System and Method for Continuous Non-Invasive Blood Pressure Estimation Using Dual-Site Photoplethysmography Pulse Transit Time Measurement Across Spatially Separated Smart Glasses and Smartwatch Wearables with Personalized Arterial Model Calibration

Abstract

Disclosed is a system and method for continuous, non-invasive estimation of arterial blood pressure using pulse transit time (PTT) measured between photoplethysmography (PPG) sensors embedded in two wearable devices worn simultaneously at anatomically distinct arterial sites: smart glasses with a PPG sensor positioned over the superficial temporal artery at the temple, and a smartwatch with a PPG sensor positioned over the radial artery at the wrist. The arterial path between the superficial temporal artery and the radial artery traverses approximately 70-90 cm of elastic arterial conduit, including segments of the external carotid, common carotid, subclavian, axillary, brachial, and radial arteries. Pulse transit time across this path is inversely related to pulse wave velocity (PWV), which is governed by the Moens-Korteweg equation linking PWV to arterial wall elastic modulus, a quantity that increases monotonically with transmural blood pressure due to the non-linear stress-strain relationship of the arterial wall. The system achieves sub-millisecond cross-device time synchronization using Bluetooth Low Energy (BLE) connection event timestamps with linear clock drift correction, enabling PTT resolution of approximately 0.3 ms at 1 kHz PPG sampling. A personalized calibration model, trained on periodic oscillometric cuff reference measurements taken by the user every 2-4 weeks, maps measured PTT to systolic and diastolic blood pressure with an initial calibration accuracy meeting the AAMI/ANSI/ISO 81060-2:2018 standard (mean error ≤ ±5 mmHg, standard deviation ≤ 8 mmHg). On-device recalibration compensates for slow arterial compliance changes from aging, medication, and hydration state. The system supports beat-to-beat blood pressure estimation during activities of daily living, exercise, sleep, and postural transitions, providing continuous hemodynamic monitoring without the discomfort of inflatable cuffs or the limitations of single-site PPG morphology analysis. 14 claims.

Field of the Invention

This invention relates to non-invasive continuous blood pressure monitoring using wearable photoplethysmography sensors, specifically to a system that exploits the natural spatial separation between smart glasses worn on the head and a smartwatch worn on the wrist to measure arterial pulse transit time as a surrogate for blood pressure, with personalized calibration against an oscillometric cuff reference.

Background

Hypertension affects 1.28 billion adults worldwide and is the single largest modifiable risk factor for cardiovascular disease, stroke, and chronic kidney disease. The 2017 ACC/AHA hypertension guidelines lowered the Stage 1 hypertension threshold to 130/80 mmHg, reclassifying an estimated 31 million previously normotensive Americans as hypertensive. Despite this, blood pressure remains the most poorly monitored major vital sign. The standard of care is sporadic oscillometric cuff measurement during clinic visits, which captures a single snapshot influenced by white-coat effect (estimated 15-30% prevalence) and misses nocturnal dipping patterns, morning surges, and exercise-related transients that carry independent prognostic value. Ambulatory blood pressure monitoring (ABPM), which inflates a cuff every 15-30 minutes over 24 hours, remains the clinical reference but is uncomfortable enough that patient compliance for repeat monitoring is below 60%, according to a 2013 meta-analysis in the Journal of Clinical Hypertension.

Cuff-less blood pressure estimation has been a research target for over two decades. The dominant approaches fall into three categories:

• ECG + single-site PPG (pulse arrival time). This approach measures the time from the R-wave of the electrocardiogram to the arrival of the PPG pulse at a peripheral site (typically the finger or wrist). The Samsung Galaxy Watch 4/5/6/Ultra and the Biobeat BB-613WP use this method with FDA 510(k) clearance. The limitation is that pulse arrival time (PAT) includes both the pre-ejection period (PEP, the isovolumetric contraction time of the left ventricle) and the true pulse transit time. PEP varies with sympathetic tone, cardiac contractility, and preload, introducing confounds that are not purely blood-pressure-related. PEP accounts for approximately 30-80 ms of a typical 200-300 ms PAT, meaning 15-35% of the measured interval carries no blood pressure information. Separating PEP from PTT requires either a seismocardiography (SCG) sensor on the chest or a second PPG site closer to the aorta.
• Single-site PPG waveform morphology. Machine learning models trained on PPG waveform features (pulse width, dicrotic notch position, area ratio, second derivative peaks) attempt to estimate blood pressure from a single PPG sensor. The Aktiia device and several research prototypes use this approach. The fundamental limitation is that PPG morphology at a single site reflects local vascular impedance and peripheral resistance, not central arterial compliance. A 2021 review in IEEE Reviews in Biomedical Engineering found that single-site PPG morphology methods achieved standard deviations of 9.4-14.2 mmHg for systolic BP in independent validation, consistently exceeding the 8 mmHg threshold required by AAMI/ANSI/ISO 81060-2:2018.
• Tonometry and volume clamp. Finapres-type devices (Finapres NOVA, ClearSight by Edwards Lifesciences) continuously measure finger arterial pressure using an inflatable finger cuff with servo-controlled counter-pressure. These achieve clinical-grade accuracy but are bulky, restrict hand use, and cost $5,000-15,000. The Caretaker Medical wireless device uses a finger cuff with a piezoelectric sensor but still requires a dedicated finger-worn device. None are suitable for continuous wear during normal daily activities.

The gap in the art is a system that measures true pulse transit time (not pulse arrival time, which includes PEP) between two anatomically separated PPG sites, using devices the user already wears for other purposes (smart glasses and smartwatch), without requiring an ECG electrode, a chest sensor, or any dedicated medical hardware. The spatial separation between the temple and the wrist provides an arterial path long enough (70-90 cm) to produce PTT values of 60-120 ms that are measurable with consumer-grade time synchronization, while avoiding the PEP confound because both measurement sites are downstream of the aortic valve.

Detailed Description

1. Anatomical Basis for Dual-Site PTT Measurement

The superficial temporal artery is a terminal branch of the external carotid artery. It emerges from the parotid gland anterior to the ear, crosses the zygomatic process of the temporal bone, and divides into frontal and parietal branches. At the temple, the artery lies within 1-2 mm of the skin surface and is easily accessible to optical sensors. The temple has been used for PPG measurement in pulse oximeters (Masimo Radius PPG, Nonin forehead reflectance sensors) and is an established site for photoplethysmographic signal acquisition.

The radial artery runs along the lateral aspect of the forearm and is the standard site for wrist-based PPG sensors in consumer smartwatches (Apple Watch, Samsung Galaxy Watch, Garmin, Fitbit, Withings). It lies 2-4 mm below the skin surface at the wrist and produces robust PPG signals across skin tones when illuminated with green (525 nm) LEDs.

The arterial path from the temporal PPG site to the radial PPG site runs: superficial temporal artery → external carotid artery → common carotid artery → brachiocephalic trunk (right side) or aortic arch (left side) → subclavian artery → axillary artery → brachial artery → radial artery. For a 175 cm adult, this path length is approximately 75-85 cm. Because both measurement sites are peripheral to the aortic valve, the measured PTT is a true vascular transit time that does not include the pre-ejection period. This eliminates the primary confound that limits ECG-to-PPG pulse arrival time methods.

The direction of pulse propagation creates a predictable arrival-time ordering: the pulse arrives at the temporal site before the radial site, because the temporal artery branches from the common carotid artery approximately 40-50 cm proximal (closer to the heart) along the arterial tree compared to the radial artery's branching point from the subclavian artery via the brachial artery. The expected PTT for a normotensive adult (PWV approximately 7-9 m/s in the elastic arteries) is 80-120 ms. For a hypertensive adult with stiffer arteries (PWV 10-14 m/s), PTT shortens to 55-85 ms. This 25-65 ms range across the clinically relevant blood pressure spectrum (90/60 to 180/120 mmHg) is well within the resolution achievable by BLE-synchronized consumer wearables.

2. Smart Glasses PPG Sensor Design

The PPG sensor is integrated into the temple arm (earpiece) of the smart glasses at the point where the arm contacts the temporal region of the skull, approximately 1-2 cm anterior to the ear. The sensor module consists of:

• Light source: Two green LEDs (525 nm, e.g., Osram SFH 2201) and one infrared LED (940 nm, e.g., Vishay VSMD66694) provide dual-wavelength illumination. Green light penetrates 1-2 mm into tissue and is absorbed primarily by oxyhemoglobin and deoxyhemoglobin in the dermal capillary bed and superficial arteries. Infrared provides a second wavelength for motion artifact rejection via adaptive noise cancellation (the infrared AC/DC ratio differs from green, enabling spectral separation of motion and pulsatile components).
• Photodetector: A silicon photodiode (e.g., ams OSRAM TSL2591, active area 3.6 mm²) with a transimpedance amplifier converts reflected light to an analog voltage. A narrowband optical filter centered at 525 nm with 30 nm FWHM bandwidth rejects ambient light outside the measurement band.
• Analog front-end: A low-noise analog-to-digital converter (24-bit, 1 kHz sampling rate, e.g., TI ADS1292 or Analog Devices MAX86141) digitizes the photodetector output. LED drive current is modulated at 1 kHz with 50% duty cycle, and lock-in detection (synchronous demodulation at the LED modulation frequency) rejects ambient light interference including 100/120 Hz flicker from artificial lighting.
• Skin contact sensor: A capacitive proximity sensor (e.g., Microchip AT42QT1010) detects whether the glasses are being worn and the temple arm is in contact with the skin. PPG acquisition activates only when skin contact is confirmed, conserving battery power.

The power consumption of the PPG module is approximately 3 mW during active measurement (2 mA green LED drive at 1.8V plus 0.5 mA analog front-end) and less than 10 μW in standby. For a smart glasses battery of 150-450 mAh (typical for current smart glasses such as the Ray-Ban Meta at 154 mAh or the Xreal Air 2 Ultra at 180 mAh), continuous PPG measurement adds approximately 0.8 mA average current draw (accounting for 50% duty cycle), reducing battery life by roughly 5-8%. Intermittent measurement (10 seconds every 5 minutes) reduces this to less than 0.3% battery impact.

3. Cross-Device Time Synchronization

Accurate PTT measurement requires sub-millisecond time synchronization between the glasses and the watch. The system uses Bluetooth Low Energy (BLE) 5.0+ connection events as a shared time reference. In a BLE connection, the central device (smartwatch) sends a connection event packet at a fixed interval (connection interval, configurable from 7.5 ms to 4 seconds). The peripheral device (smart glasses) receives each packet and records the arrival timestamp on its local clock. Both devices maintain a synchronized sample counter anchored to BLE connection events.

The synchronization protocol operates as follows:

1. The smartwatch establishes a BLE connection to the smart glasses with a connection interval of 7.5 ms (the minimum allowed by BLE specification, providing 133 synchronization points per second).
2. At each connection event, the smart glasses record the local timestamp of the received packet. The smartwatch records the local timestamp of the transmitted packet. Both devices know the connection event counter (a 16-bit value incremented at each event).
3. Once per second, the devices exchange their local timestamps for the most recent 10 connection events. A least-squares linear fit estimates the clock offset and drift rate between the two local oscillators. Consumer MEMS oscillators (32.768 kHz crystals in wearable devices) have typical drift rates of ±20-50 ppm, corresponding to 20-50 μs of drift per second. With per-second recalibration, the residual synchronization error after drift correction is approximately 50-150 μs (0.05-0.15 ms), well below the 1 ms resolution needed for PTT measurement.
4. Each PPG sample on both devices is timestamped using the drift-corrected synchronized clock. PTT is computed as the difference between corresponding pulse feature timestamps (foot of the pulse, systolic peak, or maximum first derivative) on the two devices.

An alternative synchronization method uses IEEE 802.11mc Fine Timing Measurement (FTM), available on WiFi 6E chipsets, which provides sub-nanosecond time-of-flight measurement. If both devices support WiFi FTM, synchronization accuracy of approximately 100 picoseconds is achievable, far exceeding the requirement. However, WiFi radios consume 50-100 mW during active operation, making BLE synchronization preferable for continuous monitoring on battery-constrained wearables.

4. Pulse Transit Time Extraction Algorithm

PTT is extracted from the synchronized PPG signals using a multi-step algorithm:

Step 1: Preprocessing. The raw PPG signal from each device is bandpass-filtered (0.5-8 Hz, 4th-order Butterworth, zero-phase) to remove baseline drift from respiration (0.15-0.4 Hz) and high-frequency noise while preserving the fundamental pulse frequency (0.67-3.33 Hz, corresponding to 40-200 bpm) and its first three harmonics. An adaptive motion artifact filter using the dual-wavelength (green + IR) signal from the glasses sensor and the accelerometer signal from both devices suppresses motion-induced baseline shifts using a normalized least-mean-squares (NLMS) adaptive filter with a forgetting factor of 0.999.

Step 2: Pulse detection. Individual cardiac pulses are detected in each filtered PPG signal using the first derivative (dPPG/dt). The systolic upstroke of each pulse produces a positive peak in the first derivative. Pulse boundaries are defined by zero crossings of the first derivative preceding and following each positive peak. A minimum pulse interval constraint (200 ms, corresponding to 300 bpm maximum) rejects spurious detections from noise.

Step 3: Fiducial point identification. Three fiducial points are extracted from each detected pulse on each device:

• Pulse foot: The minimum of the PPG signal preceding the systolic upstroke. Identified as the point of minimum value in the 100 ms window before the first derivative peak. The foot represents the earliest arrival of the pressure wavefront at the measurement site.
• Maximum slope point: The time of maximum first derivative (dPPG/dt)max during the systolic upstroke. This corresponds to the point of maximum rate of change of arterial volume and is the most robust fiducial against waveform morphology variation across subjects and measurement sites.
• Systolic peak: The maximum PPG amplitude during systole. Located at the zero crossing of the first derivative following the upstroke. The peak is influenced by reflected pressure waves and is less reliable for PTT estimation than the foot or maximum slope, but provides a complementary measurement for multi-fiducial averaging.

Step 4: PTT computation. For each cardiac cycle, PTT is computed as the time difference between corresponding fiducial points on the temple PPG and the wrist PPG. Three PTT values are obtained per beat (foot-to-foot, slope-to-slope, peak-to-peak). The primary PTT estimate uses the maximum-slope fiducial because it is least affected by low-frequency baseline drift and pulse amplitude variation. A weighted average of all three fiducials, with weights inversely proportional to their inter-beat variability over the previous 10 beats, provides the final PTT estimate per cardiac cycle.

Step 5: Outlier rejection. Beat-to-beat PTT values that deviate by more than 3 standard deviations from the running median (computed over the previous 30 beats) are rejected as motion-corrupted or arrhythmia-related outliers. The signal quality index (SQI) from each PPG channel, computed as the cross-correlation between the current pulse template and the running average template, must exceed 0.85 for a PTT measurement to be accepted. Beats with SQI below threshold on either channel are discarded.

5. Blood Pressure Estimation Model

The relationship between PTT and blood pressure is modeled using the Moens-Korteweg equation and an empirical exponential arterial compliance model. Pulse wave velocity is related to arterial properties by:

PWV = √(E·h / 2·ρ·r)

where E is the elastic modulus of the arterial wall (Pa), h is wall thickness (m), ρ is blood density (approximately 1060 kg/m³), and r is the internal vessel radius (m). Since PTT = L / PWV, where L is the arterial path length (m):

PTT = L · √(2·ρ·r / E·h)

The elastic modulus E depends on transmural pressure P through an exponential relationship first characterized by Hayashi et al. (1980) in the Journal of Biomechanics:

E(P) = E₀ · exp(α · P)

where E₀ is the zero-pressure elastic modulus (a property of the arterial wall composition that changes slowly with aging, atherosclerosis, and medication) and α is the pressure coefficient of stiffness (typically 0.016-0.018 /mmHg for the human brachial artery). Substituting into the PTT equation and taking the natural logarithm yields a linear relationship between ln(PTT) and blood pressure:

ln(PTT) = a - b · P

where a = ln(L) + 0.5·ln(2ρr/E₀h) and b = α/2. For systolic blood pressure (SBP) estimation, the model uses PTT measured at the maximum-slope fiducial (which corresponds temporally to the systolic pressure maximum). For diastolic blood pressure (DBP) estimation, the model uses PTT measured at the pulse foot fiducial (which corresponds temporally to the diastolic pressure minimum, immediately before the next systolic upstroke).

The personalized calibration procedure determines coefficients a and b for each user by fitting the ln(PTT) vs. BP relationship to at least 5 paired measurements taken with an oscillometric cuff (AAMI-validated device such as the Omron Platinum BP5450 or Withings BPM Connect) at different times of day over 1-3 days, capturing the user's natural blood pressure range. A minimum range of 15 mmHg in systolic pressure across calibration points is recommended for robust regression. The calibration model is a ridge regression (L2 regularization, λ = 0.01) on the feature vector [ln(PTT_slope), ln(PTT_foot), heart_rate, PTT_slope/PTT_foot_ratio], providing 4 degrees of freedom to capture subject-specific arterial geometry, compliance, and reflected-wave effects.

6. Calibration Drift Compensation

The personalized calibration parameters drift over weeks to months as the user's baseline arterial compliance changes from aging, medication changes (ACE inhibitors, ARBs, calcium channel blockers, and beta-blockers all affect arterial stiffness), hydration state, and fitness level. The system detects calibration drift using two mechanisms:

• Internal consistency monitoring: The standard deviation of PTT-estimated BP over a 24-hour period is compared to the standard deviation observed during the calibration period. If the 24-hour standard deviation exceeds 1.5 times the calibration-period standard deviation for 3 consecutive days, the system flags a potential calibration drift and prompts the user for a recalibration cuff measurement.
• Physiological plausibility checking: If the estimated blood pressure exceeds the physiological range (SBP below 70 or above 200 mmHg, DBP below 40 or above 130 mmHg) for more than 5% of measurements during a waking period, or if the nocturnal dipping ratio (nighttime SBP / daytime SBP) falls outside the 0.70-1.05 range for 3 consecutive nights, a recalibration is requested.

When a recalibration cuff measurement is taken, the system updates the calibration model incrementally using recursive least squares (RLS) with a forgetting factor of 0.98, which weights recent calibration points more heavily than older ones while retaining long-term trending information. The forgetting factor of 0.98 means that a calibration point's effective weight halves approximately every 35 new calibration points. At a recommended cadence of one cuff measurement every 2 weeks, the effective memory of the calibration model spans approximately 16 months, matching the timescale of typical arterial aging processes.

7. Motion Artifact Handling and Activity Context

Both devices contain 3-axis accelerometers and gyroscopes (6-axis IMU, e.g., Bosch BMI270, TDK InvenSense ICM-42688). The system classifies the user's activity state into categories: stationary, walking, running, cycling, driving, and sleeping. Activity classification uses a lightweight random forest model (50 estimators, 8 features: 3-axis accelerometer RMS, 3-axis gyroscope RMS, step frequency, and wrist orientation) running on the smartwatch processor, achieving greater than 95% accuracy across categories with less than 1 ms inference latency.

During different activity states, the system adjusts its measurement strategy:

• Stationary/sleeping: Continuous measurement at full 1 kHz sampling. All three fiducial-based PTT estimates are used. Signal quality is highest; SQI threshold is set to 0.90.
• Walking: Step-frequency artifacts (1.5-2.5 Hz) overlap the PPG fundamental frequency. The accelerometer-referenced adaptive filter removes step-correlated motion artifacts. PTT measurements are accepted only during the stance phase of gait (0-60% of stride cycle, identified by vertical accelerometer pattern) when wrist motion is minimal. SQI threshold is lowered to 0.80.
• Running: Arm swing introduces large-amplitude motion artifacts at 2.5-4.0 Hz. The system reduces PPG sampling to 10-second bursts every 5 minutes and uses only the maximum-slope fiducial (most motion-robust). SQI threshold is set to 0.75; measurements below threshold are discarded rather than corrected.
• Sleep: Measurement cadence is reduced to 30-second windows every 10 minutes to conserve battery. Nocturnal blood pressure trending requires only sparse sampling because blood pressure changes slowly during sleep (typical variation: 5-15 mmHg over 6-8 hours).

8. Postural Correction

Hydrostatic pressure differences between the two measurement sites affect the measured PTT. When the user is standing, the temporal artery is approximately 35-45 cm above the heart, while the radial artery is approximately 0-15 cm below the heart (depending on arm position). This creates a hydrostatic pressure difference of approximately 35-55 mmHg between the two sites (using ρgh with ρ = 1060 kg/m³ and the height difference). The hydrostatic effect changes the local transmural pressure at each site independently, affecting the local arterial compliance and therefore the local PWV.

The system corrects for posture using the 3-axis accelerometers on both devices. The gravitational acceleration vector measured by each device determines the device's orientation relative to the gravitational vertical, from which the height of each sensor relative to the heart is estimated. The heart position is modeled as a fixed point on the body approximately 15 cm below the suprasternal notch (which is approximately at the height of the ears when the head is level, conveniently close to the glasses IMU). The hydrostatic correction adjusts the PTT-to-BP mapping by adding the computed hydrostatic pressure difference at each measurement site as a covariate in the calibration model. Calibration measurements taken in both seated and standing positions improve the model's postural generalization.

9. Arrhythmia Detection and Exclusion

Cardiac arrhythmias, particularly atrial fibrillation (AF) and premature ventricular contractions (PVCs), produce irregular pulse intervals that confound PTT-based blood pressure estimation. The system detects arrhythmia using the pulse interval time series from the PPG signal:

• Atrial fibrillation: Detected by irregularity metrics including the root mean square of successive pulse interval differences (RMSSD > 50 ms sustained over 30 beats), the Shannon entropy of the pulse interval distribution (entropy > 2.0 bits), and the absence of respiratory sinus arrhythmia (RSA) in the pulse interval power spectrum (loss of the 0.15-0.4 Hz spectral peak). When AF is detected, blood pressure estimation is suspended and the user is notified.
• Premature ventricular contractions: Identified by short-long pulse interval pairs (a short interval followed by a compensatory pause, with the sum of the two intervals approximately equal to two normal intervals). PVC beats and post-PVC beats are excluded from PTT computation, but measurement continues during intervening normal sinus beats.

10. Multi-Day Blood Pressure Trending and Clinical Reporting

The system aggregates beat-to-beat blood pressure estimates into clinically meaningful summary statistics stored on the smartwatch and synced to a companion smartphone application:

• Daytime average: Mean SBP and DBP computed from all accepted measurements between 06:00 and 22:00 local time (user-adjustable). AAMI/ANSI/ISO 81060-2 compliance requires daytime ambulatory average accuracy of ±5 mmHg mean error with ≤ 8 mmHg standard deviation.
• Nighttime average: Mean SBP and DBP from 22:00 to 06:00 (or from accelerometer-detected sleep onset to wake).
• Nocturnal dipping ratio: Nighttime SBP / Daytime SBP. Normal dipping is 0.80-0.90 (10-20% decrease). Non-dipping (ratio > 0.90) and reverse dipping (ratio > 1.00) are associated with increased cardiovascular risk and are clinically significant findings detectable only with continuous monitoring, not office measurements.
• Morning surge: Maximum SBP in the 2 hours after waking minus the minimum SBP during sleep. A morning surge exceeding 35 mmHg is associated with increased stroke risk according to the Jichi Medical School ABPM study.
• Blood pressure variability: Coefficient of variation (SD/mean) of SBP over 24 hours. Elevated variability (CV > 10%) is an independent cardiovascular risk factor.
• Exportable PDF report: A structured clinical report in a format compatible with the American Heart Association's 2019 self-measured blood pressure monitoring consensus statement, including time-stamped readings, averages, dipping pattern, and a 7-day trend graph suitable for sharing with a clinician.

11. Privacy and On-Device Processing

All PPG signal processing, PTT computation, and blood pressure estimation run entirely on-device (split between the smartwatch application processor and the glasses companion processor). No raw PPG waveforms are transmitted to cloud servers. The BLE link between devices carries only PTT fiducial timestamps (8 bytes per beat, approximately 600 bytes per minute at 75 bpm) and synchronization packets. Summary blood pressure statistics may optionally be synced to a health platform (Apple HealthKit, Google Health Connect, Samsung Health) via the smartphone companion app, under user control. The system complies with HIPAA de-identification requirements (45 CFR § 164.514) when aggregated statistics are shared, and with GDPR Article 9 special category data protections when deployed in EU markets.

Claims

1. A system for continuous non-invasive blood pressure estimation comprising: a first wearable device configured to be worn on the head (smart glasses) containing a first photoplethysmography sensor positioned to measure pulsatile blood volume changes in the superficial temporal artery; a second wearable device configured to be worn on the wrist (smartwatch) containing a second photoplethysmography sensor positioned to measure pulsatile blood volume changes in the radial artery; a time synchronization module that establishes a common time reference between the two devices with sub-millisecond accuracy; and a processor executing software that computes pulse transit time as the difference between corresponding pulse feature arrival times measured by the first and second PPG sensors, and estimates arterial blood pressure from the computed pulse transit time using a personalized calibration model.
2. The system of claim 1, wherein the first PPG sensor is integrated into the temple arm of the smart glasses at a position that contacts the skin overlying the superficial temporal artery when the glasses are worn, and comprises at least one green LED (520-530 nm), one infrared LED (930-950 nm), and a silicon photodiode with a transimpedance amplifier sampling at a rate of at least 500 Hz.
3. The system of claim 1, wherein the time synchronization module uses Bluetooth Low Energy connection event timestamps exchanged between the two devices, with linear clock drift correction computed by least-squares fitting of timestamp pairs over a sliding window, achieving synchronization accuracy of less than 200 microseconds.
4. The system of claim 1, wherein pulse transit time is computed using at least one of three fiducial points extracted from each PPG signal: the pulse foot (minimum value preceding systolic upstroke), the maximum first-derivative point (maximum rate of change during systolic upstroke), and the systolic peak (maximum PPG amplitude), with a weighted combination of fiducial-based PTT estimates providing the final measurement.
5. The system of claim 1, wherein the personalized calibration model maps pulse transit time to systolic and diastolic blood pressure using a regression model calibrated against at least three paired measurements from a validated oscillometric cuff reference device, with calibration coefficients updated incrementally via recursive least squares with a forgetting factor when new reference measurements are provided.
6. The system of claim 1, further comprising a postural correction module that uses 3-axis accelerometers in both devices to estimate the height difference between the two PPG sensor sites relative to the heart, and applies a hydrostatic pressure correction to the blood pressure estimate based on the computed height difference and blood density.
7. The system of claim 1, further comprising an activity classification module that determines the user's activity state from accelerometer and gyroscope data on both devices, and adjusts the PPG measurement strategy including sampling duty cycle, motion artifact filter parameters, signal quality thresholds, and accepted fiducial types based on the classified activity state.
8. The system of claim 1, wherein the blood pressure estimation model uses a feature vector comprising the natural logarithm of PTT at the maximum-slope fiducial, the natural logarithm of PTT at the foot fiducial, heart rate, and the ratio of slope-based PTT to foot-based PTT, capturing both arterial compliance and reflected wave effects in a subject-specific regression.
9. The system of claim 1, further comprising an arrhythmia detection module that identifies atrial fibrillation from elevated pulse interval irregularity metrics (RMSSD, Shannon entropy) and premature ventricular contractions from short-long interval patterns, and excludes arrhythmia-affected beats from PTT computation while continuing measurement during intervening normal sinus beats.
10. The system of claim 1, further comprising a calibration drift detection module that monitors the statistical consistency of estimated blood pressure over multi-day windows and prompts the user for a recalibration cuff measurement when the estimated blood pressure distribution deviates from expected parameters, including 24-hour standard deviation exceeding 1.5 times the calibration-period standard deviation for 3 consecutive days or nocturnal dipping ratio falling outside the 0.70-1.05 range for 3 consecutive nights.
11. The system of claim 1, wherein the PPG sensor in the smart glasses employs dual-wavelength illumination (green and infrared) with an adaptive noise cancellation filter that exploits the wavelength-dependent ratio of motion artifact to pulsatile signal to separate motion-induced baseline shifts from arterial pulsation.
12. The system of claim 1, further comprising a clinical reporting module that computes daytime average blood pressure, nighttime average blood pressure, nocturnal dipping ratio, morning surge magnitude, and blood pressure variability coefficient of variation from continuous measurements, and generates a structured clinical report exportable to health record platforms.
13. The system of claim 1, wherein all PPG signal processing, pulse transit time computation, and blood pressure estimation execute entirely on the processors of the two wearable devices without transmitting raw PPG waveform data to external servers, and wherein only PTT fiducial timestamps are transmitted over the BLE link between devices.
14. The system of claim 1, wherein the time synchronization module alternatively uses IEEE 802.11mc Fine Timing Measurement when both devices are equipped with WiFi 6E or later chipsets, achieving sub-nanosecond synchronization accuracy and enabling PTT resolution of less than 0.1 milliseconds for high-precision hemodynamic waveform analysis including estimation of arterial stiffness indices and reflected wave timing.

Limitations and Engineering Challenges

The most significant limitation is that PTT-based blood pressure estimation requires personalized calibration against a reference cuff device and periodic recalibration. The system cannot provide absolute blood pressure values out of the box, unlike a validated oscillometric cuff. This means the first blood pressure reading requires the user to own or borrow a cuff device and complete a calibration procedure taking 15-30 minutes. For population-scale deployment, this barrier limits adoption among users who do not already monitor their blood pressure.

Skin tone affects PPG signal quality at both measurement sites. Melanin absorbs green light (525 nm), reducing the pulsatile signal amplitude in darker skin tones (Fitzpatrick types V-VI). The infrared channel (940 nm) is less affected by melanin absorption and can serve as the primary measurement wavelength for darker-skinned users, though at reduced penetration depth. Adaptive LED drive current (automatic gain control increasing LED current for lower-amplitude signals up to a thermal safety limit of 30 mW/cm² per IEC 62471) partially compensates for skin tone variation.

The temporal artery site on smart glasses requires consistent skin contact pressure. Unlike a wristwatch with a strap that maintains contact, glasses rest on the temples by spring pressure from the frame. Variations in frame fit, head movement, and hair between the sensor and skin can introduce intermittent signal loss. Frame designs with adjustable temple arm pressure (spring-loaded or memory metal temple tips) and hair-penetrating sensor protrusions (comb-like extensions around the PPG sensor) mitigate contact issues but add design complexity.

Blood pressure estimation accuracy degrades in patients with significant atherosclerotic disease that creates non-uniform arterial stiffness along the measurement path. The Moens-Korteweg model assumes a homogeneous elastic tube; focal stenoses or calcified plaques create local impedance mismatches that alter pulse wave propagation in ways not captured by a simple PTT-to-BP mapping. For patients with known peripheral artery disease, the system should display a clinical disclaimer noting reduced accuracy.

Prior Art References

1. 35 U.S.C. § 102(a)(1) - Prior art statutory basis for defensive disclosure
2. Hayashi et al. (1980) Journal of Biomechanics - Exponential elastic modulus-pressure relationship in human arteries (E = E₀·exp(αP))
3. 2017 ACC/AHA Hypertension Clinical Practice Guidelines - Stage 1 hypertension threshold lowered to 130/80 mmHg
4. Journal of Clinical Hypertension meta-analysis (2013) - ABPM compliance rates below 60%
5. IEEE Reviews in Biomedical Engineering (2021) - Single-site PPG morphology BP estimation review, SD 9.4-14.2 mmHg
6. Jichi Medical School ABPM Study - Morning BP surge > 35 mmHg and stroke risk
7. AHA 2019 Self-Measured Blood Pressure Monitoring Consensus Statement - Clinical reporting format for ambulatory BP
8. AAMI/ANSI/ISO 81060-2:2018 - Non-invasive sphygmomanometer validation standard (mean error ≤ ±5 mmHg, SD ≤ 8 mmHg)
9. US10667706B2 - Samsung: Apparatus and method for estimating blood pressure using PPG (single-site, ECG-dependent PAT)
10. US20210030283A1 - Apple: Blood pressure monitoring using ultrasonic transducer on wearable (tonometric, not PTT-based)
11. Wikipedia: Moens-Korteweg equation - Pulse wave velocity and arterial wall mechanics
12. IEC 62471 - Photobiological safety of lamps and lamp systems (optical radiation limits for skin-contact LEDs)
13. Bluetooth SIG Core Specification v5.3, Vol. 6, Part B - Link Layer connection event timing and synchronization
14. 45 CFR § 164.514 - HIPAA de-identification standard for protected health information