IEEE TRANSACTIONS ON BIOMEDICAL CIRCUITS AND SYSTEMS,VOL.6, NO. 3, JUNE 2012
`
`269
`
`A Wireless Reflectance Pulse Oximeter With Digital
`Baseline Control for Unfiltered Photoplethysmograms
`
`Kejia Li, Student Member, IEEE, and Steve Warren, Member, IEEE
`
`Abstract—Pulse oximeters are central to the move toward
`wearable health monitoring devices and medical electronics either
`hosted by, e.g., smart phones or physically embedded in their
`design. This paper presents a small, low-cost pulse oximeter
`design appropriate for wearable and surface-based applications
`that also produces quality, unfiltered photo-plethysmograms
`(PPGs) ideal for emerging diagnostic algorithms. The design’s
`“filter-free” embodiment, which employs only digital baseline
`subtraction as a signal compensation mechanism, distinguishes
`it from conventional pulse oximeters that incorporate filters for
`signal extraction and noise reduction. This results in high-fidelity
`PPGs with thousands of peak-to-peak digitization levels that are
`sampled at 240 Hz to avoid noise aliasing. Electronic feedback con-
`trols make these PPGs more resilient in the face of environmental
`changes (e.g., the device can operate in full room light), and data
`stream in real time across either a ZigBee wireless link or a wired
`USB connection to a host. On-board flash memory is available for
`store-and-forward applications. This sensor has demonstrated an
`ability to gather high-integrity data at fingertip, wrist, earlobe,
`palm, and temple locations from a group of 48 subjects (20 to 64
`years old).
`Index Terms—Filter-free, high-fidelity photoplethysmogram,
`low cost, pulse oximeter, reflectance sensor, surface biosensor,
`wearable, wireless.
`
`I.
`
`INTRODUCTION
`
`EALTH problems such as cardiovascular disease, hyper-
`
`H tension,diabetes, andcongestiveheartfailure continueto
`
`plague society [1]. These conditions are primary drivers for the
`development of wearable and mobile health monitoring tech-
`nologiesthat offer the potential to (a) increase the quality oflife
`for individuals that already suffer from these health conditions
`and (b) prevent or mitigate the onset of disease in thosethat are
`at risk to acquire these health issues [2]. Of the array of medical
`devices that can be brought to bear for wearable/mobile appli-
`cations that address these diseases, pulse oximeters offer signif-
`icantrelative promise because they provide twoclinically rele-
`vant health parameters [heart rate (HR) and blood oxygen sat-
`uration (SpO+)], they do not require electrical contactto tissue,
`
`Manuscript received December 03, 2010; revised April 24, 2011 and July
`18, 2011; accepted August 15, 2011. Date of publication November 04, 2011;
`date of current version May 22, 2012. This work was
`in part by the
`National Science Foundation under Grants BES-0093916, BES-0440183, and
`CNS-0551626, and by the Kansas State University Targeted Excellence Pro-
`gram. Opinions, findings, conclusions, or recommendations expressed in this
`material are those of the author(s) and do not necessarily reflect the viewsof the
`NSF.This paper was recommended by Associate Editor E. Jovanov.
`The authors are with the Departmentof Electrical and Computer Engineering,
`Kansas State University, Manhattan, KS 66506 USA (e-mail: kejiali@ksu.edu;
`swarren@ksu.edu).
`Color versions of one or more ofthe figures in this paper are available online
`at http://ieeexplore.icee.org.
`Digital Object Identifier 10.1109/TBCAS.2011.2167717
`
`and they can operate at very low power[3], [4]. Additionally,
`the pulsatile plethysmographic data offered by this light-based
`sensing technique (which are usually discarded by commercial
`pulse oximeters after being used to calculate the parameters
`for the front panel display) can help to ascertain hemodynamic
`information that is well-suited for the assessment of the dis-
`
`ease states listed above [5]-[8]. This information includes blood
`pressure [9], [10], arterial compliance [6], [11], [12], pulse wave
`velocity (PWV) [2], [13], stroke volume (and therefore cardiac
`output) [14], and other vascular parameters [7], [8], [15], [16].
`Other relevant quantities include respiration rate [17], [18], pa-
`tient motion [16], and even patient authentication [19]-[21].
`However, low-cost pulse oximeter designs are unavailable
`that provide (a) quality, unfiltered PPGs ideally suitable for re-
`search and education toward the realization of new PPG diag-
`nostics and (b) positional flexibility suitable for mobile and sur-
`face-based applications. While PPGsare often accessible from
`commercial desktop units via serial ports, these data have been
`filtered in proprietary ways to stabilize HR and SpO, calcula-
`tions. Further, due to their clinical prevalence, pulse oximetry
`and PPG analysis deserve coverage in biomedical instrumenta-
`tion laboratories offered in secondary education curricula, yet
`low-cost pulse oximeters that provide reasonable-quality PPGs
`are not a staple in off-the-shelf educational kits.
`types of
`Regarding ambulatory pulse oximeters, several
`wearable designs exist. Some of these use ring form factors,
`and others use finger clips. These designs use predominantly
`transmission-modesensors. For broader use with wrist watches,
`head bands, socks, sensor “Band Aids’, and other wearableplat-
`forms that are unobtrusive and well suited for mobility, it makes
`sense to consider reflectance-mode layouts. This is especially
`true when one contemplates the immense potential of ‘surface
`biosensors’ (SBs): medical sensors embedded in the surface
`of everyday consumerelectronics such as hand-held personal
`device assistants (PDAs), cell phones, smart phones, tablet PCs,
`head-mounted displays, etc. In this paradigm, physiological
`sensors will be accessible and signals will be easy to obtain,
`as human factors considerations for the overall product design
`will drive ease of use for the integrated biosensors. Addition-
`ally, each SB will utilize its host device’s processor, memory,
`display, and wireless communication resources to provide user
`services typically unavailable in wearable platforms [22]. E.g.,
`consider a reflectance pulse oximeter embedded on the back
`side of a cell phone alongside a built-in camera. As the user
`holds their finger against the reflectance sensor, these data will
`be processed by the microprocessorin the cell phone, and the
`LCDscreen will display the signals and parameters.
`In this domain, a reflectance sensor can employ a single small
`photodiode [23] as in most transmittance sensors. However,
`
`1932-4545/$26.00 © 2011 IEEE
`Authorized licensed use limited to: Fish & Richardson PC. Downloaded on July 21,2023 at 03:50:33 UTC from IEEE Xplore. Restrictions apply.
`
`APPLE 1051
`Apple v. Masimo
`IPR2022-01291
`
`1
`
`APPLE 1051
`Apple v. Masimo
`IPR2022-01291
`
`

`

`270
`
`TRANSACTIONS ON BIOMEDICAL CIRCUITS AND SYSTEMS, VOL. 6, NO. 3, JUNE 2012
`
`
`
`Multiple body locations; Warious vascular
`profiles and perfusionlevels
`Low-profile reflectance layout adaptable for
`Wearability
`wearable and SB applications
`Cost Low (<$100)
`
`
`
`
`
`TABLE I
`tissue is highly forward scattering, so the relative number of
`WIRELESS REFLECTANCE PULSE OXIMETER DESIGN REQUIREMENTS
`
`remitted photons detected in reflectance modeis low, yielding
`
`lower quality PPGs [5]. Improved sensor configurations are
`Category
`Requirements
`Signal
`therefore often adopted to better acquire the radial reflectance
`profile, including a ring-shaped photodiode design [24], [25], a
`Integrity
`Unfiltered data with an optimal SNR
`Precision
`Thousands of peak-to-peak digitization levels
`photodiode array around the central LEDs [26], and conversely
`> 240 Hz to minimize PPG/noise aliasing
`Sampling frequency
`an LED array arounda central photodiode [27]. These designs
`Baseline subtraction
`Digital andfilter-free
`generally employ cascaded high pass and low pass filters to
`
`Dataavailability
`Full access to all pulsatile/baseline data
`
`extract the PPGs [28]. Such analogfilters inevitably alter and
`Sensor
`even distort the signals of interest. These alterations are visibly
`Radial arrangement, large area, and 3-5 mm
`LED/detector geometry
`obvious in some papers, and cycle-to-cycle inconsistencies can
`source/detector separation
`
`be significant. For this reason, a filter-free design is desirable.
`Ambient light operation|Adjustable gain and reference baseline
`Functionali
`
`Finally, in a low-cost wearable sensor with a limited voltage
`Communication
`Wireless (10 m range) and USB
`range (e.g., [0, 3.3] V if battery powered) and a low-precision
`Local storage
`Onboard flash memory
`analog-to-digital converter (e.g., 8 to 12 bits), maintaining a
`
` Battery USB-rechargeable; Multi-daylifetime
`sensible vertical resolution, number ofdigitization levels, and
`Client Software
`Visualization and control panel
`
`Application
`sampling frequency for a PPG can be difficult without flexible
`Measurement sites
`baseline subtraction and PPG amplification strategies.
`In summary, the desire to extract additional physiological
`information from PPGs acquired with reflectance-mode sen-
`sors imposes design constraints with respect to signal quality.
`This paper presents the design of a low-cost, wireless, re-
`flectance-mode pulse oximeter suitable for these needs. It is
`initially housed on the surface of a printed circuit board but
`can be easily migrated to other surface-based applications.
`Here, a unique filter-free circuit (that digitally extracts the
`PPG waveform) and a two-stage, feedback-loop-driven con-
`trol system enable the acquisition of unfiltered PPGs with
`2! = 4096 levels of precision from varied body locations. An
`optimized LED/detector configuration promotes surface use,
`and the device signal quality and cost enhance its potential for
`integration into SB-based consumer devices.
`
`are undetected. Photons collected at greater radial distances
`are more likely to have traveled deeper into blood-perfused
`tissue and contain a greater percentage of AC data. Given the
`increased sensing area in a large-area detector,
`the control
`circuitry must easily compensate for baseline changes due to
`ambient light, tissue perfusion, respiration depth, etc.
`Fig. 1 shows the block diagram for the pulse oximeter cir-
`cuitry; a brief description was also included in [30]. The LED,
`sensor array, and operational amplifier (OPA)circuitry are coor-
`dinated by a Jennic JN5139 microcontroller. The intensity and
`timing of the bi-color LED are controlled by a digital-to-analog
`converter (DAC) and digital input/output ports (DIOs), respec-
`tively. A signal from the sensor array (four photodiodes sur-
`rounding the central bi-color LED)is first buffered and then fed
`to a differential OPA circuit. The buffered signal, designated
`here as the first-stage PPG signal (entire AC + DC contribu-
`tion), is sampled by an analog-to-digital converter (ADC). An-
`other ADC collects the second-stage PPG signal(the AC portion
`only) from the output of the differential OPA circuit that has a
`positive input from another microcontroller DAC.
`Nofilters are used in the signal acquisition process, whose
`elements will be introduced in Part C. Closed-Loop System.
`The battery (unstable power source) is isolated from the PPG
`excitation and collection circuitry, since it is powered by the
`microcontroller’s analog peripheral regulator (APR). Normally,
`the pulse oximeter uses a wireless link to communicate with
`a receiver on a PC, and data are stored on the PC through a
`MATLAB graphical user interface (GUI). A mini-USB connec-
`tion can provide a wired interface to the PC while thebattery is
`recharged. If neither the wireless link nor the USB connection
`is available, sampled data will be temporarily stored on the flash
`memory module(e.g., for store-and-forward applications).
`
`The design requirements are outlined in Table I. Signal re-
`quirements include quality, unfiltered PPGs whosebaselines are
`digitally removed, consistent with the discussion in Section I.
`INTRODUCTION.The high sampling rate ensures that (a) primary
`signal and noise components are adequately sampled without
`aliasing and (b) secondary noise harmonics, e.g., 120 Hz up to
`several kHz from fluorescent lighting, are not aliased on top of
`the signal components of interest.
`Regarding sensor requirements, the photodetectors are ide-
`ally distributed radially around the central excitation LEDs
`to maximize the number of photons collected. Further, an
`LED/detector separation of 3 to 5 mm is appropriate at these
`wavelengths, as it maximizes the AC/DCratio for each sensor
`channel, as verified experimentally [26], [29] and with Monte
`Carlo simulations [21]. In other words, reflectance photons
`Thefirst-stage PPG is characterized by a large DC portion
`that contain DC information from shallow, poorly perfused
`and a small AC portion, as in Fig. 2. The goal is to extract the
`epidermal layers reflect near the central excitation LEDs and
`Authorized licensed use limited to: Fish & Richardson PC. Downloaded on July 21,2023 at 03:50:33 UTC from IEEE Xplore. Restrictions apply.
`
`B. AC Extraction and Drift Resistance
`
`Tl. METHODS
`
`This section presents a design for a filter-free, reflectance
`pulse oximeter that combines many desirable features into a
`single platform. Implementation hardware is unspecified here;
`board-level components are detailed in Section III. DEVICE
`PROTOTYPE.
`
`A. Requirements and Device Layout
`
`2
`
`2
`
`

`

`LI AND WARREN: WIRELESS REFLECTANCE PULSE OXIMETER
`
`271
`
`a
`
`VCC
`
`Microcontroller (JN5139)
`
`
`32-bit
`RISC CPU
`
`Sensor Array (PDV-C173SM)
`
`OPA (OPA2336)
`
`Fig. 1.
`
`Circuit-level system layout. Coordinated by the microcontroller, signal baselines are digitally extracted as an alternative to conventional filtering.
`
`13" 555-
`
`pin of the OPA, effectively inverting the AC signal amplitude
`prior to digitization.
`$1 is naturally unstable, as both its AC and DC levels are
`influenced by changesin intrinsic blood flow, extrinsic motion,
`respiration, backgroundlight, etc. These factors cause drifting
`in 52. The input voltage range, or digitization range, of the
`12-bit ADC is set to [0, 2.4] V, so one digitization level is
`2.4 V/4095 levels = 0.586 mV. For example, given a gain
`G = 30 and a constant V,.;/, one digitization-level increment
`in the DC signal results in a decrement of 30 digitization levels
`in S» according to (1). As in Fig. 2, Sy may drift 0.3 V (512
`digital levels) in 10 seconds, which is unacceptable because the
`signal will eventually clip at the lower bound of the sampling
`range, and clipped data mean signal corruption. To address this
`issue, (1) implies that one can adjust one or more elements on
`the right side to adjust the value of Sy on theleft. In this effort,
`a Vey adjustment is employed to resist 52 drifting, since V,..5
`is an output of the DAC and can be easily updated. V,.;/ is
`defined as
`
`Vreg = MA(t) + Va
`
`(2)
`
`2nd Stage
`
` 2
`os}LA
`seoAA AM
`A
`
`AVN WA
`WAU AA
`oo
`
`
`ad
`
`0)
`
`_ wn r Vref
`
`1st Stage
`Sign
`
`T
`
`
`
`Amplitude(V)
`
`20
`Time (sec)
`
`Fig. 2. A differential amplifier with gain G compares the first-stage PPG (51)
`to a DC reference voltage to obtain the second-stage PPG (.S.).
`
`second-stage AC signal by eliminating the DC component. (In
`many systems, a high pass filter extracts the AC signal.) If the
`DC portion instead remains, then obtaining hundreds to thou-
`sands of digitization levels in the AC portion over its small
`voltage range requires an ADC of very high precision (e.g.,
`16-bit), which is inappropriate for a low-cost, low-power-con-
`sumption device. This extraction, or DC removal, process is ex-
`ecuted by the OPA unit. Its role is expressed as
`
`where JfA(t), the estimator of the DC component, Vnc, is a
`W-point (e.g., W = 256) moving average of the first-stage
`signal over the time interval that ends at t. V, (the adjustable
`term) is added to A/_A(¢) to ensure that Vp makes Sz in (1)
`positive.
`V,.7 usually varies slowly (several seconds perdigitization
`level change, in an environment with minimal motion and am-
`bient noise), and the V;.,; adjustment leads to a discontinuity
`in S». Hence, the V,,.- data must also be transmitted or stored
`where 5 and 5%are the first-stage and second-stage signals, re-
`along with the digitized second-stage data in order to restore the
`spectively, G is the gain of the OPA, and V,..¢ is a user-defined
`original PPGs, a process called “compensation.”
`Fig. 3 shows a data set from the palm. Collected data are
`reference voltage that functionally equates to the DC signal
`compensated to remove discontinuities caused by V,.¢ Jumps,
`level. To show an upward-oriented PPG peak during systole as
`with a blood pressure curve, V,.¢ is connected to the positive
`or immediate value changes, in the pulsatile waveform using
`Authorized licensed use limited to: Fish & Richardson PC. Downloaded on July 21,2023 at 03:50:33 UTC from IEEE Xplore. Restrictions apply.
`
`So =Gx (Vref - 51)
`
`(1)
`
`3
`
`T
`
`‘
`
`T
`
`T
`
`T
`
`T
`
`T
`
`24
`
`3
`
`

`

`272
`
`TRANSACTIONS ON BIOMEDICAL CIRCUITS AND SYSTEMS, VOL. 6, NO. 3, JUNE 2012
`
`
`
`Vref
`
`Collecteddata
`|
`
`i
`
`yi
`
`
`
`O45;
`
`
`
`Amplitude(Vv) O0.5}-
`
`Compensated signal
`
`\/Vref jump mark
`
`“4s
`
`20
`
`AC
`
`
`
`
`
`oooo
`
`| ImportantSignal
`[J] Feedback algorithm
`(Closed-Loop
`
`Fig. 4. Pulse oximeter controlflow that illustrates how the first-stage PPG (5)
`can be used to create the second-stage PPG (52), where both signals provide
`feedback to stabilize the acquisition process.
`
`an optimal level within the active range of the photodiode, inde-
`pendentof a subject’s vascular and perfusion profiles [30]. The
`closed loopin the upper right prevents $2 from saturation, since
`the compensation method described in (6) requires an unsatu-
`tated second-stage signal. Upon detecting saturation onset, the
`Saturation Inhibitor adjusts the V; componentof V,.;, which
`leads to a corresponding change in 52 according to (1).
`To maintain signal quality, the Intensity Regulator sensitivity
`should be minimized. When the regulator affects changes in
`LEDexcitation level, the influence on the first-stage signal con-
`verted by the photodiode sensor array will be hard to predict be-
`cause the blood-perfused tissue between the LED and the sensor
`is an unknown system. Conversely, the sensitivity of the Satu-
`ration Inhibitor should be set high to ensure a rapid response to
`signal drift. Since this adjustment only influences V,..;, the na-
`tive PPG is uncontaminated, and the second-stage signal can be
`adjusted using (6).
`In a controlled scenario, ambient noise variations can be ig-
`nored. If the desired signal intensity increases as the LED in-
`tensity increases, the signal-to-noise ratio (SNR) will improve.
`However, this implies a saturation risk due to a second-stage
`signal with too large of a magnitude within a fixed digitization
`range, in spite ofthe aforementioned drift-resistant method. Ad-
`ditionally, a more intense LED consumes more power. So, an
`optimized intensity level should be empirically predetermined
`as an Intensity Regulator reference.
`
`D. Removable Noise
`
`Fig. 3. Palm PPG data before (blue) and after (red) compensation.
`
`Time (sec)
`
`the following method.Inserting V;-.s from (2) into (1) and then
`rearranging the result isolates the first-stage signal, 51
`
`$1 = MA(t) + V4 - 22,
`The compensated second-stage signal, So, can be represented
`
`(3)
`
`as
`
`So = G x (Voc + Vi — $1)
`
`(4)
`
`where \,. is added to Vpto ensure a positive So. Substituting
`(3) into (4) yields
`
`So = S2—-G x (MA(t) — Voc).
`
`(5)
`
`Typically, Vpc is an unknown constant, butit is sensible to
`initially set Vp¢ = MA(to) at time tp and define Vjump =
`MA(t) — M-A(to} at time t (t > to) so that (5) becomes
`
`Sy = So — GX Vjump-
`
`(6)
`
`With this method, each PPG can be restored as long as the
`second-stage signal is unsaturated. The V,.- adjustmenteffec-
`tively resists first-stage-signal drifting. For example, in Fig. 3,
`the compensated signal drifts below 0 V after 6 seconds. If no
`Ve adjustment occurs, the subsequentsignal is sampled as 0 V.
`To calculate blood oxygensaturation, V;.¢ is usually considered
`equal to Vpe.
`
`In the U.S.A, ambient light often includes a 60 Hz compo-
`nent and the associated harmonic noise, e.g., 120 Hz flicker
`from full-wave-rectified fluorescent room lights plus higher-fre-
`quency harmonics. Most physiological information in a PPG re-
`C. Closed-Loop System
`sides in the range of 0-20 Hz. From the Nyquist-Shannon sam-
`The V,..¢ adjustment mechanism not only helps to realize the
`pling theorem, the lowest sampling frequency, f,, should then
`be 40 Hz, but to prevent ambient noise aliasing, sampling fre-
`AC extractiontask; it also results in resilience in the PPG signal.
`quencies ofat least 240 Hz are needed.
`In the control system, asillustrated in Fig. 4, two closed loops
`Fig. 5 depicts the magnitude spectrum of a PPG containing
`provide stability for the whole data acquisition process. The
`closed loop in the lower left maintains the 5S, value in a prede-
`ambient noise. The heart rate component is 1.329 Hz, andits
`harmonics dominate in the frequency band below 20 Hz. At
`termined range, whichis set by the Intensity Regulator that con-
`greater frequencies, noise is apparentat 60.02 Hz, 84.43 Hz (un-
`trols the led intensity via a DAC. The physical function of this
`clear source), and 119.9 Hz. Mostof this noise is removable
`control loop is to maintain the number ofreflected photons at
`Authorized licensed use limited to: Fish & Richardson PC. Downloaded on July 21,2023 at 03:50:33 UTC from IEEE Xplore. Restrictions apply.
`
`4
`
`4
`
`

`

`LI AND WARREN: WIRELESS REFLECTANCE PULSE OXIMETER
`
`273
`
`
`
`back-forth
`
`©
`up-down
`
`®
`
`w
`
`3?
`
`2F
`
`q
`
`®
`60
`
`a
`= 0
`g= 27
`&e
`=
`
`Motion Artifact Types
`
`QO
`
`@
`
`r 1
`
`
`
`Amplitude(V4
`
`0
`
`2
`
`4
`
`100
`
`1
`
`1
`
`8
`6
`Time (sec)
`(a)
`1
`
`1
`
`19.9 Hz
`
`(dB) og
`Magnitude
`
`20
`
`40
`
`60
`Frequency (Hz)
`(b)
`
`100
`
`120
`
`Fig. 5. An example of removable noise. (a) Compensated PPG corrupted by
`ambient noise. (b) Frequency spectrum of these data sampled at 240 Hz.
`
`by post-processing as long as the sampling frequency is high
`enough that these noise components do notalias into the fre-
`quency range of the signal components of interest. Note that the
`raw signal exhibits a low SNR compared to PPGs from pulse
`oximeters that employfilters, but all signal components are in-
`tact and many can be removedto create a high SNR (see Fig.12).
`
`E. Motion Artifact
`Motionartifact is an issue for a pulse oximeter, especially in
`reflectance mode [5]. Existing literature focuses on signal pro-
`cessing to reduce motionartifact and restore PPGs [31]. Most
`methods assume that enough information exists in the corrupted
`signal for PPG recovery. However, if motion is severe, satura-
`tion occurs frequently and lasts for some time, leading to data
`loss. With this in mind, this development considered motion ar-
`tifact to be a type of signal drift that can be partially addressed
`with a drift-resistant method (V,..¢ adjustment); the design does
`not address motion extraction.
`
`Motionartifact can be classified into two categories: slight
`and severe. Fig. 6 demonstrates the severe condition character-
`ized by three axes of hand motion, where the sensor is taped to
`the finger. Movements are within a 10 cm range and occur at a
`rate of ~1 Hz. The PPG is severely corrupted (the fundamental
`frequency is 1.028 Hz), and it is clipped at the upper and lower
`bounds ofthe digitization range; many AC segments are lost and
`unrecoverable.
`
`(a) PPG severely corrupted by hand motion along three axes.
`Fig. 6.
`(b) Frequency spectrum of the 28 seconds of data sampled at 100 Hz.
`
`
`08
`
`W:256 A:240 Hz
`
`
` Amplitude(¥) oo>oo
`
` 1 1
`
`10
`12
`Time (sec)
`(a)
`
`
`'
`
`T
`
`T
`T
`v
`W:256 A:0.1 Hz
`
`T
`
`T
`
`1
`
`-
`
`-
`
`0.8
`= 08
`aco
`
`==E<
`
`o mn
`
`t
`
`Amplitude(Vv) o >
`
` 1
`
`a
`
`2
`
`4
`
`6
`
`8
`
`10
`
`14
`
`i
`12
`
`Time (sec)
`(c)
`
`Fig. 7. Three uncompensated PPGs acquired at f,; = 240 Hz under similar
`slight-motion conditions but with different parameter pairs (WV. .A).
`
`The slight condition refers to, e.g., slow body movements,
`where a PPG retains its general shape but contains spurious
`components relative to a still condition. To counteractthis type
`of artifact, the shift-resistant method is promising and relies on
`the signal sinceA(t) = 5, and consequently 55=G x V4
`the setting of an optimal assignmentrate and window width for
`V,-.¢ adjustment. The DACassigns the V,.,¢ value to the positive
`according to (1) and (2).
`amplifier pin, and that voltage remains constant until the next
`As an illustration of slight motion response, Fig. 7 shows
`Vref assignment to the DAC. The window size of the moving
`three experimental records acquired under similar conditions
`average filter (the DC estimation time delay, or count) and the
`(exaggerated deep respiration activity), where a different
`rate of assigning V;..¢ to the DAC (notthe rate of V,..¢ varia-
`moving-average window width, W’,and V,.¢ assignment rate,
`A, are employed in each case. Only subplot (a) offers a reason-
`tion) influence the second-stage signal. An extreme case occurs
`when the window size W = 1| data point and the rate of as-
`able representation of the PPG. A lower assignmentrate (b) or
`signing V,. to the DAC is A = f,: motion will never influence
`wider window (c) causes the signal to drift severely, and some
`Authorized licensed use limited to: Fish & Richardson PC. Downloaded on July 21,2023 at 03:50:33 UTC from IEEE Xplore. Restrictions apply.
`
`5
`
`o =
`
`
`i
`q0
`
`5
`
`10
`
`15
`
`20
`
`3
`
`+
`1,028 Hz
`
`Time (sec)
`(a)
`+
`
`™* 20.01 Hz
`
`
`
`
`1
`
`|
`30
`
`7
`
`50
`
`o
`
`mn
`
`s
`
`i
`
`36
`
`40
`
`45
`
`a
`25
`w
`Frequency (Hz)
`(b)
`
`5
`
`

`

`274
`
`
`
` TRANSACTIONS ON BIOMEDICAL CIRCUITS AND SYSTEMS, VOL. 6, NO. 3, JUNE 2012
`
`25
`DATA CLEAR
`
`
`Fig. 8. Pulse oximeter MATLAB GUL.In this example, a series of calibration
`coefficients (lower right) is extracted from the current data (upper right).
`
`segments are nearly saturated. The empirical parameter pair
`(W= 256, A = f;) was adopted for V,..¢ adjustment.
`
`FEF. MATLABInterface
`A MATLAB GUI allowsa user to set/view communication
`parameters, visualize PPGs, processthese data in real time (e.g.,
`digitally filter a signal with a linear-phase filter), and store raw
`data to files, making it a helpful developmenttool (see [16] for a
`full description). Fig.8 illustrates an example data set obtained
`by this GUI, where acquisition options (e.g., Serial Port, Sam-
`pling Rate, Signal Channel, and Signal Processing Type) are
`specified on the left control panel. The upper axes display the
`raw PPG andbaselinefor the near-infrared channel, whereas the
`lower axes show the real-time calibration coefficient, , calcu-
`lated from the magnitudes of the fundamental red/infrared fre-
`quency components using a Fourier transform method [21]. R
`is updated every 0.5 seconds using the previous 4 seconds of
`PPG data. An overall SpO,valueis achieved by calculating the
`median or mean of 40 consecutive Jt values (in a 20-second seg-
`ment) and inserting the result into a pre-determined linear cali-
`bration equation.
`
`Phi
`
`Fig. 10. Bottom view of the wireless reflectance pulse oximeter.
`
`peripherals (including four 12-bit ADCs and two 11-bit DACs),
`and up to 21 DIO ports. The wireless link requires the most
`current, with a TX (transmitter) current draw of 38 mA and
`an RX (receiver) current draw of 37 mA. The CPU consumes
`7.75 mA at full speed, and the current required by the pe-
`ripherals (ADC, DAC, UART, Timer, etc.) is less than 1 mA
`in aggregate. The JN5139 sleep current (with an active sleep
`timer) is only 2.6 jnA.
`The excitation LED module uses a low-cost Marubeni
`
`III. DEVICE PROTOTYPE
`
`SMT660/910 bi-color LED with a typical forward current of
`20 mA and forward voltages of 1.9 V and 1.3 V for the 660 nm
`and 910 nm sources, respectively. The 11-bit DAC output
`(0-2.4 V) provides excitation signal modulation by managing
`Figs. 9 and 10 contain top and bottom views of the pulse
`the power supply for the excitation LED module.
`oximeter prototype, which consists of four main modules: mi-
`The signal sampling module consists of OPA circuitry con-
`crocontroller module, excitation LED module, signal sampling
`nected to the sensor array. Four API PDV-C173SM high-speed
`module, and power management module. The main printed cir-
`photodiodes are connected in parallel;
`their responsivity to
`cuit board is 41 mm by 36 mm,excluding the antenna board.
`wavelengths above 650 nm is more than 0.3 A/W. The pho-
`This hardware combines functionality from the Jennic JN5139-
`todiodes are arranged radially around the central LEDs and
`EK020 developmentkit with lessons learned from an earlier re-
`maintain a source/detector separation of 3-5 mm. The OPA
`flectance pulse oximeter design [29].
`chip contains two amplifier units. The sensor array signal is
`A microcontroller module is the prototype kernel. The
`buffered at the first unit and amplified by the second unit.
`JN5139 wireless module, designed for robust and secure
`The power management module includes two chips:(a) a Sil-
`low-power wireless applications,
`integrates a 32-bit RISC
`icon Labs CP2102 USB-to-UART bridge that powers the pulse
`processor with a 2.4 GHz IEEE 802.15.4 (ZigBee) transceiver,
`oximeter when the USB connection is detected and bridges
`192 kB of ROM, 96 kB of RAM,a mix of analog and digital
`Authorized licensed use limited to: Fish & Richardson PC. Downloaded on July 21,2023 at 03:50:33 UTC from IEEE Xplore. Restrictions apply.
`
`6
`
`
`BB Reftectance PutceOximeter[Client ID: 0015800000040273]
`File Tools Help
`/4000-
`3500 F
`/a000 F}
`\2eqo||
`2000
`
`|{
`
`‘Sexial Port:
`‘Semple Rate:
`
`COM?
`100He
`
`‘SignalOh
`rare 819 nm
`vires ct:
`[Detaut
`Remeshinote:
`[20s
`START
`SignalProcessing: ¢Oxygen Level
`ow
`a
`[ venate
`FilterOption:—Low Pass
`
`Lowes: Frequency (Hz
`
`
`UpperFreauency(Hzi Spo?t=
`
`6
`
`

`

`LI AND WARREN: WIRELESS REFLECTANCE PULSE OXIMETER
`
`275
`
`
`
`Vref ofInfrared Channel
`
`ab
`
`= 14-
`o
`
`:
`
`E41
`*
`
`05
`
`Ob
`50
`
`|
`
`|
`
`i
`
`~~
`
`AC of Infrared Channel
`L
`1
`65
`70
`
`|
`
`1
`55
`
`L
`60
`
`=Alii
`
`4
`
`!
`
`4
`75
`
`
`
`
`4
`=}_TEMENA
`
`
`
`
`
`
`
`
`
`“ve
`
`
`
`Time (sec)
`(a)
`
`
`2500
`
`Fundamental Frequency
`
`“
`
`4
`
`|
`
`2000 }
`
`@ 1500
`Ccunoc
`= 1000
`
`==‘
`
`500
`
`
`Frequency (Hz)
`(c)
`
`Oo
`
`20
`
`40
`
`60
`Frequency (Hz)
`(d)
`
`80
`
`100
`
`120
`
`Fig. 11. Fingertip results: 25 seconds of(a) near-infrared and (b) red PPG data, accompaniedby their (c) near-infrared and (d) red magnitude spectra.
`
`A. Fingertip Data
`
`data communication to the host and (b) an STMicroelectronics
`L6924D battery charger system with an integrated power
`switch for lithium-ion batteries which charges the battery when
`the USB connectionis detected. An LIR2477 3.6 V lithium-ion
`
`rechargeable button cell with a capacity of 180-200 mAh
`serves as the power source when the USB connectionis absent.
`Memory chips, indicators, and buttons are also housed on the
`board. Two Numonyx M25PX64 64-Mbit flash memory chips
`with SPI bus interfaces provide storage space when the pulse
`oximeter works in offline mode; each consumes 20 mA of cur-
`rent while being accessed.
`
`IV. RESULTS AND DISCUSSION
`
`Fig. 11 illustrates 25 seconds of representative fingertip data
`from a 24-year-old subject. Both channels of PPG data, red
`and near-infrared, are uncompensated. The AC values of the
`near-infrared channel [Fig. 11(a)] offer 1.2 V peak-to-peak (i.e.,
`2048 digitization levels), and the AC values of the red channel
`[Fig. 11(b)] offer fewer digitization levels: about half compared
`to the near-infrared channel. Even without the use of analog or
`digital filters, the signal demonstrates distinguishable period and
`amplitude information useful for HR and SpO, determination.
`The SNRsof the raw near-infrared and red PPGsare 8.0, and
`3.0, respectively. As shown in Fig. 11(c) and (d), up to seven
`harmonics reside in the spectrum of the near-infrared data (the
`inset shows frequency components above 5 Hz), and six dis-
`The pulse oximeter prototypes were used to acquire hundreds
`tinguishable harmonics reside in the spectrum ofthe red data.
`of PPG records from 48 different subjects that are 20 to 64 years
`Additionally, the PPG information and noise components (e.g.,
`old. Experimental results in this section were acquired in an
`60 Hz and 120 Hz grid noise) are clearly separated in the fre-
`indoor environment. The results are categorized according to
`quency domain. To further refine the signal, a properly designed
`conventional location (fingertip) versus other locations (wrist,
`digital band passfilter can be applied.
`earlobe, temple,etc.).
`Authorized licensed use limited to: Fish & Richardson PC. Downloaded on July 21,2023 at 03:50:33 UTC from IEEE Xplore. Restrictions apply.
`
`7
`
`7
`
`

`

`276
`
`IEEE TRANSACTIONS ON BIOMEDICAL CIRCUITS AND SYSTEMS, VOL. 6, NO. 3, JUNE 2012
`
`
`[teeaion|
`
`
`Location 1
`
`Fig. 13. Wrist PPGs corresponding to the placement locations in Fig. 14.
`
`he er
`Pipi
`Radial Artery Pa
`
`Location 3
`
`Location 2
`
`A
`
`
`
`O5}
`
`1
`i
`i
`i
`4
`133.5
`134
`134.4
`135
`135.5
`
`
`L
`6245
`
`T
`
`1
`63
`
`T
`
`i
`63.5
`T
`
`1
`64
`