveform (FIG. 6B).

[0029] FIG. 7 shows the cross-correlation function of the chest and neck waveform.

[0030] FIG. 8 shows the evolution of the cost function for systolic blood pressure (SBP).

[0031] FIG. 9 is the error histogram plot for SBP.

[0032] FIG. 10 shows the evolution of the cost function for diastolic blood pressure (DBP).

[0033] FIG. 11 is the error histogram plot for DBP.

[0034] FIG. 12 is a picture of the experimental measurement setup.

[0035] FIG. 13 shows a leaky-wave antenna unit cell.

[0036] FIGS. 14A-14B show the Rotman lens (RL) as designed (FIG. 14A) and fabricated (FIG. 14B).

[0037] FIG. 15 shows the S-parameters at the Rotman lens input ports. FIGS. 16A-16C illustrate the design steps for the quasi-holographic leaky-wave antenna (HLWA).

[0038] FIG. 17 shows the interference pattern at 28 GHz.

[0039] FIGS. 18A-18B are dispersion diagrams of the leaky-wave antenna unit cell showing the normalized phase constant (FIG. 18A) and the attenuation for different Hs heights.

[0040] FIGS. 19A-19C show the quasi-HLWA as designed (FIGS. 19A-19B) and fabricated (FIG. 19C).

[0041] FIG. 20 illustrates the gain comparison with and without the 25° slope transmission line for smooth and efficient power transfer between the Rotman lens and the quasi-HLWA.

[0042] FIG. 21 shows 2D plots of the scanning performance of the quasi-HLWA antenna.

[0043] FIG. 22 shows the measured results for the main beam direction for each port.

[0044] FIG. 23 shows a six-port receiver implementing an analog neural network (ANN) layer.

[0045] FIG. 24 shows a 3-layer analog neural network.

[0046] FIG. 25 is a block diagram showing the analog neural network optimization.

[0047] FIG. 26 shows a six-port receiver.

[0048] FIGS. 27A-27B show the S-parameters (phase) at port 1 (FIG. 27A) and port 2 (FIG. 27B) of the six-port receiver.

[0049] DETAILED DESCRIPTION OF THE INVENTION

[0050] As used herein, the articles “a” and “an” when used in conjunction with the term “comprising” in the claims and / or the specification, may refer to “one”, but it is also consistent with the meaning of “one or more”, “at least one”, and “one or more than one”. Some embodiments of the invention may consist of or consist essentially of one or more elements, components, method steps, and / or methods of the invention.

[0051] As used herein, the term “or” in the claims refers to “and / or” unless explicitly indicated to refer to alternatives only or the alternatives are mutually exclusive, although the disclosure supports a definition that refers to only alternatives and “and / or”. As used herein, the terms “comprise” and “comprising” are used in the inclusive, open sense, meaning that additional elements may be included. Correspondingly, the terms “consists of’ and “consisting of’ are used in the exclusive, closed sense, meaning that additional elements may not be included.

[0052] As used herein, the term “subject” or “target” are used interchangeably and refers to a person, individual or patient receiving blood pressure monitoring as described herein.

[0053] In one embodiment of the present invention, there is provided a system for continuously monitoring blood pressure in at least one subject, comprising a passive beamforming radar with a beam switching mechanism and a beam scanning mechanism; a receiver configured to receive signals reflected from the at least one subject; means for processing signals continuously that are received from the passive beamforming radar and are processed to measure blood pressure in the at least one subject; and a display on which to continuously show blood pressure values.

[0054] In one aspect of this embodiment, the means for processing signals may be a computer having a memory and a processor that tangibly stores at least one machine learning algorithm. In another aspect of this embodiment, the means for processing signals may be an analog neural network.

[0055] In this embodiment and aspects thereof, both systolic blood pressure values and diastolic blood pressure values may be displayed. Also in this embodiment and aspects thereof the passive beamforming radar may be a mm-wave frequency modulated continuous wave (mm-wave FMCW) radar. The mm-wave FMCW radar may comprise a Rotman lens at a front end, a quasi-holographic leaky-wave antenna (quasi-HLWA) disposed thereon and a 25° linear tapered substrate configured to improve power transfer between the Rotman lens and the quasi-HLWA. Particularly, the Rotman lens may be configured for beam steering in an elevation (0) plane via a switch among input ports and the quasi-HLWA is configured for frequency-controlled beam scanning in an azimuth (< ) plane. In addition, the receiver may be a six-port receiver configured for phase detection of the signals.

[0056] In another embodiment of the present invention, there is provided a method for non-invasive monitoring of blood pressure for at least one subject in real time, comprising the steps of focusing continuously at least one beam from the passive beamforming radar comprising the system, as described supra, at a distance from and sequentially onto a plurality of areas on the at least one subject; capturing signals reflected from each of the plurality of areas on the at least one subject; passing the signals through the receiver; inputting the signals received from the receiver into at least one machine learning algorithm or into an analog neural network; and outputting continuously measured values for blood pressure processed from the inputted signals for the at least one subject. Further to this embodiment, the method comprises displaying the measured values for blood pressure.

[0057] In both embodiments, the passive beamforming radar may be a mm-wave frequency modulated continuous wave (mm-wave FMCW) radar comprising a Rotman lens at a front end configured for beam steering, a quasi-holographic leaky-wave antenna (quasi-HLWA) disposed thereon configured for beam scanning and a 25° linear tapered substrate configured to improve power transfer between the Rotman lens and the quasi-HLWA; where the focusing step may comprise scanning the at least one beam over the plurality of areas on the at least one subject. The receiver may be a six-port receiver, where the passing step may comprise detecting a phase of the signals passing through the six-port receiver. In both embodiments, the measured values for blood pressure may be for systolic blood pressure or diastolic blood pressure or a combination thereof.

[0058] In both embodiments, the plurality of areas on the subject may comprise at least two of a chest, a neck, a wrist, a leg or any other body part. Particularly, the plurality of areas may comprise at least two of a heart in the chest, a carotid artery in the neck, a femoral artery in the leg, or a pulse in the wrist. More particularly, the plurality of areas may be the heart in the chest and the carotid artery in the neck.

[0059] In yet another embodiment of the present invention, there is provided a system for continuously monitoring blood pressure in multiple targets, comprising a mm-wave frequency modulated continuous wave (mm-wave FMCW) radar with a Rotman lens at a front end configured for beam steering, a quasi-holographic leaky-wave antenna (quasi-HLWA) disposed thereon configured for beam scanning and a 25° linear tapered substrate configured to improve power transfer between the Rotman lens and the quasi-HLWA; a six-port receiver configured for phase detection of signals; an analog neural network configured to process continuously signals received as input from the mm-wave FMCW radar after phase detection, where the signals are processed as output as blood pressure values in each subject in the plurality; and a display on which to continuously show the outputted blood pressure values of the multiple targets.

[0060] In this embodiment, the Rotman lens may be configured for beam steering in an elevation (0) plane via a switch among input ports and the quasi-HLWA may be configured for frequency-controlled beam scanning in an azimuth (c|)) plane. In this embodiment, the blood pressure values are systolic blood pressure values and diastolic blood pressure values.

[0061] In yet another embodiment of the present invention, there is provided a non- invasive method for tracking blood pressures of multiple targets in real time, comprising scanning continuously a plurality of beams from the mm-wave frequency modulated continuous wave (mm-wave FMCW) radar comprising the system of claim 15 at a distance from and onto a plurality of areas on the multiple targets; capturing signals reflected from each of the plurality of areas on each of the multiple targets; inputting the signals into an analog neural network; processing the inputted signals as measured values for blood pressures; outputting continuously the measured values for blood pressures; and displaying the measured values for blood pressures for the multiple targets as they are outputted.

[0062] In this embodiment, the plurality of areas on each of the multiple targets may comprise at least two of a chest, a neck, a wrist, or a leg. Also, the plurality of areas on the subject comprises at least two of a chest, a neck, a wrist, a leg, an arm, or a stomach or lower abdomen. Particularly, the plurality of areas comprise at least two of a heart in the chest, a carotid artery in the neck, a radial artery in the wrist, a femoral artery in the leg, a brachial artery in the arm, or an iliac artery in the stomach or lower abdomen.

[0063] In yet another embodiment of the present invention there is provided a non- invasive method for detecting changes in cognitive and behavioral states in a subject, comprising scanning continuously a plurality of beams from the mm-wave frequency modulated continuous wave (mm-wave FMCW) radar comprising the system as described supra at a distance from and onto a plurality of areas on the subject; capturing signals reflected from each of the plurality of areas on the subject; inputting the signals into an analog neural network; processing the inputted signals as blood pressure values; and correlating the blood pressure values with physiological changes associated with the cognitive and behavioral states in the subject. Further to this embodiment the method comprises measuring heart rate or respiratory rate or a combination thereof; and integrating values for the heart rate or values for the respiratory rate with the blood pressure values prior to the correlating step.

[0064] Provided herein are a system and method for non-contact monitoring that enables continuous blood pressure monitoring in real time without direct contact, solving problems related to inconvenience and discomfort associated with traditional blood pressure monitoring methods. The systems and methods provided herein are non-invasive, non-contact and leverage the use of beamforming to use one single radar to measure blood pressure. The non-contact blood pressure monitoring method uses a single beamforming radar which captures phase differences at two different locations, for example, the chest, leg and neck of at least one subject, resulting from pulse pressure, such as from, the carotid artery, the femoral artery and the pulse at the wrist from one subject or simultaneously from multiple subjects.

[0065] The system provided herein comprises a mm-wave frequency modulated continuous wave (mm-wave FMCW) radar with a Rotman lens, a quasi-holographic leaky-wave antenna (quasi-HLWA) and a receiver, for example, a six-port receiver, to perform multi-target tracking whereupon the signal is passed through the six-port receiver for phase detection. After phase detection, the signal serves as input to the analog neural network to measure blood pressure or to obtain measured blood pressure values in real-time, which overcomes a limitation in cuff-based systems. The system has a radar front-end such that power consumption is reduced by utilizing the Rotman Lens combined with the Holographic Leaky-Wave Antenna. The signals captured by the beam forming radar are processed continuously as input and displayed continuously as blood pressures as output. For example, the signals may be input and processed by using, preferably, an analog neural network, analog neural network / machine learning methods or, alternatively, by at least one machine learning algorithm tangibly stored on a computer having a memory and a processor.

[0066] Particularly, structurally, the quasi-HLWA has a novel leaky-wave unit cell design that enables both forward and backward radiation thereby enhancing scanning capabilities. Holographic principles are implemented for optimal unit cell placement. Moreover, a 25° linear tapered substrate that improves power transfer between the Rotman lens and the quasi-HLWA.

[0067] The non-contact blood pressure monitoring method uses a single beamforming radar which captures phase differences at two different locations, for example, at the chest, neck, wrist, leg, arm, or lower abdomen or stomach of at least one subject, resulting from heart beat and pulse pressure, such as from, the heart, the carotid artery, the radial artery, the femoral artery, the brachial artery and the the iliac artery from one subject or simultaneously from multiple subjects.

[0068] Also provided are methods and systems to correlate blood pressure dynamics with cognitive and behavioral states by utilizing blood pressure pulse obtained from radar sensors and advanced machine learning algorithms and by detecting subtle physiological changes that could correspond to psychological stress, emotional arousal, and cognitive load or other complex cognitive and behavioral states. Heart rate and respiration rate may be integrated to obtain a more comprehensive and robust model of an individual’s mental state, recognizing that a nuanced assessment requires the synthesis of multiple physiological parameters.

[0069] The system and methods provided herein are useful for the following.

[0070] Non-Contact Health Kiosks: Malls, airports, and public places could have kiosks where people may quickly gauge their blood pressure without any physical contact, ensuring hygiene and ease of use.

[0071] Vehicle Integration: Car manufacturers may integrate this non-contact system into car seats or steering wheels. Monitoring drivers’ blood pressures may be part of a broader safety system, alerting drivers if their blood pressures indicate potential health risks. Additionally, the non-contact system may be integrated into smart health devices, such as smartphones and wearables to enhance vehicular safety through health monitoring.

[0072] Workplace Health: Offices may set up non-contact stations where employees may routinely check their blood pressure, especially in high-stress environments, promoting workplace wellness without the need for direct contact.

[0073] Remote Patient Monitoring Systems: For patients quarantined or in isolation rooms, especially in cases such as contagious diseases, where the non-contact feature ensures that healthcare workers can monitor blood pressure without direct patient contact, reducing the risk of transmission.

[0074] Elderly Care Facilities: Given that elderly individuals might have fragile skin or conditions that make contact-based measurements uncomfortable, non-contact systems are perfect for routine checks in nursing homes or assisted living facilities. Other Applications: The non-contact system is useful for sleep monitoring, general health monitoring, for example, but not limited to, during sleep, and fitness tracking. The non-contact system is useful for monitoring respiration, blood pressure and / or pulse rate in a subject during sleep, while exercising or going about their daily routine as an indicator of health, status of an existing condition, such as a sleep disorder, or the possible onset of a disease or disorder.

[0075] The following examples are given for the purpose of illustrating various embodiments of the invention and are not meant to limit the present invention in any fashion.

[0076] EXAMPLE 1

[0077] Materials and Methods

[0078] System Setup

[0079] The system uses Tl’s IWR1443BOOST FMCW radar-based evaluation module (EVM) operating at a frequency of 76-81 GHz. The system uses a combination of 2 IWR sensors, each consisting of 4 receiving antennas (Rx) and 3 transmitting antennas (Tx) contained in the RF subsystem. The subsystem also consists of an ADC buffer which stores the ADC outputs of received IF signals. These outputs are then processed by extracting the phase change of the range bins from the 2D FFT. The OMRON device is used as a measurement baseline to compare the blood pressure results from the radar, as illustrated in FIG. 1.

[0080] Principles of FMCW mm-wave radar

[0081] FMCW radars transmit a frequency-modulated signal continuously in order to measure the range, velocity, and angle of the target. The transmitted signal is expressed as: where AT is the amplitude of the chirp signal, fcis the starting frequency of the chirp signal, B is the bandwidth of the chirp, and Tcis the chirp duration. After transmission, the receiver signal is reflected and produces a time-delayed signal which is represented as: where td is the time delay of the signal, td = 2R(t) / c , where R(t) is the range of the target and c is the speed of light. Both the transmitter and receiver signals are mixed and produce an intermediate frequency signal after the simplification of the in-phase signal, which is expressed as: where AR is the received signal amplitude, fb is the beat frequency, fb = 2BR(t) / Tcc, and the phase of the beat signal is given by:

[0082] The residual phase noise in y(t) can be ignored for short-range applications. So, we use the range (< 1 .5 m) for vital signs detection. Here, the phase signal reflects due to small-scale movements of the heartbeat where phase can be measured using FFT and computing the phase shift of the target range bin. After neglection, the beat signal for the nth ADC sample and mth chirp is represented as: where Tf is the fast time axis after ADC sampling and Tsis the slow time axis. As this radar is capable of detecting multiple targets, we use a single target at each time in stationary conditions for detecting the vital signs.

[0083] Workflow for vital signs detection

[0084] The flow chart for obtaining a pulse waveform is illustrated in FIG. 2. Each chirp consists of 100 ADC samples with a duration of 50ps based on the Intermediate frequency (IF) sampling rate of 2 MHz. To select the target range-bin, Fast Fourier transform (FFT) is performed on each chirp to get the Range FFT. Then it processes the phase signal and extracts the phase unwrapping from the range bin along the slow time axis, and hence the process for vital signs extraction is obtained.

[0085] The phase extraction step obtains phase values from this selected range bin, which are subsequently unwrapped beyond their inherent (-IT, IT) limitations by adding or subtracting 2TT when phase discontinuities occur, thus accurately representing subtle displacements of the chest and neck during respiration and cardiac activity. A phase difference operation is then performed on these unwrapped values to eliminate phase drift and enhance the heartbeat signal, followed by impulse noise removal to filter out artifacts and errors. The refined signal undergoes bandpass filtering through serially cascaded Bi-Quad HR filters that operate in real-time on the continuous data stream, separating the respiratory and cardiac components. The final stage employs spectral estimation techniques — including FFT, peak interval analysis, and autocorrelation — to precisely calculate breathing and heart rates from their respective waveforms, with all results ultimately displayed on the Graphical User Interface (GUI) for clinical evaluation and monitoring.

[0086] Deep learning model with optimization

[0087] We propose a deep neural network (DNN) model tuned by BO for predicting the SBP and DBP values. In (35), an intelligent heuristic optimization algorithm called bonobo optimizer (BO) was proposed. Bonobos follow a fission-fusion social style where they form different sizes of groups and move in different territories. Later, they are observed to merge again with social members performing various activities. Bonobos have adopted four different reproductive strategies which include restrictive mating, promiscuous mating, extra-group mating, and consortship mating, and these strategies are modeled and proposed as a BO algorithm. Here, the optimization algorithm chooses the alpha bonobo (abonobo) as the highest rank value, and parameters are initially set for the optimization. The bonobo optimization process begins with the initialization of BO controlling parameters randomly. Then, the bonobos are selected through the fission-fusion process and participate in mating processes for the creation of a new bonobo. During the positive phase, a new bonobo is created through promiscuous or restrictive mating. The creation of a new bonobo is represented as follows: new_bonobo‘- = bonobo;

[0088] ( bonobo'- - bonobo'’ 1

[0089] (6), where new_bonoboj,j varies from 1 to d d is the number of variables; bonobo'j denotes the jth variable of the / th bonobo; ri is the random number created in the range (0, 1 ); scab and scsb are the sharing parameters of the bonobo and alpha bonobo; abonoboj represents the / th variable of alpha bonobo; bonobopj is the / th variable of the pth bonobo. During the negative phase, a new bonobo is created through promiscuous or extra-group mating. The variable boundary conditions are implemented on new bonobos. Then, the fitness of new bonobos is evaluated. The acceptance criteria are applied on the new population of bonobos. Afterwards, the controlling parameters are updated. From the selection process to parameter updates, the process continues in a cyclic way until the algorithm satisfies the stopping criterion. By using the BO algorithm, a deep learning model is tuned. In the input layer, the linear type of activation function has been considered. However, in case of hidden layers and an output layer, either log sigmoid or tan sigmoid activation function can be applied. Therefore, the developed model has 54 inputs, 3 hidden layers each consisting of 15 neurons and 1 output.

[0090] Method for blood pressure prediction

[0091] The system is shown in FIG. 3. Briefly radar A and radar B are placed in front of the subject where subjects are in a stationary condition. Then, the radar detects the chest and neck pulse waveforms from the subject with post signal processing methods and values are recorded. In like manner, datasets have been collected from 53 subjects and features are extracted. Afterward, the SBP feature data and DBP feature data are given to the DNN model tuned by BO for training. Finally, SBP and DBP values are predicted.

[0092] EXAMPLE 2

[0093] Results

[0094] Signal processing and feature extraction

[0095] Chest and neck pulse waveforms are detected in which the radar is adjusted parallel to the subject and pointed towards the chest and neck with the effect of orientation, range, and movements. Each observation is recorded for 128 data points, and therefore the recorded samples are built up to the model.A Graphical user interface (GUI) were generated to monitor the heart rate, breathing rate values using MATLAB 2021 b. Datasets have been collected from 53 subjects and are processed with signal processing. A total of 54 features are extracted. Features such as age, gender, and BMI are considered. First, a sampling frequency fs = 5Hz is set for the signal to be processed. The ECG signals are extracted using statistical features based on RR intervals for peak-to-peak heartbeat detection. Moreover, the noise signal is removed by applying the frequency spectral domain called FFT to the chest and neck pulse waveform which is shown in FIGS. 4A-4B.

[0096] The max and min function is applied for both chest and neck waveforms Power spectral density (PSD) function. FIGS. 5A-5B show the PSD of chest and neck waveforms. It can be clearly seen that at the peak of the PSD, the frequency is around 1 .5 Hz for both the chest and the neck, which agrees well with typical values of BPM.

[0097] The autocorrelation and cross-correlation functions are utilized using the maximum and minimum peak magnitude value, and the frequency with maximum and minimum magnitude. FIGS. 6A-6B shows the autocorrelation function of chest and neck waveforms. The transmitted signal being a chirp, the sine pattern can clearly be seen when performing the autocorrelation. The cross-correlation plot of chest and neck waveform is illustrated in FIG. 7.

[0098] Additionally, PTT based features such as kurtosis, skewness, rms are obtained as a major feature. The PTT is obtained by initialising the radars at the exact same time and measuring the time difference between the radar radiating at the chest and at the neck. Furthermore, the normalization technique has been applied to compute the mean, variance, standard deviation of the chest and neck pulse waveforms. The above features are highly important to build machine learning models. Hence, a total of 525A-54 data are obtained, i.e., 525 rows and 54 columns (features).

[0099] Training of DNN model

[0100] SBP and DBP prediction

[0101] The training of the deep neural network model for blood pressure prediction demonstrates significant efficacy in the current research. A DNN architecture comprising an input layer with 54 features, three hidden layers (each with 15 neurons), and an output layer was implemented. Training of the SBP model proceeded through 10,000 iterations, achieving an RMSE of 0.30689, which represents great prediction accuracy. FIG. 8 depicts the convergence behavior of the cost function during SBP model training, with the error histogram plot (FIG. 9) confirming a concentration of errors near zero. For the DBP prediction model, similar training parameters were employed over 10,000 iterations, resulting in an RMSE of 0.2281. The evolution of the cost function for the DBP model, illustrated in FIG. 10, shows comparable convergence characteristics.

[0102] The error histogram plot for DBP (FIG. 11) further validates the model’s precision. The utilization of the BO proved instrumental in parameter tuning, with the input layer employing linear activation functions while hidden and output layers implemented either log sigmoid or tan sigmoid functions.

[0103] This activation function configuration effectively captured the complex nonlinear relationships inherent in the cardiovascular system. The dataset of 525 samples with 54 features provided sufficient training material for the model to learn the intricate patterns between the extracted radar features and corresponding blood pressure values.

[0104] Testing the model

[0105] The trained deep learning model was tested on 15 subjects. Measurements taken from the OMRON BP monitor device were used to validate the predicted BP values using the proposed method. We conducted two measurements in our experiment: one detecting the chest and neck pulse waveforms for 5 pulses at 1 m (distance between radar and subject) and another detecting the chest and neck pulse waveforms for 1 pulse (30 seconds) at 0.5-1 m. For every subject, values were recorded from the radar and features were extracted. Both SBP and DBP were predicted with high accuracy. Table 1 shows the testing results of SBP and DBP, comparing the proposed method and the OMRON device for 5 pulses in terms of root mean square error (RMSE), mean absolute error (MAE), mean square error (MSE), standard deviation (SD), and R-squared metrics. Similarly, SBP and DBP comparison results for 1 pulse are illustrated in Tables 1 and 2.

[0106] TABLE 1

[0107] 5 pulses at 1 m

[0108] When combining the overall results of 5 pulses and 1 pulse, the RMSE for SBP was calculated as 0.669, and RMSE for DBP was calculated as 0.747. The mean squared error for SBP was calculated as 0.899 and for DBP as 0.571. The mean absolute error and standard deviation for SBP were calculated as 2.267 ± 1.340 mmHg, and for DBP as 2.433 ± 1 .616 mmHg. The R value of SBP was calculated as 0.964 and for DBP as 0.840. Moreover, our proposed models for SBP and DBP were trained in the ranges (90,160) mmHg and (51 ,95) mmHg, respectively.

[0109] TABLE 2

[0110] 1 pulse at 0.5-1m

[0111] When combining the overall results of 5 pulses and 1 pulse, the RMSE for SBP was calculated as 0.669, and RMSE for DBP was calculated as 0.747. The mean squared error for SBP was calculated as 0.899 and for DBP as 0.571. The mean absolute error and standard deviation for SBP were calculated as 2.267 ± 1.340 mmHg, and for DBP as 2.433 ± 1 .616 mmHg. The R value of SBP was calculated as 0.964 and for DBP as 0.840. Moreover, our proposed models for SBP and DBP were trained in the ranges (90,160) mmHg and (51 ,95) mmHg, respectively.

[0112] Comparison with other models

[0113] During the training process, the deep neural network (DNN) model was compared with other machine learning models for both SBP and DBP. A regression learner app available in MATLAB was used to train all the machine learning models. Tables 3 and 4 describe the training performance of the DNN (last rows) for systolic (SBP) and diastolic (DBP) blood pressure. TABLE 3

[0114] Comparison of all machine learning models with our model for SBP

[0115] TABLE 4 Comparison of all machine learning models with our model for DBP

[0116] From Tables 3 and 4, the DNN model for SBP and DBP achieved better accuracy compared to other machine learning models in terms of RMSE, MAE, MSE, and R2.

[0117] Measurement results

[0118] FIG. 12 displays the radar sensors mounted on a tripod, positioned to capture the heart rate where the upper radar targeting the carotid artery in the neck and the lower radar directed at the heart for chest pulse waveforms. A cuff-based OMRON blood pressure monitor serves as a reference device to validate measurements from the non-contact radar system. The GUI with multiple data visualization panels shows real-time measurements, including breathing rate and heart rate monitoring plots.

[0119] The measurement results presented in Tables 5-6 demonstrate the comparative accuracy of the radar-based blood pressure monitoring system against the traditional OMRON device. Table 5 shows the Systolic Blood Pressure (SBP) and Diastolic Blood Pressure (DBP) measurements with 5 pulses at 1 meter distance for 15 subjects.

[0120] TABLE 5

[0121] SBP and DBP results with 5 pulses for 1 meter The SBP values measured by radar ranged from 94 to 148 mmHg, while the OMRON device showed values between 97 and 144 mmHg. The Mean Squared Error (MSE) for SBP measurements varied from 0 to 16, with a Mean Absolute Error (MAE) between 0 and 4 mmHg. For DBP, the radar measurements ranged from 63 to 88 mmHg compared to OMRON’s 64 to 84 mmHg, with MSE values between 0 and 16 and MAE values of 0 to 4 mmHg.

[0122] Table 6 displays results for the more challenging scenario of single-pulse measurements at 0.5-1 meter distance. In this case, SBP measurements by radar ranged from 95 to 137 mmHg versus 96 to 137 mmHg by OMRON, with slightly higher errors (MSE up to 36, MAE up to 6 mmHg).

[0123] TABLE 6

[0124] SBP and DBP results with 1 pulse for 0.5-1 meter

[0125] Similarly, DBP measurements in the single-pulse scenario showed radar values between 63 and 88 mmHg compared to OMRON’s 66 to 86 mmHg, with MSE values up to 36 and MAE values up to 6 mmHg. These results confirm that the proposed radarbased method achieves clinically acceptable accuracy for blood pressure monitoring, with slightly better performance in the 5-pulse scenario compared to the single-pulse measurements. Thus, there is close agreement between the OMRON and radar sensor measurement results with a mean absolute error of approximately ±5 mmHg for each. These results confirm the feasibility of accurately measuring BP for a single target.

[0126] EXAMPLE 3

[0127] 3D-orinted mmWave quasi-holoqraohic antenna for 2D beamforminq

[0128] Provided is a 2D-scanning antenna design that combines a 3D-printed Rotman lens with a quasi-holographic leaky-wave antenna (HLWA). This integration enables beam-scanning capabilities in both elevation and azimuth planes. The Rotman lens provides beam steering in the elevation plane by switching between input ports, while the quasi-HLWA achieves frequency-controlled beam scanning in the azimuth plane. The 3D-printed mmWave quasi-holographic antenna for 2D beamforming may be scaled to 77 GHz.

[0129] Leaky-Wave Antenna (LWA) Unit Cell

[0130] A unit cell simulation is made using CST MWS (11 , 13, 14). FIG. 13 shows the design of the UC and Table 7 presents its design parameters. The dispersion diagram of the unit cell reveals the propagation constant (P) of the leaky-wave mode as a function of frequency. Periodic LWAs often exhibit right-hand (RH) and left-hand (LH) behavior in their dispersion characteristics (15). This unique property enables the LWA to radiate in both the forward and the backward directions, depending on the frequency range.

[0131] TABLE 7

[0132] Rotman Lens

[0133] A Rotman lens is a passive, linear, and true-time-delay (TTD) beamforming network that can feed a linear array antenna to allow beam steering (21 , 22). It consists of a parallel plate waveguide with beam ports along the input contour, array ports along the output contour, and dummy ports to absorb residual power.

[0134] By stimulating different beam ports, the Rotman lens generates distinct time- delayed signals at the array ports, which feed the antenna elements to steer the beam in the desired direction. This enables it to scan in the 0 dimension. FIG. 14A depicts the designed Rotman lens and FIG. 14B its fabricated counterpart. The design consists of 9 input ports (P1-P9), 8 output ports (P10-P17), and 4 dummy ports (P18- P21 ). Given the large number of ports and the symmetry of the lens, only the S parameters of input port 5 are shown in FIG. 15. The reason why the S55 return loss is higher is due to the straight path that the EM waves propagate inside the lens, as well as potential phase errors (23). The Rotman lens is designed with an increment scanning step size of 14° per port, from -56° to 56°. Adding more absorber ports would further decrease the S55 to satisfy a response below -10 dB, however, it was found that this decreased the output power going out of the guided wave launchers (GWL) which allows for more overall gain.

[0135] 3D manufacturing process

[0136] The proposed holographic antenna consists of two main components: a Rotman lens and a quasi-HLWA. The entire antenna structure, including the Rotman lens, quasi-HLWA, and transmission line, is fabricated using advanced 3D printing technology known as lights-out digital additive manufacturing (LDM) (24). This is an ink-based additive manufacturing method in the material jetting family (ISO / ASTM D52900) designed for electronic component prototyping and requires minimal human operator involvement. Its accuracy and repeatability are similar to those seen with traditional PCB manufacturing. This technique employs two print heads: one for depositing AgCiteTM conductive ink containing silver nanoparticles suspended in a solvent and another for printing dielectric polymer ink. The LDM process allows for the simultaneous printing of conductive and dielectric materials with high precision, achieving layer thicknesses of 1.9 to 3.2 pm for conductive ink and 0.27 to 0.44 pm for dielectric ink. The substrate material used is lnk1092, a proprietary material developed by Nanodimensions specifically for 3D printing RF components. LDM technology enables the production of complex antenna geometries, achieving a minimum particle size of 80 nanometers and surface roughness less than 2 pm on the top surface and 0.25 pm on the bottom surface (24).

[0137] Antenna design process

[0138] The 2D beamformer design process is illustrated in FIGS. 16A-16C. In the first step (FIG. 16A), the output of the Rotman lens is extended to provide a larger surface area for the placement of the quasi-HLWA. Next, in step (FIG. 16B), the calculated interference pattern (FIG. 17) is superimposed on the extended output surface of the Rotman lens to convert guided waves to radiating waves. Finally, in step (FIG. 16C), the leakywave antennas are placed on the interference pattern. This placement ensures that leaky wave antennas are excited by the waves generated by the Rotman lens, resulting in the desired beam steering capabilities in the <|) plane. The combination of of the Rotman lens for elevation (0) steering and the quasi-HLWA for azimuth (c|)) steering enables the realization of a 3D printed 2D scanning antenna. The design presented here is classified as a quasi-holographic antenna, not merely a patchfed array. This classification is justified by the implementation of holographic principles in determining unit cell placement through calculated interference patterns.

[0139] The Rotman lens enables beam steering in the elevation (0) plane by switching between the input ports. The quasi-HLWA, on the other hand, achieves beam steering in the azimuth (c|)) plane by varying the frequency, as demonstrated by the dispersion diagram presented in FIG. 18A. FIGS. 19A-19C depict the quasi-holographic leaky- wave antenna (quasi-HLWA) design. FIGS. 19A-19B shows the antenna structure with dimensions. The total size of the antenna is Wh = 160 mm °Q Lh = 140.8 mm. The height of the Rotman lens substrate is Hr = 0.508 mm, and the quasi-HLWA substrate height is Hs = 1.008 mm. Note that the height of the quasi-HLWA substrate is the same as the height of a unit cell in the antenna. FIG. 19C displays the fabricated antenna implemented using the 3D printing technology with an lnk1092 substrate and silver conductive traces.

[0140] To enhance the overall gain of the antenna, the substrate height of the quasi- HLWA is increased. However, this introduces a trade-off between gain and scanning range, as a thicker substrate may limit the scanning range due to the excitation of surface waves and the onset of grating lobes (19, 20). To ensure smooth and efficient power transfer between the Rotman lens and the quasi-HLWA, a 25° slope transmission line is used. The 25° angle represents the best trade-off between scanning range and gain. This transition accommodates the difference in substrate heights and minimizes reflections and losses (25), as depicted in FIG. 20.

[0141] Also, keeping the Rotman lens substrate height at Hr and the HLWA substrate height at HI ensures maximum gain performance and maintains a wide bandwidth (26-28). FIG. 20 also illustrates the gain comparison with the 25° slope (Hh = 1.008 mm), without the slope (i.e., Hh = Hr), and with a -25° slope (Hh = 0.308 mm). The average gain from 20 to 40 GHz for the design with the -25° slope is 9.8 d Bi, without the slope is 14.5 dBi, while with the 25° slope is 16.5 dBi.

[0142] This particular gain is calculated by taking the maximum value of the gain at port 5 of the Rotman lens. It can be observed that the overall gain of the antenna across the especially at 26 GHz, where there is a noticeable gain increase of 7 dBi. At some frequencies, such as 32 GHz, there is a slight trade-off in gain, which is due to the phase errors introduced by the Rotman lens at this frequency. The reason why there is a height difference between the Rotman lens and the quasi-LWA is such that maximum performance is obtained from both. For the quasi-HLWA, it is derived from the the dispersion diagrams of FIGS. 18A-18B. For the Rotman lens, it is shown that a too-thin or too-thick substrate affect its performance, as the width of the transmission lines at the input and output ports is directly proportional to the substrate thickness (28). Too-thin transmission lines may lead to higher loss while too-thick transmission lines may lead to coupling between ports.

[0143] 2D beam scanning

[0144] The 2D beam scanning capability is achieved through two complementary mechanisms: (1) azimuth plane (<|)) scanning is controlled by varying the frequency from 24 GHz to 32 GHz, leveraging the frequency-dependent beam steering properties of the quasi-holographic leaky-wave antenna array, and (2) elevation plane (0) scanning is controlled by switching between the nine excitation ports (P1-P9) of the Rotman Lens. This dual-dimension scanning approach eliminates the need for active phase shifters, resulting in a purely passive beamforming system.

[0145] The 2D scanning performance is verified through measurement results presented in FIG. 21. Moving horizontally across rows demonstrates the frequency scanning effect (24-32 GHz) in the azimuth plane with a scanning range of -28° to 28°, while moving vertically through columns shows the beam steering achieved by switching between different input ports of the Rotman Lens, providing a scanning range of -54° to 54° in the elevation plane. The system achieves a maximum measured gain of 21.3 dBi at 28 GHz with an average radiation efficiency of 60%. Measurements carried out on the fabricated prototype demonstrate coherence with the simulated performance (FIG. 22). With this antenna, it is now possible to enable multi-target BP tracking.

[0146] EXAMPLE 4

[0147] Analog neural network

[0148] A key advancement in our blood pressure monitoring system is the replacement of the digital deep neural network (D-DNN) with a novel analog neural network (ANN) designed specifically for real-time BP prediction with minimal power consumption. Unlike its digital counterpart that requires periodic measurements every 30 seconds and consumes significant power, our analog implementation offers continuous monitoring with substantially reduced energy requirements. Present analog neural networks require lots of components to be able to synthesize an NxN weight matrix. In our approach, an NxN weight matrix can be realized with a Rotman Lens along with varactor elements which act as tunable weights of the analog neural network.

[0149] FIG. 23 represents a 6A~6 weight matrix realized by the Rotman lens and varactors, which implements a single layer of a neural network. After the Rotman lens, a series of power-based diodes realizes the non-linear activation function in the analog domain. The weight matrix is characterized by the S-parameters of the Rotman lens and is dynamically reconfigurable through the varactors. Implementing a fully analog neural network simply requires cascading multiple weight matrices as shown in FIG. 24, which illustrates a 3-layer analog neural network.

[0150] Optimizing the varactor values requires an algorithm such as a stochastic gradient descent (SGD), to get as close as the desired weight matrix as possible. FIG. 25 illustrates the block diagram in order to optimize the varactor weights to get the desired weight matrix.

[0151] The analog neural network improves BP readings by enabling continuous, realtime signal processing, unlike digital systems that work in intervals. The tunable weights allow it to adapt quickly to individual physiological differences, making the predictions more accurate and robust, especially in dynamic or noisy environments.

[0152] EXAMPLE 5

[0153] Plasmonic six-port receiver

[0154] A six-port is a passive interferometer that measures the relative amplitude and phase difference between two input signals by producing four output signals that are combined in specific ways. The conventional six-port receivers, while highly accurate for phase displacement detection essential in blood pressure monitoring, face fundamental limitations due to the diffraction limit that constrains their sensitivity at millimeter-wave frequencies. FIG. 26 shows the six-port receiver, consisting of 1 power divider and 3 quadrature hybrids.

[0155] To overcome this limitation, a novel Spoof Surface Plasmon Polariton (SSPP) six-port receiver is designed that exploits plasmonic principles to break the diffraction barrier. This receiver utilizes artificially engineered metallic structures with subwavelength features that support surface waves analogous to those found in optical surface plasmon polaritons but operating at microwave and millimeter-wave frequencies. By confining electromagnetic energy to dimensions significantly smaller than the wavelength, the SSPP six-port receiver achieves enhanced phase sensitivity below the traditional A 2 limit. The device incorporates corrugated structures and metamaterial-inspired elements that control the dispersion characteristics and coupling efficiency of these surface waves. The phase of the signal is extracted by letting: where P1 and P2 are the input signals, coming from the transmitter and the receiver of the FMCW radar. These signals go into 4 outputs (P3-P6) and undergo specific phase shift in order to extract the phase difference <t> 1 — <t>2. Therefore, after each output port, a power diode (B3-B6) is placed which equals:

[0156] B3= |P3|2= 0.25 |Pi + jP2\2, (9)

[0157] B4= |P4|2= 0.25 L / P2 + P2I2, (10)

[0158] 85 = |P5|2= 0.25 l / Pi +;P2|2, (11) BQ = |P6|2= 0.25 | Pi - P2I2. (12)

[0159] Expanding the expressions using trigonometry gives:

[0160] Then the differences are done and then the division, which are made either on IC or by using a differential amplifier.

[0161] 83 - 84 = A1A2 sin(AG). (18) where A© = 61 - O2 = 01 - 02 = Ao. We finally find: arctan(tan(A0)) = AS. (20)

[0162] The plasmonic six-port receiver has the potential of detecting subwavelength phase changes below the diffraction limit (less than A / 10 ).

[0163] FIGS. 27A-27B describe the simulation results of the six-port receiver and show good agreement with the described equations. A SSPP six-port receiver brings the accuracy up to A / 30000, which is significant to detect phase changes induced by the skin displacement caused by the BP wave.

[0164] References

[0165] 1. Rantz, et al. (2015). Gerontology, 61 (3), 281-290.

[0166] 2. Malwade, et al. (2018). Computer Methods and Programs in Biomedicine, 161 :233-237.

[0167] 3. Hawthorne, et al. (2022). JMIR Human Factors, 9(1 ):e30091.

[0168] 4. Adami, et al. (2021 ). Sensors, 21 :517.

[0169] 5. Goodman, C.T.D. and Kitchen, G.B. (2021). Anaesthesia and Intensive Care Medicine, 22:49-53.

[0170] 6. Kaufmann, et al. (2020). Journal of Critical Care, 57:118-123.

[0171] 7. Sheridan, et al. (1994). Burns, 20(5):451-452. 8. Quan, et al. (2021). Advances in Non-lnvasive Blood Pressure Monitoring. Sensors, 21(13):4273.

[0172] 9. Pawlowski, et al. (2017). Medical Studies, 3(3):222-226.

[0173] 10. Dias, D. and Cunha, J.P.S. (2018). Sensors, 18(8):2414.

[0174] 11. Lu, et al. (2020). JMIR mHealth and uHealth, 8(11):e18907.

[0175] 12. Canzanello, et al. (2001). Archives of Internal Medicine, 161(5):729-731 .

[0176] 13. Schaefer, et al. (2021). Journal of Aging and Health, 33(7-8 suppl):40S- 50S.

[0177] 14. Yang, etal. (2021). A Method of UWB Radar Vital Detection Based on Time Extraction of Strong Vital Signs. Journal of Sensors. 2021 , Article ID 7294604.

[0178] 15. Hu, X. and Jin, T. (2016). Sensors, 16(12):2025.

[0179] 16. Yang, et al. (2017). In 2017 IEEE / CIC International Conference on Communications in China (ICCC) (pp. 1-5).

[0180] 17. Ren, et al. (2016). Journal of Computer and Communications, 4(3):1-7.

[0181] 18. Nguyen, V.H. and Pyun, J.Y. (2015). Sensors, 15(3), 6740-6762.

[0182] 19. Park, et al. (2019). Electronics, 8(5):561.

[0183] 20. Munoz-Ferreras, et al. (2018). In 2018 2nd URSI Atlantic Radio Science Meeting (AT-RASC) (pp. 1-4). IEEE.

[0184] 21. Wang, et al. (2020). In 2020 IEEE Asia-Pacific Microwave Conference (APMC) (pp. 1104-1106). IEEE.

[0185] 22. Purnomo, et al. (2021). Sensors, 21(9):3172.

[0186] 23. Hu, Y. and Toda, T. (2022). IEEE Access, 10:78219-78230.

Claims

WHAT IS CLAIMED IS:

1. A system for continuously monitoring blood pressure in at least one subject, comprising: a passive beamforming radar with a beam switching mechanism and a beam scanning mechanism; a receiver configured to receive signals reflected from the at least one subject; means for processing signals continuously that are received from the passive beamforming radar and are processed to measure blood pressure in the at least one subject; and a display on which to continuously show blood pressure values.

2. The system of claim 1 , wherein said means for processing signals is a computer having a memory, a processor that tangibly stores the at least one machine learning algorithm.

3. The system of claim 1 , wherein said means for processing signals is an analog neural network.

4. The system of claim 1 , wherein systolic blood pressure values and diastolic blood pressure values are displayed.

5. The system of claim 1 , wherein the passive beamforming radar is a mm- wave frequency modulated continuous wave (mm-wave FMCW) radar.

6. The system of claim 5, wherein the mm-wave FMCW radar comprises a Rotman lens at a front end, a quasi-holographic leaky-wave antenna (quasi-HLWA) disposed thereon and a 25° linear tapered substrate configured to improve power transfer between the Rotman lens and the quasi-HLWA.

7. The system of claim 6, wherein the Rotman lens is configured for beam steering in an elevation (0) plane via a switch among input ports and the quasi-HLWA is configured for frequency-controlled beam scanning in an azimuth (< ) plane.

8. The system of claim 1, wherein the receiver is a six-port receiver configured for phase detection of the signals.

9. A method for non-invasive monitoring of blood pressure for at least one subject in real time, comprising the steps of: focusing continuously at least one beam from the passive beamforming radar comprising the system of claim 1 at a distance from and sequentially onto a plurality of areas on the at least one subject; capturing signals reflected from each of the plurality of areas on the at least one subject; passing the signals through the receiver; inputting the signals received from the receiver into at least one machine learning algorithm or into an analog neural network; and outputting continuously measured values for blood pressure processed from the inputted signals for the at least one subject.

10. The method of claim 9, further comprising displaying the measured values for blood pressure.11 . The method of claim 9, wherein the passive beamforming radar is a mm- wave frequency modulated continuous wave (mm-wave FMCW) radar comprising a Rotman lens at a front end configured for beam steering, a quasi-holographic leaky- wave antenna (quasi-HLWA) disposed thereon configured for beam scanning and a 25° linear tapered substrate configured to improve power transfer between the Rotman lens and the quasi-HLWA; said focusing step comprising scanning the at least one beam over the plurality of areas on the at least one subject.

12. The method of claim 9, wherein the receiver is a six-port receiver, said passing step comprising detecting a phase of the signals passing through the six-port receiver.

13. The method of claim 9, wherein the measured values for blood pressure are for systolic blood pressure or diastolic blood pressure or a combination thereof.

14. The method of claim 9, wherein the plurality of areas on the subject comprises at least two of a chest, a neck, a wrist, a leg, an arm, or a stomach or lower abdomen.

15. The method of claim 14, wherein the plurality of areas comprise at least two of a heart in the chest, a carotid artery in the neck, a radial artery in the wrist, a femoral artery in the leg, a brachial artery in the arm, or an iliac artery in the stomach or lower abdomen.

16. The method of claim 15, wherein the plurality of areas is the heart in the chest and the carotid artery in the neck.

17. A system for continuously monitoring blood pressure in multiple targets, comprising: a mm-wave frequency modulated continuous wave (mm-wave FMCW) radar with a Rotman lens at a front end configured for beam steering, a quasi-holographic leaky-wave antenna (quasi-HLWA) disposed thereon configured for beam scanning and a 25° linear tapered substrate configured to improve power transfer between the Rotman lens and the quasi-HLWA; a six-port receiver configured for phase detection of signals; an analog neural network configured to process continuously signals received as input from the mm-wave FMCW radar after phase detection, said signals processed as output as blood pressure values in each subject in the plurality; and a display on which to continuously show the outputted blood pressure values of the multiple targets.

18. The system of claim 17, wherein the Rotman lens is configured for beam steering in an elevation (0) plane via a switch among input ports and the quasi-HLWA is configured for frequency-controlled beam scanning in an azimuth (< ) plane.

19. The system of claim 17, wherein the blood pressure values are systolic blood pressure values and diastolic blood pressure values.

20. A non-invasive method for tracking blood pressures of multiple targets in real time, comprising: scanning continuously a plurality of beams from the mm-wave frequency modulated continuous wave (mm-wave FMCW) radar comprising the system of claim 15 at a distance from and onto a plurality of areas on the multiple targets; capturing signals reflected from each of the plurality of areas on each of the multiple targets; inputting the signals into an analog neural network; processing the inputted signals as measured values for blood pressures; outputting continuously the measured values for blood pressures; and displaying the measured values for blood pressures for the multiple targets as they are outputted.

21. The method of claim 20, wherein the plurality of areas on each of the multiple targets comprises at least two of a chest, a neck, a wrist, or a leg.

22. The method of claim 21 , wherein the plurality of areas on each of the multiple targets comprises at least two of a heart in the chest, a carotid artery in the neck, a femoral artery in the leg, or a pulse in the wrist.

23. The method of claim 21 , wherein the plurality of areas is the heart in the chest and the carotid artery in the neck. to correlate blood pressure dynamics with cognitive and behavioral states by utilizing blood pressure pulse obtained from radar sensors and advanced machine learning algorithms and by detecting subtle physiological changes that could correspond to psychological stress, emotional arousal, and cognitive load or other complex cognitive and behavioral states. Heart rate and respiration rate may be integrated to obtain a more comprehensive and robust model of an individual’s mental state, recognizing that a nuanced assessment requires the synthesis of multiple physiological parameters.

24. A non-invasive method for detecting changes in cognitive and behavioral states in a subject, comprising:scanning continuously a plurality of beams from the mm-wave frequency modulated continuous wave (mm-wave FMCW) radar comprising the system of claim 15 at a distance from and onto a plurality of areas on the subject; capturing signals reflected from each of the plurality of areas on the subject; inputting the signals into an analog neural network; processing the inputted signals as blood pressure values; and correlating the blood pressure values with physiological changes associated with the cognitive and behavioral states in the subject.

25. The method of claim 24, further comprising: measuring heart rate or respiratory rate or a combination thereof; and integrating values for the heart rate or values for the respiratory rate with the blood pressure values prior to the correlating step.

Images