Mobile Monitoring
`with Wearable
`Photoplethysmographic
`Biosensors
`
`WearableTechnology
`
`©EYEWIRE
`
`H. HARRY ASADA,
`PHILLIP SHALTIS,
`ANDREW REISNER,
`SOKWOO RHEE, AND
`REGINALD C. HUTCHINSON
`
`Technical and Clinical Aspects of a Ring Sensor for
`Ambulatory, Telemetric, Continuous Health Monitoring
`in the Field, in the Hospital, and in the Home
`
`W earable biosensors (WBS) will permit contin-
`
`uous cardiovascular (CV) monitoring in a
`number of novel settings. Benefits may be real-
`ized in the diagnosis and treatment of a number
`of major diseases. WBS, in conjunction with appropriate
`alarm algorithms, can increase surveillance capabilities for
`CV catastrophe for high-risk subjects. WBS could also play a
`role in the treatment of chronic diseases, by providing infor-
`mation that enables precise titration of therapy or detecting
`lapses in patient compliance.
`WBS could play an important role in the wireless surveil-
`lance of people during hazardous operations (military,
`fire-fighting, etc.), or such sensors could be dispensed during
`a mass civilian casualty occurrence. Given that CV physio-
`logic parameters make up the “vital signs” that are the most
`important information in emergency medical situations, WBS
`might enable a wireless monitoring system for large numbers
`of at-risk subjects. This same approach may also have utility
`in monitoring the waiting room of today’s overcrowded emer-
`gency departments. For hospital inpatients who require CV
`monitoring, current biosensor technology typically tethers pa-
`tients in a tangle of cables, whereas wearable CV sensors
`could increase inpatient comfort and may even reduce the risk
`of tripping and falling, a perennial problem for hospital
`patients who are ill, medicated, and in an unfamiliar setting.
`On a daily basis, wearable CV sensors could detect a
`missed dose of medication by sensing untreated elevated
`blood pressure and could trigger an automated reminder for
`the patient to take the medication. Moreover, it is important
`for doctors to titrate the treatment of high blood pressure,
`since both insufficient therapy as well as excessive therapy
`(leading to abnormally low blood pressures) increase mortal-
`ity. However, healthcare providers have only intermittent val-
`ues of blood pressure on which to base therapy decisions; it is
`possible that continuous blood pressure monitoring would
`permit enhanced titration of therapy and reductions in mortal-
`ity. Similarly, WBS would be able to log the physiologic sig-
`nature of a patient’s exercise efforts (manifested as changes in
`heart rate and blood pressure), permitting the patient and
`healthcare provider to assess compliance with a regimen
`proven to improve health outcomes. For patients with chronic
`cardiovascular disease, such as heart failure, home monitoring
`employing WBS may detect exacerbations in very early (and
`
`often easily treated) stages, long before the patient progresses
`to more dangerous levels that necessitate an emergency room
`visit and costly hospital admission.
`In this article we will address both technical and clinical is-
`sues of WBS. First, design concepts of a WBS will be pre-
`sented, with emphasis on the ring sensor developed by the
`author’s group at MIT. The ring sensor is an ambulatory, tele-
`metric, continuous health-monitoring device. This WBS com-
`bines miniaturized data acquisition features with advanced
`photoplethysmographic (PPG) techniques to acquire data re-
`lated to the patient’s cardiovascular state using a method that is
`far superior to existing fingertip PPG sensors [1]. In particular,
`the ring sensor is capable of reliably monitoring a patient’s
`heart rate, oxygen saturation, and heart rate variability. Techni-
`cal issues, including motion artifact, interference with blood
`circulation, and battery power issues, will be addressed, and ef-
`fective engineering solutions to alleviate these problems will be
`presented. Second, based on the ring sensor technology the
`clinical potentials of WBS monitoring will be addressed.
`
`WBS System Paradigm
`For novel healthcare applications to employ WBS technology,
`several system criteria must be met. The WBS hardware solu-
`tion must be adequate to make reliable physiologic measure-
`ments during activities of daily living or even more
`demanding circumstances such as fitness training or military
`battle. There must exist data processing and decision-making
`algorithms for the waveform data. These algorithms must
`prompt some action that improves health outcomes. Finally,
`the systems must be cost effective when compared with less
`expensive, lower technology alternatives.
`
`WBS Design Paradigm
`The monitoring environments for out-of-hospital, wearable
`devices demand a new paradigm in noninvasive sensor design.
`There are several design requirements central to such devices.
`Compactness, stability of signal, motion and other distur-
`bance rejection, durability, data storage and transmission, and
`low power consumption comprise the major design consider-
`ations. Additionally, since WBS devices are to be worn with-
`out direct doctor supervision, it is imperative that they are
`simple to use and comfortable to wear for long periods of
`time. A challenge unique to wearable sensor design is the
`
`28
`
`IEEE ENGINEERING IN MEDICINE AND BIOLOGY MAGAZINE
`
`0739-5175/03/$17.00©2003IEEE
`
`MAY/JUNE 2003
`
`Apple Inc.
`APL1005
`U.S. Patent No. 8,989,830
`
`FITBIT, Ex. 1005
`
`

`

`WBS solutions, in various stages of
`technologic maturity, exist for measuring
`established cardiopulmonary “vital signs”:
`heart rate, arterial blood pressure, arterial
`oxygen saturation, respiratory rate,
`temperature, and even cardiac output.
`
`trade-off between patient comfort, or long-term wearability,
`and reliable sensor attachment. While it is nearly needless to
`say that WBS technology must be safe, it should be noted that
`there have been tragic reports of serious injury resulting from
`early home monitoring technology [2]. Evolving regulatory
`guidelines for hospital and home monitoring technology can
`be found in the U.S. National Fire Protection Association
`Health Care Facilities Handbook.
`At the same time, the physiologic information generated
`by WBS technology must trigger some appropriate system ac-
`tion to improve health outcomes. Abnormal states must be ef-
`ficiently recognized while false alarms are minimized. This
`requires carefully designed WBS devices, as well as innova-
`tive postprocessing and intelligent data interpretation. Post-
`processing of sensor data can improve usability, as illustrated
`by recent improvements in pulse oximetry technology [3]-[5].
`Data interpretation can occur in real time (as is necessary for
`detecting cardiovascular-related catastrophes) or offline (as is
`the standard-of-care for arrhythmia surveillance using Holter
`and related monitoring). Real-time alarm “algorithms” using
`simple thresholds for measured parameters, like heart rate and
`oxygen saturation, have demonstrated high rates of false
`alarms [6], [7]. Algorithms for off-line, retrospective data
`analysis are also in a developmental stage. Studies of novel
`automated “triage” software used to interpret hours of contin-
`uous noninvasive ECG data of monitored outpatients suggest
`that the software’s diagnostic yield is not equal to a human’s
`when it comes to arrhythmia detection [8], [9]. It will presum-
`ably require further
`improvements in WBS hardware,
`middleware, and software in order to fully exploit the promise
`of wearable ambulatory monitoring systems.
`It is important to bear in mind the present limitations of the
`technology, such as reliability, system complexity, and cost,
`but there is a wide scope of exciting healthcare applications
`available for this technology, as will be discussed later in this
`article. WBS technology is a platform upon which a new para-
`digm of enhanced healthcare can be established. Considering
`that hardware solutions will
`inevitably become smaller,
`cheaper, and more reliable, and diagnostic software more so-
`phisticated and effective, it seems more a matter of when cost
`effectiveness will be achieved for WBS solutions, not if.
`
`Available WBS Monitoring Modalities
`WBS solutions, in various stages of technologic maturity, exist
`for measuring established cardiopulmonary “vital signs”: heart
`rate, arterial blood pressure, arterial oxygen saturation, respira-
`tory rate, temperature, and even cardiac output. In addition,
`
`there are numerous WBS modalities that can offer physiologic
`measurements not conventional in contemporary medical mon-
`itoring applications, including acoustic sensors, electrochemi-
`cal
`sensors,
`optical
`sensors,
`electromyography
`and
`electroencephalography, and other bioanalytic sensors (to be
`sure, some of these sensors have well-established medical util-
`ity, but not for automated surveillance). These less established
`WBS modalities are outside the scope of this review.
`Wearable electrocardiogram systems represent the most
`mature WBS technology. Holter and related ambulatory
`electrophysiologic monitoring solutions have established util-
`ity in the diagnosis of cardiac arrhythmias. There has been
`substantial examination of this technology in the medical lit-
`erature, with excellent reviews available [10]. Temperature is
`technically trivial to measure using WBS, but the continuous
`monitoring of body temperature is only a soft surrogate for
`perfusion, and it lacks established utility outside of traditional
`clinical settings [11], [12]. There is not a satisfactory ambula-
`tory solution for cardiac output measurement; it has been
`shown that cardiac output can be extracted from thoracic
`bioimpedance measurements, although speaking and irregu-
`lar breathing, as well as posture changes and ambulation, can
`corrupt this signal. In the future, bioimpedance is likely to
`prove a powerful WBS modality, since the signal carries in-
`formation about pulsatile blood volumes, respiratory vol-
`umes, intracellular and extracellular fluid balances, and has
`been shown to enable tomographic imaging. Respiration can
`be measured using bioimpedance, chest wall geometry, and
`acoustic means. While the basic sensor technology exists for
`monitoring respiratory rate, it requires the conversion of a
`continuous waveform into an integer (breaths per unit time),
`or the imprecise conversion of the measured parameter into an
`estimated volumetric rate (liters of gas per unit time).
`Ambulatory systems for arterial blood pressure measure-
`ment exist. The portapres, employing the volume clamp tech-
`nique for measuring ABP, offers a continuous waveform. The
`technology encumbers a finger and the wrist of the subject, is
`somewhat uncomfortable, and requires some expertise to set
`up for a subject (for instance, the finger cuff size must be care-
`fully matched to the finger). A more common WBS solution
`for 24-hour monitoring of ABP involves a portable version of
`the common oscillometric cuff that fits around the upper arm.
`This solution requires that the patient keep the monitored arm
`immobile while the cuff inflates for measurements. By report,
`this solution has been known to interfere with the sleep and
`other activities of monitored subjects (and has been reported
`to cause bruising of the arm at the cuff site) [13].
`
`IEEE ENGINEERING IN MEDICINE AND BIOLOGY MAGAZINE
`
`MAY/JUNE 2003
`
`29
`
`FITBIT, Ex. 1005
`
`

`

`small size, rings are generally worn without removal more of-
`ten than watches. Additionally, recent studies have indicated
`that the finger is one of the best places for WBS sensor attach-
`ment [16]. The primary vasculature of the finger is located
`near the surface and therefore makes it optimal for monitoring
`arterial blood flow using noninvasive optoelectronic sensors.
`Thus, a ring is ideal for long-term measurements. As will be
`illustrated in the following sections, the development of the
`ring sensor has stressed first an understanding of and then the
`subsequent elimination of front-end signal artifacts. By im-
`plementing a mechanical design that is sensitive to the true
`causes of signal corruption, significant improvements in over-
`all signal quality can be achieved and sensor effectiveness for
`various environments can be improved.
`Figure 1 shows the typical waveform of a photoplethys-
`mograph signal obtained from a human subject at rest. The
`signal comprises a large segment of dc signal and a small-am-
`plitude ac signal. The dc component of photon absorption re-
`sults from light passing through various nonpulsatile media,
`including tissue, bones, venous blood, and nonpulsatile arte-
`rial blood. Assuming that these are kept constant, a bandpass
`filter can eliminate the dc component. However, wearable
`PPG sensors do not meet this premise since, as the wearer
`moves, the amount of absorption attributed to the nonpulsatile
`components fluctuates. Power spectrum analysis reveals that
`this motion artifact often overlaps with the true pulse signal at
`a frequency of approximately 1 Hz. Therefore, a simple noise
`filter based on frequency separation does not work for PPG
`ring sensors to eliminate motion artifact.
`Furthermore, wearable PPG sensors are exposed to diverse
`ambient lighting conditions, ranging from direct sunlight to
`flickering room light. In addition, wearable PPG sensors must
`be designed for reduced power consumption. Carrying a large
`battery pack is not acceptable for long-term applications. The
`whole sensor system must run continually using a small bat-
`tery. Several ways to cope with these difficulties are:
`➤ secure the LEDs and the photodetector (PD for short) at a
`location along the finger skin such that the dc component
`may be influenced less by the finger motion
`➤ modulate the LEDs to attenuate the influence of
`uncorrelated ambient light as well as to reduce power
`consumption
`➤ increase the amplitude of the ac component so that the
`signal-to-noise ratio may increase
`➤ measure the finger motion with another sensor or a sec-
`ond PD and use it as a noise reference for verifying the
`signal as well as for canceling the disturbance and noise.
`In the following sections these methods will briefly be dis-
`cussed, followed by specific sensor designs and performance
`tests. There are other techniques for reducing motion artifact
`for general-purpose PPG. These, however, are mostly signal
`processing techniques applicable to PPG intended for short-
`term use. The motion artifact problem we are facing in wear-
`able PPG design is different in nature; the source signal qual-
`ity must be improved before applying signal processing.
`Therefore, the focus must be placed on basic sensor design.
`
`Techniques for Reduced Motion Artifact
`Sensor Arrangement
`The location of the LEDs and a PD relative to the finger is an
`important design issue determining signal quality and robust-
`ness against motion artifact. Figure 2 shows a cross-sectional
`
`No fully satisfactory WBS solution exists for ABP monitor-
`ing. Because this physiologic parameter has been the corner-
`stone of many decades of clinical and physiology practice, it
`will be important to develop future WBS solutions for monitor-
`ing ABP. It is also worth investigating surrogate measures of
`ABP that prove easier to measure, such as pulse wave velocity
`(which correlates well with degree of hypertension [14], [15])
`and the second-derivative of the photoplethysmograph. This ar-
`ticle focuses on a wearable ring pulse-oximeter solution, which
`measures the PPG as well as the arterial oxygen saturation. The
`PPG contains information about the vascular pressure wave-
`forms and compliances. Efforts to extract unique circulatory in-
`formation, especially an ABP surrogate, from the PPG
`waveform are discussed later in this article. The PPG provides
`an effective heart rate (measuring heart beats that generate
`identifiable forward-flow), useful for circulatory consider-
`ations though less useful for strict electrophysiologic consider-
`ations. For instance, the PPG signal may reveal heart rate
`variability, provided ectopic heart beats, which corrupt the as-
`sociation with autonomic tone, can be excluded.
`
`Development of a Wearable Biosensor—
`The Ring Sensor
`Technical Issues of PPG Ring Sensors
`Central
`to the ring sensor design is the importance of
`long-term wearability and reliable sensor attachment. Since
`continuous monitoring requires a device that must be
`noninvasive and worn at all times, a ring configuration for the
`sensor unit is a natural choice. Because of the low weight and
`
`Pulsatile Arterial Blood
`Nonpulsatile Arterial Blood
`
`Venous Blood
`
`Tissue, Bone, Etc...
`
`AC
`
`DC
`
`LightAbsorption
`
`Time
`
`Fig. 1. Illustrative representation of the relative photon
`absorbance for various sections of the finger. The dc compo-
`nent is significantly larger than the ac component.
`
`Control Volume 2
`
`Digital Artery
`
`Bone
`
`Bone
`
`21
`
`21
`
`Photodiode
`
`Control Volume 1
`
`LED
`
`(a)
`
`(b)
`
`Fig. 2. (a) For the reflective illumination method, movement of
`the photodiode relative to the LED (position 1 to position 2)
`leads to a photon path that no longer contains the digital ar-
`tery. (b) For the transmittal illumination method, movement of
`the photodetector relative to the LED still contains photon
`paths that pass through the digital artery.
`
`30
`
`IEEE ENGINEERING IN MEDICINE AND BIOLOGY MAGAZINE
`
`MAY/JUNE 2003
`
`FITBIT, Ex. 1005
`
`

`

`view of the finger with the ring sensor. The LEDs and PD are
`placed on the flanks of the finger rather than the dorsal and
`palmar sides. These locations are desirable for two reasons:
`➤ both flanks of fingers have a thin epidermal tissue layer
`through which photons can reach the target blood vessels
`with less attenuation
`➤ the digital arteries are located near the skin surface paral-
`lel to the length of the finger.
`It should be noted that an arterial pulsation is not only
`greater in magnitude than cutaneous pulsations but is also less
`susceptive to motion due to the naturally higher internal pres-
`sure. While the capillary collapses with a small external pres-
`sure on the order of 10~30 mmHg, the artery can sustain an
`external pressure up to 70~80 mmHg [17], [18]. Therefore,
`light static loads, such as contact with the environment, may
`not disturb the arterial pulsation.
`For these reasons, at least one optical device, either the PD
`or the LED, should be placed on one lateral face of the finger
`near the digital artery. The question is where to place the other
`device. Figure 2 shows two distinct cases. One case places both
`the PD and the LED on the same side of the fin-
`ger-base, and the other places them on opposite
`sides of the finger. Placing both the PD and the
`LED on the same side creates a type of reflective
`PPG, while placing each of them on opposite
`sides makes a type of transmittal PPG. In the fig-
`ure the average pathway of photons is shown for
`the two sensor arrangements. Although the exact
`photon path is difficult to obtain, due to the heter-
`ogeneous nature of the finger tissue and blood, a
`banana-shaped arc connecting the LED and PD,
`as shown in the figure, can approximate its aver-
`age path [19]. Although these two arrangements
`have no fundamental difference from the optics
`point of view, their practical properties and per-
`formance differ significantly with respect to mo-
`tion artifact, signal-to-noise ratio, and power
`requirements [20]-[22].
`Reflective PPG needs more secure attach-
`ments of the LED and PD to the skin surface,
`when compared to transmittal PPG. Once an air
`gap is created between the skin surface and the
`optical components due to some disturbance, a
`direct optical path from the LED to the PD may
`be created. This direct path exposes the PD di-
`rectly to the light source and consequently
`leads to saturation. To avoid this short circuit,
`the LED light beam must be focused only in the
`normal direction, and the PD must also have a
`strong directional property (i.e., polarity), so
`that it is sensitive to only the incoming light
`normal to the device surface. Such strong di-
`rectional properties, however, work adversely
`when a disturbance pressure acts on the sensor
`bodies, since it deflects the direction of the
`LED and PD leading to fluctuations in the out-
`put signal. As a result, reflective PPG configu-
`rations are more susceptive to disturbances.
`In contrast, transmittal PPG configurations
`do not have the short circuit problems, since the
`LED and PD are placed on the opposite sides of
`the finger; no direct path through the air can be
`
`created. Additionally, this design allows us to use devices hav-
`ing a weak polarity, which is, in general, more robust against
`disturbances. Furthermore, transmittal PPG is less sensitive to
`local disturbances acting on the finger, since the LED irradi-
`ates a larger volume of the finger. In the transmittal PPG con-
`figuration, the percentage of the measured signal does not
`significantly change although some peripheral capillary beds
`are collapsed. The percentage change is greater for reflective
`PPG, since this volume is smaller.
`Figure 3 shows an experimental comparison between trans-
`mittal and reflective PPGs. Two sets of PPG sensors, one re-
`flective and one transmittal, were attached to the same finger.
`Both were at rest initially, and then shaken. The transmittal
`PPG was quite stable, while the reflective PPG was susceptive
`to the motion disturbances.
`
`Lighting Modulation
`As is the case with most other WBS technologies, on-board
`power is an extremely important design consideration and is of-
`ten the limiting factor in design size, function, and flexibility.
`
`Reflective
`
`Start of
`Applied Motion
`
`Transmittal
`4
`6
`
`2
`
`8
`
`10
`
`12
`
`14
`
`16
`
`18
`
`20
`
`Time (s)
`
`10
`
`8 6 4 2 0
`
`0
`
`Voltage(V)
`
`Fig. 3. Corruption of a continuous PPG waveform during the application of a
`simple chopping motion. Note that while the motion corrupts the reflective
`sensor signal, the transmittal sensor signal remains unaffected.
`
`LED Brightness
`
`Photodetector Output
`
`Data Sampling
`(a)
`
`T1
`
`High-Speed
`Photo-
`detector
`
`Short, Bright
`LED Pulse
`
`Ts
`
`(b)
`
`Fig. 4. (a) The slow response time of the photodetector meant that the LED
`had to be modulated at lower frequencies for data sampling. (b) A faster
`photodetector response time makes it possible to increase the modulation
`frequency of the LED.
`
`IEEE ENGINEERING IN MEDICINE AND BIOLOGY MAGAZINE
`
`MAY/JUNE 2003
`
`31
`
`FITBIT, Ex. 1005
`
`

`

`described above is an effective method for preventing these
`types of injuries.
`In addition to saving power, the modulation of LED light-
`ing provides an effective means for reducing ambient light
`disturbances. Reading the PD output while the LED is turned
`off yields the baseline PPG level attributed to the ambient
`light alone. Subtracting this reading from the one acquired
`with the LED illuminated gives the net output correlated with
`the LED lighting. More sophisticated modulation schemes
`can be applied by controlling the LED brightness as a periodic
`time function. Computational power requirements often pro-
`hibit complex modulation, however. Design trade-offs must
`be considered to find the best modulation scheme.
`
`40
`
`50
`
`60
`
`Transmural Pressure
`Increasing the detected amplitude of arterial pulsations (i.e.,
`the ac component in Figure 1) improves the signal-to-noise ra-
`tio of PPG. It is well understood that the application of an ex-
`ternal pressure on the tissue surrounding the artery will
`increase the pulsatile amplitude. Such a pressure reduces the
`transmural pressure; that is, the pressure difference between
`inside and outside of the blood vessel. The pulsatile amplitude
`becomes a maximum when the transmural pressure ap-
`proaches zero, since the arterial compliance becomes maxi-
`mal with zero transmural pressure
`[25], [26]. Applying a pressure, how-
`ever, may interfere with tissue perfu-
`sion. Since the device is worn for long
`periods of time, the pressure must be
`kept such that it does not exceed levels
`that could damage other vasculature
`[27]. Thus, the mechanism for holding
`the LED and PD must be designed
`such that it provides a safe level of
`continuous pressure, well below the
`established clinical threshold.
`Figure 5 shows the pulsatile ampli-
`tude of a finger base PPG for varied
`pressures generated by a finger cuff. As
`the cuff pressure increases, the PPG
`amplitude increases until it reaches a
`maximum. As the pressure keeps in-
`creasing further,
`the amplitude de-
`creases due to occlusion of the blood
`vessels. The cuff pressure yielding the
`largest PPG amplitude, generally near
`the mean arterial pressure [28], is too
`high to apply for a long period of time.
`But, to prevent the capillary beds from being collapsed, the cuff
`pressure must be on the order of 10 mmHg, which is too low to
`obtain a sufficient PPG amplitude.
`A solution to this problem is to apply the pressure only at a
`local spot near the photodetector. When using a cuff or any of
`the devices that provide uniform surface pressure onto the fin-
`ger or the arm, it constricts the blood vessels, thus limiting or
`significantly impeding the amount of blood supplied down-
`stream. However, by providing a local, noncircumferential in-
`crease in pressure near the sensor’s optical components, it is
`possible to amplify the plethysmograph waveform while
`avoiding the potentially dangerous situation of long-term
`flow obstruction. As shown in Figure 6, the tissue pressure in
`the vicinity of one of the arteries can be increased with use of a
`
`40
`
`50
`
`60
`
`To keep the overall unit small, the ring sensor design demands a
`power source that is no larger than the coin batteries used for
`wristwatches. Despite the superior stability and robustness,
`transmittal PPG consumes more power. According to the Lam-
`bert-Beer law, the brightness decreases exponentially as the
`distance from the light source increases. Transmittal PPG must
`have a powerful LED for transmitting light across the finger.
`This power consumption problem can be solved with a lighting
`modulation technique using high-speed devices. Instead of
`lighting the skin continually, the LED is turned on only for a
`short time, say 100 ~ 1000 ns, and the signal is sampled within
`this period. High-speed LEDs and PDs, which have become
`available at low cost in recent years, can be used for this pur-
`pose. Figure 4 shows a schematic of high-frequency, low-duty
`cycle modulation implemented to minimize LED power con-
`sumption. Utilizing fast rise-time optical detectors, it is possi-
`ble to incorporate a modulation frequency of 1 kHz with a duty
`ratio of 0.1%, a theoretical power usage that is 1,000 times less
`than conventional full-cycle modulation methods [23].
`Use of a strong light source needed for transmittal PPG
`may cause a skin-burning problem. As reported in [24], if the
`sensor is attached for a long time the heat created by a power-
`ful LED may incur low-temperature skin burning. The afore-
`mentioned
`high-frequency,
`low-duty
`rate modulation
`
`2.5
`
`−2.5
`
`0
`
`300
`
`10
`
`20
`
`30
`Time (s)
`(a)
`
`0
`
`0
`
`10
`
`20
`
`30
`Time (s)
`(b)
`
`Voltage(V)
`
`Pressure(mmHg)
`
`Fig. 5. (a) PPG signal amplitude. (b) Pressure at the photodetector.
`
`Finger
`
`Sensor
`
`Locally
`Pressurized
`
`Photo-
`detector A
`
`Vein
`
`Photo-
`detector B
`
`LED
`
`Artery
`
`Pusher
`
`Fig. 6. The schematic of a locally pressurized sensor band.
`
`32
`
`IEEE ENGINEERING IN MEDICINE AND BIOLOGY MAGAZINE
`
`MAY/JUNE 2003
`
`FITBIT, Ex. 1005
`
`

`

`special mechanism pushing Photodetector A toward the skin.
`This mechanism, which is attached to the sensor band, would
`change the pressure distribution such that the transmural pres-
`sure of one of the arteries could be high enough to obtain a
`large pulsatile signal while keeping the pressure low else-
`where to allow for sufficient blood perfusion. As long as the
`pressurized area is small enough to perfuse it from the sur-
`rounding tissue, the local pressurization causes no major
`complication although the pressure is applied for many days.
`
`Noise Reference
`Wearable sensors are to be used with no supervision by medi-
`cal professionals, as mentioned previously. It is therefore im-
`portant to monitor whether signals have been obtained under
`proper conditions. Although the techniques described above
`are effective for reducing motion artifact, it is still necessary
`to verify the signal before sending it out for clinical diagnosis.
`Questionable data can be rejected if the wearable sensor has a
`means to monitor the hand motion and other sources of distur-
`bances. In the following section, a novel method for detecting
`finger motion for the verification of signal reliability as well
`as for recovering the correct signal from a distorted signal will
`be described.
`The motion of the finger can be measured with an acceler-
`ometer attached to the body of the ring.
`MEMS accelerometers are now avail-
`able at low cost, but they are still too
`bulky and/or consume too much power
`to use for the ring sensor. There is no
`commercially available product satis-
`fying both the power limit and form
`factor requirements. Instead of using a
`standard sensor dedicated for motion
`measurement, the PPG optical sensor
`can be used as a motion sensor. The fact
`that PPG is susceptive to motion distur-
`bances implies that the sensor has the
`potential to be an effective detector of
`motion. With minor modifications to
`the original PPG design, the PPG mo-
`tion detector would have a high sensi-
`tivity to detect
`the nonpulsatile dc
`component shown in Figure 1. The
`techniques developed for reducing mo-
`tion artifact are to be reversed in order
`to increase the motion sensitivity. First,
`the reflective PPG arrangement should
`be used, so the distance between the
`LED and PD must be shortened. The
`pressure should be kept low so that less
`pulsatile signals may be observed. The
`location of the PD should be away from
`the arteries and close to a vein instead. In addition, the wave-
`length of the LED should be selected such that it is more sensi-
`tive to the reduced hemoglobin (i.e., approximately 660 nm),
`since the nonpulsatile vein is filled mostly with the reduced he-
`moglobin. Figure 6 shows a desirable location for
`the
`photodetector detecting the finger motion. Photodetector B in
`the figure is placed close to the LED as well as to a vein on the
`low-pressure side. Figure 7 shows an experimental result of the
`reflective PPG exposed to hand motion. It is clear that the PPG
`signal has a strong correlation with the acceleration of the hand.
`
`Vertical
`Motion
`
`Horizontal
`Motion
`
`Motion Detection PD
`
`The motion detector can be used not only for monitoring
`the presence of motion but also for canceling noise. By us-
`ing PD-B as a noise reference, a noise cancellation filter
`can be built to eliminate the noise of PD-A that correlates
`with the noise reference signal. Assuming that the hemo-
`dynamic process observed by PPG is stationary and that the
`noise is additive, adaptive noise canceling methods, such as
`the classical Widrow method [29], can be applied in order
`to recover the true pulsation signal from corrupted wave-
`forms. As shown in Figure 8, the noise-canceling filter
`combines two sensor signals; one is the main signal cap-
`tured by PD-A and the other is the noise reference obtained
`by PD-B. The main signal mostly consists of the true pulsa-
`tile signal, but it does contain some noise. If we know the
`proportion of the noise contained in the main signal, we can
`generate the noise of the same magnitude by attenuating the
`noise reference signal and then subtract the noise from the
`main signal to recover the true pulsatile signal. If the noise
`magnitude is not known a priori, it must be determined
`adaptively during the measurement. Various algorithms for
`adaptive filtering can be applied to tune the filter in real
`time. Some can determine optimal filter gains and parame-
`ters based on the evaluation of the recovered signal, as
`shown in Figure 8 by the feedback from the output to the
`
`Acceleration Applied
`
`No
`Accel
`
`Medium
`Accel
`
`No
`Accel
`
`Low
`Accel
`
`No
`Accel
`
`High
`Accel
`
`0
`
`10
`
`20
`
`30
`Time (s)
`(a)
`
`40
`
`50
`
`60
`
`0
`
`10
`
`20
`
`30
`Time (s)
`(b)
`
`40
`
`50
`
`60
`
`2 1 0
`
`−1
`−2
`
`2 1 0
`
`−1
`−2
`
`Voltage(V)
`
`Voltage(V)
`
`Signal
`Source
`
`Noise
`Source
`
`Photodetector A
`Main
`Signal
`
`+
`
`−
`
`Noise
`Reference
`
`Adaptive
`Filter
`
`Photodetector B
`
`Fig. 8. Block diagram of adaptive noise cancellation using
`second PPG sensor as noise reference.
`
`Fig. 7. Reflective PPG used as a motion detector: PPG output has a strong correlation
`with the vertical and horizontal acceleration of the finger. (a) Photoplethysmogram
`with vertical acceleration. (b) Photoplethysmogram with horizontal acceleration.
`
`IEEE ENGINEERING IN MEDICINE AND BIOLOGY MAGAZINE
`
`MAY/JUNE 2003
`
`33
`
`FITBIT, Ex. 1005
`
`

`

`Since continuous monitoring requires
`a device that must be noninvasive
`and worn at all times, a ring
`configuration for the sensor unit
`is a natural choice.
`
`adaptive filter block. Details of this adaptive filtering
`method are beyond the scope of this article. The dual
`photodetector design shown