Brief Communication

PersonalDistributedExposimeter forRadioFrequencyExposureAssessment inReal

Environments

ArnoThielens,1* HansDeClercq,2 SamAgneessens,3 Jeroen Lecoutere,2

LeenVerloock,1FrederickDeclercq,3 G€unterVermeeren,1EmmericTanghe,1

HendrikRogier,3 Robert Puers,2 LucMartens,1andWout Joseph11Wireless&CableGroup, Department of InformationTechnology, GhentUniversity/

iMinds, Ghent, Belgium2Microelectronics andSensorsGroup, Department of ElectrotechnicalEngineering,

CatholicUniversity ofLeuven, Heverlee, Belgium3ElectromagneticsGroup, Department of InformationTechnology, GhentUniversity,

Ghent, Belgium

For the first time, a personal distributed exposimeter (PDE) for radio frequency (RF) measurementsis presented. This PDE is designed based on numerical simulations and is experimentally evaluatedusing textile antennas and wearable electronics. A prototype of the PDE is calibrated in an anechoicchamber. Compared to conventional exposimeters, which only measure in one position on the body,an excellent isotropy of 0.5 dB (a factor of 1.1) and a 95% confidence interval of 7 dB (a factor of5) on power densities are measured. Bioelectromagnetics 34:563–567, 2013.© 2013 Wiley Periodicals, Inc.

Key words: dosimetry; radio frequency exposure; textile antennas; PIFA; wearableelectronics

Assessing the typical range of exposures to radiofrequency (RF) radiation is important to study compli-ance with international guidelines, such as those fromthe International Commission on Non-Ionizing Radia-tion Protection (ICNIRP) [ICNIRP, 1998] and nationallegislation. Moreover, assessing this range of exposureis also crucial for environmental safety and epidemio-logical studies that quantify potential health effects ofRF radiation. Personal, daily electromagnetic fieldexposure of a subject is usually assessed by means ofexposimeters or personal dosimeters, for which aprotocol has been developed [Röösli et al., 2010]. Thecurrently existing exposimeters are frequently used inmeasurement campaigns of exposures to RF radiation[Neubauer et al., 2007; Joseph et al., 2008, 2010;Knafl et al., 2008; Röösli et al., 2008; Frei et al., 2009;Viel et al., 2009; Bolte and Eikelboom, 2012]. Howev-er, they are faced with very large uncertainties due toshadowing of the body and variability of their positionon the body [Blas et al., 2007; Bahillo et al., 2008;Neubauer et al., 2010; Iskra et al., 2010, 2011; Bolteet al., 2011]. RF electromagnetic fields are incident onthe human body and will be reflected and absorbed.

This causes the electric field (strength) at the positionof an exposimeter to be different from the incidentelectric field (strength). This reflection and absorptionwill be dependent on the subject’s morphology [K€uhnet al., 2009; Neubauer et al., 2009, 2010; Bakkeret al., 2010; El Habachi et al., 2010]. Consequently,an exposimeter will measure the total electric fields(i.e., including scattering of the body) instead ofthe incident fields, which are necessary for the

Grant sponsor: Research Foundation—Flanders (FWO-V);grant number: 3G004612.

W. Joseph and E. Tanghe are Post-Doctoral Fellows of the FWO-V.

*Correspondence to: Arno Thielens, Wireless & Cable Group,Department of Information Technology, Ghent University/iMinds,Ghent, Belgium.E-mail: arno.thielens@intec.ugent.be

Received for review 29 October 2012; Accepted 3 April 2013

DOI: 10.1002/bem.21793Published online 6 June 2013 in Wiley Online Library(wileyonlinelibrary.com).

Bioelectromagnetics 34:563^567 (2013)

� 2013Wiley Periodicals, Inc.

quantification and comparison of exposure levels[ICNIRP, 1998]. Current exposimeters also show anunwanted dependence on the polarization of theincident electric fields [Bolte et al., 2011]. Otheruncertainties of the current exposimeters are out-of-band detections, a possible nonlinear power response,and technical defects [Bolte et al., 2011].

For the first time, an on-body worn personaldistributed exposimeter (PDE) containing multiple RFacquisition nodes is proposed. The nodes are optimallydistributed on the body and carefully calibrated in ananechoic chamber using a real human subject. Inaddition, numerical simulations of a heterogeneoushuman body phantom in realistic electromagneticenvironments are carried out in order to determine anoptimal placement of the nodes for measurement ofthe incident electric fields. This PDE is demonstratedat 950 MHz, the Global System for Mobile Communi-cations (GSM) downlink frequency that is present inmost environments [Joseph et al., 2010] and for whichincident fields can be described using stochasticparameters [Vermeeren et al., 2008; Iskra et al., 2011].A further novelty of this exposimeter is the use oftextile antennas, which are lightweight, flexible, andhave comparable performance to their rigid counter-parts [Locher et al., 2006; Hertleer et al., 2007;Agneessens et al., 2012], and wearable electronics[Carta et al., 2009; Jourand et al., 2010], which canboth be unobtrusively integrated in clothing in order tomaximize wearability of the PDE [Roelenset al., 2008; Reusens et al., 2009].

The antenna chosen for this application is aquarter-wavelength planar inverted F antenna (PIFA)that offers a good trade-off between antenna perfor-mance and size [Declercq et al., 2011]. The antennais constructed entirely from lightweight, flexible andbreathable materials in order to maximize wearability.Furthermore, the conductive ground plane, inherentto the antenna topology, minimizes the influence ofthe body on the antenna performance. The antennacovers the GSM 950 MHz downlink frequency bandand has a 60 MHz bandwidth. The simulated gainand beam width at half maximum are 2.9 dBi and110�, respectively, while the antenna efficiency is76.6%.

Each textile antenna is connected to an RFexposure acquisition system. These RF acquisition nodescontain a commercially available receiver that is tunedfor a 950 MHz link (CC1100E, Texas Instruments,Dallas, TX), and a microcontroller (PIC18f14k22, Mi-crochip, Chandler, AZ) for data management. In this firstprototype, the RF exposure data are communicated viaan inter-integrated circuit (I2C) to a main unit thatinterfaces with a personal computer using a USB. The

architecture is modular such that the amount of nodes iseasily extendable and other frequency bands can beexplored. Acquisition parameters such as sample rate andfrequency channel can be adjusted during measurements.Since the antennas, acquisition nodes, and all intercon-nections are all flexible and lightweight, they can becomfortably worn by volunteers without impeding bodymovement.

The design of the distributed exposimeter isbased on finite-difference time-domain (FDTD) simu-lations using the Virtual Family Male (VFM) [Christet al., 2010] (grid step ¼ 1.5 mm) in an uprightanatomical position. This is a heterogeneous humanbody model consisting of 81 different tissues basedupon magnetic resonance imaging of a healthy volun-teer with a body mass index (BMI) of 22.3 kg/m2. Thedielectric properties assigned to the phantom’s tissuesare taken from the Gabriel database [Gabrielet al., 1996]. The electric fields surrounding thephantom can be determined by means of FDTDsimulations and the methods presented in Vermeerenet al. [2008, 2012] and Iskra et al. [2011] enable thedetermination of these fields for realistic environ-ments.

To model the PDE, potential positions to deploythe wearable antennas (measurement positions) aredetermined at 1 cm from the phantom’s upper body. Inaddition, the required number of antennas (N) for aPDE is determined. In the procedure, it is taken intoaccount that realistic antennas have a finite surface andmay not overlap. In cylindrical coordinates (seeFig. 1a) a discretization step in the z-coordinate (alongthe body’s main axis) of 10 cm is chosen. For theazimuth angle w ¼ 0 : Dw : 2p; a step of Dw ¼ p=6 ischosen. Furthermore, antennas may not be positionedon the phantom’s face.

A linear regression model based on the simula-tion results is constructed. This enables us to deter-mine the incident root-mean-squared (RMS) electricfields Efree

RMS, using the electric fields recorded by thePDE Ebody

RMS;i, with i the ith selected measurementposition. The model is described by the followingequation:

EfreeRMS ¼ bo þ

XNi¼1

biEbodyRMS;i þ err ð1Þ

where bi (i ¼ 0, …, N) are the regression coef-ficients, and err is the residual. This regression modelwas built for the ideal case of ‘perfect’ sensors that canrecord the actual Ebody

RMS;i value in position i, and for therealistic case of linearly polarized antennas that onlyrecord one component in the plane tangential to the

564 Thielens et al.

Bioelectromagnetics

body, which is the case for the textile antennas usedthroughout this paper. We have constructed a step-wise algorithm that selects a set of measurementpositions (and polarizations), minimizing the error-on-prediction for a given posture of a selected phantom.

Figure 1 shows the results of our step-wisealgorithm using 4000 observations and 1000 controlvalues in the (realistic) indoor pico-cell environment[Vermeeren et al., 2008; Iskra et al., 2011]. The error-on-prediction of the linear regression model is shownin Figure 1b as a function of the number of measure-ment positions. Both perfect sensors and linearlypolarized textile antennas are considered. The allowedorientations of the linearly polarized antennas in thetangent plane to the body are multiples of p/9. Someaccuracy is lost (1.7% on average) when only onecomponent of the field can be measured, compared toa perfect measurement of the full RMS values. Using aresolution of 20� (p/9), we can estimate the incidentelectric fields with an average error of 14% using 3measurement positions (i.e., 3 antennas on the body),and 9.2% using 10 measurement positions. We thusprove that it is possible to accurately predict theincident electric fields using only a few measurementpositions on the body. Figure 1a shows the threepositions (blue ellipses) that, when combined, have thelowest error-on-prediction found by the model. Theapproximate polarizations are shown by black arrowsand a possible location for the processing unit isshown in red.

Using a set of three textile antennas placed onthe positions predicted by the step-wise algorithm(Fig. 1a), a first prototype of the exposimeter isconstructed. This model is calibrated in an anechoicchamber on a human subject with a BMI comparableto that of the VFM (�1 kg/m2). The subject is placedin an anechoic chamber in an upright anatomicalposition in the far field of a dipole radiating at950 MHz, and is rotated over 360� in azimuth angle wfor two orthogonal polarizations of the dipole (hori-zontal and vertical, perpendicular and parallel to thesubject’s axis of rotation, respectively), which emits ata constant output power. The subject is rotated in wbecause in reality the azimuthal angle of incidence of ameasured incident plane wave is unknown. Eachantenna (i) will receive a certain power Pbody

r;i as afunction of w. To determine an average measuredresponse Rmeas (dB), these received powers areaveraged over w and divided by the received power ofthe antennas in free-space Pfree

r , averaged over thesubject’s rotation axis [CENELEC, 2008]:

Rmeas ¼ 10� log< ð1=NÞPN

i¼1 Pbodyr;i Þ > w

Pfreer

!ð2Þ

An average of the received powers of thedifferent antennas is made in a PDE; this reduces thevariance on the measured data. Figure 2 shows theresponse Rmeas corresponding to the lowest 95%

Fig.1. Positions, linearpolarizations, anderror-on-predictionofa personal exposimeter. a: Optimal positions determined by thestep-wise algorithm (indicated by blue ellipses), possible loca-tion of the central processing unit (red rectangle), and linearpolarizations (black arrows) of the textile antennas at1 cm fromtheVFM.b:Averageerror-on-predictionofthepersonaldistribut-edexposimeterasafunctionofthenumberofmeasurementposi-tions (number of on-body antennas) for 4000 observations and1000controlvalues.Errorsaregivenforperfect sensorsandline-arlypolarized textileantennasat1 cm fromthephantom; resolu-tionoftheantennas’linearpolarizationis20�.

Personal Distributed Exposimeter 565

Bioelectromagnetics

confidence interval (c95) as a function of the numberof antennas in the PDE (one antenna corresponds to asingle exposimeter). For horizontal polarization, a6.5 dB reduction (division by a factor of 4.5) in c95 forthree antennas is measured on an initial c95 of 13.5 dB(factor of 22.4) for the best measurement with onetextile antenna (a single exposimeter). A 5 dB reduc-tion (division by a factor of 3) on 12.4 dB (factor of17.4) is measured on the c95 for vertically polarizedincident plane waves. This results in a final c95 of7 dB (factor of 5 for horizontal polarization) and7.4 dB (factor of 5.5 for vertical polarization).

In a previous study by Bolte et al. [2011], acommercial, single exposimeter was worn on the righthip of a subject rotated over 360� under exposure to aGSM downlink signal. An interquartile distance of6.5 dB (factor of 4.5) and 15.5 dB (factor of 35) weremeasured for incident horizontal and vertical polar-izations, respectively. In Neubauer et al. [2010],different possible locations of an exposimeter on thehuman body were investigated on a human bodyphantom in a simulated multipath environment at946 MHz. This led to an interquartile distance of 8 dB(factor of 6) and a c90 of 18 dB (factor of 62). Ourmeasurements using three antennas show an interquar-tile distance of 4.5 dB (factor of 2.8) and theaforementioned c95 of 7 dB (factor of 5) and 7.5 dB(factor of 5.5) for horizontally and vertically polarizedsources, which are considerably better. A c95 of18.5 dB (factor of 70.8) on the body is estimated forcommercial exposimeters at 900 MHz in realisticenvironments [Iskra et al., 2011]. The FDTD simu-lations at 950 MHz show a c95 of 25 dB (factor of

316) on the VFM for a single dosimeter in the sameexposure conditions as the calibration, and 22 dB(factor of 158) in realistic exposure conditions. Allthese values are much larger than the c95s measured inthis study.

The measurements show that a huge improvementin variance and thus accuracy can be obtained usingjust three antennas. Moreover, the PDE also exhibitsexcellent performance in terms of isotropy (I), definedas the ratio of Rmeas between two orthogonal polar-izations. Figure 2 shows that for the combination of thethree textile antennas, I ¼ 0.5 dB (factor of 1.1), whichis much better than the I of 6.4 dB (factor of 4.4) forcommercial exposimeters located on the body [Bolteet al., 2011]. The important quantity for exposureassessment is the ratio R of the power densities S:

R ¼ 10� log< ð1=NÞPN

i¼1 Sbodyr;i > w

Sfree

!¼ Rmeas þ A

ð3Þwhere Sbodyi is the power density on the location

of antenna i, Sfree is the free-space power density, andA represents a logarithmic correction factor. Since anR of zero is desired for accurate measurements, A isdetermined to be 15.4 dB (factor of 34.7).

In general, one can conclude that a personaldistributed exposimeter (PDE) with good accuracy canbe constructed using a limited number of antennas. Wehave developed a regression model to construct such aexposimeter based on numerical simulations, and havecalibrated a preliminary model of the exposimeterconsisting of 1, 2, or 3 antennas in an anechoicchamber using a real human. It is shown that theprototype of the PDE performs much better thancommercially available exposimeters.

Future research will consist of extending the PDEto other frequencies and calibrating it using multiplesubjects. The concept can also be used to distinguishbetween near-body and far-from-body sources.

REFERENCES

Agneessens S, Van Torre P, Declercq F, Spinnewyn B, StockmanG, Rogier H, Vande Ginste D. 2012. Design of a wearable,low-cost, through-wall Doppler radar system. Int J AntennasPropagation 12: Article ID 840924, 9 pp, DOI: 10.1155/2012/840924

Bahillo A, Blas J, Fernández P, Lorenzo RM, Mazuelas S, AbrilEJ. 2008. E-field assessment errors associated with RFdosemeters for different angles of arrival. R Radiat ProtDosimetry 132:51–56.

Bakker JF, Paulides MM, Christ A, Kuster N, van Rhoon GC.2010. Assessment of induced SAR in children exposed toelectromagnetic plane waves between 10 MHz and5.6 GHz. Phys Med Biol 55:3115–3130.

1 2 3−25

−20

−15

−10

−5

0

Number of antennas

R mea

s (dB

)

Horizontal polarizationVertical polarization

Fig. 2. Angular averaged response (Rmeas) with smallest 95%confidence interval as a function of the number of combined an-tennasonahumansubject; 95%confidenceintervalsareshownasblackbars.

566 Thielens et al.

Bioelectromagnetics

Blas J, Lago FA, Fernández P, Lorenzo RM, Abril EJ. 2007.Potential exposure assessment errors associated with body-worn RF dosimeters. Bioelectromagnetics 28:573–576.

Bolte JFB, Eikelboom T. 2012. Personal radiofrequency electro-magnetic field measurements in the Netherlands: Exposurelevel and variability for everyday activities, times of dayand types of area. Environ Int 48:133–142.

Bolte JFB, Van der Zande G, Kamer J. 2011. Calibration anduncertainties in personal exposure measurements of radio-frequency electromagnetic fields. Bioelectromagnetics 32:652–663.

Carta R, Jourand P, Hermans B, Thone J, Brosteaux D, Vervust T,Bossuyt F, Axisa F, Vanfleteren J, Puers R. 2009. Designand implementation of advanced systems in a flexible-stretchable technology for biomedical applications. SensorActuat A Phys 156:79–87.

Christ A, Kainz W, Hahn EG, Honegger K, Zefferer M, Neufeld E,Rascher W, Janka R, Bautz W, Chen J, Kiefer B, Schmitt P,Hollenbach H, Shen J, Oberle M, Szczerba D, Kam A,Guag JW, Kuster N. 2010. The Virtual Family—Develop-ment of surface-based anatomical models of two adults andtwo children for dosimetric simulations. Phys Med Biol 55:N23–N38.

Declercq F, Georgiadis A, Rogier H. 2011. Wearable aperture-coupled shorted solar patch antenna for remote tracking andmonitoring applications. Proceedings of the 5th EuropeanConference on Antennas and Propagation (EUCAP). Rome,Italy.

El Habachi A, Conil E, Hadjem A, Vazquez E, Wong MF, Gati A,Fleury G, Wiart J. 2010. Statistical analysis of whole-bodyabsorption depending on anatomical human characteristicsat a frequency of 2.1 GHz. Phys Med Biol 55:1875–1887.

European Committee for Electrotechnical Standardization (CEN-ELEC). 2008. Basic standard for the in-situ measurementof electromagnetic field strength related to human exposurein the vicinity of base stations. Brussels. Belgium: EN50492:2008.

Frei P, Mohler E, Neubauer G, Theis G, Burgi A, Frohlich J,Braun-Fahrlander C, Bolte J, Egger M, Roösli M. 2009.Temporal and spatial variability of personal exposure toradiofrequency electromagnetic fields. Environ Res 109:779–785.

Gabriel C, Gabriel S, Corthout E. 1996. The dielectric propertiesof biological tissues. Phys Med Biol 41:2231–2293.

Hertleer C, Tronquo A, Rogier H, Vallozzi L, Van Langenhove L.2007. Aperture-coupled patch antenna for integration intowearable textile systems. IEEE Antenn Wirel PropagationLett 6:392–395.

International Commission on Non-Ionizing Radiation Protection(ICNIRP). 1998. Guidelines for limiting exposure to time-varying electric, magnetic, and electromagnetic fields (up to300 GHz). Health Phys 74:494–522.

Iskra S, McKenzie R, Cosic I. 2010. Factors influencing uncertain-ty in measurement of electric fields close to the body inpersonal RF dosimetry. Radiat Prot Dosimetry 140:25–33.

Iskra S, McKenzie R, Cosic I. 2011. Monte Carlo simulations ofthe electric field close to the body in realistic environmentsfor application in personal radiofrequency dosimetry. RadiatProt Dosimetry 147:517–527.

Joseph W, Vermeeren G, Verloock L, Heredia MM, Martens L.2008. Characterization of personal RF electromagnetic fieldexposure and actual absorption for the general public.Health Phys 95:317–330.

Joseph W, Vermeeren G, Verloock L, Martens L. 2010. Estimationof whole-body SAR from electromagnetic fields usingpersonal exposure meters. Bioelectromagnetics 31:286–295.

Jourand P, DeClercq H, Puers R. 2010. Robust monitoring of vitalsigns integrated in textile. Sensor Actuat A Phys 161:288–296.

Knafl U, Lehmann H, Riederer M. 2008. Electromagnetic fieldmeasurements using personal exposimeters. Bioelectromag-netics 29:160–162.

K€uhn S, Jennings W, Christ A, Kuster N. 2009. Assessment ofinduced radio-frequency electromagnetic fields in variousanatomical human body models. Phys Med Biol 54:875–890.

Locher I, Klemm M, Kirstein T, Troster G. 2006. Design andcharacterization of purely textile patch antennas. IEEE TransAdv Packag 29:777–788.

Neubauer G, Feychting M, Hamnerius Y, Kheifets L, Kuster N,Ruiz I, Sch€uz J, Uberbacher R, Wiart J, Röösli M. 2007.Feasibility of future epidemiological studies on possiblehealth effects of mobile phone base stations. Bioelectromag-netics 28:224–230.

Neubauer G, Preiner P, Cecil S, Mitrevski N, Gorter J, Garn H.2009. The relation between the specific absorption rate andelectromagnetic field intensity for heterogeneous exposureconditions at mobile communications frequencies. Bioelec-tromagnetics 30:651–662.

Neubauer G, Cecil S, Giczi W, Petric B, Preiner P, Fröhlich J,Röösli M. 2010. The association between exposure deter-mined by radiofrequency personal exposimeters and humanexposure: A simulation study. Bioelectromagnetics 31:535–545.

Reusens E, Joseph W, Latré B, Braem B, Vermeeren G, Tanghe E,Martens L, Blondia C, Moerman I. 2009. Characterizationof on-body communication channel and energy efficienttopology design for wireless body area networks. IEEETrans Inf Technol Biomed 13:933–945.

Roelens L, Joseph W, Reussens E, Vermeeren G, Martens L. 2008.Characterization of scattering parameters near a flat phan-tom for wireless body area networks. IEEE Trans Electro-mag Compat 50:185–193.

Röösli M, Frei P, Mohler E, Braun-Fahrländer C, Burgi A,Fröhlich J, Neubauer G, Theis G, Egger M. 2008. Statisticalanalysis of personal radiofrequency electromagnetic fieldmeasurements with nondetects. Bioelectromagnetics 29:471–478.

Röösli M, Frei P, Bolte J, Neubauer G, Cardis E, Feychting M,Gasjek P, Heinrich S, Joseph W, Mann S, Martens L,Mohler E, Parslow RC, Poulsen AH, Radon K, Sch€uz J,Thuroczy G, Viel J, Vrijheid M. 2010. Conduct of apersonal radiofrequency electromagnetic field measurementstudy: Proposed study protocol. Environ Health 9:23.

Vermeeren G, Joseph W, Olivier C, Martens L. 2008. Statisticalmultipath exposure of a human in a realistic electromagneticenvironment, Health Phys 94:345–354.

Vermeeren G, Joseph W, Martens L. 2012. Statistical multi-pathexposure method for assessing the whole-body SAR in aheterogeneous human body model in a realistic environ-ment. Bioelectromagnetics 34:240–251.

Viel JF, Cardis E, Moissonnier M, deSeze R, Hours M. 2009.Radiofrequency exposure in the French general population:Band, time, location and activity variability. Environ Int35:1150–1154.

Personal Distributed Exposimeter 567

Bioelectromagnetics