0% found this document useful (0 votes)
35 views21 pages

PPG Theroy

Uploaded by

lorenzodamico163
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
35 views21 pages

PPG Theroy

Uploaded by

lorenzodamico163
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 21

Feature

©istockphoto.com/pandpstock001
Towards a Continuous
Non-Invasive Cuffless
Blood Pressure Monitoring
System Using PPG:
Systems and Circuits Review
Guoxing Wang, Mohamed Atef, and Yong Lian

such as motion artifacts (MAs), noise, low PPG signal amplitude


Abstract with large DC components, and possible solutions are discussed.
This paper presents an overview of the cuffless, continuous time, Techniques and circuits for MAs reduction, DC photocurrent can-
non-invasive blood pressure measuring devices (cNIBP) based on cellation, and sensitivity improvements are presented. To enhance
photoplethysmography (PPG). The cNIBP fundamental challenges, the BP system performance, both the optimization of the integrated
PPG receiver design and the reduction of the LED driver power
Digital Object Identifier 10.1109/MCAS.2018.2849261 consumption are introduced. Useful design considerations are pro-
Date of publication: 15 August 2018 posed for designers to improve the PPG-based BP system accuracy.

6  IEEE circuits and systems magazine 1531-636X/18©2018IEEE THIRD quarter 2018


Authorized licensed use limited to: Chulalongkorn University provided by UniNet. Downloaded on October 21,2024 at 09:03:19 UTC from IEEE Xplore. Restrictions apply.
I. Introduction The tonometry method is suitable for measuring in-

H
ypertension (or high blood pressure) is a major stantaneous BP. It applies a small force perpendicular
risk factor for dangerous health conditions such to a superficial artery [9], [10]. A pressure transducer
as heart attack, stroke, chronic heart failure, and measures the pressure value at the contact point. The
kidney disease [1]. High Blood Pressure (BP) generally external force must be small to not dramatically reduce
develops over many years without any symptoms, and the blood flow through the artery. The contact is main-
could cause damage to blood vessels and heart continu- tained, and the transducer measures the blood pres-
ously. According to WHO, worldwide high blood pressure sure wave. The tonometry method needs a continuous
causes 7.5 million deaths, about 12.8% of the total of all control for the contact pressure and the transducer,
deaths [2]. Among people with high blood pressure, only which increases the system complexity and cost. The
54% of them have their condition under control [3]. Thus, accuracy is affected by the motion artifact. The BP
high blood pressure is one of the most important chal- monitoring becomes uncomfortable as the measure-
lenges in public healthcare globally. ment time increases.
To diagnose and manage high blood pressure, it is Pulse wave velocity (PWV) is a cuffless method that
important to measure BP routinely. Although the tra- can measure beat-to-beat BP over a long period [11].
ditional clinical BP measurement provides information on PWV refers to the velocity of the pressure pulses gen-
one’s BP level, it comes with several shortfalls, e.g. incon- erated during left ventricular ejection, after opening of
venience, sparse BP readings, and white coat syndrome. the aortic valve. The device normally measures pulse
Studies show that risks associated with hypertension can transit time as the time delay between the carotid artery
be better managed through frequent BP monitoring. BP pulse wave and the femoral artery pulse wave and use
monitoring at home not only provides better prognostic that to calculate the PWV. Unfortunately, it requires a
value but also prevents from serious organ damage com- trained medical expert to locate the arteries used for
pared with clinic BP monitoring [4]. This is because in- the BP measurements. The accuracy is significantly
home BP monitoring gives more frequent BP readings than affected by the motion during BP measurement. Also,
clinic BP monitoring. The frequent measurement leads to the distance between locations to determine the arrival
more accurate BP data than sparse clinic BP measurement. time of the pulse wave must be measured precisely and
Additional benefit of home BP monitoring is that it helps to manually [12].
identify white coat and masked hypertension. Compared Ideally the best method for continuous non-invasive
with ambulatory BP monitoring, in-home BP monitoring blood pressure (cNIBP) monitoring should offer high ac-
costs less and is easy to implement. curacy, remain comfortable for long time measurements
There are two different methods for BP monitoring, (i.e. cuffless measurement), require no medical supervi-
one is cuff-based and the other is cuffless. Commercially sion, and automatically store the data for later usage. The
available BP devices based on the oscillometric method recent development in wearable wireless cuff-less contin-
use an inflatable cuff during measurement. However, the uous BP monitoring devices meet most of these require-
inflatable cuff is not feasible for long-term BP measure- ments and could become the best candidate of cNIPB.
ments due to discomfort caused by repeated inflation Due to the emerging interest in recent years in de-
and deflation. If used during sleep, it can cause sleep veloping cNIBP measuring devices, IEEE introduced a
disturbances [5], [6]. cuffless BP measurement standard, i.e. IEEE standard
Volume clamp-based BP devices can measure contin- 1708–2014, which standardizes “Wearable, Cuffless Blood
uous BP using an inflatable finger cuff with a built-in PPG Pressure Measuring Devices” [5]. Table I summarizes the
(photoplethysmography) sensor and a pressure con- recommended accuracy levels that are equivalent to the
troller unit [7]. When the volume under the finger cuff American Association for the Advancement of Medical In-
becomes constant (constant PPG signal), the BP value strumentation (AAMI) requirements [13] and the BHS pro-
can be evaluated as the pressure value in the finger cuff. tocols that are recommended for cuff-based devices [5].
However, the vessel under the finger cuff is pressed, and The manufacturer should determine the grade level of the
the oxygen level decreases in these vessels, which can Device Under Test (DUT) according to the accuracy lev-
cause tissue hypoxia during continuous measurements. els indicated in Table I. The BP mean absolute difference
Variations in finger positioning within the cuff influence (MAD) is calculated using the DUT and a reference de-
the BP measurement, reducing its precision [8]. vice for both systolic blood pressure (SBP) and diastolic

G. Wang, M. Atef, and Y. Lian are with the School of Microelectronics, Shanghai Jiao Tong University, No. 800 Dongchuan Road, Minhang District, Shang-
hai, China, e-mail: [email protected]. M. Atef is also with the Electrical Engineering Department, Assiut University, Assiut, Egypt, e-mail:moh_atef@
aun.edu.eg.

THIRD quarter 2018 IEEE circuits and systems magazine 7


Authorized licensed use limited to: Chulalongkorn University provided by UniNet. Downloaded on October 21,2024 at 09:03:19 UTC from IEEE Xplore. Restrictions apply.
blood pressure (DBP). The British Hypertension Society This paper presents the current challenges and the
(BHS) requires the accuracy for BP MAD to be less than possible solutions in the development of cNIBP devices.
±5 mmHg, with a standard deviation less than 8 mmHg. The available sub-blocks and the algorithms developed
Most of wearable BP devices under development are for cNIBP devices are discussed. Recommendations and
based on PPG or PTT (Pulse Transit Time) techniques, guidelines are provided for the improvement of each
thus require no cuff and no medical supervision. They sub-block toward a high-performance cNIBP device.
are low cost and comfortable to use. These devices could The rest of the paper is organized as follows. The PPG
pave the way for wide adoption of home BP monitoring. sensing principles of operation are introduced in Section II.
The PPG/PTT BP-based methods still have some chal- The different BP models and algorithms for calculating BP
lenges with respect to their accuracy and the need for values are presented in Section III. The PPG sensor cir-
recalibration. They cannot reach Grade A level for BP cuitries are discussed in Section IV. Commercially available
standard requirements. More hardware and software im- cNIBP devices are provided in Section V and conclusion
provements are still needed for a wearable cNIBP de- remarks are drawn in Section VI.
vice that can function in daily life. The main challenges
and issues like motion artifacts (MA), large background II. Photoplethysmography (PPG) Signal Acquisition
light, in-band noise, and low-power operation affects the Two types of cuffless BP measurement technique are
BP measurement accuracy and portability. widely used in wearable BP devices, i.e. PPT based
or PPG-only techniques. For PTT based
devices, both ECG and PPG signals are
Table I. recorded simultaneously. Rich features
MAD accuracy level with comparisons to the ANSI/AAMI SP10 can be extracted from the combination
and BHS evaluation systems [5].
of ECG and PPG signals, which increases
ANSI/AAMI Recommended BP model accuracy. More details about
MAD(mmHg) SP10 BHS Grading BP modeling using PTT will be covered in
≤4 pass Grade A A Section III A.
4–5 pass mostly in Grade A, A Figure 1 shows the block diagram of
less in Grade B a typical cNIBP system. This hardware
5–6 pass or fail mostly in Grade B unit acquires the ECG and PPG signals of
B, less in Grade A, the patient via a PPG sensor and an ECG
extremely less in sensor [14]. The MCU processes the input
Grade C and Grade D
signals to calculate the BP. During the cali-
6–7 mostly fail, mostly in Grade C, C bration phase, the MCU acquires an addi-
less pass less in Grade B and
tional signal from a reference commercial
Grade D
blood pressure monitoring device. There
≥7 fail worse than Grade C D
is a communication channel between the

Power Manager

Synchronized
Analog ECG and PPG Waves’s Peaks
Signal Detection/PTT
PPG Bluetooth Calculation/
Sensor Blood Pressure
Control Wireless
Module Transmission Display
Analog Digital
Input (MCU) Output
Analog
Module Module
Signal Serial Port
ECG Initialization
Sensor and Debugging

Figure 1. Block diagram of cNIBP system [14].

8  IEEE circuits and systems magazine THIRD quarter 2018


Authorized licensed use limited to: Chulalongkorn University provided by UniNet. Downloaded on October 21,2024 at 09:03:19 UTC from IEEE Xplore. Restrictions apply.
cNIBP and PC/smartphone via serial port/wireless so tionally, fingertip sensors are affected by daily activities.
that the ECG and PPG signals and BP values from the Thus, the reflectance signal from the wrist or forehead
user can be collected by the phone/PC for monitoring would be a better solution for wearable devices. How-
and storage. The smartphone can transmit the collected ever, the signal received from reflectance mode is easily
data and report it to a medical care unit. affected by MAs and pressure disturbances, which limit
A PPG sensor is required to sense the light signal the measurement accuracy [17], [18].
modulated by the BP wave traveling through the arter-
ies. The PPG sensor consists of two components, a light
emitted diode (LED) as a light source and an optical LED
receiver. There are two modes for PPG sensing, trans-
mission mode and reflectance mode as shown in Fig. 2.
Light passing though human tissue can be absorbed
by skin, bone, and arterial and venous blood, as shown
in Fig. 3. The optical receiver detect the changes in the PD LED PD
reflected light intensity due to the change in the blood (a) (b)
flow volume. The incident light is amplitude-modulat-
Figure 2. Position of light-emitting diode and photodetector
ed by the changes in blood volume during the cardiac in different photoplethysmography modes: (a) transmission
cycle, with maximum blood volume occurring in the mode and (b) reflectance mode.
systolic phase and a minimum blood volume at the dia-
stolic phase. The systolic peak (primary peak) is com-
ing from the direct pressure wave traveling from the
left ventricle to the periphery of the body, the diastolic
Light Source
peak (secondary peak) is the effect of the pressure
wave reflections by arteries of the lower body [15], as Other Tissues
shown in Fig. 4. The received optical signal has both AC Venous Blood
and DC components. The AC component has an ampli- Non-Pulsatile
tude that depends on the amount of light absorbed by Component of
Artery Blood
the change in the blood volume and has a fundamental
frequency related to the heart rate. The other detected
optical signal component has a constant value (DC) de- Systolic Diastolic Time
Phase Phase
pending on the tissue structure thickness, skin color,
and the average blood volume. This fixed tissue struc- ac: Pulsatile
Waveform

ture absorbs a constant amount of light and does not Part


PPG

change with time [16]. dc: Steady


Part

A. The Selection of PPG Sensing Mode Time


As mentioned earlier, there are two modes for PPG mea-
surement. The transmission mode requires a higher LED Figure 3. DC and AC components of the PPG signal due to
variation in light absorption from blood volume and tissue [17].
light intensity and works well for thin body parts, where-
as reflectance mode shows a higher detected signal level
compared to transmission mode at low transmitted light
intensity in thicker body parts. In general, reflectance
100
mode can give the same signal quality as transmission
Tp
mode using lower LED power [18]. As the probability is 80 Primary Peak
much higher for weak light photons to be reflected than
Amplitude

60 Secondary Peak
transmitted through thick tissue, reflectance mode can
First
be used in any part of the human body, while transmis- 40 Pulse
sion mode is limited to peripheral parts such as the fin- Foot
20
gertip, foot, and earlobe. For example, PPG signals from Dicrotic Notch
the front of the head and chest can only be detected in 0
0 0.2 0.4 0.6 0.8 1 1.2 1.4
reflectance mode. It is worth noting that fingertips and Time/s
earlobes have limited blood perfusion such that these
Figure 4. PPG waveform with basic BP features indicated.
sites are susceptible to low ambient temperatures; addi-

THIRD quarter 2018 IEEE circuits and systems magazine 9


Authorized licensed use limited to: Chulalongkorn University provided by UniNet. Downloaded on October 21,2024 at 09:03:19 UTC from IEEE Xplore. Restrictions apply.
To minimize the loss in thick tissues, the tissue thickness
at the PPG sensor position should not be greater than
the penetration depth of the light wavelength.

B. Light Wavelength PD and the LED are, the better shallow surface light can
The light penetration depth is inversely related to its ab- be detected. The practical distance between the PD and
sorption coefficient (a) . The greater the absorption, the LED is from 4 to 6 mm [18]. Green light has better immuni-
smaller the light penetration depth in tissue. To minimize ty to MAs than red and IR lights, as the IR light penetrates
the light loss in thick tissues, the tissue thickness at the more deeply and is affected by all surface and deep tissue
PPG sensor position should not be greater than the pen- movements [20]. The PPG signal from the green light is re-
etration depth of the used light wavelength. For visible duced sharply by increasing contact pressure compared
light, the penetration depth through tissue under the skin with the IR PPG signal. This is an indication of the occlu-
surface ranges from 100 nm to 1200 nm [19]. A reflectance sions of superficial skin blood vessels. The PPG second-
PPG sensor located in thin tissue regions like fingertips, ary peak and dicrotic notch are more structured with red
earlobe, and the forehead results in stronger signals than than with green light (Fig. 4) [19].
in thicker tissue regions such as the chest. Red and IR
light can penetrate deeper into the tissue than green light C. PPG Sensor Position
wavelengths can; these wavelengths can thus give better There are different possible sites for PPG sensor place-
results for deep-tissue blood flow in the muscles. In ad- ment, such as the fingertip, forehead, earlobe, chin, and
dition, the photodetector responsivity of the red and IR wrist. Fingertips give the strongest PPG signal, the fore-
wavelengths is greater than that of the green light. Green head and earlobe have good signal strength, and the
light, on the other hand, is absorbed near the surface, wrist has a low signal amplitude. Although the fingertip
so it is suitable for the shallow blood flow directly under produces the strongest PPG signal, the MAs associated
the skin. Short wavelengths like blue and green (405 to with routine activities produce a great deal of noise,
520 nm) have stronger blood absorption than red and IR which reduces the signal quality. Wrist and earlobe loca-
light (650 to 850 nm). Therefore, the signal-to-noise ratio tions also suffer from MAs due to hand and neck move-
(SNR) for a green light source is better than that of an IR ments, respectively. The forehead has a flat and large
source when used for superficial blood flow in the skin. region for sensor placement, easy attachment by using a
The received signal at a certain light penetration depth head bandage, and is a stable region with low motion ar-
can be controlled by adjusting the distance between the tifact effects. It also produces a strong PPG signal. How-
LED and the photodiode. By increasing the distance be- ever, as forehead PPG sensors work only in reflectance
tween the photodiode (PD) and the LED, the PD will de- mode, dirt and sweat may affect measurement accuracy.
tect the light rays with deeper penetration. The closer the The chin region has a fair PPG signal, but it suffers from
growing hairs, dirt and sweat [18], [21], [22]. The work in
[18] studied the different locations with respect to PPG
signal strength, attachment, artifact, and portability.
Circuit
Board (CPU, Outer Ring They introduced a decision matrix to determine the pre-
RF Transmitter, (Housing) ferred sensor locations, and the forehead received the
etc.) highest score.

D. PPG Sensor Structures


A PPG sensor position and structure should be comfort-
able enough for use during daily activities. Fingers can
be easily accessed, and they provide a strong PPG sig-
nal. One of the finger-based structures is a ring sensor,
Flexible Wires
Connecting the Inner Ring which can be attached to the finger base, as shown in
Inner Ring and (With Optical Fig. 5 [23]. To reduce the MAs, a double ring structure
the Outer Ring Sensor Unit)
was developed to reduce the effect of outside forces.
A wristwatch-type PPG sensor was introduced in
Figure 5. PPG ring sensor [23].
[24]. This design collects the PPG signal from both the

10  IEEE circuits and systems magazine THIRD quarter 2018


Authorized licensed use limited to: Chulalongkorn University provided by UniNet. Downloaded on October 21,2024 at 09:03:19 UTC from IEEE Xplore. Restrictions apply.
Main Device Watch Strap
Main Device
Watch Strap
(b) Top View

Photo Transistor Watch Strap


IR Diode
(Conductive
Sensor Fabric)
Holder
(a) Side View (c) Bottom View

Figure 6. The wristwatch-type PPG array sensor module [24].

wrist radial and ulnar arteries. The wristwatch is com- III. Blood Pressure Modeling and Algorithms
fortable to wear, but the PPG signal coming from the There are several methods to estimate the BP depending
wrist area is weak. An array of phototransistors and on the sensing and calculation techniques. One method
LEDs were placed to improve the PPG sensor sensitiv- is to calculate BP based on the PTT. The Another meth-
ity and its accuracy. A conductive fiber wristband was od uses PPG signal only to calculate the BP in order to
employed to reduce external noise. An armband-based reduce the hardware complexity.
wearable monitoring device was developed in [25] us- There are two main directions for BP estimation
ing a near-infrared LED and a phototransistor in reflec- using PPG only. The first one involves the parametric
tive mode. models, which try to extract specific parameters such
A Nellcor forehead reflectance-mode sensor [26] se- as instantaneous heart rate, systolic, and diastolic pe-
cures the sensor by means of a headband. Integrating riods from individual PPGs. Then, the BP can be pre-
the PPG sensor into preexisting equipment like a sol- dicted using these extracted parameters and a para-
dier’s helmet provides a comfortable and acceptable at- metric model (Section III B). The second direction is
tachment for the user [18]. the non-parametric models. Non-parametric BP meth-
In [22], an ear-worn reflective PPG sensor with im- ods try to extract specific features (time domain or
proved surface coupling was proposed. The sensor has frequency domain features) with strong relations to BP
been proven to be robust against pressure-induced sen- from the PPG waveform.
sitivity degradation [22]. A wireless earpiece PPG sensor The main steps to calculate the BP from PPG signals
with a magnetic earring was introduced in [27]. The sensor are listed in [32]; some steps can be merged from differ-
can be comfortably worn on the earlobe, and it contains an ent algorithms into the following steps:
embedded accelerometer to provide MA noise reference 1) Preprocessing: deals with noises caused by power
for adaptive noise cancellation. However, the earlobe be- line interface, baseline wandering [33], motion arti-
comes susceptible to inaccuracies due to peripheral vaso- facts [34]–[37], and signal quality assessment [38],
constriction and perfusion. To overcome these limitations, [39]. Classification is performed to discriminate be-
the external auditory canal has been proposed as an alter- tween motion artifacts and clean PPG. A band pass
native monitoring site, on the hypothesis that this central filter is used to remove the high-frequency noise
site will be better perfused [28]. above 5 Hz and reject the DC and low-frequency
For low cost and high integration ability, blood pres- noise component below 0.5 Hz.
sure can be estimated using a smartphone camera.
Initial BP screening with a low-cost solution using a
simple and portable device is the aim for such smart- Smartphone
phone health-care applications [29]–[31]. The PPG sig- LED PPG ANN
nal is processed by an artificial neural network (ANN)
built in a smartphone to evaluate the BP, as shown in Display
Fig.7 [31]. The on-line evaluation of the experimental Camera Acquired
Frames Signal
BP results of the proposed method shows a maximum
Processing
error of 12 mmHg [31]. This reported BP determina-
tion accuracy using a smartphone is still worse than
Figure 7. Capturing and processing PPG data using smart-
the 8 mmHg required for it to be used as a commercial
phone [31].
healthcare device.

THIRD quarter 2018 IEEE circuits and systems magazine 11


Authorized licensed use limited to: Chulalongkorn University provided by UniNet. Downloaded on October 21,2024 at 09:03:19 UTC from IEEE Xplore. Restrictions apply.
2) Feature Selection: selects a set of features that are sion [14], machine learning [43], [44], the Windkessel
the most relevant and least redundant for BP, to model [45], or the BP transport model [46], etc.
enhance the performance of the BP classification
model and reduce the algorithm complexity by se- A. PTT-based Blood Pressure Models
lecting a smaller number of efficient features [40]. Pulse Transit Time (PTT) is the time it takes the Pulse
3) Feature Extraction: extracts signal features from Pressure waveform to propagate through a certain length
time and frequency domains [30], [41], [42]. Time of the arterial tree. The PAT is the time delay that takes
domain characteristics can be defined as (a) pulse for the pulse wave to travel along the artery and arrive
amplitude; (b) pulse length (ON time); (c) pulse at the periphery in the same cardiac cycle. The PAT can
rate (peak-to-peak time); (d) valley depth differ- be measured as the time delay between the R-wave peak
ence (voltage difference between two minimums); of the ECG signal and a particular point of the PPG signal
and (e) peak height difference (between two maxi- (foot PATf, peak PATp or maximum slope point PATs) as
mums) [32], as shown in Fig. 10 and Fig 8. Frequen- shown in Fig. 9 [47]. The PAT is equal to the summation of
cy domain characteristics of the PPG signal can the PTT and the pre-ejection period (PEP). The PEP is the
be extracted by applying the Discrete-time Short time to open the aortic valve to start pushing blood out
Time Fourier Transform (STFT). of the ventricle. As the PEP is difficult to measure from
4) Regression: estimates the blood pressure with a the PPG signal, the PAT is acceptably approximated to be
proper model, such as parametric PTT-based regres- measured as the PTT. Both PPG and ECG waveforms are
required for a PTT-based BP device.
PWV has been found to be an important parameter in
the analysis of arterial blood pressure. The propagation
1.2 Ts Td
of blood in the artery is very similar to that of the propa-
Approximate Interior Angle
1 gation of a compressible fluid. The relationship between
Normalized Amplitude

Td
T 90 of Main Peak
vessel elasticity and the wave speed of a pressure pulse
0.8 Maximum First Derivative: in a thin-walled vessel can be described by the Moens-
Tsmaxdrv
T maxdrv1
Ts 1
Kortweg equation ([48]):
0.6 Valley of
Dicrotic Notch L = hE 0 exp (gP)
0.4 Peak of PWV = , (1)
Dicrotic Notch PTT tR
Tdn Valley
0.2
Tdn Peak
where PWV is the pulse wave velocity, L is the length of
0 the vessel, PTT is the pulse transit time, h is the thick-
0 0.2 0.4 0.6 0.8 1 ness of the vessel wall, R is the radius of the vessel, t
Time (s)
is the blood density, E 0 is the zero-pressure modulus in
Figure 8. PPG waveform with important BP features indicated. mmHg, and g is a constant that depends on the particu-
lar vessel (typically 0.016 mmHg - 1 to 0.018 mmHg - 1 ).
Under the assumption that the overall percentage
change in the arterial wall thickness (h) and arterial ra-
dius (R) are small over the physiological pressure range,
R
equation (1) can be simplified to relate BP to PTT [49], i.e.:

ECG 2
2RtL
P = 1 c ln c m - 2 ln (PTT) m = k 1 ln (PTT) + k 2, (2)
p hE 0

where k 1 and k 2 are patient-dependent coefficients.


According to the works of [50], within a small pres-
sure range, the ln() function can be linearized, and the
PPG BP equation becomes the following:
2
2RtL A
PATf P = 1 ec m 1 - 1o =
hE 0 c PTT 2 m
, (3)
PATs
p PTT 2 + B
PATp
where A and B are patient-dependent constants.
By using series expansion, the ln(PTT) function in
Figure 9. Pulse arrival time measured from the R-wave ECG equation (2) can be approximated by the linear model in
signal and a certain point in the PPG signal [47].
equation (4), which has also had been proved e
­ mpirically

12  IEEE circuits and systems magazine THIRD quarter 2018


Authorized licensed use limited to: Chulalongkorn University provided by UniNet. Downloaded on October 21,2024 at 09:03:19 UTC from IEEE Xplore. Restrictions apply.
[51]. Because of its simplicity and low computational ear regression can simulate the arterial blood pressure
complexity, it has become a widely used model [52], [53]. equations (7) and (8) [45].
P = a· PTT + b, (4) I 0 Ts CrR 2 ^
1 + e - RC h, (7)
Ts Ts
Ps = Pt | t = Ts = Pts e - RC +
It was observed in [47] that BP is highly correlated T 2s + C 2 r 2 R 2
Td
not only with PTT but also with heart rate (HR); hence, Pd = P ^t t = Td h = Ptd e - RC , (8)
they proposed the following model:
where Pts and Ptd are the initial values of systolic Ps and
P = a 1 PTT + b 1 HR + c 1, (5)
diastolic Pd blood pressure, respectively, and I 0 is the
A non-linear approach is proposed by [54], [55]. It is maximum value of the blood pumped by the heart dur-
based on the non-linear relationship between the pres- ing systole. Similarly, R and C can also be expressed as
sure and the vascular volume. a function of Ps, Pd, Ts, Td and I 0 .
Shown in (6) at the bottom of the page, where L is the The parametric models can achieve good prediction
distance traveled by the pulse, b is a patient-dependent results an individual [45], [46], [57], but the accuracy de-
coefficient, and P hydro = tgh. creases over time. Moreover, these over-simplified para-
There are numerous models proposed by the research- metric models require initial calibration and frequent
ers. However, the accuracy of these models very much de- recalibration for each person.
pends on how accurately the unknown model parameters The second direction is the non-parametric mod-
can be estimated. Most of these unknown parameters are els which try to extract specific features, with strong
patient dependent. Hence, the model should be tailored relations to BP, from the PPG waveform, refer to Fig. 8
for each patient in order to monitor his or her blood pres- and Table II. Then, the BP can be predicted using these
sure continuously and accurately. In addition, for a given
patient, these model parameters may change with time,
Table II.
and thus, subsequent calibration may be needed to main-
Important PPG BP features indicated in Fig. 8.
tain accuracy over time [56].
Features Definition
B. PPG-only Blood Pressure Models f1 Ts systolic time
The PPG-only models reduce the hardware complexity for f2 Td diastolic time
cuff-less BP estimation. There are two main directions in us-
f3 Tdn Peak time distance from onset to Dicrotic
ing PPG-only approach. The first one involves the paramet- Notch
ric models, which try to extract specific parameters such
f4 Td90 the time distance from onset to the point
as instantaneous heart rate, systolic, and diastolic periods
with 90% of the maximum height within
from individual PPGs. Then, the BP can be predicted using diastolic time(DN)
these extracted parameters and a parametric model. Some
f5 Ts max drv1 time distance from onset to the point
examples of the parametric methods are the Windkessel
with maximum first derivative within Ts
model [45], regression of long-term and short-term features
f6 amplitude at valley of DN
[57], and the pulse transport theory-based model [46].
Instead of directly relating systolic and diastolic BP f7 amplitude at peak of DN
values with PPG features, the Windkessel model first f8 approximate interior angle of main peak
maps PPG features to specific body parameters and f9 mean value of whole pulse
then derives BP values from these parameters [45]. The f10 mean value of pulse within Ts
2-element Windkessel model estimates the total periph-
f11 mean value of pulse before valley of DN
eral resistance (R) and arterial capacitance(C ) of the
body using PPG features. Then, a BP model using lin- f12 mean value of pulse after valley of DN

Z
] L tb
] 1 + exp (bPint) , h = 0
]
]
P = [ 2L · ln  (6)
] gh tb
] exp [b (Pint - Phydro)] + exp (- bPhydro) - exp (- bPhydro)
]] ,h ! 0
exp [b (Pint - Phydro)] + 1 - 1
\

THIRD quarter 2018 IEEE circuits and systems magazine 13


Authorized licensed use limited to: Chulalongkorn University provided by UniNet. Downloaded on October 21,2024 at 09:03:19 UTC from IEEE Xplore. Restrictions apply.
Table III.
Comparison of different PPG-based BP models and algorithms.
Algorithms Dataset Calibration Accuracy(mmHg)
[58] Non-linear Regression, 5 Case A certain time for ME: SBP(5.4%), DBP(7.7%)
(ECG + PPG) each person
[59] Linear Regression, 10 subjects Every 3.3 hours STD: SBP(1.69), DBP(1.04)
(ECG + PPG), adaptive
Kalman filter
[37] ACF, linear regression, MIMIC, 100 A certain time for ME ± STD: ABP(3.06 ± 3.69)
(ECG + PPG) Cases each person
[34] Iterative linear, (ECG + PPG) MIMIC, 6 A certain time for ME ± STD: SBP(0.13 ± 3.73),
Cases each person DBP(0.22 ± 4.33 mmHg)
[25] Moens-Korteweg, 10 Cases Once for every 12 ME ± STD: SBP(2.4 ± 5.7)
(ECG + PPG) hours
[41] SVR, (PPG) MIMIC Never MAD ± STD: SBP(12.38 ± 16.17),
MAP(7.52 ± 9.54 mmHg),
DBP(6.34 ± 8.45 mmHg)
[42] ANN (PPG) MIMIC, 5000 Never MAD ± STD: SBP(3.24 ± 4.75 mmHg),
Cases MAP(2.16 ± 3.14), DBP(1.79 ± 2.70)
[30] SVM, (PPG) UQ, 32 Cases Never Classification(20 interval): SBP(98,81
%), DBP(98.21 %)
[43] Random Forest, (PPG) 410 Cases Never MAD (#5,10,15): SBP(62.43%, 86.34%,
90.37%), DBP(79.02%, 90.24%, 93.90%)
[60] Linear regression, (PPG) 18 subjects A certain time for MAD ± STD: SBP(sleeping: 2.68 ± 0.39
one person mmHg, sleepy: 4.37 ± 1.20
[45] Windkessel model, (PPG) UQ, 32 Cases Once for all MAD ± STD: SBP(0.78 ± 13.1 mmHg),
DBP(0.59 ± 10.2)
[46] Harmonic Balancing model, 5 Cases A certain time for MAD ± STD: SBP(1.37 ± 7.61 mmHg),
(PPG) one person DBP(−1.40 ± 6.00)

port vector machine (SVM). The estimation accuracy can


1 reach 98.81% and 98.21% for SBP and DBP, respectively.
0.9 Pulse Pulse Height
Pulse Rate Reference [41] gets its features from both the PPG and ECG
0.8 Amplitude Difference
0.7 signals. Using support vector regression (SVR), it reaches
Amplitude

0.6 grade B for both DBP and mean blood pressure (MBP)
0.5 estimations and reaches grade C for SBP according to the
0.4 BHS protocol. In [42], an ANN-based learning process is
0.3
0.2 Trought Depth Difference used to train the features from the PPG signal only. The
Pulse Length
0.1 results can fulfill the AAMI requirement. Using only the
0 2 4 6 8 10 12 14
PPG signal, [43] implements a random forest (RF)-based
Time/s
prediction model and achieves grade B for both SBP and
Clean PPG Artifact
DBP by BHS. Even though the non-parametric BP models
exhibit better accuracy than the parametric models, they
Figure 10. Time domain features of the PPG signal and mo-
suffer from heavy calculations.
tion artifact effects [32].
Table III compares the different PPG-based BP models
and algorithms. The first five methods in Table III [25],
­ xtracted features. Non-parametric models try to solve
e [34], [37], [58], [59] use both ECG and PPG signals with lin-
the low accuracy of parametric models. ear or nonlinear regression. These methods have higher
Examples of non-parametric learning methods can be accuracy, but they require both ECG and PPG signals and
found in [30], [41]–[43]. Reference [30] tries to classify the frequent calibration. The methods in [30], [41]–[43] do
SBP and DBP within different prediction ranges with 19 not need frequent calibration, and only PPG information
features extracted from the PPG signal using only a sup- is needed. However, they require excessive training and

14  IEEE circuits and systems magazine THIRD quarter 2018


Authorized licensed use limited to: Chulalongkorn University provided by UniNet. Downloaded on October 21,2024 at 09:03:19 UTC from IEEE Xplore. Restrictions apply.
calculations for the SVM, SVR, and
ANN algorithms and show higher Adaptive Noise Cancellation
error than the other methods us- Signal Source

"
ing both ECG and PPG signals. The x PPG y=x+m + X
(Body Volume
last three methods [45], [46], [60] Changes) Sensor Restored
m – Signal
are only based on the PPG signal. Motion

"
Artiface m
They use linear regression and
Estimated
non-parametric models that do Motion
not need as much training and Artifact
calculation as the SVM methods. Noise Source Accelerometer a Adaptive
(Body Motion) (Noise Reference) Filter
However, they do require calibra-
tion and have the highest error.

C. Motion Artifact (MA)


Cancellation Figure 11. Block diagram of adaptive noise cancellation (ANC) to remove MA. The
MAs introduce a disturbance to the noise reference is generated by an accelerometer to sense the movement [27].
PPG signal. One of the examples is
as shown in Fig. 10. The time do-
main parameters like pulse amplitude, pulse length, pulse
rate, valley depth difference, and peak height difference, LED
PD
will vary too much at strong MA events. A wearable BP de- Blood
od
vice must be able to reduce/cancel artifacts generated by Vessel
sel
Motion Artifact
movements. The high-frequency MA components can M
Motion Artifact
Sensor
be canceled by a 5 Hz LPF, but the in-band MA noise can- Signal
LED u (n
(n)
not be removed by a simple filter. BVP
Advanced signal processing techniques are required to Normal PPG d (n) Component
Sensor
deal with MA disturbances. If the motion-induced noise sig- PD
nal is known, the desired noise-free signal can be obtained
Skin Tissue
by subtracting the MA noise from the noisy output of the
PPG sensor, as shown in Fig. 11. The adaptive filters require
Figure 12. MA cancellation using a secondary LED/PD to
a MA reference noise signal that has a strong correlation to generate the MA noise [63].
the MA and is uncorrelated with the MA-free PPG signal. A
synthetic MA noise reference signal can be generated from
the MA-corrupted PPG signal [61]. Reference MA noise strong voltage. The main requirements of a PPG pream-
signals can be obtained from an additional transducer at- plifier are high sensitivity, high gain, wide dynamic range,
tached to sense the movement (accelerometer) [27], [62] or and the ability to reject ambient light (DC photocurrent).
by employing an additional reflectance-type optoelectronic A low power consumption is essential to prolong the bat-
sensor, as shown in Fig. 12 [63]. Fig. 12 shows the principle of tery life for portability. As the transmitted optical power is
MA cancellation using a secondary LED/PD to generate the limited by safety regulations, the photodiode current can
MA noise. The main sensor LED/PD contacts the skin to de- be as small as a few tens of nano-amperes. A high sensi-
tect the blood volume pulse (BVP) component (inevitably tivity (low noise) optical receiver is required to deal with
MA affected). The secondary LED/PD does not contact the this weak light signal. Depending on the relative positions
skin (7.5 mm air gap) to detect only MA [63]. Then, the MA of the receiver and the LED and the sensing location, the
is canceled by a non-linear adaptive canceler filter using a signal strength can vary dramatically from a few nA to sev-
recursive least squares (RLS) algorithm. eral n A. A wide dynamic range is a key parameter in order
to accommodate the different signal levels. The light
IV. PPG Sensing Circuitries penetrates through the human tissues and is absorbed
The PPG receiver receives the weak optical signal, con- by the blood, and the light is then reflected after it has
verts it, and amplifies it to a strong electrical signal. The been modulated by the blood volume change. The modu-
optical signal is converted to electrical current by means lation depth is very small, and the AC to DC ratio can vary
of a large-area photodiode (a few mm2) to collect most of from 0.1% to 3% depending on skin color, thickness, sens-
the PPG optical signal. A low-noise, high-gain preampli- ing position and the light wavelength. A DC photocurrent
fier follows the PD to amplify the weak photocurrent to a rejection is needed to prevent PPG receiver saturation.

THIRD quarter 2018 IEEE circuits and systems magazine 15


Authorized licensed use limited to: Chulalongkorn University provided by UniNet. Downloaded on October 21,2024 at 09:03:19 UTC from IEEE Xplore. Restrictions apply.
A. Transimpedance Amplifier (TIA) a second technique using signal level compression. An
The TIA is the most important block in the PPG receiver AGC is needed to improve the TIA linearity; for large in-
as it determines the receiver sensitivity and bandwidth. put signals, the gain is reduced to keep the TIA working in
The major limitation of the TIA is the high parasitic capac- the linear region. For a shunt-feedback TIA, the dynamic
itance associated with the large-area photodiode, which range can be extended by using a variable resistor chang-
deteriorates the gain, bandwidth, power consumption, ing with the input optical signal power, shown in Fig. 13.
and noise performance of the preamplifier. Shunt-feed- The variable feedback resistor can be implemented with
back TIAs have succeeded in achieving high bandwidth a MOSFET operating in the linear region, connected in
and good sensitivity. However, they suffer from trade-offs parallel with a fixed r­ esistor to improve the linearity and
between bandwidth, transimpedance gain, power, and to limit the maximum resistance (Figure 13). The AGC
sensitivity. Open-loop TIAs input stages such as common output Vc is a DC signal with a value depending on the in-
gate (CG) [64] and regulated cascode (RGC) show low put signal strength. The AGC can be implemented with a
­input impedance [65], which can reduce the effect of the circuit that determines the output signal strength (peak
high input capacitance. However, the noise performance detector or average power) to control the gate voltage of
of these topologies is worse than that of feedback TIAs. the MOSFET (a low-pass filter can perform the averaging
Several techniques have been demonstrated in [66]–[68] process). The replica TIA generates a reference voltage
to relax the effect of the large photodiode capacitance on for the difference amplifier. The difference amplifier sub-
the TIA performance. tracts the reference voltage from the averaged signal to
The dynamic range of a TIA is defined by its overload give the control signal Vc [69]. The stability of the AGC-
current and its sensitivity. A TIA with a fixed transimped- TIA is an important issue. When the feedback resistance
ance gain is not sufficient for the wide dynamic range (R F ) varies while keeping the open-loop gain fixed, both
required. There are two techniques to extend the TIA dy- the bandwidth and the quality factor will change. If R F
namic range, including automatic gain control (AGC) and is reduced, the frequency of the first open-loop pole at
the input 1/ (R fb .C T ) increases, and it will come closer to
the second pole at the output l/ ^ R out .C out h, which may
lead to peaking at a given fixed loop gain. The bandwidth
RF of the AGC-TIA tends to increase with the magnitude of
Repllca TIA

Difference the input signal. The low cutoff frequency is determined



Amplifier + by the integration time of the integrator. The low cutoff
frequency should be as low as 0.5 Hz not to lose any of
+
– the PPG signal features. So, the capacitor C AGC and the
VC
RAGC
CAGC

VPD VDD
+

MAGC
AGC
RF M2
Low-Pass Filter Vbias2 M3
– Peak
+ AGC Circuitry M7 Detector
Vout

Figure 13. Transimpedance amplifier with automatic gain Vout


M6
control [69]. Iin M1 M9

Vbias3 Vbias1
M8
TIAF M4 M5
SI
STIA
Cint
Iph Rf M10
– VTIA Rint
A – Log Amp
+0 Vint Vcnt
Cd A1
OTAF +
Vph
REF Vref ≈ avg (VTIA)
Figure 15. Logarithmic TIA with ambient light rejection and
Figure 14. TIA and variable-gain switched-integrator (SI) [70]. AGC loops [67].

16  IEEE circuits and systems magazine THIRD quarter 2018


Authorized licensed use limited to: Chulalongkorn University provided by UniNet. Downloaded on October 21,2024 at 09:03:19 UTC from IEEE Xplore. Restrictions apply.
resistor R AGC should have a high
value, which could be limited by Logarithmic Digital-to-Resistance Converter (DRC)
the chip area.
Binary to Digital
The receiver in Fig. 14 is an ex- ∆–Σ
Thermometer
ample of using AGC. The first block INDRC[k]
is a conventional TIA with 500 KΩ RDRC(t ) CLK(32 kHz)
Three-Stage LIQAF
resistive feedback [70] followed by

Combiner
Register
Accumulator OUT [k]
a voltage mode switched-integrator Vd (t ) LCD
Iphoto(t ) Kaccum
(SI). This block serves multiple pur­­ 45 –1
1–z
poses, acting as a variable-gain am- Rpar
plification stage that enables imple-
CLK(32 kHz)
mentation of large on-chip gains Off-Chip
exceeding 50 MΩ, which can be eas- Photodiode
ily tuned via the digitally controlled LIQAF1 LIQAF2 LIQAF3
integration time.
A logarithmic TIA depicted in Figure 16. Light-to-Digital Converter architecture [71].
Fig. 15 uses compression tech-
niques to reduce the signal level at
high input current [67]. Additionally, an AGC loop is intro- rejection is needed to prevent PPG receiver saturation.
duced to further extend the dynamic range. M1 acts as the An optical preamplifier with ambient photocurrent re-
logarithmic amplifier transistor. M6, M7 and M9 provide jection is shown in Fig. 17 [72]. The transimpedance
the feedback resistors, their resistances are controlled amplifier converts the current from the photodiode into
by the AGC signal. The transimpedance gain is specified a differential voltage. Surrounding the amplifier is an
by the current mirror ratio ^ g m5 /g m4 h and the feedback re- outer feedback loop comprising an error amplifier and
sistance. M9 acts as a feedback transistor placed directly a transistor. The rejection circuit operates as follows:
across the input and output terminals of the current mir- The DC component of the photocurrent produces an
ror (M4, M5). Using this direct feedback topology, M9 re- offset in the average levels of the differential outputs.
duces the input impedance seen at the photodiode, im- This offset is integrated over time by the error amplifier.
proving TIA speed by the cost of the lower output swing. Transistor M ctl acts as a variable current sink that, at
In [71] the traditional TIA is replaced by a digital-to- steady-state, eliminates the average photocurrent from
resistance converter (DRC) consisting of a logarithmic the signal path. The average photocurrent consists of
resistor network and a digital multi-bit RD modulator as the ambient photocurrent and the DC component of the
shown in Fig. 16. A parallel resistor, R par, improves sen- signal. The transistor M 10 in Fig. 15 [67] has the same
sitivity and dynamic range by extending the minimum function as M ctl . For the pulsed PPG signal used to re-
value of I bias that is supported. The combination of the duce sensor power consumption, the traditional method
DRC and R par enable continuous operation of the Light-
to-Digital Converter (LDC) over approximately a three-
order-of-magnitude range of I bias . The DRC is controlled
Vref
through feedback, which consists of a digital accumula-
tor and a nonuniform quantizer based on a laddered in-
verter quantizer /amplifier /filter (LIQAF) [71]. The vari- Transimpedance
Cd ′ Rf Preamplifier
able resistance R DRC (t) is achieved with a set of PMOS
switches connected to taps of a polysilicon resistor +
+ –
string ladder, with the PMOS switches being controlled is + Idc is Vout
by the output of a RD digital modulator that provides – +
Idc 0
A
high-resolution resistance (32 steps from 64 K to 10 MΩ)
Rf –
control through dithering. In contrast to the previous
Mctl
AGC approaches, the LDC system offers a high digital in- +
Error
tegration that minimizes analog complexity.
Amplifier

B. DC Photocurrent Rejection Circuits


The PPG signal has a large DC component, which can Figure 17. Differential output TIA with DC photocurrent rejec-
tion loop [72].
reach 99% of the total signal value. A DC photocurrent

THIRD quarter 2018 IEEE circuits and systems magazine 17


Authorized licensed use limited to: Chulalongkorn University provided by UniNet. Downloaded on October 21,2024 at 09:03:19 UTC from IEEE Xplore. Restrictions apply.
Using a differential signal path helps maximize
the preamplifier immunity to noise from
the power supply and substrate.

[68], [72] is not accurate. A DC photocurrent rejection as adding an on-chip tuning capacitor may be required
TIA with sample-and-hold (S/H) in the feedback loop to improve matching.
was introduced in [73]. The additional S/H block helps A block diagram of the implemented PPG SoC system
to detect the pulsed signal average accurately. Using a introduced in [74] is shown in Fig. 18. It consists of an in-
differential signal path helps maximize the preamplifier tegrate-and-hold amplifier, a dynamic range enhancement
immunity to noise from the power supply and substrate. (DRE) current source, a 9-bit successive approximation
The photodiode, however, is connected to only one ter- (SAR) ADC, and a digital core to supply feedback and con-
minal. This creates an asymmetry at the input of the dif- trol signals to the analog blocks (not shown for simplic-
ferential structure. As a result, an additional capacitor is ity), such as the sampling switch at the negative terminal
required at the other input of the transimpedance ampli- of the op-amp and the reset switch across the CF [74].
fier in order to match the photodiode capacitance and to The DRE circuit generates a current (Ist), which subtracts
rebalance the differential circuit. Perfect matching en- the static component of the photocurrent (Iph), and the
sures that noise injected at the bias voltage appears as residue (Iph-Ist) is integrated by the integrate-and-hold
a common-mode signal that is effectively rejected by the amplifier. Clean PPG data can be accessed from the SoC
differential structure. In practice, however, some mis- by using a standard SPI protocol, allowing for easy inter-
match can be expected, and additional measures such facing with low-power microcontrollers. However, this
DRE technique is very complex and requires more area
for the digital part compared to the analog solutions.

DRE Digital
C. PPG Low Frequency Contents
Feedback
VCMI The PPG signal has a low frequency of about 0.5 Hz and
a high frequency up to 5 Hz. Therefore, the AGC and DC
Ist
+ Vout photocurrent cancellation circuits need to be designed
ADC
– with large time constants ^T = 1/RC h to have less than
IRes Rf
a 0.5 Hz low cutoff frequency. In addition, the bandwidth
Iph of the PPG amplification stage should be around 5 Hz to
Cf
reduce the out of band noise. This implies that large ca-
pacitances and large resistances must be implemented.
The total area of the resistor and capacitor will be quite
Figure 18. A block diagram of the system on chip (SoC) with
large. Some special techniques have been explored to
DRE for static photocurrent rejection [74].
solve the above problems.

1) MOS pseudo-resistor
A resistance of hundreds of mega ohms can be achieved
Vctrl1 Vctrl2
Vctrl when a MOSFET is biased in the subthreshold region. This
Ma Mb Ma Mb type of MOS resistor is commonly used together with a ca-
Va Vbulk Vb Va Vc Vb pacitor in biomedical applications for low cutoff frequency
[75]. Since tuning a resistor by trimming is inconvenient
(a) (b)
and costly, a tunable MOS pseudo-resistor was proposed
Ma Mb [76], as shown in Fig. 19(a). The tolerable DC input voltage
Va Vb range was measured to be 0.5 V. Considering that this MOS
(c) resistor exhibits asymmetric and nonlinear resistance as
the voltage varies across it, [77] proposed a balanced tun-
Figure 19. MOS pseudo-resistors: (a) common tunable MOS able MOS pseudo-resistor in their programmable biomedi-
pseudo resistor [76], (b) balanced tunable MOS pseudo re- cal sensor interface chip, as illustrated in Fig. 19(b). The
sistor [76], and (c) MOS-bipolar pseudo resistor [78].
two transistors can be turned on in an ­alternating m­ anner

18  IEEE circuits and systems magazine THIRD quarter 2018


Authorized licensed use limited to: Chulalongkorn University provided by UniNet. Downloaded on October 21,2024 at 09:03:19 UTC from IEEE Xplore. Restrictions apply.
to achieve a symmetric incremental resistance curve. The 4) a -block
resistance was claimed to be almost constant when the To design a filter architecture with an ultra-low cutoff fre-
voltage across it changed. In addition, a MOS-bipolar de- quency, robust performance and compactness, active-
vice, which is a drain-gate shorted MOSFET, has also been RC implementations are preferred. The architecture of
proposed for implementing large resistances. Using a the first order current-steering low-pass filter (CS-LPF)
PMOSFET as an example, with negative voltage, it functions is given in Fig. 22, and the a -block is enclosed within the
as a diode-connected PMOSFET; with positive voltage, the
parasitic BJT is turned on and works as a diode-connected
VB
BJT. [78] employed this technique to design a circuit with a
low cutoff frequency. Two MOS-bipolar devices connected VA VB VA
in series were used to reduce the distortion of large output
signals, as shown in Fig. 19(c). The cutoff frequency of the
amplifier was 0.025 Hz, in which the equivalent resistance
of the device was larger than 10 13 X [75]. However, stan-
(a) (b)
dard MOS-bipolar pseudo resistors suffer from process,
voltage, and temperature (PVT) variations in addition to Figure 20. (a) switched-capacitor implementation [79], (b)
possible light and EMI sensitivities, leading to variations in switched capacitor implementation with 10-times larger ef-
fective resistance than (a) [80].
cutoff frequency.

2) Switched capacitors
Switched capacitors can be used to implement on-chip r01
Vin Vout
PVT-insensitive high resistance as shown in Fig. 20(a) –Av1 –Av2
[79]. In this topology, the switching frequency fs and the Y
capacitor in the middle determine the resistance precisely
as 1/fs C . The switched-capacitor resistor in Fig. 20(b) miti- VDD
gates manufacturability and interference issues by realiz- Two-Stage MX4
MY4 Vb3
ing a tenfold resistance increase by charge sharing in the Miller
Cm1
switched-capacitor circuits [80]. Unfortunately, the switch- Capacitance
Multiplier MX3
ing produces additional noise along with the weak analog MY3 Vb2
signal in the front end, decreasing the receiver sensitivity. Rm2 Cm2 X Vb
r0x
r0y
MX2
3) Miller capacitance multiplier MY2 Vb1
A Miller capacitance multiplier is used to implement a Rb
large effective capacitance value using a small value ca-
MCN2 MCN1
pacitor [81], [82]. A two-stage Miller capacitance multipli-
er for compensation is proposed to implement the high GND

equivalent capacitance required for an ultra-low frequen-


cy biomedical amplifier [81]. In [83], using a capacitance Figure 21. Schematic diagram of two-stage Miller capaci-
multiplier, the TIA achieves a passband gain of 89 dBV, tance multiplier [83].
a DC current rejection ranging from 0.5 to 85 n A, and a
23 dB high pass cutoff frequency of 0.5 to 9.5 Hz. This
corresponds to an improvement of six times the DC cur-
Rf
rent rejection range compared with previous techniques
[83]. The current through C m1 is converted to voltage by Vb1 Cf
R b, amplified and fed back to node Y, as shown in Fig. 21.
The equivalent capacitance seen from node Y can be ex- Ri
M1 – Vout
pressed as follows:
Vb2 +
g mcn 1 rox g mcn 2 R b A v2 M2
C eq = · C m 1, (9) Vref
1 + sC m1 R b
α-Block
A disadvantage of such capacitance multiplication
techniques is the limited linear range they possess, which
Figure 22. Schematic of CS-LPF [73].
is inversely proportional to the multiplication ratio.

THIRD quarter 2018 IEEE circuits and systems magazine 19


Authorized licensed use limited to: Chulalongkorn University provided by UniNet. Downloaded on October 21,2024 at 09:03:19 UTC from IEEE Xplore. Restrictions apply.
and M 2 in the saturation region.
VDD MBP MBP MBP Both of them have identical aspect
VB1 VB3
ratios but different gate voltages
MM M1 V1 Vcm V2 MM to steer current (shown in Fig. 22),
+ where a is given by Id1/Id2. In
Vin+ MN VSGMR iR MN
MR MR M1 Vin– [84], a low cutoff frequency control
MC1 block is used, which consists of a
– transconductance block gm 2 and
i0
a current-to-voltage block 1/gm 1,
V0 where a is given by gm 2 /gm 1 1 1.
VB2 MC2

MBN MBN
5) Small transconductance OTA
The LPF can be designed using a
VSS
Gm-C technique. A reduced trans-
conductance (g m) OTA suitable
Figure 23. Low g m OTA for low-frequency applications [86].
for low-frequency applications is
required. Several techniques have
been proposed to implement small
g m for designing these filters with low cutoff frequencies
VDD [75]. The current division technique has been proposed
M3A
M3B to reduce the small signal transconductances in voltage-
IBias to-current converters [85]. The currents of two MOSFETs
M2B
are partially canceled at the output. The small signal
M2D +

Vin Vin transconductance of the OTA can be reduced to 10 - 8 A/V.
The current division method is insufficient to implement
M1A M1B transconductances at the nV/A level. To achieve an even
Mu
1: N2 M2C M2A N 2 : 1 smaller transconductance, the above methods can be
Mu combined [86], refer to Fig. 23. The transconductance of
the OTA can be reduced to 10 - 9 A/V. The drain currents
N of M 1 and M N are partially canceled at the output of the
OTA, leading to very large transconductance reductions.
Figure 24. PMOS-input symmetrical OTA with series-paral-
lel current division to reduce transconductance without loss
It can be shown that the resulting small signal transcon-
in a linear range [87]. ductance becomes the following:

i0 N- 1 g
Gm = = mMR (12)
v1 - v2 M+ N+ 1
dotted line [73]. Here, the virtual ground at the negative
input of the opamp has been extended to node X since In equation (12), M is defined as the ratio of transcon-
the a -block is enclosed within a loop. For PPG sensing ductances of M M and M 1, while N is the ratio of trans-
application, a single-ended version is sufficient. The conductances of M N and M 1 .
new transfer function is given by the following: Another technique to reduce the OTA g m is the se-
ries-parallel (SP) current division structure, which re-
- R F /R i
A (s) = , (10) duces the transconductance based on a SP current mir-
1 + SR F C ror. The transconductance of the OTA can be reduced to
a
a11 10 - 12 A/V [87], indicated in Fig. 24. For the NMOS current
mirrors in Fig. 24, N unit transistors are placed in series
a
f= , (11) and in parallel to achieve an effective output transcon-
2 rR f C
ductance G m = g m1 /N 2, where g m1 is the gate transcon-
With this technique, CS-LPF is implemented using a ductance of the transistors M 1 .
38 pF capacitor to achieve a cutoff frequency from 62 to
1.8 Hz (and even lower). The effective capacitive reduc- D. Modulation
tion is from 500 to 12500 times [73]. Sources of additive noise to the PPG signal include system-
The a -block consists of two transistors, with M 1 op- atic circuit offsets, flicker noise, external light sources inter-
erating at the triode region and acting as a linear resistor ference, and 50 Hz power line noise. As PPG signals have a

20  IEEE circuits and systems magazine THIRD quarter 2018


Authorized licensed use limited to: Chulalongkorn University provided by UniNet. Downloaded on October 21,2024 at 09:03:19 UTC from IEEE Xplore. Restrictions apply.
With a modulated light source, the receiver can effectively
suppress the interference from ambient and 1/f noise.

low bandwidth (5 Hz), higher frequency additive noise can any in-band transmitted noise reducing the SNR. There-
be removed with low-pass filtering. However, static (DC) fore, low-noise and low-ripple LED driver is required.
offsets and low frequency noise (also known as fixed pat- For wearable/mobile devices, the battery lifetime is a
tern noise), which occupy the same spectrum as the PPG critical issue. The battery lifetime can be extended by
signal, require different rejection techniques. One solution reducing the LED ON time; also, the LED current should
is to modulate the light source to shift it to a higher spec- be reduced as much as possible, as shown in Fig. 27. The
trum range where there is less noise content, as shown sampling rate needs to be high enough not to miss the
in Fig. 25. With a modulated light
source, the receiver can effectively
suppress the interference from am-
bient and 1/f noise. Demodulation Modulation Demodulation
envelope detection is attractive
due to the simplicity of its imple-
mentation and low power use, but + + + +
it is prone to distortion by addi- Vin Va Vb LPF Vout
tive noise. The system illustrated – – – –
+
in Fig. 25 uses decimation by 4 FIR
filters for the demodulation. Down-
Aggressors
sampling the signal by a factor M re- (a)
sults in placing a copy of the signal
Vin Signal Va Vb Vout
back in the baseband, completing Aggressores
the demodulation [74]. For demodu-
lation using multiplications inside
the MCU [88], as shown in Fig. 26, 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4
the MCU spent much time in full- f/fch f/fch f/fch f/fch
(b)
power processing mode, which led
to an average high power consump- Figure 25. (a) Block diagram and (b) frequency-domain illustration of the chopper [74].
tion. To reduce power consumption
in the MCU, the percentage of time
that the processor core is in a
sleep or power-down state should
be increased. Processing time can Analogue Domain Digital Domain
be reduced by choosing an alter-
native MCU core that supports (1 + m [n])/2 Local Oscillator
LED
hardware multiplication [88]. Driver PWM
m [n]
Even though light modulation Homodyne
will save the MCU and LED trans- Bias Midrange
LPF
Output
mitter power dissipation, the detec- Stabilisation Setpoint
Fifth Order
+
tion with the pre-amplifier requires LPF Product
DAC Detector Baseband
a higher bandwidth and hence First Order –
LPF Output
higher bias current than if modula- ADC
Vout[n] Fifth Order
tion was not employed. Photodide
Amplifier
E. LED Driver
A key characteristic for the PPG
LED driver circuit is its low noise; Figure 26. A block diagram illustration of a single channel PPG homodyne detection
system with current offset compensation [88].
the optical receiver will amplify

THIRD quarter 2018 IEEE circuits and systems magazine 21


Authorized licensed use limited to: Chulalongkorn University provided by UniNet. Downloaded on October 21,2024 at 09:03:19 UTC from IEEE Xplore. Restrictions apply.
fast features of the PPG signal. Increasing the pulse rep- Typical systems sample with rates between 100 Hz and
etition rate of the LED current results in better signal ac- 1000 Hz with a pulse width ranging from 10 n s up to 100 n s.
curacy but at higher energy consumption. The LED cur- A study was introduced in [89] to select the optimum
rent pulse repetition rate and pulse width are a trade-off LED duty cycle to reduce the power consumption for a
between signal quality and the total consumed energy. certain SNR.
A LED driver is proposed in [90] with 89% efficiency. A
low supply voltage and low average current are desired
RED On for devices such as a PPG heart rate monitor, as shown
220 µs RED Off
1,780 µs in Fig. 28. The LED current and as a result the transmit-
ted optical power is controlled by pulse density modu-
lation (PDM). The integrated digital blocks for varying
RED LED the LED optical power help to reduce the chip area. The
(660 nm) LED driver implemented in 180 nm CMOS along with an
external inductor has a controllable current range from
Figure 27. Pulse train to control the LED driver. 28 n A up to 1.3 mA using a 0.5 to 0.6 V supply voltage.
The LED driver was designed as 8-bit, wide-swing
cascode current mirrors, as shown in Fig. 29 [74]. This
L topology was chosen to maintain the high linearity of
Vout
Iout LED Current the output current across a large variation of the cas-
VDD Iout
+ code drain voltage. Voltage variation occurs because
– PDM the forward LED voltage can be as small as 1.1 V for an IR
LED and as large as 1.8 V for a red LED. Additionally, for
t (s) operation with a small coin-cell battery, the voltage can
be as high as 3 V when the battery is fully charged and
Figure 28. LED driver based on pulse density modulation as low as 2.3 V when the battery discharges. The LSB of
(PDM) [90]. the LED driver DAC is 100 n V with a maximum current
of 25.6 mA. To reduce the voltage drop along the high
current path, the DAC is toggled by switches connect-
VDD ing the gates of the cascode transistors to either VTOP
ILED
10Ibias 2Ibias 8Ibias and or VBOT to ground (not shown). A digitally controlled
Ibias
VTOP NW/L fast startup current reference for duty cycling the LED
CSb
10 W/L 2 W/L driver bias is also utilized to conserve the power. The
Current duty cycle is 0.7%, which corresponds to a square LED
0.25 W/L VCASV DAC current pulse that is 40 n s wide every 60 ms.
CD
VBOT
10 W/L NW/L
F. PPG Integrated Sensor
CD
A wearable cNIPB requires both low power and light
CSb weight/small area. The integration of the PPG receiver
and the LED driver reduces the number of off-chip el-
Figure 29. The wide-swing cascode LED driver DAC [74].
ements to obtain a compact portable system. Table IV

Table IV.
Recently published integrated PPG sensors.
Process Biasing Noise Rejection Integrated LED Driver
[84] 0.35 μm 2.5 V/58 μA 40.8 nA 15 μA, 165 pF on chip No
[73] 0.35 μm 2.5 V/240 μA 3.53 nA 43.9 μA, 2.2 nF Off chip No
[71] 180 nm 0.5 V/8 μA 4 nA 3.5 μA logarithmic DRC No
[74] 180 nm 2.3 V/425 μW 2.5 nA 100 μA, DRE current source LSB 100 μA maximum current
of 25.6 mA
[70] 0.35 μm 3.3 V/1.88 mW 445 μV No 1.3 mW, Feedback path to
control the LED current

22  IEEE circuits and systems magazine THIRD quarter 2018


Authorized licensed use limited to: Chulalongkorn University provided by UniNet. Downloaded on October 21,2024 at 09:03:19 UTC from IEEE Xplore. Restrictions apply.
shows recently published PPG sensors. The PPG sensors to overcome the low wrist PPG signal. Light modulation
in [71], [73], [74], [84] have a DC Photocurrent Rejection can enhance the PPG receiver sensitivity at the expense
loop. The record of 100 n A DC photocurrent rejection of higher power consumption. Low power operation and
was achieved in [74] because of the (DRE) current small system area are necessities for a wearable device
source. Two recent PPG sensors [70], [74] have also inte- for cNIBP monitoring. Much power can be saved on the
grated the LED driver on the same chip. transmitting side by transmitting less optical power us-
ing a highly sensitive PPG receiver. The LED driver pulse
V. Commercial cNIBP Devices repetition rate and pulse width should be optimized to
There are many cNIBP devices based on PPG that are have good signal quality and acceptable power consump-
trying to find their way to the market. The ViSi Mobile tion. To reduce the system area, the large capacitance/
system enables cNIBP of patients based on the PAT resistances required for the LPF and DC photocurrent
[91]. The SOMNOtouch NIBP simultaneously records cancellation can be integrated on chip by using several
the ECG and finger PPG, and the PTT is calculated. techniques such as a Gm-C LPF with nA/V transconduc-
By means of a one-time calibration, the systolic and tance OTA, capacitance multiplier, pseudo resistor, and
diastolic blood pressure can be calculated [92]. These a -block. The integration of the photodiode with the TIA
commercial devices have an acceptable accuracy un- and an LED driver using standard CMOS technology will
der small motion (rest condition) to reduce the motion reduce the system area and cost. Both smartwatches
artifact effect on the PPG signal. The processing is and smartphones are perfect candidate devices to have
done in a separate PC unit to perform the BP analy- cNIBP measuring capability.
sis and calculations. The Simband smart watch from
Samsung generates 8 PPG streams; based on the con- Acknowledgments
fidence indicator of the PPG channel, an algorithm may This research was supported by the Natural Science
choose the best PPG. The ECG and xyz accelerometer Foundation of China under Grant 61474074.
signals are also available. The PAT is computed and
can be used to calculate blood pressure. The Monitor- Guoxing Wang (M’06, SM’13) re -
ing mode consumes a lot of power (400 mW peak pow- ceived his PhD in electrical engineer-
er), as all the sensors and the processor are ON. This is ing from the University of California
not feasible in a system that need to work for 24 hours at Santa Cruz, US, in 2006. He was a
[93]. The LED ON time can be controlled to reduce the Member of the Technical Staff at Age-
average power to 50 mW. The Project Zero Wrist Blood re Systems, San Jose, California,
Pressure Monitor (BP6000) is a cuffless blood pressure from 2006 to 2007. In 2007–2009, he joined Second
smart watch developed by Omron [94]. The BP6000 re- Sight Medical Products, Sylmar, California, where he
sults were shown at the CES 2016 event, and it will be designed the integrated circuit chip that went into the
available in the market shortly after Omron receives a eyes of patients to restore vision. Currently, he is an
certification from the FDA. Associate Professor in the School of Microelectronics,
Shanghai Jiao Tong University, Shanghai, China. He has
VI. Conclusion published in various peer-reviewed journals and con-
Non-invasive, continuous time, cuffless BP devices have ferences. His current research interests include wire-
some major challenges to come to the market, with MAD less technologies and biomedical electronics. Dr. Wang
levels required to be less than "5 mmHg. A calibration currently serves as Deputy Editor in Chief for IEEE
process is needed to reach that accuracy. More research Transactions on Biomedical Circuits and Systems (TBi-
is needed on MA cancellation algorithms to reach a true oCAS) (2016–2017). He is a member of the IEEE Biomed-
wearable, continuous time, cuffless BP device. The ex- ical Circuits Systems Technical Committee (BioCAS).
isting capabilities for the expected commercial PPG- He served as an Associate Editor for IEEE Transactions
based BP devices have acceptable performance under on Circuits and Systems II (2012–2015), Guest Editor for
only very small motion effects. For beat-to-beat BP mea- IEEE Journal on Emerging and Selected Topics in Cir-
surements, real-time processing is needed, which will cuits and Systems (JETCAS) and Guest Editor for IEEE
require developing accurate BP algorithms with fewer Transactions on Biomedical Circuits and Systems, both
operations and lower complexity. The most mature com- in 2014. He served as the technical program chair for
mercial BP devices deploy a wristwatch structure be- the IEEE Conference on Biomedical Circuits and Sys-
cause it is comfortable and easier to wear, but the wrist tems in 2016. He was the local chair for the first IEEE
PPG signal is weak. Therefore, more effort should be Green Circuits and Systems (ICGCS) in 2010 and for
spent toward developing a high-sensitivity PPG receiver the second Asia Pacific Conference on Postgraduate

THIRD quarter 2018 IEEE circuits and systems magazine 23


Authorized licensed use limited to: Chulalongkorn University provided by UniNet. Downloaded on October 21,2024 at 09:03:19 UTC from IEEE Xplore. Restrictions apply.
­ esearch in Microelectronics & Electronics (PrimeA-
R Award (2013), Chen-Ning Franklin Yang Award in Science
sia), in 2010. and Technology for New Immigrant (2014), and Design
Contest Award in 20th International Symposium on Low
Mohamed Atef (M’08, SM’12) received Power Electronics and Design (2015). As an educator, he
the B.Sc. and M.Sc. degrees in electri- received the University Annual Teaching Excellent Award
cal engineering, electronics and com- in two consecutive academic years, from 2008 to 2010,
munications from Assiut University, and many other teaching awards from the Faculty of En-
Egypt, in 2000 and 2005, respectively. gineering of NUS. Under his guidance, his students re-
From 2006 to 2007, he was a research- ceived many awards including the Best Student Paper
er at the Czech Technical University in Prague, Depart- Award in ICME 2007, winner of 47th DAC/ISSCC Student
ment of Microelectronics, working on the improvement Design Contest in 2010, Best Design Award in A-SSCC
of quantum dot optical properties. He received his PhD 2013 Student Design Contest. He is the President-Elect
in 2010 from Vienna University of Technology, Institute of of the IEEE CIRCUITS AND SYSTEMS (CAS) Society,
Electrodynamics, Microwave and Circuit Engineering, Member of IEEE Fellow Committee, and Steering Com-
and then worked as a post-doctoral researcher until mittee Member of the IEEE TRANSACTIONS ON BIO-
the end of 2012. Mohamed Atef is currently an Associ- MEDICAL CIRCUITS AND SYSTEMS. He was the Editor-
ate Professor at Assiut University, Egypt. He has been in-Chief of the IEEE TRANSACTIONS ON CIRCUITS AND
visiting the School of Microelectronics, Shanghai Jiao SYSTEMS—II: EXPRESS BRIEFS for two terms from 2010
Tong University since 2015. His research interests are to 2013. He was the Guest Editor for eight special is-
in the areas of integrated circuits, optoelectronic inte- sues in the IEEE TRANSACTIONS ON CIRCUITS AND
grated circuits, and communications over plastic opti- SYSTEMS—I: REGULAR PAPERS, the IEEE TRANSAC-
cal fiber. He is an author of two Springer books, “Opti- TIONS ON BIOMEDICAL CIRCUITS AND SYSTEMS, and
cal Communication over Plastic Optical Fibers: Journal of Circuits, Systems Signal Processing. He was
Integrated Optical Receiver Technology” and “Opto- the Vice President for Publications of the IEEE CAS Soci-
electronic Circuits in Nanometer CMOS Technology”. ety from 2013 to 2015, Vice President for the Asia Pacific
Furthermore, he is an author and co-author of more Region of the IEEE CAS Society from 2007 to 2008, Ad-
than 60 scientific publications. Since 2012, he has been Comm Member of the IEEE Biometrics Council from
a senior member of IEEE. 2008 to 2009, CAS Society Representative to the Bio-
Technology Council from 2007 to 2009, Chair of the Bi-
Yong Lian (M’90, SM’99, F’09) received oCAS Technical Committee of the IEEE CAS Society
the B.Sc. degree from the College of from 2007 to 2009, Chair of DSP Technical Committee
Economics and Management, Shang- of the IEEE CAS Society from 2010 to 2011, Member of
hai Jiao Tong University, Shanghai, the IEEE Medal for Innovations in Healthcare Technol-
China, in 1984 and the Ph.D. degree ogy Committee from 2010 to 2012, and a Distinguished
from the Department of Electrical En- Lecturer of the IEEE CAS Society from 2004 to 2005. He
gineering, National University of Singapore (NUS), Singa- is the Founder of the International Conference on
pore, in 1994. He spent nine years in industry and joined Green Circuits and Systems, the Asia Pacific Confer-
NUS in 1996, where he was the Deputy Department Chair ence on Postgraduate Research in Microelectronics
for Research, Area Director for IC and Embedded Sys- and Electronics, and the IEEE BIOMEDICAL CIRCUITS
tems in the Department of ECE, member of University AND SYSTEMS CONFERENCE. He is a Fellow of the
Tenure and Promotion Committee, and member of Sen- Academy of Engineering Singapore.
ate Delegacy. In 2011, he was appointed as the first Pro-
vosts Chair Professor in the Department of Electrical and References
[1] (2017). [Online]. Available: https://www.cdc.gov/bloodpressure/
Computer Engineering, NUS. He is currently a Professor facts.htm
in York University, Toronto, ON, Canada. His research in- [2] (2017, May). [Online]. Available: http://www.who.int/
terests include biomedical circuits and systems and sig- [3] R. Merai, C. Siegel, M. Rakotz, P. Basch, J. Wright, B. Wong, and P.
Thorpe, “CDC grand rounds: A public health approach to detect and
nal processing. Dr. Lian received many awards including control hypertension,” Morbidity Mortality Wkly. Rep., vol. 65, no. 45, pp.
the IEEE Circuits and Systems Societys Guillemin-Cauer 1261–1264, Nov. 2016.
[4] G. S. Stergiou, K. K. Argyraki, I. Moyssakis, S. E. Mastorantonakis,
Award (1996), the IEEE Communications Society Multi- A. D. Achi-mastos, V. G. Karamanos, and L. G. Roussias, “Home blood
media Communications Best Paper Award (2008), Institu- pressure is as reliable as ambulatory blood pressure in predicting tar-
tion of Engineers Singapore Prestigious Engineering get-organ damage in hypertension,” Amer. J Hypertension, vol. 20, pp.
616–621, 2007.
Achievement Award (2011), Hua Yuan Association/Tan [5] IEEE Standard for Wearable, Cuffless Blood Pressure Measuring De-
Kah Kee International Society Outstanding Contribution vices, IEEE Standard 1708–2014, 2014.

24  IEEE circuits and systems magazine THIRD quarter 2018


Authorized licensed use limited to: Chulalongkorn University provided by UniNet. Downloaded on October 21,2024 at 09:03:19 UTC from IEEE Xplore. Restrictions apply.
[6] L. A. Geddes, Handbook of Blood Pressure Measurement. Humana blood pressure and ECG from photoplethysmograph using smart
Press, 1991. phones,” in Proc.11th ACM Conf. Embedded Networked Sensor Systems,
[7] R. Raamat, K. Jagomagi, and J. Talts, “Continuous recording of the 2013, pp. 88–89.
oscillometric mean arterial pressure by the differential servo system [30] A. Visvanathan, R. Banerjee, A. D. Choudhury, A. Sinha, and S.
with two photoplethysmographic sensors,” in Proc. 8th IEEE Int. Conf. Kundu, “Smart phone based blood pressure indicator,” in Proc. 4th ACM
BioInformatics and BioEngineering, 2008, p. 4696827. MobiHoc Workshop Pervasive Wireless Healthcare, 2014, pp. 19–24.
[8] S. K. Lal, A. Mihailidou, M. Cejnar, R. Henderson, M. Jones, and S. Hu- [31] F. Lamonaca, K. Barbe, Y. Kurylyak, D. Grimaldi, W. V. Moer, A.
nyor, “Continuous, non-invasive volume-clamp blood pressure: Determi- Furfaro, and V. Spagnuolo, “Application of the artificial neural network
nants of performance,” J. Hypertension, vol. 11, no. 12, pp. 1413–1422, 1993. for blood pressure evaluation with smartphones,” in Proc. 7th IEEE Int.
[9] G. M. Drzewiecki, J. Melbin, and A. Noordergraaf, “Arterial tonom- Conf. Intelligent Data Acquisition and Advanced Computing Systems Tech-
etry: Review and analysis,” J. Biomech., vol. 16, no. 2, pp. 141–152, 1983. nology and Applications, 2013, pp. 408–412.
[10] G. L. Pressman and P. M. Newgard, “A transducer for the continu- [32] R. Couceiro, P. Carvalho, R. P. Paiva, J. Henriques, and J. Muehlsteff,
ous external measurement of arterial blood pressure,” IEEE Trans. Bio- “Detection of motion artifacts in photoplethysmographic signals based
Med. Electron., vol. 10, no. 2, pp. 73–81, 1963. on time and period domain analysis,” in Proc. Annu. Int. Conf. IEEE Engi-
[11] J. Solà, M. Proença, and O. Chételat, “Wearable PWV technologies neering Medicine and Biology Society, 2012, pp. 2603–2606.
to measure blood pressure: Eliminating brachial cuffs,” pp. 4098–4101, [33] K. Duan, Y. Hu, Z. Qian, and G. Wang, “An FPGA-based morphologi-
July 2013. cal filter for baseline wandering correction in photoplethysmography,”
[12] L. Petera, N. Nouryb, and M. Cerny, “Review of methods for non- in Proc. IEEE BioMedical Circuits and Systems Conf., Shanghai, China,
invasive and continuous blood pressure monitoring: Pulse transit time 2016.
method is promising?” Innovation Res. BioMed. Eng., vol. 35, no. 5, pp. [34] R. Wang, W. Jia, Z. Mao et al., “Cuff-free blood pressure estimation
271–282, 2014. using pulse transit time and heart rate,” in Proc. IEEE 12th Int. Conf.
[13] American National Standard for Manual, Electronic, or Automated Signal Processing, 2014, pp. 115–118.
Sphygmomanometers, ANSI/AAMI Standard SP10:2002, 2002. [35] R. Krishnan, B. Natarajan, and S. Warren, “Two-stage approach for
[14] M. Atef, L. Xiyan, G. Wang, and Y. Lian, “PTT based continuous time detection and reduction of motion artifacts in photoplethysmographic
non-invasive blood pressure system,” in Proc. IEEE 59th Int. Midwest data,” IEEE Trans. Biomed. Eng., vol. 57, no. 8, pp. 1867–1876, 2010.
Symp. Circuits and Systems, Abu Dhabi, 2016, pp. 333–336. [36] R. Yousefi, M. Nourani, S. Ostadabbas et al., “A motion-tolerant
[15] S. D. Kalkhaire and V. G. Puranik, “Remote detection of photople- adaptive algorithm for wearable photoplethysmographic biosensors,”
thysmographic signal and SVM based classification,” pp. 128–132, Dec. IEEE J. Biomed. Health Inform., vol. 18, no. 2, pp. 670–681, 2014
2016. [37] X. He, R. A. Goubran, and X. P. Liu, “Secondary peak detection of
[16] E. D. Chan, M. M. Chan, and M. M. Chan, “Pulse oximetry: Under- PPG signal for continuous cuffless arterial blood pressure measure-
standing its basic principles facilitates appreciation of its limitations,” ment,” IEEE Trans. Instrum. Meas., vol. 63, no. 6, pp. 1431–1439, 2014.
Respiratory Med., vol. 107, no. 6, pp. 789–799, 2013. [38] Q. Li and G. Clifford, “Dynamic time warping and machine learning
[17] T. Tamura, Y. Maeda, M. Sekine, and M. Yoshida, “Wearable pho- for signal quality assessment of pulsatile signals,” Physiol. Meas., vol.
toplethysmographic sensors: Past and present,” MDPI Electron., vol. 3, 33, no. 9, pp. 1491–1501, 2012.
no. 2, pp. 282–302, 2014. [39] I. SilvaI, J. Lee, and R. G. Mark, “Signal quality estimation with
[18] C. J. Pujary, “Investigation of photodetector optimization in reduc- multichannel adaptive filtering in intensive care settings,” IEEE Trans.
ing power consumption by a noninvasive pulse oximeter sensor,” M.S. Biomed. Eng., vol. 59, no. 9, pp. 2476–2485, 2012.
thesis, Worcester Polytechnic Inst., 2004. [40] K. Duan, Z. Qian, M. Atef, and G. Wang, “A feature exploration meth-
[19] J. Spigulis, L. Gailite, A. Lihachev, and R. Erts, “Simultaneous re- odology for learning based cuffless blood pressure measurement using
cording of skin blood pulsations at different vascular depths by mul- photoplethysmography,” in Proc. 38th Annu. Int. Conf. IEEE Engineering
tiwavelength photoplethysmography,” Appl. Opt., vol. 46, no. 10, pp. Medicine and Biology Society, 2016.
1754–1759, 2007. [41] M. Kachuee, M. Kiani, H. Mohammadzade et al., “Cuff-less high-
[20] J. Lee, K. Matsumura, K. Ichi Yamakoshi, P. Rolfe, S. Tanaka, and accuracy calibration-free blood pressure estimation using pulse
T. Yamakoshi, “Comparison between red, green and blue light reflec- transit time,” in Proc. IEEE Int. Symp. Circuits and Systems, 2015,
tion photoplethysmography for heart rate monitoring during motion,” pp. 1006–1009.
in Proc. IEEE 35th Annu. Int. Conf. Engineering Medicine and Biology So- [42] Y. Kurylyak, K. Barbe, F. Lamonaca et al., “Photoplethysmogram-
ciety, 2013, pp. 1724–1727. based blood pressure evaluation using Kalman filtering and neural net-
[21] V. Mecnika, E. Kviesis-Kipge, I. Krievins, Z. Marcnikevics, and A. works,” in Proc. IEEE Int. Symp. Medical Measurements and Applications,
Schwarz, “Wearable PPG sensor matrix for cardiovascular assess- 2013, pp. 170–174.
ment,” Mater. Sci. Textile Clothing Technol., vol. 8, pp. 76–82, 2013. [43] E. Monte-Moreno, “Non-invasive estimate of blood glucose and
[22] J. A. C. Patterson, D. G. McIlwraith, and G.-Z. Yang, “A flexible, low blood pressure from a photoplethysmograph by means of machine
noise reflective PPG sensor platform for ear-worn heart rate monitor- learning techniques,” Artif. Intell. Med., vol. 53, no. 2, pp. 127–138,
ing,” in Proc. 6th Int. Workshop Wearable and Implantable Body Sensor 2011.
Networks, 2009, pp. 288–291. [44] B. liang, K. Duan, Q. X. M. Atef, Z. Qian, G. Wang, and Y. Lian, “Live
[23] S. Rhee, B.-H. Yang, and H. H. Asada, “Artifact-resistant, power- demonstration: A support vector machine based hardware platform for
efficient design of finger-ring plethysmographic sensors,” IEEE Trans. blood pressure prediction,” in Proc. IEEE BioMedical Circuits and Sys-
Biomed. Eng., vol. 48, pp. 795–805, 2001. tems Conf., Shanghai, China, 2016.
[24] Y. Lee, H. Shin, J. Jo, and Y.-K. Lee, “Development of a wristwatch- [45] A. Choudhury, R. Banerjee, A. Sinha et al., “Estimating blood pres-
type PPG array sensor module,” in Proc. IEEE Int. Conf. Consumer Elec- sure using windkessel model on photoplethysmogram,” in Proc. 36th
tronics, 2011, pp. 168–171. Annu. Int. Conf. IEEE Engineering Medicine and Biology Society, 2014, pp.
[25] Y.-L. Zheng, B. P. Yan, Y.-T. Zhang, and C. C. Y. Poon, “An armband 4567–4570.
wearable device for overnight and cuff-less blood pressure measure- [46] S. Huang, P. H. Hung, C.-H. Hong et al., “A new image blood pressure
ment,” IEEE Trans. Biomed. Eng., vol. 61, no. 7, pp. 2179–2186, 2014. sensor based on PPG, RRT, BPTT, and harmonic balancing,” IEEE Sen-
[26] [Online]. Available: http://www.medtronic.com/ sors J., vol. 14, no. 10, pp. 3685–3692, 2014.
[27] M.-Z. Poh, N. C. Swenson, and R. W. Picard, “Motion-tolerant mag- [47] F. S. Cattivelli and H. Garudadri, “Noninvasive cuffless estimation
netic earring sensor and wireless earpiece for wearable photoplethys- of blood pressure from pulse arrival time and heart rate with adaptive
mograph,” IEEE Trans. Inf. Technol. Biomed. (1995–2012), vol. 14, no. 3, calibration,” in Proc. 6th Int. Workshop Wearable and Implantable Body
pp. 786–794, 2010. Sensor Networks, 2009, pp. 114–119.
[28] K. Budidha and P. A. Kyriacou, “The human ear canal: Investiga- [48] R. Carretta, “McDonald’s blood flow in arteries: Theoretical, ex-
tion of its suitability for monitoring photoplethysmographs and arterial perimental and clinical principles,” J. Hypertension, vol. 16, no. 4, p. 553,
oxygen saturation,” Physiol. Meas., vol. 35, no. 2, pp. 111–128, 2014. 1998.
[29] A. D. Choudhury, A. Visvanathan, R. Banerjee, A. Sinha, A. Pal, [49] D. B. McCombie, A. T. Reisner, and H. H. Asada, “Adaptive blood
C. Bhaumik, and A. Kumar, “Demo abstract: Heartsense estimating pressure estimation from wearable PPG sensors using peripheral ­a rtery

THIRD quarter 2018 IEEE circuits and systems magazine 25


Authorized licensed use limited to: Chulalongkorn University provided by UniNet. Downloaded on October 21,2024 at 09:03:19 UTC from IEEE Xplore. Restrictions apply.
pulse wave velocity measurements and multi-channel blind identifica- photoplethysmographic heart-rate sensor,” IEEE J. Solid-State Circuits,
tion of local arterial dynamics,” in Proc. 28th Annu. Int. Conf. IEEE Engi- vol. 49, no. 1, 2014.
neering Medicine and Biology Society, 2006, pp. 3521–3524. [72] K. Phang and D. A. Johns, “A CMOS optical preamplifier for wireless
[50] L. Xu, X. Guo, F. Yang, S. Yin, X. Zhang, and M. Q. H. Meng, “Imple- infrared communications,” IEEE Trans. Circuits Syst. II, vol. 46, no. 7, pp.
mentation of cuff-less continuous blood pressure measurement system 852–859, 1999.
based on android,” in Proc. Int. Conf. Information and Automation, 2012, [73] A. K. Y. Wong, K.-P. Pun, Y.-T. Zhang, and K. N. Leung, “A low-power
pp. 552–556. CMOS front-end for photoplethysmographic signal acquisition with ro-
[51] L. Geddes, M. Voelz, C. Babbs, J. Bourland, and W. Tacker, “Pulse bust dc photocurrent rejection,” IEEE Trans. Biomed. Circuits Syst., vol.
transit time as an indicator of arterial blood pressure,” Psychophysiol- 2, no. 4, pp. 280–288, 2008.
ogy, vol. 18, no. 1, pp. 71–74, 1981. [74] E. S. Winokur, T. ODwyer, and C. G. Sodini, “A low-power, dual-wave-
[52] K. Chan, K. Hung, and Y. Zhang, “Noninvasive and cuffless mea- length photoplethysmogram (PPG) SOC with static and time-varying
surements of blood pressure for telemedicine,” in Proc. Annu. Reports interferer removal,” IEEE Trans. Biomed. Circuits Syst., pp. 581–589, 2015.
Res. Reactor Institute, vol. 4, pp. 3592–3593, 2001. [75] C. Yan Li, C. Y. Poon, and Y.-T. Zhang, “Analog integrated circuits
[53] H. Y. Xiang, Y. Y. Liu, Y. F. Qin, W. Pan, and M. S. Yu, “Calibration of design for processing physiological signals,” IEEE Rev. Biomed. Eng.,
pulse wave transit time method in blood pressure measurement based vol. 3, pp. 93–105, 2010.
on the Korotkoff sound delaytime,” in Proc. World Congr. Medical Physics [76] R. H. Olsson, D. L. Buhl, A. M. Sirota, G. Buzsaki, and K. D. Wise,
and Biomedical Engineering, Beijing, China, 2013, pp. 426–429. “Band-tunable and multiplexed integrated circuits for simultaneous
[54] P. Shaltis, A. Reisner, and H. Asada, “A hydrostatic pressure ap- recording and stimulation with microelectrode arrays,” IEEE Trans.
proach to cuffless blood pressure monitoring,” in Proc. 26th Annu. Int. Biomed. Eng., vol. 52, no. 7, pp. 1303–1311, 2005.
Conf. IEEE Engineering Medicine and Biology Society, 2004, pp. 2173– [77] X. D. Zou, X. Y. Xu, L. B. Yao, and Y. Lian, “A 1 V 450 nW fully inte-
2176. grated programmable biomedical sensor interface chip,” IEEE J. Solid-
[55] C. C. Y. Poon, Y. T. Zhang, and Y. Liu, “Modeling of pulse transit time State Circuits, vol. 44, no. 4, pp. 1067–1077, 2009.
under the effects of hydrostatic pressure for cuffless blood pressure [78] R. R. Harrison and C. Charles, “A low-power low-noise CMOS ampli-
measurements,” in Proc. 3rd IEEE Engineering Medicine and Biology Soci- fier for neural recording applications,” IEEE J. Solid-State Circuits, vol.
ety Int. Summer School Medical Devices and Biosensors, 2006, pp. 65–68. 38, no. 6, pp. 958–965, 2003.
[56] D. Buxi, J.-M. Redout, and M. R. Yuce, “A survey on signals and [79] T. Denison, K. Consoer, W. Santa, A.-T. Avestruz, J. Cooley, and A.
systems in ambulatory blood pressure monitoring using pulse transit Kelly, “A 2 μW 100 nV/rtHz chopper-stabilized instrumentation amplifier
time,” Inst. Phys. Eng. Med. Physiol. Meas., vol. 36, pp. R1–R26, 2015. for chronic measurement of neural field potentials,” IEEE J. Solid-State
[57] R. Kondo, S. Bhuiyan, H. Kawanaka, and I. M. Koji Oguri, “Separate Circuits, vol. 42, no. 12, pp. 2934–2945, 2007.
estimation of long-and short-term systolic blood pressure variability [80] N. Verma, A. Shoeb, J. Bohorquez, J. Dawson, J. Guttag, and A. P.
from photoplethysmograph,” in Proc. 36th Annu. Int. Conf. IEEE Engi- Chandrakasan, “A micro-power EEG acquisition SOC with integrated
neering Medicine and Biology Society, 2014, pp. 1851–1854. feature extraction processor for a chronic seizure detection system,”
[58] F. A. F. Marques, D. Ribeiro, and M. F. M. Colunas, “A real time, wear- IEEE J. Solid-State Circuits, vol. 45, no. 4, pp. 804–816, 2010.
able ECG and blood pressure monitoring system,” in Proc. 6th Iberian [81] G. A. Rincon-Mora, “Active capacitor multiplier in miller-compen-
Conf. Information Systems and Technologies, 2011, pp. 1–4. sated circuits,” IEEE Trans. Solid-State Circuits, vol. 35, no. 1, pp. 26–32,
[59] A. Jadooei, O. Zaderykhin, and V. I. Shulgin, “Adaptive algorithm for 2000.
continuous monitoring of blood pressure using a pulse transit time,” in [82] J. Aguado-Ruiz, A. J. Lopez-Martin, and J. Ramirez-Angulo, “Three
Proc. IEEE 33rd Int. Scientific Conf. Electronics and Nanotechnology, 2013, novel improved CMOS C-multipliers,” Int. J. Circuit Theory Appl., vol. 40,
pp. 297–301. pp. 607–616, 2012.
[60] E. C. P. Chua, S. J. Redmond, G. McDarby et al., “Towards using [83] Y. Li, A. Wong, and Y.-T. Zhang, “Fully-integrated transimpedance
photo-plethysmogram amplitude to measure blood pressure during amplifier for photoplethysmographic signal processing with two-stage
sleep,” Ann. Biomed. Eng., vol. 38, no. 3, pp. 945–954, 2010. miller capacitance multiplier,” Electron. Lett., vol. 46, no. 11, 2010.
[61] M. R. Ram, K. V. Madhav, E. H. Krishna, N. R. Komalla, and K. A. [84] A. K. Y. Wong, K. N. Leung, K.-P. Pun, and I. Yuan-Ting Zhang, “A
Reddy, “A novel approach for motion artifact reduction in PPG signals 0.5-Hz high-pass cutoff dual-loop transimpedance amplifier for wear-
based on AS-LMS adaptive filter,” IEEE Trans. Instrum. Meas., vol. 61, no. able NIR sensing device,” IEEE Trans. Circuits Syst. II, vol. 57, no. 7,
5, pp. 1445–1457, 2012. 2010.
[62] P. Gibbs and H. H. Asada, “Reducing motion artifact in wearable [85] J. Silva-Martinez and J. Salcedo-Suner, “IC voltage-to-current trans-
bio-sensors using MEMS accelerometers for active noise cancellation,” ducers with very-small transconductance,” Analog Integr. Circuits Signal
in Proc. American Control Conf., 2005, pp. 1581–1586. Process., vol. 13, no. 3, pp. 285–293, 1997.
[63] T. Shimazaki and S. Hara, “Breathing motion artifact cancellation in [86] S. Solis-Bustos, J. Silva-Martinez, F. Maloberti, and E. Sanchez-Si-
PPG-based heart rate sensing,” in Proc. 9th Int. Symp. Medical Informa- nencio, “A 60 dB dynamic-range CMOS sixth-order 2.4 Hz low-pass filter
tion and Communication Technology, 2015, pp. 200–203. for medical applications,” IEEE Trans. Circuits Syst. II, vol. 47, no. 12, pp.
[64] S. M. Park and C. Toumazou, “Low noise current-mode CMOS tran- 1391–1398, 2000.
simpedance amplifier for Giga-bit optical communication,” in Proc. IEEE [87] A. Arnaud, R. Fiorell, and C. Galup-Montoro, “Nanowatt, sub-ns
Int. Symp. Circuits and Systems, Monterey, CA, 1998, pp. 293–296. otas, with sub-10-mv input offset, using series-parallel current mir-
[65] S. M. Park and H. J. Yoo, “1.25-Gb/s regulated cascode CMOS tran- rors,” IEEE J. Solid-State Circuits, vol. 41, no. 9, pp. 2009–2018, 2006.
simpedance amplifier for gigabit ethernet applications,” IEEE J. Solid- [88] J. A. C. Patterson and G.-Z. Yang, “Dual-mode additive noise rejec-
State Circuits, vol. 39, no. 1, pp. 112–121, 2004. tion in wearable photoplethysmography,” in Proc. 9th Int. Conf. Wear-
[66] M. Atef and H. Zimmermann, Optoelectronic Circuits in Nanometer able and Implantable Body Sensor Networks, 2012, pp. 97–102.
CMOS Technology. Switzerland: Springer International Publishing, 2016. [89] D. A. Peláez and E. R. Villegas, “LED power reduction trade-offs
[67] E. Kamrani, F. Lesage, and M. Sawan, “Fully on-chip integrated pho- for ambulatory pulse oximetry,” in Proc. IEEE 29th Annu. Int. Conf.
todetector front-end dedicated to real-time portable optical brain imag- Engineering Medicine and Biology Society, Lyon, France, 2007, pp.
ing,” Opt. Photonics J., vol. 2, pp. 300–313, 2012. 2296–2299.
[68] M. Tavakoli, L. Turicchia, and R. Sarpeshkar, “An ultra-low-power [90] W. Saadeh, T. Tekeste, and M. H. Perrott, “A ≥ 89% efficient LED
pulse oximeter implemented with an energy-efficient transimpedance driver with 0.5V supply voltage for applications requiring low aver-
amplifier,” IEEE Trans. Biomed. Circuits Syst., vol. 3, no. 1, pp. 27–38, 2010. age current,” in Proc. IEEE Asian Solid-State Circuits Conf., 2013,
[69] B. Razavi, Design of Integrated Circuits for Optical Communications. pp. 273–276.
New York: McGraw-Hill, 2003. [91] (2016). [Online]. Available: http://www.soterawireless.com/visi-
[70] K. N. Glaros and E. M. Drakakis, “A sub-mW fully-integrated pulse mobile/
oximeter front-end,” IEEE Trans. Biomed. Circuits Syst., vol. 7, no. 3, [92] (2016). [Online]. Available: http://somnomedics.eu/products/long-
pp. 363–375, 2013. term-blood-pressure-somnotouchtm-nibp/
[71] M. Alhawari, N. A. Albelooshi, and M. H. Perrott, “A 0.5 V ≤ 4 μW [93] (2016). [Online]. Available: https://www.simband.io/
CMOS light-to-digital converter based on a nonuniform quantizer for a [94] (2016). [Online]. Available: https://omronhealthcare.com/

26  IEEE circuits and systems magazine THIRD quarter 2018


Authorized licensed use limited to: Chulalongkorn University provided by UniNet. Downloaded on October 21,2024 at 09:03:19 UTC from IEEE Xplore. Restrictions apply.

You might also like