An Integrated Wearable Wireless Vital Signs Biosensor for … · 2021. 1. 9. · VySys, includes two compact wearable wireless biosensor devices for continuous vital signs capturing

448 IEEE SENSORS JOURNAL, VOL. 20, NO. 1, JANUARY 1, 2020

An Integrated Wearable Wireless VitalSigns Biosensor for Continuous

Inpatient MonitoringDavid Liang Tai Wong , Senior Member, IEEE, Jufeng Yu, Member, IEEE,

Yongfu Li, Senior Member, IEEE, Chacko John Deepu, Senior Member, IEEE,Duy Hoa Ngo, Chongyu Zhou, Member, IEEE, Shashi Raj Singh,

Alvin Koh, Rachel Hong, Bharadwaj Veeravalli , Senior Member, IEEE,Mehul Motani , Fellow, IEEE, Kee Chaing Chua, Yong Lian, Fellow, IEEE,

and Chun-Huat Heng , Senior Member, IEEE

Abstract—A compact, light-weight and low-power wirelessvital signs monitoring system based on Wireless Body AreaNetwork (WBAN) protocol has been developed. The system,VySys, includes two compact wearable wireless biosensordevices for continuous vital signs capturing and transmis-sion, a gateway to relay the message collected from thebiosensors to cloud, and finally a client apps to access anddisplay the stored data in the cloud. Both biosensor devicescan last for 24 hours and weigh less than 22 g and 44 g,respectively.They consist of proprietary in-house bio-sensingintegrated circuit (IC) and commercial off-the-shelf BluetoothLow Energy (BLE) module. VySys has been deployed inclinical trials with 14 subjects. From the studies, the accuracyand advantage of VySys are evaluated and the five vital signs captured (heart rate (HR), respiration rate (RR), temperature(TMP), oxygen saturation (SpO2) and systolic blood pressure (SBP)) are benchmarked against a commercial medical-grade device. The results show strong statistical correlation (r > 0.68). In terms of clinical significance, all its meandifference are within limits of accepted clinical discrepancies. In terms of efficacy by comparing against the best knownreported results, (1) VsSys is more precise by 28.2%, 36.2%, 70.0%, 37.6% and 34.4% for HR, RR, TMP, SpO2 and SBP,respectively and (2) has a narrower 95% limit of agreement (LoA) by 24.5%, 23.9%, 50.6%, 37.4% and 34.4% for HR, RR,TMP, SpO2 and SBP, respectively.

Index Terms— Wearable, low-power, wireless, sensor, aging population.

Manuscript received August 6, 2019; revised September 9, 2019;accepted September 9, 2019. Date of publication September 18, 2019;date of current version December 20, 2019. This work was supportedin part by the National Research Foundation, Singapore, under GrantNRF-CRP8-2011-01. The associate editor coordinating the review ofthis article and approving it for publication was Dr. Jürgen Kosel.(Corresponding author: Chun-Huat Heng.)

D. L. T. Wong, J. Yu, D. H. Ngo, C. Zhou, S. R. Singh, B. Veeravalli,M. Motani, K. C. Chua, and C.-H. Heng are with the Department ofElectrical and Computer Engineering, National University of Singapore,Singapore 117583 (e-mail: [email protected]).

Y. Li and Y. Lian are with the Department of Micro-Nano Electronicsand MoE Key Lab of Artificial Intelligence, Shanghai Jiao Tong University,Shanghai 200240, China.

C. J. Deepu is with the School of Electrical and Electronic Engineering,University College Dublin, Dublin 4, Ireland.

A. Koh and R. Hong are with the Medical Engineering Researchand Commercialization Initiative (MERCI), Department of Surgery,Yong Loo Lin School of Medicine, National University of Singapore,Singapore 119228.

Digital Object Identifier 10.1109/JSEN.2019.2942099

I. INTRODUCTION

DEVELOPED countries with ageing population are facingincreasing healthcare cost. Statistically, elderly are four

times more likely to get admitted and tend to stay twotimes longer than the younger population [1]–[3]. Continuousmonitoring of vital signs is crucial to assess the conditionof a patient and to avoid adverse events. This has been thenorm for patients under the intensive care unit (ICU). Even fornormal ward, it is common to monitor patient’s five vital signsduring every interval of 4 to 6 hours [4]. This is usually laborintensive, and accounts for about 30% of a nurse’s workload.

The advancement in wireless and sensor technology hasbrought an opportunity to mitigate the above-mentioned issue.In terms of sensing using fabric materials, [5], [6] have broughtabout techniques to measure vital signs using textile mate-rials embedded onto clothing which can potentially improve

1558-1748 © 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

Authorized licensed use limited to: Shanghai Jiaotong University. Downloaded on March 25,2020 at 08:53:59 UTC from IEEE Xplore. Restrictions apply.

https://orcid.org/0000-0002-0737-6719

https://orcid.org/0000-0001-9000-1813

https://orcid.org/0000-0003-3262-0207

https://orcid.org/0000-0002-5696-8403

WONG et al.: INTEGRATED WEARABLE WIRELESS VITAL SIGNS BIOSENSOR FOR CONTINUOUS INPATIENT MONITORING 449

patient compliance. However, they are limited in capturinga few vital signs such as heart rate (HR) and respirationrate (RR). In addition, these technologies are not very matureand are yet to appear in the commercial space. For commercialwearable light-weight wireless patches, the reported results areusually with fewer than two vital signs, such as HR and RRwhich are still insufficient to serve as a complete monitoringtool for inpatient [7]–[10].

There were also research attempts where certain vital signssuch as HR, oxygen saturation (SpO2), blood pressure (BP)and temperature (TMP) are derived from a single contin-uous sensor such as an accelerometer and PPG [11], [12]but only remained in the research settings. In recent papers,there were works to increase the number of vital signs thatcan be captured but it still lacks the complete five vitalsigns and it is bulky without wireless capabilities [13], [14].In [15], significant effort has been made to miniaturizean ‘all-in-one’ vital signs measurement device to reduceits size and weight. However, RR monitoring is absentand it is not suitable for wearable and continuous wirelessmonitoring.

There were also research done to increase the comfort ofpatients while being monitored at home by means of a wear-able wireless monitoring with specialized anomaly detectionbut they lack in providing all the five vital signs [16], [17].In [18], a complete five vital signs monitoring system was pro-posed but it lacks in terms of weight, dimension, continuousoperating duration and accuracy. In addition, the article gavea high-level overview of the wireless performance but madeno performance analysis or comments on data loss duringmultiple concurrent sensors transmission. Although they havebeen many reported compact and low power wireless sensordevice targeting for such application, we have yet seen a fullyintegrated, wearable continuous, light-weight, end-to-end fivevital signs system reported to date. There is also a lack ofstudies on the feasibility and effectiveness of such wirelesssystem.

In this work, we developed Vital Signs Systems (VsSys)to investigate the feasibility and effectiveness of such systemwhile addressing some limitations of the existing systems.VsSys is the most comprehensive and light-weight end-to-endwireless vital signs monitoring platform reported to date.It provides up to five real-time vital signs monitoring, i.e.electrocardiogram (ECG) with HR, RR, TMP, SpO2 andderived systolic blood pressure (SBP). It leverages on cloudinfrastructure to enable centralization of data storage and easydata access on demand. More importantly, client apps withgraphical user interface (GUI) are tailored based on the inputof healthcare personnel.

To target for wearability and long hour of continuous mon-itoring, the wireless wearable biosensors are miniaturized andoptimized for low-power consumption. This is made feasiblewith our own proprietary low-power bio-medical integratedcircuit (IC) [19]–[25], which can provide continuous 3-leadECG and RR monitoring at nano-watt level. In addition, cus-tomized data packet and protocol are employed to maximizethe energy efficiency of the adopted Bluetooth Low Energy(BLE) module.

Fig. 1. End-to-end wearable vital signs monitoring system consistingof the wearable biosensors worn on a subject, the gateway, Cloud withAnalytics for signal processing and Client Apps for displaying real-timewaveform and vital signs.

In this paper, we will report the results obtained using thedeveloped system through clinical trials on 14 subjects. Theseresults are benchmarked against a commercial instrument, andshine light on its feasibility and effectiveness.

The contributions of this work are summarized as follows:1. Design and implementation of a light-weight, low-

power, wireless end-to-end continuous five vital signsmonitoring system supporting up to 24 hours.

2. Reported on the efficacy or accuracy of the wirelessend-to-end continuous monitoring system for 14 healthysubjects.

3. Reported on the robustness of the wireless end-to-endsystem under 24-hour continuous and simultaneoustransmission for 4 subjects.

4. Comparison methodology at 12-second interval enablinglarger measurement pairs to be captured in contrast tomanual methods of data collection.

5. BLE configuration of the connection interval and slavelatency while buffering and packing of raw signalson application data frame before transmission enablingreliable continuous transmission and power reduction onthe biosensors.

In Section II, we introduce the system architecture forthe overall end-to-end integrated system of VsSys. This isfollowed by Section III where we discuss the clinical trialsetup and its results. In Section IV, we compare the resultsof VySys against other devices, the effects of motion artifactsand the trends of wearable sensors. Finally, we conclude inSection V.

II. SYSTEM ARCHITECTURE

VsSys consists of three key components, i.e. wirelessvital signs biosensors for data capture, gateway and cloudinfrastructure for data storage and analytics processing, andclient apps for displaying clinically vital information as shownin Fig. 1.

A. Biosensors

There are two wearable biosensors for vital signs monitoringas shown in Fig. 2: 1) NUS_TRE which captures ECG, RR



Fig. 2. Our wearable wireless device worn on the body The NUS_TREis attached onto the chest using 6 electrodes while the PPG (SpO2)biosensor is clipped onto the finger.

TABLE ISPECIFICATIONS OF NUS_TRE

and TMP signals through on-body electrodes; 2) NUS_PPGwhich captures photoplethysmography (PPG) signal throughfinger clips.

The NUS_TRE has 6 snap-in electrodes. Four of them areused for 3-lead ECG, positioned at the left-arm (LA), right-arm(RA), left-leg (LL), and right-leg (RL) electrodes. The othertwo are used for RR, positioned across the chest at LL and RA,respectively. Proprietary low-power integrated 2-channel ECGand 2-channel bio-impedance (IMP) analog front end (AFE)chip [19]–[25] is used to amplify and digitize these capturedsignals. The acquired data is then sent to a microcontrollerunit (MCU), CC2540, via serial peripheral interface (SPI).

For body TMP sensing, the NUS_TRE employs an externalthermistor which is placed under the armpit of test subject. Theresulting voltage information is then captured by an analog-to-digital converter (ADC) embedded within the MCU.

The MCU also control a BLE transceiver to link up thebiosensor with the gateway for data transmission. Designefforts through component selection and printed circuit board(PCB) design optimization have led to device miniaturizationdown to 32×2.9×0.9 mm as shown in Fig. 3. The detaileddesign specifications are listed in Table I.

Fig. 3. Low-power 2 channel ECG and bio-impedance AFE die layout(top) contained at the back of PCB of our wearable wireless NUS_TRE.The front contains the temperature circuit, BLE System-On-Chip (SoC),wireless antenna design and AFE interfaces.

Fig. 4. The front and back view of our PCB integrating the PPG clip-onsensor. The front contains the AFE to control the LEDs, BLE modulefor data transmission and the MCU for overall device coordination. Thecomponents on the front is powered by the power regulator at the backof the PCB.

The NUS_PPG employs transmissive light-based technol-ogy to measure the rate of blood volume changes and oxygensaturation (SpO2). A finger is placed between two emittingLED diodes (red and infrared) and a receiving photodiode.The MCU then controls the LEDs emission, and samplesthe received illumination intensity of the photodiode [26].Similarly, a BLE transceiver is used to send the data to thegateway as shown in Fig. 4. The detailed specifications arelisted in Table II.

Wireless module consumes the most energy of the wholesystem. Hence, the data packet and wireless communicationprotocol of the BLE module employed have been redesignedto achieve low power consumption with high data rate trans-mission. Much design and implementation effort are put intotuning the BLE’s generic access profile (GAP) to attain a suit-



TABLE IISPECIFICATIONS OF NUS_PPG

able trade-off between the power consumption and robustnessduring each rapid processing and transmission cycles. In addi-tion, at the BLE’s generic attributes (GATT) layer, we madethe data packaging customizable to support additional through-put, making it easier to incorporate more sensor data inthe future. The connection interval and slave latency in theGATT are configured to 7.5 ms and 0 ms, respectively. Theconnection interval is set to the lowest supported value andno connection events can be skipped enabling a more reliabletransmission at the cost of increased power consumption.To reduce the power consumption, the MCU buffered theacquired raw sensor data and packed 20 bytes (maximumallowable application data frame for BLE) before the physicalwireless transmission. This has led to about 43.7% powerreduction. The resulting biosensors can operate continuouslyfor about 24 hours with a single battery charge.

B. Gateway and Cloud With Analytics

The gateway is employed as a conduit between the biosen-sors to the Cloud [27]. To conserve the biosensor powerconsumption, it is unrealistic to expect long distance withcontinuous transmission from the biosensor device. The low-power wide-area network (LPWAN) such as LoRA and Sig-fox, supports low-power and long-range transmission but itis unable to support the bandwidth required for stream-ing continuous raw signals due to its inherent low datarate [28], [29]. After much investigation, the low-powerBLE interface with its ability to support higher data rate at ashorter distance is chosen as the wireless interface between thebiosensor and the gateway as shown in Fig. 1. The gateway andthe Cloud interface can be connected via Ethernet or Wi-Fi.Each subject, wearing a NUS_TRE and NUS_PPG, is pairedto a specific gateway. The gateway can come in the formof a laptop, smart phone or embedded computer such asRaspberry PI. The design rationale of one subject paired toone gateway enables: 1) Stability: computational workload onthe gateway is predictable and tractable leading to timelinessand better wireless reliability. 2) Robustness: failure of asingle gateway is contained to only a single subject and doesnot affect two or more subjects. 3) Mobility: if the gatewayis a portable smart phone, the subject can dock or takethe gateway along to experience seamless continuousmonitoring.

The Cloud should be able to receive, store, process andretrieve the biosensor data. Given the processing power

of Cloud, it makes sense to incorporate signal processingalgorithms to extract key biometrics for the biosensor data,such as HR, RR, TMP, and SpO2. Signal processing algorithmscan also be employed to mitigate the commonly observedmotion artifacts.

The HR is calculated based on the time interval betweentwo successive R-R peaks in the ECG waveform. Techniquessimilar to [20] have been employed, such as smoothing filter,integration, thresholding, and averaging. The RR is calculatedbased on the time interval between two successive crests inthe IMP waveform. Eight successive intervals are observed toaverage the RR. The SpO2 is calculated based on the modula-tion ratio between the received red and infrared signals of thePPG waveform. We used the Contec MS100 Simulator alongwith a high dependency (HD) medical grade predicate (gold-standard) device to calibrate our SpO2 through a modulationplot [26], [30]. The relationship between the measured ratio tothe corresponding SpO2 level is stored in the Cloud as a look-up table (LUT) to ease the data extraction by the client apps.Similarly, calibrated thermistor reading to the correspondingTMP is stored as LUT to extract the correct TMP reading forclient apps [31].

C. Client Apps

The client apps are tailor-made based on the survey resultscollected from hospital healthcare personnel. This is critical asit shows very similar user interface as existing hospital medicalinstrument, and also incorporates features that are considereduseful by the main user. For example, it provides both wardview and bed view as shown in Fig. 5. The overall wardview summarized the key vital signs and biometrics for allpatients while the individual bed view displays more detailedand enlarged vital signs for a specific patient. Furthermore,it also allows query for past collected data from the Cloudwhich is helpful for diagnosis.

III. CLINICAL RESULTS

Our biosensors have undergone preliminary (pre-scan)IEC medical safety test (IEC60601-1), electromagnetic dis-turbance and radiation immunity test (IEC60601-1-2 andIEC61000-4-2). The IEC60601-1-2 recommends CISPR11 standard for electronic emission test on industrial scientificand medical (ISM) devices, and our biosensors have passedthe emission test [32]. The rationale of the emission test is toensure that the biosensors do not generate unwanted emissionabove a certain threshold as this could potentially disrupt theoperation of the existing medical devices in the hospital. Twotrials, i.e. controlled clinical trial and the extended clinicaltrial, using our developed system were carried out afterobtaining ethics approval from institutional review board (IRB)and domain specific review board (DSRB) approval underthe reference number IRB–B-16-264 and 2017/00605, respec-tively. After the clinical trials, we evaluated the data collec-tion overhead and robustness of wireless packet transmission.We also benchmarked the efficacy of VsSys.



Fig. 5. Client Apps – a) Ward view (Top): Overview of 6 patientview where 4 slots are active where the top most row shows the real-time ECG data followed by the bio-impedance and the PPG waveform.b) Bed view (Bottom): Enlarged real-time plot of ECG, bio-impedanceand PPG alongside with HR, RR, TMP and SpO2 for a single subject.

A. Controlled Clinical Trial

The objectives of the controlled clinical trial are to evaluatethe safety of the biosensor device, efficacy and the robustnessof the wireless vital sign monitoring system. Besides sub-jecting our biosensor to IEC test as mentioned earlier, thebiosensor is also closely monitored for its warmness and wear-ability throughout the trial. In terms of efficacy, the measuredresults from our system are compared against the predicatedevice used in the hospital. In terms of robustness, the gatewayand cloud are closely monitored to observe any drop-in datapacket. In this trial, a total of 10 participants were recruitedand monitored continuously for 8 hours. With 2 participantsinvolved each day, the whole trial duration lasted for 5 days.The details of the cohort are tabulated in Table III.

To evaluate the robustness of the wireless communicationlink, participants were allowed to leave for toilet and lunchbreaks. Upon returning, it was found that our biosensors canautomatically be probed to re-establish the BLE connectionwithout any human intervention. The reliable communicationrange of the biosensors with the gateway is also established tobe around 3 meters. Beyond which, the packet loss becomesintolerable. In addition, to mitigate the wireless interferenceissue due to the crowded spectrum at 2.4-GHz industry,

TABLE IIISUBJECT IN CONTROLLED CLINICAL TRIAL

Fig. 6. Clinical trial in a ward setting for 4 healthy subjects for 24 hoursat the investigational medicine unit (IMU) of NUS. Subjects were restedon the bed positioned at Head of Bed (HOB) 45◦ throughout the trial.

TABLE IVSUBJECT IN EXTENDED CLINICAL TRIAL

ISM radio band, we have configured the Wi-Fi router tooperate at 5-GHz ISM band.

B. Extended Clinical Trial

After the controlled clinical trial on healthy subjects,we proceed with the extended clinical trial under hospital wardenvironment as shown in Fig. 6. Similar objectives such asefficacy and robustness are evaluated in this trial. However,the trial duration has been extended to 24 hours to furtherstress on the system’s performance (biosensor, gateway, cloudand client apps) of the end-to-end system. The cohort sizewas 4 and the statistics are summarized in Table IV. Unlike thecontrolled clinical trial, a medical practitioner is present onsiteto access the subject’s eligibility to wear our biosensor for longhour. In addition, 4 subjects are on trial concurrently instead



TABLE VPERCENTAGE OF DATA LOSS

of 2 to evaluate the capacity of our system. From the trial, allthe biosensors are found to operate beyond 24 hours, and noserious adverse events have been observed during the trial.

C. Data Collection Overhead

For manual vital signs monitoring, it takes a nurse roughly10 minutes for each patient, and is repeated every 4 hours.This accounts for around 30% of her time. In comparison,the VySys requires only a one-time installation of about10 minutes per patient, which results in significant time savingfor such monitoring activities.

D. Robustness of the Wireless End-to-End

Wireless based monitoring always raise concerns on therobustness of the wireless link, which could incur critical dataloss. In our studies, we investigated the data loss by trackingthe data packet sequence number incorporated within eachtransmission packet. The results of the data loss based oneach subject during the 24-hour extended clinical trial aretabulated in Table V. The ‘30 Mins’ refers to a 30 minutedata capture that is used for benchmarking for the first foursubjects in the next sub-section. The results of the dataloss for entire trial is depicted in the ‘24 Hours’ row. The24-hour data loss is shown in Fig. 7. The mean data loss is0.0657%, 0.0496% and 0.0112% for ECG, IMP and PPG,respectively. For ECG, this accounts to about a single sampleloss for every 6 complete ECG cycles with HR of 60 bpm(beats per minute) which is considered to be negligible forHR estimation. For IMP, this accounts to about two sampleloss for 5 complete breathing cycle of 20 brpm (breathsper minute) which is considered to be negligible. For PPG,the loss accounts to about a single sample loss every 70seconds which is negligible for SpO2 calculation. The shorterdata length contributes to the lower data loss in NUS_PPG.TMP has no data loss throughout the whole trial. This ismainly because the frequency of TMP transmission is at least1000 times lesser in comparison to ECG and IMP allowingmore opportunities for re-transmission in case data loss occur.

Furthermore, BLE adopts the contention-less frequencyhopping spread spectrum (FHSS) technology in the mediumaccess control (MAC) layer, where each sensor node is

Fig. 7. Packet Loss Rate – a) NUS_TRE (Top): ECG packet loss for24 hours for the 4 subjects. b) NUS_PPG (Bottom): PPG packet lossfor 24 hours for the 4 subjects. The data loss is mainly contributed bycollision, obstruction or signal attenuation due to distance between thebiosensor and the gateway.

assigned a hopping pattern [33]. Collision, potentially leadingto data loss, can happen when two or more biosensorsconcurrently transmits on the same frequency band. Withan increasing number of biosensors in the same location orvicinity, there will be a higher probability of collision. Anyblockage of transmission between the biosensor device andthe gateway could also lead to data loss. In addition, the weakwireless coverage between the biosensor and gateway duringin-ward toilet breaks has also contributed to data loss.

E. Benchmark and Efficacy

As mentioned earlier, it is important to evaluate the effi-cacy of VySys by benchmarking our results against predicatedevice. We chose a 30-minute period for the comparison,which statistically provides sufficient number of measurementpairs from both VsSys and the predicate, and is more thanmost clinical evaluation [13]–[15], [18]. In this evaluation,we employ different techniques such as overlaying vital signsfrom different devices, and statistical methods.

For the first approach, 2 complete ECG cycles of LeadII from both VsSys and the predicate device are overlaidtogether as shown in Fig. 8. The observed difference isinsignificant, and the subtle morphological differences couldbe due to the small difference in electrode placement. Theprominent PQRST waveform can be clearly observed for bothcases. The small differences between them did not impact theHR estimation.

As the predicate device also produces 12 seconds intervalof HR, RR, TMP, SpO2 in the report, we also overlay themwith the results collected from VsSys for a combined periodof 30 minutes (150 time units ∗ 12 seconds) as shown



Fig. 8. Overlaying our ECG waveform against the predicate device forthe 4 subjects.

Fig. 9. Overlaying HR, RR, TMP and SpO2 against predicate overa period of 30 minutes where each time instance T is a multipleof 12 seconds. ‘bpm’ and ‘brpm’ denotes beats per minute and breathsper minute, respectively.

in Fig. 9. Similarly, there is a high degree of correlation forthe signals with similar trending patterns. This includes HR(top-left of Fig. 9) ranging from about 60 to 95 bpm indicatingthe heart rate variability (HRV), RR (top-right of Fig. 9)ranging from 12 to 23 brpm, TMP (bottom-left of Fig. 9)ranging from about 36.7 ◦C to 36.9 ◦C, SpO2 (bottom-right) ranging from 98% to 100%. The HRV is due to theregulation of the autonomic nervous system [34].

For the statistical method, we used scatter plot to visu-alise the relationship between the measurement pairs (thepredicate is represented on the x-axis and VsSys on they-axis) and to estimate the correlation coefficient, r, whichdetermines the closeness of the measurement pairs betweenboth devices. Referring to the Taylor’s systems [35], r values< 0.35 are categorized as weak correlation, 0.36 to 0.67 asmoderate correlation and 0.68 to 0.89 as strong correlation andvalues > 0.9 as very high correlation.

However, a strong correlation alone does not mean goodagreement on the measurements between both devices. Hence,we also applied the Bland-Altman analysis [36], [37] toobtain two levels of gauge, which are the mean differenceand the 95% limits of agreement (LoA) for each vital sign.Mean difference is defined as the average measurement dif-ference between the measurement pairs. The mean differenceshows how bias VsSys is against the predicate, where a

Fig. 10. The scatter and Bland-Altman plot of HR measured among14 subjects for a period of 30 minutes. ‘bpm’ denotes beats per minute.

positive/negative value indicates that VySys produced readingsthat are higher/lower than the predicate. The LoA tells usthat 95% of the data are within the specified range of themean difference, indicating the certainty of the deviation.The LoA is composed of the upper bound (+1.96SD) andthe lower bound (−1.96SD) of the mean difference. Theupper/bound is obtained by adding the mean difference withthe multiplication of positive/negative 1.96 by its standarddeviation, respectively. We compare the results obtained fromthe Bland-Altman against the acceptable discrepancies metricused in various clinical trials as tabulated in Table VI [18].Vital signs within the acceptable discrepancies are deemedto be clinical significant. The scatter and the Bland-Altmanplot of 14 subjects for HR, RR, TMP, SpO2 and SBP fromboth the controlled and extended clinical trial are shownin Figs. 10 to 14, respectively. The resolutions in the scatterand Bland-Altman plot appears discrete when the measure-ment pairs are plotted directly, given that the predicate deviceproduces vital signs resolution in whole number (except forTMP in one decimal point). As such, both the measurementpairs were randomized via a uniform de-rounding distributionfunction to its desired decimal point (HR, RR, TMP and SpO2to one decimal place and TMP to two decimal places). Forexample, the RR value of 15 brpm can be randomized and takea value from 14.5 to 15.4 brpm and TMP value of 36.5 ◦C canbe randomized and take a value from 36.45 ◦C to 36.54 ◦C.In Figs. 10 to 14, ‘n’ is defined as the sum of an equal numberof measurement pairs from each subject for 30 minutes at12 seconds interval, ‘r’ is defined as the correlation coefficient.In Fig. 10, the HR has 2100 measurement pairs with a veryhigh correlation score of 0.98, one standard deviation of±1.58 and mean difference of −0.37 with LoA of −3.48 to2.73 bpm. The RR has 2100 measurement pairs with a strongcorrelation score of 0.80, one standard deviation of ±2.19 andmean difference of −0.67 with LoA of −4.97 to 3.63 brpm asshown in Fig. 11. The TMP has 2100 measurement pairs withcluttered measurement pairs since there are minimal variationof TMP within the 30 minute period as shown in Fig. 12.In Fig. 13, the SpO2 has 2100 measurement pairs within therange from 92% to 100%.

In Fig. 14, for SBP, we have limited blood pressure pointsfrom the predicate device as it takes about 3 minutes forgenerating a single value and the subject may not feel comfort-able using the cuff-based predicate device throughout the trial.



Fig. 11. The scatter and Bland-Altman plot of RR measured among14 subjects for a period of 30 minutes. ‘brpm’ denotes breaths per minute.

Fig. 12. The scatter and Bland-Altman plot of TMP measured among14 subjects for a period of 30 minutes.

Fig. 13. The scatter and Bland-Altman plot of SpO2 measured among14 subjects for a period of 30 minutes.

In literature, it is reported that the systolic blood pres-sure (SBP) is correlated with the pulse transit time (PTT)approach [38], [39]. Based on the collected ECG and PPGdata, we model the SBP by extracting the PTT based onthe time difference between the R-peak of the ECG andthe PPG. Given the limited points available for analysis, weperform 5-fold cross-validation over 280 samples so that wecan reuse the data that has been used for the test. The SBPhas a strong correlation score of 0.78, one standard deviationof ±7.87 and mean difference of 0.61 with a LoA of −14.81to 16.03 mm Hg.

The overall numerical metrics of the plots for each vitalsigns are summarized in Table VII. ‘n’ is defined as the total

Fig. 14. The scatter and Bland-Altman plot of SBP measured among14 subjects for 280 points.

TABLE VIACCEPTED DISCREPANCIES BETWEEN OUR

DEVICE AND PREDICATE [18], [15]

TABLE VIISTATISTICAL SUMMARY FOR COMPARED VITAL SIGNS

number of measurement pairs, ‘r’ is the correlation coefficient,‘±1SD’ is the one standard deviation variation from the mean,‘md’ is the mean difference, ‘ULoA’ and ‘LLoA’ is the upperand lower bound of LoA, respectively. In terms of efficacy withregards to the correlation r, all five vital signs possess strongcorrelation against the predicate results. In terms of clinicalsignificance with regards to Table VI, the mean differenceand one standard deviation of all vital signs are well withinlimits except for RR and SBP’s one standard deviation whichis deviated by 0.19 brpm and 2.87 mm Hg, respectively. TheLoA of HR, TMP and SpO2 are also well within limits interms of clinical significance. However, the upper and lowerLoA of RR and SBP are deviated by a margin of [−2.97, 1.63]and [−9.81, 11.03], respectively.

IV. DISCUSSION

A. Performance Evaluation

We investigated the feasibility and effectiveness of continu-ous monitoring using VySys for a cohort of healthy subjects in



Fig. 15. The impact of ambulatory movement during continuous monitoring on various actions ((a) rest, (b) walk, (c) run and (d) rest). The topdiagram shows the vital signs trend for a duration of 100 time units (20 minutes) and the bottom diagram shows a snapshot of the continuous rawsignal for 3072 samples (12 seconds). The action ‘(a) rest’ occurs from about 0 to 25 time units, ‘(b) walk’ occurs from about 25 to 50 time units,‘(c) run’ occurs from about 50 to 75 time units and ‘(d) rest’ from about 75 to 100 time units. The instantaneous vital signs from the vital signs trendat 13, 38, 63 and 88 time units are intersected by the blue, red, green and yellow lines, respectively. Each of these line points to its correspondingcontinuous vital signs as indicated by the blue, red, green and yellow boxes, respectively. Within each box, the ECG, IMP and PPG is representedby blue, red and green, respectively.

clinical trial settings. Table VIII is an extension of Table VII,and it shows the performance of VySys against other worksthat were introduced in the earlier literature survey. FromTable VIII, VsSys has demonstrated improved efficacy com-pared to other reported result, especially in one standard devi-ation (precision) and a narrower LoA. In terms of performanceefficacy against the best reported results in literature, VsSys ismore precise by 28.2%, 36.2%, 70.0%, 37.6% and 34.4% forHR, RR, TMP, SpO2 and SBP, respectively and a narrowerLoA by 24.5%, 23.9%, 50.6%, 37.4% and 34.4% for HR,RR, TMP, SpO2 and SBP, respectively. The improved efficacyis contributed by several factors such as low-noise signalacquisition of sensor, analytics and also our methodology thatcaptures the measurement pairs at similar time interval in

contrast to cases of several minutes of lapse in certain trialsdue to manual recording [15]. Furthermore, inter and intra-observer variability were eliminated since we captured thecontinuous raw signals to derive the measurements of the vitalsigns. In addition, measurements done at a shorter intervalof 12 seconds can provide better care and surveillance asvital signs can change drastically for patients under criticalconditions [40]–[42]. The comparison of sensor specificationsas listed in Table VIII are shown in Table IX. The parameterfor comparison are ‘weight’ which refers to the total weight ingrams, ‘DIM’ which refers to the dimension in terms of length(L), width (W) and height (H) in mm, ‘Hours’ which refers tothe duration (number of hours) of continuous operation, ‘Wear-able’ indicates if the sensor can be worn during monitoring,



TABLE VIIIPERFORMANCE COMPARISON

TABLE IXSPECIFICATIONS COMPARISON

‘CONT Raw’ indicates if the sensor could transmit continuousraw signals such as ECG, ‘IoT Capable’ indicates if thesensor could be connected wirelessly to the internet, ‘5 VitalSigns’ indicates if the system could provide five completevital signs during monitoring and ‘HR, RR, TMP, SpO2 and

SBP’ indicates which type of vital signs are supported bythe system. From the comparison, VySys is a relatively light-weight wearable wireless five vital signs system while beingable to achieve 24 hours of continuous operation with Internetof Things (IoT) capabilities. The closest matching continuouswearable wireless five vital signs that we could find is [18]where VsSys excels in terms of weight, dimension, durationof continuous operation and accuracy.

B. Limitations

A limitation of this study is the relatively low number ofsubjects. However, given that this was a preliminary feasibilitystudy, further clinical evaluation with a significant numberof subjects can be carried out to re-affirm the results. It isestimated that at least 385 subjects are required using theCochran’s sample size formula with z-score of 1.96, popu-lation proportion of 0.5 and margin of error at 0.05 [43].

C. Performance During Ambulatory

In the clinical trials, subjects were monitored at rest and wewere unable to make inference concerning the performancewith patients during ambulatory motion. We made a quickstudy on the impact of ambulatory artifacts on the vital signswhich are derived from the continuous raw signals (ECG, IMPand PPG). These continuous raw signals are time-aligned usingthe timestamp and signal with sampling rate below 256 Hzhas been up-sampled to 256 Hz. We introduced three actions,which are ‘rest’, ‘walk’ and ‘run’ to observe the accuracyand changes of the vital signs as shown in Fig. 15. The ‘rest’involves the subject to be sitting on a chair and remaining still.This is used as a baseline where the continuous raw signalsis free from motion artifacts. The ‘walk’ involves the subjectto be walking and ‘run’ involves the subject to be running.Each action will be performed for about 5 minutes startingwith ‘(a) rest’ followed by ‘(b) walk’, then ‘(c) run’ and thenback to ‘(d) rest’. The purpose of these sequence of actions isto induce an increasing activity and then back to stationary.

Several observations can be deduced from these plots.The HR trend of VySys is similar to the predicate. This ismainly because the R-R peaks in the ECG waveform is stillprominent and can be easily detected by our algorithm even innoisy and baseline wandering waveform. It is also observedthat the HR trend increases when the subject increases theamount of activity and begins to decrease when the subject isat ‘(d) rest’. For TMP, the trend between the predicate andVySys is correlated. It can be seen that there is a gradualincrease in TMP as the subject transit from ‘(a) rest’ to‘(b) walk’, ‘(b) walk’ to ‘(c) run’ and a gradual decrease from‘(c) run’ to ‘(d) walk’. It is also noted that the TMP is immuneto motion artifact for all cases. For SpO2, the PPG waveformappears to be oscillating more as the movement increases incontrast to ‘(a) rest’ and ‘(d) rest’. Even though the SpO2trend remains at 100% most of the time, the slight decreaseand increase in SpO2 at around the 40th time units to the65th time units shows that there exists correlation trendeven during increasing activity. The accuracy and correla-tion remains good during increasing activity because the



Fig. 16. The impact of artifacts, i.e. ‘coughing’ (top-left), ‘yawning’ (top-right), ‘talking’ (bottom-left) and ‘sneezing’ (bottom-right), on the continuousraw signals while the subject is stationary. The x-axis is represented by the sample number ranging from 0 to 8447 samples (33 seconds). ‘Coughing’happens from around 3000 samples to 7000 samples. ‘Yawning’ happens from around 3000 samples to 8000 samples. ‘Talking’ happens from around3000 samples to 8448 samples. ‘Sneezing’ happens from around 2500 samples to 8448 samples. Each red arrow indicates the approximate locationwhen the artifact happened.

modulation ratio derived from the PPG signals is independentof the oscillation in the waveform. However, there will be animpact on deriving pulse rate using the PPG peaks as the peaksof the PPG waveform (which represents the systolic beat ofthe heart) can be artificially induced by the footsteps madeduring movement.

The RR from VySys is highly correlated to the predicateduring both ‘(a) rest’ and ‘(d) rest’ as shown in Fig. 15.During ‘(b) walk’ and ‘(c) run’, the raw IMP signal appearsto be oscillating more frequently and the quick rising andfalling pattern (high-frequency component) forming the crestbecome more prominent during ‘(c) run’ as each crest followsthe footsteps. The pickup of footsteps component in the IMPwaveform is mainly due to change of IMP between theelectrode contact and the skin. This means both VySys andpredicate which uses the IMP method are not immune tofootsteps. The RR difference between the predicate and VsSysduring footsteps is mainly due to the sensitivity in generatingthe IMP waveform and the algorithm used in detecting thecrest of the IMP waveform. A better-known method to validate

RR during ambulatory is to use the capnography method as itpossesses better immunity against motion artifacts [7].

D. Performance During Stationary

We also performed another experiment to evaluate the signalacquisition and algorithm performance when the subject is sta-tionary while ‘coughing’, ‘yawning’, ‘talking’ and ‘sneezing’.The waveform of these continuous raw signals subjected tothese actions is shown in Fig. 16. The first to the fourth row ineach sub-figure represent the ECG amplitude, IMP amplitude,TMP and PPG amplitude, respectively. The x-axis, with aduration of 33 seconds, is represented by the sample numberranging from 0 to 8447 samples. The artifacts were introducedaround the 3000th sample onwards and the red arrows in eachfigure mark a more precise starting location of each artifact.From these figures, we noticed that ECG, IMP and PPG signalsare vulnerable to noise artifacts. Our analytics is still able todetect all R peaks of the ECG as indicated by the asteriskwhen subjected to these noise artifacts. This can be seen inFig. 16 ‘Coughing’, at the 2nd, 3rd and 4th cough where the



additional spikes that resemble the R peaks of the ECG werenot detected, thus reducing false positives rates.

We also observed that the ECG signals are corrupted bymuscle noise (electromyography (EMG)) except for ‘talking’.The EMG in the ECG can be seen as the darkening of theECG signals and affect the morphology of the ECG waveform.Given that EMG is a wideband noise, filtering techniquessuch as Kalman filter is an effective method to remove suchtypes of artifacts [44]. Furthermore, all IMP crest can also bedetected as indicated by the asterisk on the IMP waveform.There are no significant changes in the TMP signal. However,the PPG waveform appears to experience abrupt baseline driftduring ‘coughing’ and ‘sneezing’ and this is also reflected inits corresponding IMP signals. Looking at the continuous rawsignal on each sub-figure of Fig. 16, classification of varioususer action using machine learning techniques can be furtherexplored.

E. Trends

With regards to the trend on vital signs monitoring,we observed that the interest in electric and fiber-optic sensorsembedded into textiles [5], [6] seems to be decreasing in recentyears due to the need to have a wide range of apparel or beltshapes to adjust to the body. These sensors are now beingreplaced by patch monitors [9], [10] as they provide reliabilityin signal acquisition and ease in setup. In addition, non-contactsensors embedded into a bed or seat including direct skin-contact sensors [45] that do not require additional steps inthe preparation for monitoring are also gaining traction andappealing given the minimal disruption on the user’s activity.

For future works, VsSys can also be extended to enableother healthcare application. This includes preventive health-care through wide-scale deployment in a community or homecare. The massive vital signs data collected through ourinnovative system would lead to a more affordable and betterhealthcare system. Future works can incorporate the use ofthe five vital signs into the modified early warning scoreto assists nurses and detecting patients with physiologicaldeterioration [15], [18].

V. CONCLUSION

Our VsSys features a comprehensive in-house clinical grade∼66g continuous wireless low-power wearable biosensors(3-lead ECG, RR, TMP, SpO2 and SBP) capable of 24-houroperation. The results from the system show strong correlation,within clinical accepted discrepancies for its mean differenceand a narrower LoA for all vital signs when compared againstreported results. This study proves the feasibility of employingwireless vital signs monitoring system to improve the produc-tivity of hospital staff.

ACKNOWLEDGMENT

The authors would like to thank the subjects who partici-pated during the clinical trials for their time. Dr. Lim Yinghaofor the administrative works and supporting our clinical trial.They also like to thank IMU of Yong Loo Lin School ofMedicine, National University of Singapore for reserving time

and space for us to conduct recce at the site prior to the actualextended clinical trial.

REFERENCES

[1] E. M. Crimmins, “Lifespan and healthspan: Past, present, and promise,”Gerontologist, vol. 55, no. 6, pp. 901–911, 2015.

[2] R. Suzman and J. Beard, “Global health and aging,” NIH, vol. 1, no. 4,pp. 273–277, 2011. [Online]. Available: https://www.who.int/ageing/publications/global_health.pdf

[3] J. L. Dieleman et al., “Trends in future health financing and cov-erage: Future health spending and universal health coverage in 188countries, 2016–40,” Lancet, vol. 391, no. 10132, pp. 1783–1798,May 2018.

[4] S. P. McGrath, I. M. Perreard, M. D. Garland, K. A. Converse,and T. A. Mackenzie, “Improving patient safety and clinician work-flow in the general care setting with enhanced surveillance monitor-ing,” IEEE J. Biomed. Health Inform., vol. 23, no. 2, pp. 857–866,Mar. 2019.

[5] Y. Koyama, M. Nishiyama, and K. Watanabe, “Smart textile usinghetero-core optical fiber for heartbeat and respiration monitoring,” IEEESensors J., vol. 18, no. 15, pp. 6175–6180, Aug. 2018.

[6] X. Yang et al., “Textile fiber optic microbend sensor used for heart-beat and respiration monitoring,” IEEE Sensors J., vol. 15, no. 2,pp. 757–761, Feb. 2015.

[7] M. J. M. Breteler et al., “Reliability of wireless monitoring using awearable patch sensor in high-risk surgical patients at a step-down unitin The Netherlands: A clinical validation study,” BMJ Open, vol. 8,no. 2, Feb. 2018, Art. no. e020162.

[8] M. Hernandez-Silveira et al., “Assessment of the feasibility of anultra-low power, wireless digital patch for the continuous ambulatorymonitoring of vital signs,” BMJ Open, vol. 5, no. 5, May 2015,Art. no. e006606.

[9] S. S. Lobodzinski and M. M. Laks, “New devices for very long-termECG monitoring,” Cardiol. J., vol. 19, no. 2, pp. 210–214, 2012.

[10] S. S. Lobodzinski, “ECG patch monitors for assessment of cardiacrhythm abnormalities,” Progr. Cardiovascular Diseases, vol. 56, no. 2,pp. 224–229, Sep./Oct. 2013.

[11] J. Wannenburg and R. Malekian, “Body sensor network for mobilehealth monitoring, a diagnosis and anticipating system,” IEEE Sensors J.,vol. 15, no. 12, pp. 6839–6852, Dec. 2015.

[12] T. Elfaramawy, C. L. Fall, S. Arab, M. Morissette, F. Lellouche, andB. Gosselin, “A wireless respiratory monitoring system using a wearablepatch sensor network,” IEEE Sensors J., vol. 19, no. 2, pp. 650–657,Jan. 2019.

[13] G. Sun et al., “Vital-SCOPE: Design and evaluation of a smart vital signmonitor for simultaneous measurement of pulse rate, respiratory rate,and body temperature for patient monitoring,” J. Sensors, vol. 2018,Feb. 2018, Art. no. 4371872.

[14] W. S. Renier et al., “Analytical accuracy of the handheld PICO moni-toring device during emergencies,” BMJ Innov., vol. 5, no. 1, pp. 20–27,Jan. 2019.

[15] M. Weenk, H. Van Goor, M. Van Acht, L. J. L. P. G. Engelen,T. H. Van De Belt, and S. J. H. Bredie, “A smart all-in-one device tomeasure vital signs in admitted patients,” PLoS ONE, vol. 13, no. 2,2018, Art. no. e0190138.

[16] D. L. T. Wong, S. M. Sathappan, Kumaran, J. Yu, C. H. Heng,P. Kojodjojo, and M. Feng, “Continuous ECG monitoring trial foroutpatient—Patient receptiveness and signal accuracy,” in Proc. 41stAnnu. Int. Conf. IEEE Eng. Med. Biol. Soc., Jul. 2019, pp. 1–5.

[17] M. A. Rosenberg, M. Samuel, A. Thosani, and P. J. Zimetbaum, “Useof a noninvasive continuous monitoring device in the management ofatrial fibrillation: A pilot study,” Pacing Clin. Electrophysiol., vol. 36,no. 3, pp. 328–333, Mar. 2013.

[18] M. Weenk et al., “Continuous monitoring of vital signs using wearabledevices on the general ward: Pilot study,” JMIR mHealth uHealth, vol. 5,no. 7, p. e91, 2017.

[19] C. J. Deepu et al., “An ECG-on-chip with 535 nW/channel integratedlossless data compressor for wireless sensors,” IEEE J. Solid-StateCircuits, vol. 49, no. 11, pp. 2435–2448, Nov. 2014.

[20] C. J. Deepu, X. Y. Zhang, D. L. T. Wong, and Y. Lian, “An ECG-on-chipwith joint QRS detection & data compression for wearable sensors,” inProc. IEEE Int. Symp. Circuits Syst., May 2016, p. 2908.

[21] C. J. Deepu, X. Y. Xu, D. L. T. Wong, C. H. Heng, andY. Lian, “A 2.3 µW ECG-on-chip for wireless wearable sensors,” IEEETrans. Circuits Syst., II, Exp. Briefs, vol. 65, no. 10, pp. 1385–1389,Oct. 2018.



[22] C. J. Deepu, X. Zhang, C. H. Heng, and Y. Lian, “A 3-lead ECG-on-chipwith QRS detection and lossless compression for wireless sensors,” IEEETrans. Circuits Syst., II, Exp. Briefs, vol. 63, no. 12, pp. 1151–1155,Dec. 2016.

[23] Y. Luo, K.-H. Teng, Y. Li, W. Mao, Y. Lian, and C.-H. Heng, “A 74-µW11-Mbps wireless vital signs monitoring SoC for 3-lead ECG, respirationrate, and body temperature,” IEEE Trans. Biomed. Circuits Syst., to bepublished.

[24] X. Zhang, Z. Zhang, Y. Li, C. Liu, Y. X. Guo, and Y. Lian, “A 2.89 µWdry-electrode enabled clockless wireless ECG SoC for wearable appli-cations,” IEEE J. Solid-State Circuits, vol. 51, no. 10, pp. 2287–2298,Oct. 2016.

[25] W. Mao, Y. Li, C.-H. Heng, and Y. Lian, “A low power 12-bit1-kS/s SAR ADC for biomedical signal processing,” IEEE Trans.Circuits Syst. I, Reg. Papers, vol. 66, no. 2, pp. 477–488, Feb. 2019.

[26] V. Chan and S. Underwood, “A single-chip pulsoximeter design usingthe MSP430,” Texas Instrum. Incorporated, Dallas, TX, USA, Appl.Note MSP430, 2012, pp. 1–11.

[27] S. R. Singh, J. Jayasuriya, C. Zhou, and M. Motani, “A RESTful Webnetworking framework for vital sign monitoring,” in Proc. IEEE Int.Conf. Commun., Jun. 2015, pp. 524–529.

[28] F. Adelantado, X. Vilajosana, P. Tuset-Peiro, B. Martinez,J. Melia-Segui, and T. Watteyne, “Understanding the limits ofLoRaWAN,” IEEE Commun. Mag., vol. 55, no. 9, pp. 34–40, Sep. 2017.

[29] N. Poursafar, M. E. E. Alahi, and S. Mukhopadhyay, “Long-rangewireless technologies for IoT applications: A review,” in Proc. Int. Conf.Sens. Technol. (ICST), Dec. 2017, pp. 1–6.

[30] S.-S. Oak and P. Aroul, “How to design peripheral oxygen satu-ration (SpO2) and optical heart rate monitoring (OHRM) systemsusing the AFE4403,” Texas Instrum. Incorporated, Dallas, TX, USA,Tech. Rep. SLAA655, 2015.

[31] MuRata, “NTC thermistors,” muRata, Nagaokakyo, Japan,Tech. Rep. R44E-16, 2014.

[32] Medical Electrical Equipment—Part 1 : General Requirements for BasicSafety and Essential Performance Collateral Standard: Electromagneticdisturbances—Requirements and Tests, 4.0, document IEC 60601-1 :2005, 2014.

[33] Specification of the Bluetooth System: Bluetooth Specifications 4.0,document, Jun. 2010.[Online]. Available: https://www.bluetooth.org/docman/handlers/downloaddoc.ashx?doc_id=456433

[34] H.-G. Kim, E.-J. Cheon, D.-S. Bai, Y. H. Lee, and B.-H. Koo, “Stressand heart rate variability: A meta-analysis and review of the literature,”Psychiatry Invest., vol. 15, no. 3, pp. 235–245, 2018.

[35] R. Taylor, “Interpretation of the correlation coefficient: A basic review,”J. Diagnostic Med. Sonogr., vol. 6, no. 1, pp. 35–39, 1990.

[36] J. M. Bland and D. G. Altman, “Statistical methods for assessingagreement between two methods of clinical measurement,” Lancet,vol. 327, no. 8476, pp. 307–310, Feb. 1986.

[37] J. M. Bland and D. G. Altman, “Comparing methods of measurement:Why plotting difference against standard method is misleading,” Lancet,vol. 346, no. 8982, pp. 1085–1087, Oct. 1995.

[38] A. Hennig and A. Patzak, “Continuous blood pressure measurementusing pulse transit time,” Somnologie, vol. 17, no. 2, pp. 104–110,Jun. 2013.

[39] S. S. Shahrbabaki, B. Ahmed, T. Penzel, and D. Cvetkovic, “Photo-plethysmography derivatives and pulse transit time in overnight bloodpressure monitoring,” in Proc. 38th Annu. Int. Conf. IEEE Eng. Med.Biol. Soc., Aug. 2016, pp. 2855–2858.

[40] A. C. C. Yang, “Poincare plots: A minireview, PhysioNet heart ratevariability. techniques, applications and futute directions,” PhysioNet,Cambridge, MA, USA, Tech. Rep. HRV 2006, 2006.

[41] K. Ochiai, S. Takahashi, and Y. Fukazawa, “Arrhythmia detection from2-lead ECG using convolutional denoising autoencoders,” in Proc. KDD,2018, pp. 1–7.

[42] R. Mahajan, T. Viangteeravat, and O. Akbilgic, “Improved detection ofcongestive heart failure via probabilistic symbolic pattern recognitionand heart rate variability metrics,” Int. J. Med. Informat., vol. 108,pp. 55–63, Dec. 2017.

[43] W. G. Cochran, Sampling Techniques, 3rd ed. Hoboken, NJ, USA: Wiley,1977.

[44] O. Sayadi, R. Sameni, and M. B. Shamsollahi, “ECG denoising usingparameters of ECG dynamical model as the states of an extendedKalman filter,” in Proc. Annu. Int. Conf. IEEE Eng. Med. Biol.,Aug. 2007, pp. 2548–2551.

[45] Ł. Dziuda, P. Zielinski, P. Baran, M. Krej, and L. Kopka, “A study of therelationship between the level of anxiety declared by MRI patients in theSTAI questionnaire and their respiratory rate acquired by a fibre-opticsensor system,” Sci. Rep., vol. 9, no. 1, 2019, Art. no. 4341.

David Liang Tai Wong (S’11–M’15–SM’18)received the B.Eng. (Hons.) degree in computersystems engineering from Curtin University ofTechnology, Western Australia, in 2007, andthe M.Eng. degree in integrated circuits andembedded systems from National University ofSingapore, Singapore, in 2014. He was withPanasonic Corporation as a Research andDevelopment Engineer cum Technical Lead from2008 to 2011. He is currently a Research Asso-ciate with the National University of Singapore.

His research interests include low-power embedded hardware andsoftware co-design, data communications and network, and transla-tional research from clinical needs to technical/engineering “require-ments, design, implementation, testing” for supporting clinical trials. Heis actively contributing to the Singapore Section Young ProfessionalsAffinity Group, where he served as the Secretary in 2015, Vice-Chair,in 2016 and 2019, respectively, and Chair from 2017 to 2018.

Jufeng Yu (M’15) received the B.Eng. degree inelectronic and information engineering from theNanjing University of Aeronautics and Astronau-tics, China, in 2007, and the master’s degreefrom the School of Electrical and ElectronicEngineering, Nanyang Technological University,Singapore, in 2010. He was a Research Asso-ciate with Nanyang Technological University, anda Software Development Engineer at FutureTechnology Devices International Singapore. Hewas a Research Engineer at National University

of Singapore during this work. His primary research interests includesignal processing, machine learning, wireless body sensor networks,data analysis and its application in health care systems.

Yongfu Li (S’09–M’14–SM’18) received theB.Eng. and Ph.D. degrees from the Depart-ment of Electrical and Computing Engineer-ing, National University of Singapore (NUS),Singapore.

He was a Research Engineer with NUS from2013 to 2014. He was a Senior Engineer(2014–2016), a Principal Engineer (2016–2018),and a member of the technical staff (2018–2019)with GLOBALFOUNDRIES, and as a Design-to-Manufacturing (DFM) Computer-Aided Design

(CAD) Research and Development Engineer. He is currently an Asso-ciate Professor with the Department of Micro and Nano ElectronicsEngineering and MoE Key Lab of Artificial Intelligence, Shanghai JiaoTong University, China. His research interests include analog/mixed-signal circuits, data converters, power converters, biomedical signalprocessing with deep learning technique, and DFM circuit automation.

Dr. Li received the Singapore Economic Development Board GLOB-ALFOUNDRIES Graduate Scholarship. He is actively contributing to theSingapore Section Young Professionals Affinity Group and has servedas the Secretary (2012–2013) and the Chair (2014–2015).

Chacko John Deepu (S’07–M’14–SM’15)received the B.Tech. degree in electronics andcommunication engineering from the Universityof Kerala, India, in 2002, and the M.Sc. andPh.D. degrees in electrical engineering fromNational University Singapore, Singapore,in 2008 and 2014, respectively. His researchinterests include low power biomedical circuitdesign, energy efficient signal processing, andwearable healthcare devices. He was a recipientof the Institution of Engineers Singapore

Prestigious Engineering Achievement Award (2011), the Best DesignAward at the Asian Solid State Circuit Conference (2013), and theIEEE Young Professionals, Region 10 Individual Award (2013). He hasserved as a member of the Technical Program Committee for the IEEEconferences ASICON 2017 TENCON 2016. He is a reviewer of severalIEEE journals and conferences and serves as an Associate Editor forthe IEEE TRANSACTIONS ON BIOMEDICAL CIRCUITS AND SYSTEMS.



Duy Hoa Ngo received the Ph.D. degree in com-puter science from the University of Montpellier,France. He has expertise in semantic and knowl-edge representation, clinical terminology, appliedmachine learning, deep learning, and naturallanguage processing in clinical text mining. He iscurrently working as a Research Scientist at theAustralian E-Health Research Centre, CSIRO.

Chongyu Zhou (M’16) received the B.Eng. andPh.D. degrees from the Department of Electricaland Computer Engineering, National Universityof Singapore, in 2012 and 2017, respectively.His primary research interests include machinelearning, distributed decision making and opti-mization, with applications to mobile crowd sens-ing and wireless sensor networks.

Shashi Raj Singh received the B.Tech. degreefrom the Indian Institute of Technology Kanpur,India, and the M.S. degree from the PolytechnicInstitute of New York University, New York, NY,USA, all in electrical and computer engineering.He has worked for two years as a Technical Ana-lyst at Reliance Communication, Mumbai, India.He was a Research Associate at the Departmentof Electrical and Computer Engineering, NationalUniversity of Singapore. His research interestsare in the area of wireless networks.

Alvin Koh received the B.Sci. (Hons.) degreefrom Singapore University of Social Sciencein 2007. He works as a System Principal Engi-neer at Thermo Fisher Scientific Inc. He receivedhis training in Connecticut, USA, and Shanghai,China, in 2013. His skills included integrating newmedical product design, hardware integration,and testing activities of newly designed instru-ments at various phases against product require-ments. He specializes in experiment planningand execution from the design of experiments,

root cause analysis, and other DFSS methodologies. Moreover, hecarried out quality and regulatory compliance. He was also a recipient ofthe EDB Stanford Biodesign Scholarship in 2011.

Rachel Hong received the B.Eng. (Hons.)degree in materials engineering from NanyangTechnological University (NTU), Singapore,in 2003. She is currently the Co-Directorof the Medical Engineering Research andCommercialization Initiative (MERCI), NationalUniversity of Singapore (NUS). Her expertiseand research interests are in medical devicedevelopment, where she has lead 21 novelmedical device projects, including cardiovascularimplants, minimally invasive surgical tools, andmicrosensor based devices.

Bharadwaj Veeravalli (M’98–SM’07) receivedthe B.Sc. degree in physics from Madurai-Kamaraj University, India, in 1987, the master’sdegree in electrical communication engineeringfrom the Indian Institute of Science (IISc), Ben-galuru, India, in 1991, and the Ph.D. degree fromthe Department of Aerospace Engineering, IISc,in 1994. He is currently with the Departmentof Electrical and Computer Engineering, Com-munications and Information Engineering (CIE)Division, National University of Singapore, Sin-

gapore, as a tenured Associate Professor. His mainstream researchinterests include G Cloud/grid/cluster computing (big data processing,analytics/machine learning, and resource allocation), scheduling in par-allel and distributed systems, cybersecurity (cloud security, FHE), andmultimedia computing. He is one of the earliest researchers in thefield of Divisible Load Theory (DLT). He is a member of the IEEEComputer Society (IEEE-CS). He received gold medals for his bachelordegree overall performance and for an Outstanding Ph.D. Thesis fromIISc, in 1987 and 1994, respectively. He is currently serving at theeditorial board of the IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTEDSYSTEMS as an Associate Editor.

Mehul Motani (S’92–M’01–SM’15–F’18)received the B.E. degree from Cooper Union,New York, NY, USA, the M.S. degree fromSyracuse University, Syracuse, NY, USA, and thePh.D. degree from Cornell University, Ithaca, NY,USA, all in electrical and computer engineering.

He was a Visiting Fellow at PrincetonUniversity. He was also a Research Scientist atthe Institute for Infocomm Research, Singapore,for three years, and a Systems Engineer atLockheed Martin in Syracuse, NY, USA, for

over four years. He is currently an Associate Professor with theElectrical and Computer Engineering Department, National Universityof Singapore (NUS), and also a Visiting Research Collaborator withPrinceton University. His research interests include information theoryand coding, machine learning, biomedical informatics, wireless andsensor networks, and the Internet of Things.

Dr. Motani has served on the Organizing and TechnicalProgram Committees for numerous IEEE and ACM conferences.He was a recipient of the Intel Foundation Fellowship for his Ph.D.research, the NUS Annual Teaching Excellence Award, the NUS Facultyof Engineering Innovative Teaching Award, and the NUS Faculty ofEngineering Teaching Honors List Award. He actively participates inthe Institute of Electrical and Electronics Engineers (IEEE) and theAssociation for Computing Machinery (ACM). He has served as theSecretary of the IEEE Information Theory Society Board of Governors.He has served as an Associate Editor for both the IEEE TRANSACTIONS ONINFORMATION THEORY and the IEEE TRANSACTIONS ON COMMUNICATIONS.

Kee Chaing (KC) Chua received the Ph.D.degree in electrical engineering from the Uni-versity of Auckland, New Zealand, in 1990. Hejoined the Department of Electrical Engineering,National University of Singapore (NUS), as aLecturer. He is currently a Full Professor withthe Department of Electrical and Computer Engi-neering, NUS. He has served as the Vice Deanof the Faculty of Engineering for Research twice,from 2003 to 2006 and from 2008 to 2009. From1995 to 2000, he was seconded to the Centre for

Wireless Communications (now the Institute for Infocomm Research),a National Telecommunications R&D Center funded by the SingaporeAgency for Science, Technology, and Research as its Deputy Director.From 2001 to 2003, he was on leave of absence from NUS to workat Siemens AG, where he was the Founding Head of the Mobile CoreR&D Department in Singapore, funded by Siemens’ ICM Group. From2006 to 2008, he was seconded to the National Research Foundationin the Prime Minister’s Office as the Director. He was appointed theDean of the Faculty of Engineering, NUS, in 2014, after serving as theHead of the Department of Electrical and Computer Engineering, since2009. He Chaired the World Economic Forum’s Global Agenda Councilon Robotics and Smart Devices in 2011, and spoke on the role of roboticsand smart devices in shaping new models of development at the WorldEconomic Forum in Davos in 2012. He is a recipient of an IEEE 3rd

Millennium Medal, a Fellow of the Singapore Academy of Engineering,and a Fellow of the Institution of Engineers, Singapore. His researchinterests are in communication networks.



Yong Lian (M’90–SM’99–F’09) has publishedmore than 300 papers. His research interestsinclude biomedical circuits and systems and sig-nal processing. He is the Founder of the IEEEBiomedical Circuits and Systems Conference(BioCAS) and the Asia Pacific Conference onPostgraduate Research in Microelectronics andElectronics (PrimeAsia). He is a Fellow of theAcademy of Engineering Singapore. He is Pres-ident of the IEEE Circuits and Systems (CAS)Society, Chair of the IEEE Periodicals Partner-

ship Opportunities Committee, a member of the IEEE Periodicals Com-mittee, a member of the IEEE Periodicals Review and Advisory Commit-tee, a member of the IEEE Biomedical Engineering Award Committee,and a member of the Steering Committee of the IEEE TRANSACTIONSON BIOMEDICAL CIRCUITS AND SYSTEMS. He has served as the VP forPublications and VP for Region 10 of the IEEE CAS Society, andmany other roles in the IEEE. He received more than 15 awards,including the IEEE Circuits and Systems Society’s Guillemin-CauerAward, the IEEE Communications Society Multimedia CommunicationsBest Paper Award, the Institution of Engineers Singapore PrestigiousEngineering Achievement Award, and the Winner of the Design ContestAward in ISLPED2015. Dr. Lian was the Editor-in-Chief of the IEEETRANSACTIONS ON CIRCUITS AND SYSTEMS PART II for two terms.

Chun-Huat Heng (S’96–M’04–SM’13) receivedthe B.Eng. and M.Eng. degrees from the NationalUniversity of Singapore, Singapore, in 1996 and1999, respectively, and the Ph.D. degree fromthe University of Illinois at Urbana–Champaign,Urbana, IL, USA, in 2003. From 2001 to 2004, hewas with Wireless Interface Technologies, whichwas later acquired by Chrontel. Since 2004, hehas been with the National University of Singa-pore, where he is currently an Associate Profes-sor. He is focusing on CMOS integrated circuits

involving synthesizer, delay-locked loop, and transceiver circuits. He wasa recipient of the NUS Annual Teaching Excellence Award, in 2008, 2011,and 2013, respectively, and was on the ATEA Honor Roll in 2014. He wasa recipient of the Faculty Innovative Teaching Award in 2009, and the2018 IES Prestigious Engineering Award. He was an Associate Editorof IEEE TCAS-II, and was a Technical Program Committee member ofISSCC and ASSCC. Dr. Heng currently serves as an Associate Editorfor the IEEE TRANSACTIONS ON VERY LARGE SCALE INTEGRATION (VLSI)SYSTEMS.


Documents

An Integrated Wearable Wireless Vital Signs Biosensor for … · 2021. 1. 9. · VySys, includes two compact wearable wireless biosensor devices for continuous vital signs capturing