Biosensors and Bioelectronics 87 (2017) 522–530

Contents lists available at ScienceDirect

Biosensors and Bioelectronics

http://d0956-56

n CorrE-m1 Au

journal homepage: www.elsevier.com/locate/bios

Multimodal technique to eliminate humidity interference for specificdetection of ethanol

Ahmed Hasnain Jalal a,1, Yogeswaran Umasankar b,1, Pablo J. Gonzalez a, Alejandro Alfonso a,Shekhar Bhansali a,n

a Department of Electrical and Computer Engineering, Florida International University, Miami, FL 33174, United Statesb Biomolecular Sciences Institute, Florida International University, Miami, FL 33174, United States

a r t i c l e i n f o

Article history:Received 22 June 2016Received in revised form16 August 2016Accepted 30 August 2016Available online 30 August 2016

Keywords:EthanolNafionFuel cellContinuous monitoringPotentiostatWearable

x.doi.org/10.1016/j.bios.2016.08.10663/& 2016 Elsevier B.V. All rights reserved.

esponding author.ail address: sbhansa@fiu.edu (S. Bhansali).thors with equal contribution.

a b s t r a c t

Multimodal electrochemical technique incorporating both open circuit potential (OCP) and ampero-metric techniques have been conceptualized and implemented to improve the detection of specificanalyte in systems where more than one analyte is present. This approach has been demonstratedthrough the detection of ethanol while eliminating the contribution of water in a micro fuel cell sensorsystem. The sensor was interfaced with LMP91000 potentiostat, controlled through MSP430F5529LPmicrocontroller to implement an auto-calibration algorithm tailored to improve the detection of alcohol.The sensor was designed and fabricated as a three electrode system with Nafion as a proton exchangemembrane (PEM). The electrochemical signal of the interfering phase (water) was eliminated by im-plementing the multimodal electrochemical detection technique. The results were validated by com-paring sensor and potentiostat performances with a commercial sensor and potentiostat respectively.The results suggest that such a sensing system can detect ethanol at concentrations as low as 5 ppm. Thestructure and properties such as low detection limit, selectivity and miniaturized size enables potentialapplication of this device in wearable transdermal alcohol measurements.

& 2016 Elsevier B.V. All rights reserved.

1. Introduction

Continuous wearable alcohol measurement system has beensought in numerous fields ranging from law enforcement to clin-ical monitoring to safety systems (Goodman et al., 1986; HighwayTraffic Safety Administration and Department of Transportation,2013; Room et al., 2005; WHO, 2015). Breath alcohol (BA) mea-surement devices are used by law enforcement agencies for therandom monitoring of drivers to determine whether they aredriving under influence of alcohol (Mason and Dubowski, 1974;Dubowski, 1994; Zuba, 2008; Polissar et al., 2015). Since law en-forcement involves measuring the BA content of a random popu-lation at one point in time, breath analyzers are adequate. How-ever, in a clinical application focused on understanding con-sumption and metabolism of alcohol, measurements are requiredover extended periods, ideally starting from 30 min before con-sumption and extending to 8 h after the last drink. Given that al-cohol is generally consumed in the evening and the subject willlikely sleep following his ingestion of alcohol, a continuous alcohol

sensor is preferred. The simplest approach to continuous mon-itoring of alcohol is a wearable sensor that uses alcohol vaporsemanating from the skin and secreted with sweat (Sakai et al.,2006).

After alcohol ingestion, a very small proportion (1% of overallalcohol consumption) is excreted with sweat from the exocrinesweat gland or in a diffusive manner (Marques and McKnight,2009). The sweat is mostly composed of water (�99% of overallvolume) with small amount of nitrogenous compounds, metal andnonmetal ions, metabolites, xenobiotics, organic volatile com-pounds and so on (Jadoon et al., 2015). Hence, elimination of crossselectivity with different compounds, especially with the watercontent of sweat or atmospheric humidity is challenging for anytransdermal alcohol sensors. The reaction kinetics (Zhang et al.,2008) and maximum power density (Lee and Hwang, 2009) of thefuel cell sensor are directly dependent on relative humidity. In theelectrochemical sensing, therefore, humidity has an adverse im-pact on the calibration of the sensors. Similar to humidity, othervolatile compounds released from the skin such as aldehydes andketones (Mochalski et al., 2014), will undergo reaction in the an-ode creating interference signal. Only a selective and accurate al-cohol measurement device would be the possible solution toovercome this issue. This can be achieved in two ways: firstly, todevelop a highly selective sensor that can afford accurate signal for

Fig. 1. (a) Block diagram and schematic of peripheral connections in the alcoholmonitoring device, (b) top and (c) bottom layer of the PCB board where the toplayer consists of LMP91000, microprocessor and other integrated electronic com-ponents and the bottom layer consist of Bluetooth component.

A.H. Jalal et al. / Biosensors and Bioelectronics 87 (2017) 522–530 523

only alcohol or secondly, electrically eliminate the noise signalsfollowing a robust electrochemical technique.

Current alcohol sensors in the market use five measurementtechniques for monitoring alcohol: optical (Sanford et al., 2001;Shabaneh et al., 2014; Semwal et al., 2016), colorimetric (Williamsand Reese, 1950; Lidén et al., 1998), electrochemical (Santra et al.,2016; Jiang et al., 2016; Shan et al., 2010; Kim et al., 2000; Swiftet al., 1992), biomarker sensing (An et al., 2014) or solid-statetechniques (Yu et al., 2012; Kadir et al., 2014; Kim and Lee., 2014).Most of these sensors suffer from instability, non-linearity, cross-selectivity, inaccuracy at low concentrations, and portability(Paixão and Bertotti, 2004). Electrochemical alcohol sensors havetypically been found to be most suitable for long term sensing.Among them, fuel cell sensors offer simplicity, relatively high ac-curacy, sensitivity, long working lifetime, scalability and port-ability (Marques and McKnight, 2009), hence they find use inbreathalyzers.

Wearable transdermal alcohol measurement devices have beenbuilt before e.g. ‘GinerWrisTAS’ (Injury Research Foundation, 2007).However, this device did not have a data acquisition system andrequired a humid chamber for measurements. Others (Injury Re-search Foundation, 2007) integrated the data acquisition systemand incorporated PEM as an electrolyte. These sensors sufferedfrom false positive readings due to the presence of volatile organiccompounds, lack of selectivity, and needs for frequent manualcalibrations. Both fuel cell/electrochemical breathalyzers andwearable devices measure interactions between the sensor andvolatile organic compounds (VOCs) such as acetone, acet-ophenone, isoprene, etc (Mochalski et al., 2014; Acevedo et al.,2007). The interactions lead to change in potential and generate asignal. When these devices are used to for measurement of alco-hol, the VOCs trigger false positives.

To address the needs of the target users, a complete wearablesystem needs to be developed. Such a system should be able torecalibrate and reference itself to its environment, minimizing theeffect of VOCs, thus enabling long term quantitative monitoring.The platform needs to provide real-time statistical analysis fordifferent vapor footprints, including alcohol so it can be calibratedto the needs of the user. Such systems should also be low poweredto ensure long usage life between charges. This paper presentssuch a sensing system that has the ability to communicate withexternal devices through Bluetooth for longitudinal data analysisto ensure functionality. Finally, the sensor was packaged into awristwatch format.

2. Experimental

2.1. Materials and methods

Perfluorinated Nafion424 reinforced with poly-tetra-fluoro-ethylene (PTFE) fibers (thickness 0.03 cm) as a PEM, nickel chlor-ide anhydrous, nickel sulfamate, boric acid, 95% sulfuric acid(H2SO4) and 37% hydrochloric acid (HCl) were purchased fromSigma-Aldrich. Lead and nickel sheets were purchased fromMcMaster-Carr for the electroplating process. Acetone and ethanol(95.27%) were purchased from Fisher scientific Inc. All other usedchemicals were of analytical grade. The preparation of aqueoussolutions was done with de-ionized (DI) water.

MICRO5 PID sensor purchased from BW Technologies. Po-tentiostat (CHI 1230B with MC470) purchased from CH instru-ments Inc. LMP91000 miniaturized potentiostat with analog frontend (AFE), a 16-bit ultra-low power microcontroller(MSP430F5529LP), RN42 Bluetooth chip and LP2591 power man-agement system were purchased from Texas Instruments (TI).MCP72831 charge controller, 12 bit digital-to-analog converter

with integrated electrically erasable programmable read-onlymemory (EEPROM) and an I2C compatible serial interface fromMicrochip were purchased.

2.2. Potentiostat platform and fuel cell sensor construction

The fuel cell sensor was interfaced with the AFE (LMP91000) ona wearable platform. The working electrode (WE), reference elec-trode (RE), and counter electrode (CE) of the sensor were con-nected to the corresponding pins of the AFE. The AFE was linked tothe microcontroller via inter-integrated circuit (I2C) interface. TheBluetooth module was connected via universal asynchronous re-ceiver/transmitter (UART) peripheral of the MSP430. The liquidcrystal display (LCD) used the serial peripheral interface (SPI) ofthe MSP430. The connections and their functions are discussed inSection 3. The peripheral connection is depicted in the block dia-gram as shown in Fig. 1a.

The printed circuit board (PCB) was manufactured on a two-layer board using a 1.6 mm thick FR-4 substrate with 28.35 gcopper and an area of 6.45 cm2. All the electronic componentsincluding the microcontroller, AFE, power management system,and associated circuitry were mounted on the top layer as shownin Fig. 1b. The Bluetooth chip was mounted on the bottom layerBluetooth solely (Fig. 1c).

Plated 200 mm thick micro-perforated stainless steel sheet with180 mm pore was used to form WE and CE. A non-conventionalstandardized material was used for RE. Nafion was hot-pressedbetween the electrodes to form the fuel cell. The catalyst (nickel)coating on the micro-perforated stainless steel sheet electrodeswas achieved by a five step process: (i) anodic cleaning of stainlesssheet for 5 min with 25% H2SO4 below room temperature at acurrent density of 13.94 A m�2. In electro-cleaning setup, thestainless steel served as the anode and the lead sheet served as thecathode; (ii) acid cleaning the stainless sheets at room tempera-ture with 1:10:1000 solution of HCl, H2SO4 and DI water; (iii)Wood's nickel strike using 1.5 M HCl and 1.009 M anhydrous nickelchloride solution and by applying current densities of 4.65 A m�2

and 1.4 A m�2 for 2 min each consecutively (Harding and Di Bari,

A.H. Jalal et al. / Biosensors and Bioelectronics 87 (2017) 522–530524

1987). Here, a nickel sheet was used as the anode and the targetelectrode as the cathode; (iv) galvanostatically electroplating thesheets with nickel by Watt's deposition method at 50 °C at 0.2 A ina mixture of 0.93 M nickel sulfamate, 0.025 M nickel chloride, and0.48 M boric acid solution (Harding and Di Bari, 1987); (v) bakingand drying the sheets at 190 °C for 2 h. In the above electroplatingprocess, voltage was maintained below 3 V.

Sandwiching of PEM between micro-perforated electrodes wasachieved by hot pressing the layers with the hydraulic press(model 2100 from PHI) at 75 °C and 17.237 MPa for 10 min. Beforesandwiching, the pieces were ultrasonicated for 3–5 min in a de-tergent and subsequently cleaned with DI water, then ultra-sonicated in acetone for 3–5 min to remove any organic residues.The dimension of the PEM, CE, WE, and RE were 1.5 cm�1 cm,1.5 cm�1 cm, 1.5 cm�0.8 cm, and 1.5 cm�0.2 cm, respectively.The sensor was designed in such a way that the CE and RE werepasted onto the facet of the PTFE and the WE on the Nafionmembrane.

Standard solutions of ethanol and vapor were calibrated usingthe MICRO5 PID sensor. OCP and amperometric studies of the fuelcell were with the commercial potentiostat. Ethanol vapor wasgenerated by bubbling a constant flow of air through a 15.8 Methanol solution. Ethanol vapor was passed through a custom 3Dprinted chamber (�0.7 cm3) containing standard PID sensor or theWE of developed sensor for measurements. The chamber wasdesigned in such a way that the CE was exposed to theatmosphere.

3. Results and discussions

3.1. Electrochemical mechanism of ethanol fuel cell sensor

The operation of PEM fuel cell sensor depends primarily on(i) the redox reaction at the anode and cathode surface, (ii) thePEM's hydration and its ability to transport protons from the an-ode to the cathode, and (iii) the catalyst's ability to enhance theredox reaction. The redox reactions of the ethanol fuel cell can berepresented by Eqs. (1), (2), and Fig. 2a. The overall fuel cell re-action can be represented by Eq. (3) (Friedl and Stimming, 2013;Antolini, 2007):

C2H5OHþ3H2O-12Hþþ12e�þ2CO2 (E0¼0 �085 V) (1)

3O2þ12Hþþ12e--6H2O (E0¼1 �23 V) (2)

Volta

ge (V

)

H+

H+

O2

C2H5OH H2O

CO2

6e-

4e-

a'

b'

d'

c'

(a)

Fig. 2. (a) Schematic of ethanol oxidation and oxygen reduction in a fuel cell sensor, whsensor in presence (continuous green) and absence (dashed blue) of 95% ethanol in 100reader is referred to the web version of this article.).

C2H5OHþ 3O2-2CO2þ3H2O (E0¼1 �145 V) (3)

In this reaction, protons are exchanged from the anode to thecathode through the PEM, and the electrons flow through the in-ternal circuit. The E0 values given in the Eqs. (1) and (2) are thethermodynamic standard potentials vs. standard hydrogen elec-trode (SHE), and the E0 value in Eq. (3) is the equilibrium potentialdifference, which represents 12 electrons per ethanol molecule.However, the thermodynamic values are practically of little use aspractical systems do not operate under reversible conditions. Inorder to obtain the characteristics of the constructed three elec-trode fuel cell sensor, OCP technique (time vs. voltage) was used.In these studies, to enhance the oxygen reduction and ethanoloxidation, Ni was used as a catalyst (Friedl and Stimming, 2013;Antolini, 2007). The effect of humidity on Hþ ion transport wasminimized by treating each sensor in a humid chamber for 30 minunder room temperature as it is established that the rate andamount of proton exchange depends on the water content of thePEM (Hu Junming et al., 2014; Hartnig and Roth, 2012). The OCPmeasurements were taken over a period of 1000 s in a closedchamber containing the fuel cell sensor. Even though the steadystate potential of the sensor was attained within 5 s, t¼100 s waschosen to introduce the ethanol in order to illustrate the stabilityof the baseline. The response time of the sensor can be seen to beless than 2 s (Fig. 2b). The same figure shows that the OCP ofethanol oxidation (100% humid condition) was 0.07 V. In thesensor, thermodynamic standard potential was considered insteadof equilibrium potential difference as the fuel cell sensor consistsof three electrode system and measurements were half-cell mea-surements. According to Nernst equation, OCP of ethanol shouldbe much higher than the thermodynamic standard potential ofethanol (0.085 V). This low ethanol OCP value is due to the mixedpotential generated because of both ethanol and humidity.

The experimental results in Fig. 2b also revealed that there wasa �0.2 V deviation in OCP for 100% humidity (absence of ethanol)compared to the OCP in the presence of ethanol. This deviationvaries with the percent change in humidity at the rate of 2.7 mVfor each percentage decrease in humidity (Fig. S1 in Supplemen-tary materials). Humidity level varies inconsistently in practicalconditions, and the exposure to various humidity percentage re-vealed that the reference value of the sensors' OCP signal oscil-lates. Therefore, deriving a relationship between humidity and theethanol OCP signal generation for calibration was not possiblebased on empirical results. Hence, quantifying ethanol based onthe OCP technique alone was inaccurate, even though significantsignals were measured for ethanol in OCP. However, the OCP signal

-0.3

-0.15

0

0.15

0 200 400 600 800Time (second)

(b)

ere a′, b′, c′, and d′ represents WE, CE, PEM and RE respectively. (b) OCP of fuel cell% humidity. (For interpretation of the references to color in this figure legend, the

A.H. Jalal et al. / Biosensors and Bioelectronics 87 (2017) 522–530 525

variation between ethanol and the humidity (�0.2 V) provided asignificant opportunity to design a multimodal method to elim-inate interference. Similar to humidity, any organic volatile com-pound capable of oxidizing on the anode has its own OCP sig-nature, which is an important characteristic that supports selectivedetection of desired compound in a given binary system.

3.2. A method to eliminate the interfering signal

Selectivity is the biggest challenge in the successful construc-tion of any fuel cell sensor. In diffusion control process, the OCP,due to any given reaction, is independent of its concentration.However, in the case of low concentration and low volume mea-surements, the rate of the reaction, rate of diffusion, and rate ofevaporation are the limiting parameters. If the concentration ofthe interfering compound is much higher than the ethanol, thesensor's accuracy would be low; because as discussed above, theexperimental potential of each compound varies due to the mixedpotential, which in turn affect the current generated by the fuelcell. To improve accuracy, a multimodal method containing bothOCP and amperometric techniques was employed for subsequentmeasurements.

During the redox reaction the current flow between the elec-trodes can be measured using amperometric method as faradaiccurrent. This generated current depends on: formal potential ofthe reactions, applied potential across the electrodes, and rate of

-90

-75

-60

-45

-30

-15

0

0 500 1000 1500Time (s)

Cur

rent

(µA

)

(c)No interference

Ethanol

-20

-15

-10

-5

0

5

10

0 1000 2000 3000 4000 5000 6000Time(s)

Cur

rent

(µA

)

Interference

Ethanol

(a)

Fig. 3. (a) Amperometric plot showing humidity and ethanol signals, (b) calibration p(c) amperometric studies after calibration showing only the ethanol signal, where the intthree electrode systems.

the redox reaction. For example, the amperometric measurementof ethanol at a fixed potential (�0.05 V) in the presence of highhumidity given in Fig. 3a revealed that there is a response for notonly ethanol but the humidity as well. The same figure showedthat if the applied potential was lower than the OCP of ethanol; itresulted in a negative current response and vice versa for thehumidity. The response depended on the difference (ΔV) betweenOCP and the applied potential, resulting in a variation in the out-put current. This showed that the rate of faradaic reactions on theelectrode surface could be manipulated using external voltage. Asa step towards eliminating humidity interference signals, signaldue to humidity was taken as an example for the following stu-dies. Similar method can be applied for eliminating the signals ofall other organic volatile compounds. However, the method re-ported here under is only applicable for binary chemical systemand has not been validated for multiple interfering compounds.

The selectivity of ethanol in the presence of humidity wasachieved through the following steps, (i) identifying the OCP in thepresence of the humidity, and (ii) applying the obtained OCP valueacross the WE and RE and measuring the current flow between CEand WE. In this process the current signal due to humidity waseliminated and the current flow due to ethanol oxidation for thatparticular potential was recorded. The experiments with humidity(490%) showed that the OCP of the fuel cell sensor varied be-tween �0.05 V to �0.2 V. By keeping the potential exactly at theexperimental potential (Ecell) at any given humidity level (in the

-0.4

0

0.4

0.8

1.2

-0.4 -0.2 0 0.2 0.4Voltage (V)

Cur

rent

(µA

)

(b)

-0.08

-0.04

0

0.04

0.08

0.12

0 200 400 600 800 1000

Three ElectrodeTwo Electrode

Ethanol

Volta

ge (V

)

Time(s)

(d)

lot of current vs. voltage scan in presence of humidity, shows the voltage at 0 A,erfering signal was eliminated, (d) OCP signature comparison of ethanol at two and

OCP measurement

Scan: -0.2 to 0.1 V

< -0.05 V

Measure currentScan:-0.05 ~ -0.25V

I = 0 A

Fix V, Measure I

Fit calibration

Display BAC

Measure I, Fit data

Display BAC

No

Yes

No

Fig. 4. Flow chart representing steps involved in selective ethanol detection.

A.H. Jalal et al. / Biosensors and Bioelectronics 87 (2017) 522–530526

absence of ethanol), the current flow due to humidity waseliminated.

The identification of the Ecell at any given humidity was carriedout by a series of amperometric studies, where various potentialswere applied across the electrodes and the current was measured(Fig. 3b). From the applied potential vs. current plot the exact Ecellat which the current falls to zero was identified for that particularhumidity. Even though the steady state OCP measurements can beused to measure this Ecell value, the amperometric method wasused to find the Ecell to significantly improve the accuracy. At-taining OCP, the steady state value would vary depending on theenvironment and the need for prolonged scans in real-time cali-brations. The amperometric results after calibration (Fig. 3c) showthe ethanol signal only, and the signal interference due to hu-midity was eliminated.

In a separate study, the anodic polarization of the micro-fuelcell (Fig. S2 in Supplementary materials) showed a linear increasein the current from �0.3 V to �0.08 V indicating the mixed po-tential signal of both humidity and ethanol. In the same curve,there was a change in slope in the region of �0.08 V to 0 V in-dicating the activation polarization for ethanol oxidation. The ac-tive region of the ethanol oxidation was in the range of �0.8 V to0.2 V. The region greater than 0.2 V was the concentration polar-ization, where the reaction is diffusion limited.

3.3. Improving the stability of the OCP signature

To achieve the chemical specificity, it was imperative that thesensor had a precise Ecell for each chemical compound in any givenelectrochemical system. The current fuel cell sensors in the marketconsists of two electrode setup; maintaining precise Ecell is gen-erally not possible with a two electrode system due to the po-tential drop across the cell because of electrolyte resistance andpossible polarization of the CE. It is well known that during am-perometric measurements the electrodes get polarized, which inturn perturbs the electrochemical system. However, precise Ecellcan be maintained in this process using a three electrode system inwhich the potential of WE is measured relative to the RE. Further,due to high impedance between the WE and RE the current passesbetween WE and CE avoiding polarization of the RE. The electro-motive force in a three electrode system caused by the standardpotential (E0) is given as Nernst equation (4).

Δ= − ( )GE /nF 40 0

where,ΔG0, n and F are respectively Gibb's free energy, number ofelectron and Faraday constant. The stability of the Ecell was testedusing OCP technique and the two and three electrode systemswere compared, as seen in Fig. 3d. The results showed that afterethanol addition there were erratic changes in the OCP of the twoelectrode fuel cell sensor while the three electrode configurationresulted in a stable OCP of the three electrode system.

3.4. Algorithm for the sensor auto-calibration

Based on the experimental observations reported in Section3.2, a flow chart (Fig. 4) was developed for implementing an al-gorithm that would result in chip-potentiostat to have with auto-calibration ability and selectivity to ethanol in high humid con-ditions. The nullification of current signal produced by humiditywas achieved by two following functions:

Function 1: auto-calibrate fuel cell sensor in certain intervals(which depend on the steady state response of the nickelcatalyst).Function 2: measurement of ethanol using amperometric

measurement.

The calibration (1st function) was necessary to know the hu-midity signal and nullify it in the second function. The 1st functioninvolved OCP and amperometric measurements, the 2nd functioninvolved amperometric measurements. In the 1st function if thevalue was lower than the threshold OCP, it indicated no ethanolpresence and the system proceeded to amperometric measure-ments. In this step Ecell for 0 A was identified by measuring currentacross the electrodes while scanning the potential. The obtainedEcell value was stored in the system. This stored value was the biasvoltage for amperometric measurements in the 2nd function,where the current was measured and fitted against a pre-de-termined calibration curve. In the device memory, there will bemultiple pre-determined calibration curves were stored for eachbiasing voltage. These curves were used as the current signalmagnitude in the 2nd function depends on the biasing potential.The final step is to display BAC from the calibration curve fitting.This process nullifies the interfering signal in any givenenvironment.

3.5. Configuration of the analog front end alcohol sensing device

A schematic of the potentiostat is given in Fig. 5a. The ar-rangement of the circuit is that of a non-inverting operationalamplifier. The voltage supplied by the source E, was closely

(a)

(b)

Three electrodes

fuel cell sensor

CE

RE

WE

A1

Variable Bias

VREFDivider

VREF VDD

TIAR load

R TIA

C2C1

Temp.sensor

SCL

SDA

12C Interface

and control

registers MENB

DGND

V out

AGND

+

-

+

-

Fig. 5. (a) Block diagram of a simple potentiostat where CA is the control amplifier; Ic is the current at the counter electrode, Z1 is the impedance across counter andreference electrode, Z2 is the impedance across reference and working electrode, and E is a voltage source. (b) Functional block diagram of LMP91000 AFE with the fuel cellsensor.

A.H. Jalal et al. / Biosensors and Bioelectronics 87 (2017) 522–530 527

followed by the voltage between the RE and WE terminals. Z1 andZ2 are the characteristic impedances of the fuel cell between therespective terminals. Any change in impedance due to ethanolcoming in contact with the WE was reflected by the change incurrent Ic (current at CE), as show in Fig. 2a. As described in pre-vious sections, the impedance at the negative terminal of theamplifier was very high, which made the current flowing throughthe RE negligible.

The LMP91000 can be configured to perform different types ofelectro-analytical techniques. The detection method used in thesystem was amperometric. Referring to Fig. 5b, amplifier A1 is thecontrol amplifier that implements the potentiostat circuit. Thevariable bias block of the LMP was used to provide a user config-ured potential across positive and negative terminals of A1. Thispotential was held constant between the reference and workingterminals by the potentiostat. The transition from minimal currentflow to voltage was made available by the trans-impedance am-plifier (TIA) whose forward voltage gain is dependent upon on afeedback resistor Rtia. It can be connected either internally or ex-ternally to the feedback path of the TIA as depicted in the internalblock diagram of the AFE of LMP91000 in Fig. 5b (LMP91000Sensor AFE System: Configurable AFE Potentiostat for Low-PowerChemical- Sensing Applications, 2014). It converts the currentflowing from CE and WE to a proportional voltage. Its output isconnected to Vout pin (this pin can be toggled to give output of thetemperature sensor of the LMP) and the C2 pin.

The LMP was configured via the microcontroller to performthree electrode amperometry following the functional block

diagram sketched in Fig. 5b. The microcontroller connected to theLMP via the I2C interface, as depicted in Figs. 1a and 5b. The serialclock line and serial data bus line on the I2C bus were connected to3.3Vdc from the system power management unit with one externalpull up resistor each. The schematic representation in Fig. 5b wasderived for the 3-lead amperometric cell in potentiostat config-uration (LMP91000 Sensor AFE System: Configurable AFE Po-tentiostat for Low-Power Chemical- Sensing Applications, 2014).The output voltage available at the Vout pin of the LMP91000 wasthen routed to the microcontroller general purpose input/output(GPIO), where it was conditioned by an internal analog-to-digitalconverter for interpretation. The TIA gain is adjusted to provide avoltage proportional to cell current, it can be internally pro-grammed via software for a range of 2.5kΩrRtiar350kΩ andexternally configured as required.

The internal feedback resistor was optimized for an optimal TIAamplifier gain using a value of Rtia¼120 kΩ, this provided a largeenough signal gain to manipulate the data and provided enoughheadroom for voltage swings as a result of changes in alcoholconcentration. The module enable (MENB) was tied to ground inorder to signal a communication ready status to the micro-controller. In the case of multiple alcohol sensors, the MENB line ofeach AFE sensing device on the I2C bus can be toggled through toextract data from each sensor individually. The voltage reference(VREF) to the AFE sensing device was externally provided by thedigital to analog converter (MCP4724) and was adjustable throughsoftware and can meet a wide range of supply voltages to theLMP91000 and to specify bias voltages with accuracy. The load

y = -0.0448x - 5.7322R² = 0.9396

-50

-45

-40

-35

-30

-25

-20

-15

-10

-5

0

0 200 400 600 800

Ethanol (ppm)

Cur

rent

(nA

)

Fig. 6. Amperometric data obtained in (a) commercial MC470 potentiostat and (b) LMP91000 potentiostat in the presence of ethanol and humidity (c) concentration vs.current plot showing the linear response of the sensor with the RSD of 30%.

A.H. Jalal et al. / Biosensors and Bioelectronics 87 (2017) 522–530528

resistor was set to its lowest internal resistance value ofRload¼100Ω in order to draw maximum current from the sensorand subsequently become amplified by the TIA.

3.6. Microcontroller operation and alcohol concentration measure-ment technique

MSP430F5529LP was chosen due to its small LQFP-80 pinpackaging size and low power mode settings that allow powerconsumption down to 1.4 mA in LPM3, low operating voltage rangeof ≤ ≤V V V1.8 3.7DC Batt DC and portability to other microprocessingunits if so desired due to the simplicity of the C programminglanguage. Features that made this microcontroller attractive forour application were its ability to store 128 kB of non-volatile flashmemory and 8 kB of RAM, allowing the software to execute duringpower-on and reset events, and a 12-bit ADC that was used tomeasure the analog output of the LMP.

The MCU has the capability to communicate via SPI, UART, andI2C. The alcohol sensing platform uses I2C to communicate withthe LMP91000 and MCP4724, UART to communicate with the RN-42 Bluetooth module and SPI to communicate with the mono-chrome LCD. The present configuration for the MCU was shown inFig. 1b. In this figure it can be seen that the MCU can be pro-grammed and debugged on-chip and on a minimally system in-vasive procedure simply by connecting two Spy-Bi-Wire emulatorcables. The external circuitry shown in Fig. 1b are bypass capaci-tors and ground connections required for proper MCU operation,this configuration was derived from a device specific datasheetand user guide. The algorithm for interpreting sensor data wasadapted from (Mihajlovic et al., 2014); under this configurationtwo additional cables were routed from the C1 and C2 pins to theGPIO pins. The data from C1 and C2 consists of the analog outputvoltage Vout and this voltage presents at the inverting input of theTIA, which is directly connected to the electrodes of the sensor andis the same voltage at the non-inverting input of the WE TIA, thiswas a fixed percentage of the VREF or divided reference voltage(VREFDIV

), and is chosen dependent on current flow to the WE. Thisallowed computation of current flowing at the WE of the threeelectrode system (IWE) as follows IWE¼ ( − )V V R/out REF tiaDIV

, once VREFDIVhas been established using the aforementioned procedure, Rtia waschosen depending on the sensitivity of the sensor given in ppm aspreviously discussed, the current of the sensor is then calibratedvia software as a function of all described parameters.

3.7. Power management, data transmission and user interactivity

The developed wearable platform operated on a 3.7VDC lithium-ion (Li-ion) battery capable of providing up to 1000 mAh. Battery

replacement was a trivial task in the platform. The unregulatedbattery voltage was regulated to provide a constant 3VDC source tothe system in through a battery voltage range of ≤ ≤V V V3 3.7DC DC DC ,the system has been designed to prevent Li ion battery drainagebelow a 3VDC–2.4VDC threshold as this can potentially cause anunsafe condition (Hruska, 1997) through means of a Power Man-agement Unit LP5921. The system operates with a current draw of2 mA during normal operation with Bluetooth disabled, whenBluetooth capabilities were enabled by the user, the current con-sumption of the system increased to approximately 40 mA. Asshown in the block diagram in Fig. 1a the Li-ion battery can berecharged on the wearable platform through a micro-USB deviceconnected directly to a Li-ion battery charging circuit which allowssimultaneous system operation and charging functions.

The AFE LMP91000 was chosen as the signal path solutionbetween the MSP430F5529LP and the sensor due to its ability todetect current in the nano-ampere (nA) range and provide anoutput voltage proportional to current times a gain factor. In orderto ensure reliable AFE sensing operation, automatic system cali-bration based on a particular sensor's OCP and a reliable referencevoltage to the AFE sensing device, the DAC device (MCP4725)provided configurable reference voltage to the AFE sensing device,of which a fixed percentage would be applied across the sensor forbiasing purposes as determined during calibration time. Having afixed voltage percentage that was software programmable allowsthe platform to configure itself to any sensor in order to eliminateany abnormalities that may be present. Information sent from theBluetooth module to the receiving device is encrypted using asimple algorithm to avoid data compromise and subsequentlydecrypted at authorized receiving device and displayed to the user.

3.8. Comparison of MC470 and LMP91000

To validate the device, data from the commercially availableelectrochemical device MC470 was compared with the con-structed LMP91000 device as shown in Fig. 6a and b respectively.During these experiments, both the devices were placed in anidentical setup with similar parameters as given in the experi-mental section. The concentration of ethanol used for these ex-periments was 7.272 M. The experiments were carried out for700 s, where the humidity was introduced after 50 s and theethanol was introduced after 200 s. The biasing potential was keptconstant at �0.2 V for both the experiments. Both the resultsshowed there was no interfering signal due to humidity, and haveresponse to the ethanol (Fig. 6a and b). Comparison of the resultsin the same figure shows that the LMP91000 device has 30 timeshigher current signal due to the presence of amplifier in thedevice.

A.H. Jalal et al. / Biosensors and Bioelectronics 87 (2017) 522–530 529

The effect of ethanol concentration on the fuel cell sensor wasstudied for sensor evaluation. In these experiments, there were tendifferent concentrations tested in the physiological range oftransdermal ethanol (5 ppm to 800 ppm). The optimized sensoroperating parameters were 42% humidity at 25 °C, at �0.05 Vbiasing potential. The concentration vs. current plot given Fig. 6cshows that the sensor response is linear from 50 ppm to 800 ppm,with the sensitivity of �0.23 nA ppm�1 cm�2 (RSD¼30%) withthe lowest detection limit of 5 ppm. The RSD values for con-centrations below 50 ppm was 55%, which went down to 18% forthe concentrations higher than 50 ppm. These results prove thatthe device along with the sensor has the capability to measureeven the lowest transdermal ethanol concentration (r200 ppm)(Gamella et al., 2014).

Bland-Altman plots were used to characterize the repeatabilityand reproducibility of the sensor (Fig. S3 in Supplementary ma-terials). The plots show the average and difference between twomeasurements of current on the x-axis and y-axis respectively. Inthe repeatability experiments, two sensors were studied and thedata was collected multiple times (n¼10), whereas for reprodu-cibility ten different sensors were studied and the data was col-lected twice in each sensor. The concentration of ethanol andpercentage of humidity used in these experiments were 50 ppmand 46% respectively. Bland-Altman plots shows an absence of biasin measures of repeatability (p¼0.0004) and reproducibility(p¼0.0004) were considered statistically significant. The correla-tion coefficient values for repeatability (r2¼0.8123) and reprodu-cibility (r2¼0.8102) are closer to unity indicating strong relation-ship between multiple measurements. The limits of agreement areexpressed as averages of the differences 71.96 SD. The limits ofagreement for repeatability was 21.43 and 57.60, and for re-producibility it was 251.1 and �209.9.

4. Conclusion

A miniaturized alcohol monitoring system containing micro-fuel cell sensor with a compact potentiostat has been developed.The device includes data processing and transmission units withlow power consumption which can also provide highly stablesignal. The multi-modal technique provided a pathway to design amethod to eliminate the major interferent, humidity, in the alco-hol monitoring device. An algorithm was developed to implementauto-calibration ability in the developed device to improve theselectivity towards ethanol. The method for eliminating signal dueto humidity in this work demonstrated potential pathway foreliminating any organic volatile compound interfering signal. Themodification of nickel plated electrode with the thin film catalystcould be a prospective development of the present work, wheresensitivity and detection limit can be improved several fold.

Acknowledgment

This work was supported by NSF Innovation Corps #1616196.

Appendix A. Supporting information

Supplementary data associated with this article can be found inthe online version at http://dx.doi.org/10.1016/j.bios.2016.08.106.

References

Acevedo, C.A., Sánchez, E.Y., Reyes, J.G., Young, M.E., 2007. Volatile organic com-pounds produced by human skin cells. Biol. Res. 40, 347–355. http://dx.doi.org/

10.4067/S0716-97602007000400009.An, T.K., Yun, H.-J., Narote, R., Kim, R., Lee, S.U., Kim, Y., Nam, S., Cha, H., Jeong, Y.J.,

Kim, K., Cho, S., Kwon, S.-K., Kim, Y.-H., Park, C.E., 2014. Synthesis and char-acterization of an ester-terminated organic semiconductor for ethanol vapordetection. Org. Electron. 15, 2277–2284. http://dx.doi.org/10.1016/j.orgel.2014.06.033.

Antolini, E., 2007. Catalysts for direct ethanol fuel cells. J. Power Sources 170, 1–12.http://dx.doi.org/10.1016/j.jpowsour.2007.04.009.

Dubowski, K.M., 1994. Quality assurance in breath-alcohol analysis. J. Anal. Toxicol.18, 306–311. http://dx.doi.org/10.1093/jat/18.6.306.

Friedl, J., Stimming, U., 2013. Model catalyst studies on hydrogen and ethanoloxidation for fuel cells. Electrochim. Acta 101, 41–58. http://dx.doi.org/10.1016/j.electacta.2012.12.130.

Gamella, M., Campuzano, S., Manso, J., Rivera, G.G., de, López-Colino, F., Reviejo, A.J.,Pingarrón, J.M., 2014. A novel non-invasive electrochemical biosensing devicefor in situ determination of the alcohol content in blood by monitoring ethanolin sweat. Anal. Chim. Acta 806, 1–7. http://dx.doi.org/10.1016/j.aca.2013.09.020.

Goodman, R.A., Mercy, J.A., Loya, F., Rosenberg, M.L., Smith, J.C., Allen, N.H., Vargas,L., Kolts, R., 1986. Alcohol use and interpersonal violence: alcohol detected inhomicide victims. Am. J. Public Health 76, 144–149.

Harding, W., Di Bari, G. (Eds.), 1987. Testing of Metallic and Inorganic Coatings.ASTM International, 100 Barr Harbor Drive, PO Box C700, West Conshohocken,PA 19428-2959. http://dx.doi.org/10.1520/STP947-EB.

Hartnig, C., Roth, C., 2012. Fundamentals and Performance of Low Temperature FuelCells, Woodhead Pub.

Highway Traffic Safety Administration, N., Administration, Department of Trans-portation, U., 2013. 2013 Data: Alcohol-Impaired Driving.

Hruska, L.W., 1997. Smart batteries and lithium ion voltage profiles, In: Proceedingsof The Twelfth Annual Battery Conference on Applications and Advances, IEEE,pp. 205–210. http://dx.doi.org/10.1109/BCAA.1997.574104.

Hu Junming, H., Xu Liangfei, X., Li Jianqiu, L., Ouyang, M., Cheng Siliang, C., FangChuan, F., 2014. Water management in a self-humidifying PEM fuel cell systemby exhaust gas recirculation, In: Proceedings of the 2014 IEEE Conference andExpo Transportation Electrification Asia-Pacific (ITEC Asia-Pacific), IEEE, pp. 1–6.http://dx.doi.org/10.1109/ITEC-AP.2014.6941251.

Injury Research Foundation, T., 2007. Continuous Transdermal Alcohol Monitoring:A Primer for Criminal Justice Professionals.

Jadoon, S., Karim, S., Akram, M.R., Kalsoom Khan, A., Zia, M.A., Siddiqi, A.R., Murtaza,G., Jadoon, S., Karim, S., Akram, M.R., Kalsoom Khan, A., Zia, M.A., Siddiqi, A.R.,Murtaza, G., 2015. Recent developments in sweat analysis and its applications.Int. J. Anal. Chem. 2015, 1–7. http://dx.doi.org/10.1155/2015/164974.

Jiang, G., Goledzinowski, M., Comeau, F.J.E., Zarrin, H., Lui, G., Lenos, J., Veileux, A.,Liu, G., Zhang, J., Hemmati, S., Qiao, J., Chen, Z., 2016. Free-standing functio-nalized graphene oxide solid electrolytes in electrochemical gas sensors. Adv.Funct. Mater. 26, 1729–1736. http://dx.doi.org/10.1002/adfm.201504604.

Kadir, R.A., Rani, R.A., Zoolfakar, A.S., Ou, J.Z., Shafiei, M., Wlodarski, W., Kalantar-zadeh, K., 2014. Nb2O5 Schottky based ethanol vapour sensors: effect of metalliccatalysts. Sens. Actuators B Chem. 202, 74–82. http://dx.doi.org/10.1016/j.snb.2014.04.083.

Kim, H.-J., Lee, J.-H., 2014. Highly sensitive and selective gas sensors using p-typeoxide semiconductors: overview. Sens. Actuators B Chem. 192, 607–627. http://dx.doi.org/10.1016/j.snb.2013.11.005.

Kim, K.-C., Cho, S.M., Choi, H.-G., 2000. Detection of ethanol gas concentration byfuel cell sensors fabricated using a solid polymer electrolyte. Sens. Actuators BChem 67, 194–198. http://dx.doi.org/10.1016/S0925-4005(00)00398-1.

Lee, P.H., Hwang, S.S., 2009. Performance characteristics of a PEM fuel cell withparallel flow channels at different cathode relative humidity levels. Sensors 9,9104–9121. http://dx.doi.org/10.3390/s91109104.

Lidén, H., Vijayakumar, A., Gorton, L., Marko-Varga, G., 1998. Rapid alcohol de-termination in plasma and urine by column liquid chromatography with bio-sensor detection. J. Pharm. Biomed. Anal. 17, 1111–1128. http://dx.doi.org/10.1016/S0731-7085(98)00077-6.

LMP91000 Sensor AFE System: Configurable AFE Potentiostat for Low-Power Che-mical-Sensing Applications, 2014.

Marques, P.R., McKnight, A.S., 2009. Field and laboratory alcohol detection with2 types of transdermal devices. Alcohol. Clin. Exp. Res. 33, 703–711. http://dx.doi.org/10.1111/j.1530-0277.2008.00887.x.

Mason, M.F., Dubowski, K.M., 1974. Alcohol, traffic, and chemical testing in theUnited States: a résumé and some remaining problems. Clin. Chem. 20,126–140.

Mihajlovic, Z., Milosavljevic, V., Rajs, V., Milivojcevic, F., Zivanov, M., 2014. Miniaturelow cost electrochemical sensor module for measurement of gas concentration,in: 2014 22nd Telecommunications Forum TELFOR (TELFOR), IEEE, pp. 702–705.http://dx.doi.org/10.1109/TELFOR.2014.7034505.

Mochalski, P., King, J., Unterkofler, K., Hinterhuber, H., Amann, A., 2014. Emissionrates of selected volatile organic compounds from skin of healthy volunteers. J.Chromatogr. B 959, 62–70. http://dx.doi.org/10.1016/j.jchromb.2014.04.006.

Paixão, T.R.L.C., Bertotti, M., 2004. Development of a breath alcohol sensor using acopper electrode in an alkaline medium. J. Electroanal. Chem. 571, 101–109.http://dx.doi.org/10.1016/j.jelechem.2004.04.015.

Polissar, N.L., Suwanvijit, W., Gullberg, R.G., 2015. The accuracy of handheld pre-arrest breath test instruments as a predictor of the evidential breath alcoholtest results. J. Forensic Sci. 60, 482–487. http://dx.doi.org/10.1111/1556-4029.12659.

Room, R., Babor, T., Rehm, J., 2005. Alcohol and public health. Lancet 365, 519–530.http://dx.doi.org/10.1016/S0140-6736(05)17870-2.

A.H. Jalal et al. / Biosensors and Bioelectronics 87 (2017) 522–530530

Sakai, J.T., Mikulich-Gilbertson, S.K., Long, R.J., Crowley, T.J., 2006. Validity oftransdermal alcohol monitoring: fixed and self-regulated dosing. Alcohol. Clin.Exp. Res. 30, 26–33. http://dx.doi.org/10.1111/j.1530-0277.2006.00004.x.

Sanford, C.L., Mantooth, B.A., Jones, B.T., 2001. Determination of ethanol in alcoholsamples using a modular Raman spectrometer. J. Chem. Educ. 78, 1221. http://dx.doi.org/10.1021/ed078p1221.

Santra, S., Sinha, A.K., De Luca, A., Ali, S.Z., Udrea, F., Guha, P.K., Ray, S.K., Gardner, J.W., 2016. Mask-less deposition of Au-SnO2 nanocomposites on CMOS MEMSplatform for ethanol detection. Nanotechnology 27, 125502. http://dx.doi.org/10.1088/0957-4484/27/12/125502.

Semwal, V., Shrivastav, A.M., Verma, R., Gupta, B.D., 2016. Surface plasmon re-sonance based fiber optic ethanol sensor using layers of silver/silicon/hydrogelentrapped with ADH/NAD. Sens. Actuators B Chem 230, 485–492. http://dx.doi.org/10.1016/j.snb.2016.02.084.

Shabaneh, A.A., Girei, S.H., Arasu, P.T., Rahman, W.B.W.A., Bakar, A.A.A., Sadek, A.Z.,Lim, H.N., Huang, N.M., Yaacob, M.H., 2014. Reflectance response of taperedoptical fiber coated with graphene oxide nanostructured thin film for aqueousethanol sensing. Opt. Commun. 331, 320–324. http://dx.doi.org/10.1016/j.optcom.2014.06.035.

Shan, C., Yang, H., Han, D., Zhang, Q., Ivaska, A., Niu, L., 2010. Electrochemical

determination of NADH and ethanol based on ionic liquid-functionalized gra-phene. Biosens. Bioelectron. 25, 1504–1508. http://dx.doi.org/10.1016/j.bios.2009.11.009.

Swift, R.M., Martin, C.S., Swette, L., LaConti, A., Kackley, N., 1992. Studies on awearable, electronic, transdermal alcohol sensor. Alcohol. Clin. Exp. Res. 16,721–725.

WHO, 2015. Global status report on alcohol and health 2014, WHO.Williams, M.B., Reese, H.D., 1950. Colorimetric determination of ethyl alcohol. Anal.

Chem. 22, 1556–1561. http://dx.doi.org/10.1021/ac60048a025.Yu, M.-R., Suyambrakasam, G., Wu, R.-J., Chavali, M., 2012. Performance evaluation

of ZnO–CuO hetero junction solid state room temperature ethanol sensor.Mater. Res. Bull. 47, 1713–1718. http://dx.doi.org/10.1016/j.materresbull.2012.03.046.

Zhang, J., Tang, Y., Song, C., Xia, Z., Li, H., Wang, H., Zhang, J., 2008. PEM fuel cellrelative humidity (RH) and its effect on performance at high temperatures.Electrochim. Acta 53, 5315–5321. http://dx.doi.org/10.1016/j.electacta.2008.02.074.

Zuba, D., 2008. Accuracy and reliability of breath alcohol testing by handheldelectrochemical analysers. Forensic Sci. Int. . http://dx.doi.org/10.1016/j.forsciint.2008.03.002