Corresponding author: Arno Thielens, Ghent University, Technologiepark 15, B-9052 Ghent,

Belgium, email: arno.thielens@intec.ugent.be, phone: +32 9 33 14918, fax: +32 9 33 14899

Running title: RF Exposure near an Attocell

Grant sponsor info: Arno Thielens is a post-doctoral fellow of Flanders Innovation and

Entrepreneurship under agreement no. 150752. This work has received funding from the iMinds

Internet of Things project.

Conflict of interest: none

Abstract- The wireless, radio-frequency (RF) telecommunication networks of the future will

provide users with gigabit per second data rates. Therefore, these networks are evolving

towards hybrid networks, which will include the commonly used macro- and microcells in

Radiofrequency Exposure near an Attocell as Part of an Ultra-High Density Access Network

Arno Thielens*, Günter Vermeeren*, Olivier Caytan*, Guy Torfs*, Piet Demeester*, Johan

Bauwelinck*, Hendrik Rogier*, Luc Martens *, and Wout Joseph *

(email:arno.thielens@intec.ugent.be, tel : +32 9 33 14918, fax:+32 9 33 14899)

*Department of Information Technology, Ghent University, Ghent, Belgium

Technologiepark 15, B-9052 Ghent, Belgium

combination with local ultra-high density access networks consisting of so-called attocells.

The usage of attocells requires a proper compliance assessment of the exposure to RF

electromagnetic (EM) radiation. This paper presents, for the first time, such a compliance

assessment of an attocell operating at 3.5 GHz with an input power of 1 mW, based on both

root-mean-squared electric field strength (Erms) and peak 10 g-averaged specific absorption

rate (SAR10g) values. The Erms values near the attocell are determined using finite-difference

time-domain (FDTD) simulations and measurements by a tri-axial probe. They are

compared to the International Commission on Non-Ionizing Radiation Protection’s

(ICNIRP) reference levels. All measured and simulated Erms values above the attocell are

below 5.9 V/m and lower than the reference levels. The SAR10g values are measured in a

homogeneous phantom, which resulted in a SAR10g of 9.7 mW/kg, and using FDTD

simulations, which resulted in a SAR10g of 7.2 mW/kg. FDTD simulations of realistic

exposure situations are executed using a heterogeneous phantom, which yielded SAR10g

values lower than 2.8 mW/kg. The studied dosimetric quantities are in compliance with the

ICNIRP guidelines when the attocell is fed an input power < 1 mW. The deployment of

attocells is thus a feasible solution to provide broadband data transmission without

drastically increasing the personal RF exposure.

Key wWords- attocells, radiofrequency, compliance, SAR, 5th generation networks,

INTRODUCTION

Worldwide, billions of users are connected through wireless communication [5GPP, 2014; Nokia,

2014; EricsssonDahlman et al., 20145; Samsung, 2015]. These users require a broadband service

that is ever increasing in speed and amount of communicated data. Simultaneously, the users

want to access their data, stored either locally or remotely in the cloud, or communicate with one

another on any given time and location. This ultra-broadband, wireless access is enabled by

radio-frequency (RF) electromagnetic fields (EMFs), which are interchanged between antennas.

These are on the one side part of a network operated by a telecommunication company and on the

other side embedded in a personal device operated by the user. Different technologies for

wireless communication, assigned to specific frequency bands, exist. As these technologies

evolve over time, they are commonly categorized in new generations of wireless

telecommunication systems. The current wireless technologies are referred to as 4th generation

networks [Chowdhury et al., 2010]. Each new generation is usually characterized by a new

standard, an increased bandwidth, and, consequently, a change in the type of antennas that are

being used [Chowdhury et al., 2010]. The current telecommunication networks up to the 3rd

generation relyied mainly on a heterogeneous network of RF antennas for emitting and receiving

signals. These range from large base station antennas (BSAs) [Thielens et al., 2013; Thors et al.,

2014] that cover large areas (km²) [Gupta and Kumar, 2015], and are referred to as macrocells, to

small antennas that emit lower powers, only cover areas of several square meters, and are referred

to as femtocells [Aerts et al., 2015]. operating in a single frequency band and covering large areas

[Gupta and Kumar, 2015]. Consequently, these antennas are referred to as macrocells. However,

these cellular networks also employed smaller cells, referred to as microcells and picocells

[Cooper et al., 2006], in relatively large indoor environments such as subway stations and

shopping malls. In the 4th generation of telecommunication networks, a transition was made

towards base station antennas (BSAs) that used multiple frequency bands [Thielens et al., 2013],

BSAs that communicated via multiple antennas towards multiple users who would also operate

multiple antennas simultaneously, so-called multiple-input multiple output (MIMO) [Thors et al.

2014], and a heterogeneous network where users are served by macrocells and picocells operated

by the operators, as well as cells that cover smaller areas and are operated by the users [Aerts et

al., 2015], depending on the most energy efficient solution. These antennas are commonly

referred to as femtocells.

It is expected that in future 5th generation networks, the required bandwidth and data throughput

will increase [5GPP, 2014; Nokia, 2014; Dahlman et al., 2014Ericssson, 2015; Samsung, 2015].

A potential approach to deal with this increase in demands from the network is to deploy an ultra-

high density access network consisting of large number of very small cells, so called attocells.

These are integrated in a floor and cover very small areas of only a few dm².Yet, they provide a

very high data rate in these small areas. A first concept of such a network is described in Lannoo

et al. [2015] and promises large improvements in wireless telecommunications. Since attocells

will be smaller and widely distributed in indoor environments, the antennas used in attocells will

be closer to the user than those employed in previous generations of cellular networks [Gupta and

Kumar, 2015]. This close distance might be a risk in terms of personal exposure to the RF EMFs

emitted by the antennas. Consequently, a proper compliance with existing guidelines [ICNIRP,

1998] and regulations is necessary before such a network will be commercially introduced.

However, up till now, personal exposure near attocells has not been studied.

The absorption of RF EMFs in the human body is commonly quantified using the specific

absorption rate (SAR), which is the ratio of the absorbed RF power and the mass in which it is

absorbed. Basic restrictions on the whole-body averaged SAR (SARwb) and localized, peak 10 g-

averaged SAR (SAR10g) exist [ICNIRP, 1998]. Reference levels on the incident root-mean-

squared electric (Erms) and magnetic (Hrms) field strengths are derived from the basic restrictions

[ICNIRP, 1998]. The International Electrotechnical Commission (IEC) [IEC, 2010] and the

European Committee for Electrotechnical Standardization (CENELEC) [CENELEC, 2010] have

both issued guidelines on determining compliance near antennas using measurements or

simulations of the SAR, Erms, and Hrms.

The compliance with regard to exposure guidelines has not been studied for the antennas used in

attocells. However, other types of BSAs have been studied. In Thors et al. [2008, 2014], Kos et

al. [2011], and Thielens et al. [2013], Thors et al. [2008, 2014], Kos et al. [2011], the compliance

of BSAs used in macrocells has been studied and compliance boundaries for those BSAs have

been defined. Cooper et al. [2006] provided an overview of compliance of the smaller micro- and

picocells operating at frequencies from 80 MHz to 2500 MHz. The exposure in femtocells is

compared to that in macrocells in Zarikoff and Malone [2013] and Aerts et al. [2014, 2015] and

Zarikoff and Malone [2013]. In Boursianis et al., [2012] and Aerts et al. [2014] measurements of

both the uplink of a mobile device connected to a femtocell and the downlink from the femtocell

itself showed that during calls the uplink exposure is dominant. However, a non-user near the RF

source will still receive a dominant exposure from the femtocell [Aerts et al., 2014].

The objective of this paper is to, for the first time, study the RF exposure near an attocell. To this

aim both measurements near an actual attocell configuration and numerical simulations of a

model of the same attocell are executed. First, the Erms field strengths are studied near the attocell

in its worst-case operating conditions. Second, the peak SAR10g is studied in a standardized

phantom near the antenna used in the attocell. Third, the realistic SAR10g in a heterogeneous

phantom is studied near the attocell. These results are important for the implementation of 5 th

generation networks, which require a compliance with EMF standards.

MATERIALS AND METHODS

The Attocell

The studied attocell wais the one described in Lannoo et al. [2015]. Figure 1 shows an illustration

of the implementation of the attocells in a transparent floor.

[Figure 1]

A layer of acrylic glass (thickness = 6 mm) wais supported by a wooden framework that makeds

up the different attocells. The purpose of this configuration wais to allow very fast, ultra-

broadband communication between a subject, who wouldis wearing a device and could moves

around on the floor, and the attocells, which weare contained within the floor. A subject would

thus move around on the structure shown in Figure 1 (a). Two types of wooden beams weare

used (see Fig. 1. (a) and (b)): the layer underneath the acrylic glass consisteds of parallel small

beams (dimensions of the cross section: 1.7 x 1.7 cm²), while these weare supported by a second

layer of orthogonally placed thicker beams (cross section: 4.2 x 4.2 cm²) made from the same

wood. This resulteds in square cells of 15 x 15 cm². Each cell containeds an antenna that

operateds between to 3.25 GHz and -3.75 GHz. The antennas weare placed with their slots either

parallel to the X or Y axis (see Fig. 1. (c)). , in order toThis implementation ensured polarization

diversity. Figure 2 shows an illustration of the studied antennas and their dimensions.

[Figure 2]

The antennas weare linearly polarized planar, substrate-integrated-waveguide (SIW) cavity-

backed slot antennas, made out of foam material (the substrate) and copper plated nylon (the

conducting elements). The antennas weare based on a design described in Lemey et al. [2014]

and Caytan et al. [2015]. The antenna presented in Lemey et al. [2014] hads the same design, but

useds a grounded coplanar waveguide as feeding structure instead of a through and an SMA. The

antenna presented in Caytan et al. [2015] hads the same design, but useds a cork substrate instead

of a foam substrate. The antennas studied in this paper weare supported by a layer of plastic

(thickness: 5 mm), see Figure. 1 (b), which wais attached to the bottom of the wooden beams.

The plastic wais penetrated by a Sub-Miniature A (SMA) connector, which connecteds the

antenna to opto-electronic components., These,which in their turn, are connected to a fiber-

backbone [Lannoo et al., 2015]. The antennas weare centered in the different attocells and weare

either horizontally (with their slot parallel to the X-axis) or vertically polarized (with their slot

parallel to the Y-axis), see Figure 1. The communication protocol applieds a maximum input

power of 0 dBm (1 mW) at the feed point of the antennas. These emitted at a center frequency of

3.5 GHz in a frequency band between 3.25 GHz and-3.75 GHz. This frequency band wais chosen

because it is located in the lower part of the Ultra-Wide Band (UWB), which ranges from 3.1

GHz to 10.6 GHz [Porcino and Hirt, 2003].

Lannoo et al., [2015] envisage a transmission protocol that uses so-called soft handovers, where

both localization and path prediction are used to optimize energy consumption and connectivity

during the handover between different cells, which can emit simultaneously with a total input

power < 1mW. However, in the current study we assumed a simplified The transmission protocol

of the attocells with so-called hard handovers, where the transmission between two cells wais

designed in such a way that only one antenna in the floor couldan emit at a time. [Lannoo et al.

2015]. Therefore, we only considered configurations where one antenna wais emitting RF

radiation during the simulations and measurements. For compliance purposes, the worst-case

operational conditions of 1 mW input power in the antenna weare considered at a harmonic

frequency of 3.5 GHz.

The antennas placed in the attocells weare characterized using reflection coefficient (S11)

measurements that were performed by a vector network analyzer (N5242A PNA-X, Agilent

Technologies, Santa Clara, CA, USA).

Exposure Measurements

The EMF measurements near the attocells weare performed using a DASY3 mini system

(SPEAG, Zurich, Switzerland), see Figure 3. The system, consistedting of two parts. First,: an

isotropic E-Field Probe for gGeneral nNear-fField mMeasurements (ER3DV6, SPEAG,

Switzerland), which wais able to measure in a frequency range of 40 MHz - 6 GHz, hads a

dynamic range of 2-1000 V/m, and a linearity of ±0.2 dB. The second component of the system

was , and a Pythron IXE α-C-T robot, which allows forenabled a three-dimensional translation of

the probe with an accuracy of 0.005 mm. The total expanded uncertainty (k=2) for of a typical

dosimetry the setup of this kind for free-space near-field measurements equals was 16 14 %,

based on the DASY3 mini application note [Speag]. Matlab v7.04 (Mathworks, Natick, MA,

USA) wais used for steering the robot and interfacing with DASY system and its software.

During the measurements, the antenna in the attocell wasis excited by a sinusoidal wave from a

signal generator (SMB 100A, Rhode & Schwartz, Munich, Germany) with with a power of

16 dBm and rescaled to an antenna input power of 0 dBm (1 mW) at 3.5 GHz. The field probe

wais used to scan an area of 30 cm in the X-direction and 30 cm in the Y-direction

(corresponding to 4 attocells) with a step size of 15 mm resulting in a total of 441 points in the

horizontal plane, at different separation distances in the Z-direction from the antenna. During the

measurements the separation distance was defined as the distance to the top of the antenna, which

wais placed in one of the four attocells. Measurements were performed at distances 70 mm,

100 mm and 120 mm. The antenna wais either vertically or horizontally polarized. First,

measurements weare carried out above the acrylic glass. Second, measurements along a vertical

line weare executed without the acrylic glass present, as a function of the separation distance

between the probe and the antenna. The measured field strengths weare compared to the

reference levels issued by ICNIRP [ICNIRP, 1998].

[Figure 3]

After the measurements of free-space EMFs, one of the antennas wais detached from the setup

and placed underneath a standardized oval flat phantom (ELI4, SPEAG, Zurich, Switzerland)

compatible with the IEC 62209-2 standard, see Figure 3. This phantom wais filled with tissue

simulating liquid HSL2450 (SPEAG, Switzerland) used at 3.5 GHz with a measured relative

permittivity ε r= 36.1 and conductivity σ= 2.47 S/m. Figure 3 shows an illustration of the studied

setup under study. The patch of the antenna was facing upwards towards the flat surface of the

phantom at a distance of 10 mm from the tissue simulating liquid. SAR measurements weare

executed in the liquid using an antenna calibrated for this purpose: the isotropic dosimetric probe

EX3DV4 (SPEAG, Switzerland). This probe couldan measure from 10 MHz up to 6 GHz and

hads a dynamic range of 0.010 mW/g up to 100 mW/g with a linearity of ±0.2 dB. The and a

noise level of the probe was which is typically below 0.001 mW/g. The total expanded

uncertainty (k=2) of the dosimetric setup was 16 % [Speag]. The measured SAR values weare

used to determine the peak SAR10g according to [IEC, 2010] and weare compared to the reference

levels issued by ICNIRP [ICNIRP, 1998]. The size of the area scan was 8 cm by 12 cm with a

measurement step of 10 mm. The measurements were interpolated to a homogeneous grid with a

step of 2 mm using cubic spline interpolation. The peak spatial-averaged SAR in 10 g has been

assessed in a volume of 32 mm by 32 mm by 30 mm with a grid step of 8 mm by 8 mm by 5mm

around the location of maximum peak SAR obtained from the area scan.

Numerical Simulations

Numerical simulations weare executed in order to validate the measurements and estimate the

SAR in heterogeneous phantoms. First, the used antenna wais modeled using two methods: the

method of moments (MoM) implemented in Feko (Feko, Stellenbosch, South Africa) and the

finite-difference time-domain (FDTD) method implemented in Sim4Life (Zürich Med Tech,

Zürich, Switzerland). The components of the antennas hadve the same dimensions as those

shown in Figures 1 and 2. The foam material wais assigned a measured relative permittivity ε r=

1.495 and a conductivity σ= 0.0058 S/m. A broadband simulation (3-4 GHz) wais then executed

using both simulation packages in order to determine the reflection coefficient (S11) of the

antenna. These results weare then compared to those obtained by measurements in order to

validate the created antenna model.

Second, the model of the antenna wais placed in a numerical model of the floor (see Fig. 1 (d)).

The acrylic glass wais assigned a relative permittivity ε r= 2.6 and a conductivity σ= 0.0025 S/m,

the wood hads an ε r= 1.6 and σ= 0.0059 S/m [Torgovnikov, 1993], and the plastic had an ε r=

2.25 and σ= 0.0005 S/m. All the listed dielectric parameters weare valid at 3.5  GHz. A

simulation domain bounded by perfectly matched layers (PML) wais defined around the attocell.

This simulation domain wais then discretized using a rectilinear grid. A FDTD harmonic

simulation at 3.5 GHz wais executed with and without the presence of the acrylic glass, in order

to determine the EM fields surrounding the attocell. The simulations had a total number of

3.9 x 106 cells and reached a steady state after 11 periods. The expanded uncertainty (k=2) of the

simulation equaled 12 % [Bakker et. al., 2010, 2011].

Third, the antenna model wais placed underneath the homogeneous phantom (ELI4) in the same

configuration as used in the SAR measurements. An input power of 1 mW wais fed to the

antenna at a frequency of 3.5 GHz. The shell wais assigned an ε r= 3.5 and σ= 0 S/m, while the

liquid wais assigned an ε r= 36.1 and σ= 2.47 S/m. The simulations wasare compared with the

measurements. They It was used can be applied to demonstrate determine compliance with the

ICNIRP basic restrictions [ICNIRP, 1998] according to standard IEC62209-2 [IEC, 2010]. This

simulation had a total number of 6.5 x 106 cells. The expanded uncertainty (k=2) on SAR for the

simulation equaled also 12 % [Bakker et. al., 2010, 2011].

Fourth, a heterogeneous phantom wais placed in the vicinity of the attocell containing the studied

antenna, which emitting emitted a sinusoidal wave at 3.5 GHz with an input power of 1 mW. The

antenna wais placed either horizontally or vertically polarized in the attocell. The phantom, see

Figure 4 (a), wais either standing with its right foot centered above the emitting attocell, see

Figure 4 (b), or with the back or front of its head centered on the emitting attocell, see Figure 4

(c)-(e). These configurations represented realistic exposure situations where the phantom wais

either standing on the attocell or lying on the floor. The SAR results of these simulations weare

indicative for a realistic RF exposure near an attocell. The expanded uncertainty (k=2) on SAR

for the simulations with heterogeneous human body models equaled 23 % [Bakker et. al., 2010,

2011].

[Figure 4]

The Virtual Family Male (VFM) [Christ et al., 2010b] wais selected as human body model or

phantom. This model is shown in Figure 4. The VFM waisis a 34-year-old adult male with a mass

of 72.2 kg and a height of 1.80 m. The dielectric properties of the phantom weare loaded from the

database included in Sim4life, which wais based on Gabriel [1996]. The level of the whole-body

averaged SAR in the ICNIRP basic restrictions for the general public at 3.5 GHz [ICNIRP, 1998]

waisis 0.08 W/kg. The phantom has a mass of 72.2 kg, which implies that its SARwb can couldcan

never be higher than 0.08 W/kg for an input power of 1 mW. Therefore, only the peak SAR 10g

wais studied for the heterogeneous phantom.

The peak SAR10g values can could be used to determine allowed input powers of the antennas. To

this aim, the peak SAR10g values weare determined for a certain input power Pin = 1 mW. These

values weare then compared to the basic restrictions (BR) on the peak SAR10g for the general

public, which weareis 2 W/kg for the head and trunk and 4 W/kg for the limbs. The allowed input

power (P¿allowed) wais then determined using:

P¿allowed= BR

peak SAR10 g× P¿

(1)

RESULTS

Modelling of the SIW antenna

Figure 5 shows the S11 of the studied antenna, obtained using MoM, FDTD, and measurements

with a Vector Network Analyzer (VNA). Figure 5 also shows FDTD simulations of the S11 of the

antenna placed in the attocell, with and without the acrylic glass. Dispersion of the dielectric

parameters was not taken into account in this simulation.

[Figure 5]

All the S11 values are below -10 dB between 3.25 GHz and 3.75 GHz. The -10 dB bandwidth of

the antenna is 666 MHz, 610 MHz, and 724 MHz, using FDTD, MoM, and free-space

measurements, respectively. A relatively good agreement in bandwidth between the numerical

model and the real antenna is thus obtained with relative errors of 8% and 16% for the FDTD

model and the MoM model, respectively. The frequency of the first local minimum is located at

3.38 GHz, 3.34 GHz, and 3.34 GHz, using FDTD, MoM, and measurements, respectively. The

location of this minimum is thus in very good agreement (1% deviation from the real antenna for

FDTD). The frequency of the second local minimum is located at 3.74 GHz, 3.65 GHz, and 3.77

GHz, using FDTD, MoM, and measurements, respectively. The frequency corresponding to this

minimum is again in good agreement between the simulations and the measurement (deviation <

3%). A shift towards lower frequencies of resonance and deeper peaks of resonance are observed

when then antenna is placed in the vicinity of dielectric objects. Frequency shifts of

approximately 10 MHz and 50 MHz are found for the configuration with only wood and wood

and acrylic glass, respectively, in comparison to the free-space FDTD simulation.

EMF Measurements and Numerical Simulations

Figure 6 shows the results of electric field strength measurements and FDTD simulations at a

separation distance of 8 mm above the acrylic glass, which corresponds to a distance of 7 cm

above the antenna (the closest area scan performed to the antenna), while the antenna is fed with

an input power of 1 mW at 3.5 GHz.

[Figure 6]

Figure 6 shows a larger spread of Erms in the direction orthogonal to the slot of the antenna. The

patterns in Erms have comparable shapes for the measurements and simulations. The fields above

the floor are all lower than 5.9 V/m, which is lower than the ICNIRP reference level of 61 V/m at

this frequency. The maximum Erms values are 5.9 V/m and 5.3 V/m for the antenna placed with its

slot parallel to the X-axis, determined using measurements and simulations, respectively. These

values are 5.1 V/m for both simulations and measurements for the orthogonal configuration. The

measured and simulated peak Erms values are thus in excellent agreement (< 10% relative error).

Figure 7 shows that the electric field strength decreases as a function of the distance from the

antenna.

[Figure 7]

Without the acrylic glass present, which is not achievable in operational mode, the maximally

measured electric field strength is 19 V/m (at 10 mm from the antenna) and the maximally

simulated electric field strength is 69 V/m (at the antenna’s surface). The antenna can thus emit

an electric field strength that exceeds the ICNIRP reference levels. However, it must be pointed

out that the ICNIRP reference levels are intended for far-field exposure situations. An evaluation

of RF compliance with the ICNIRP guidelines [ICNIRP, 1998] requires SAR measurements

and/or simulations to be compared with the basic restrictions. These are presented in the next

section. The three closest measurements and the simulations are in excellent agreement (average

relative error of 8%), while for separation distances higher than 30 mm from the antenna, the

decay in measured Erms follows a different trend than the simulated values.

SAR Measurements and Numerical Simulations

Figure 8 shows the simulated and measured SAR in the IEC phantom at 5 mm from the

phantom’s shell inside the liquid for the attocell whose antenna receives an input power of 1 mW

at 3.5 GHz. The distance of 5 mm is in compliance with the IEC 62209-2 standard for the area

scan since the standard requires that: “ the distance between the geometrical center of the probe

detectors and the inner surface of the phantom shall be 8 mm or less and constant within

±1,0 mm” [IEC, 2010].

[Figure 8]

The simulations and the measurements exhibit a similar SAR pattern. The peak SAR 10g is

7.2 mW/kg using simulations and 9.7 mW/kg using measurements in the phantom. These values

are more than 200 times smaller than the ICNIRP basic restriction on the SAR10g which is 2 W/kg

at 3.5 GHz. As mentioned before, these values are not representative for a realistic SAR 10g value,

since they are measured with the emitting antenna touching the phantom. In practice, the antennas

will be located at least 6 cm underneath the acrylic glass. It should be noted that due to the close

distance of the phantom to the antenna, the |S11| of the antenna will differ from the one shown in

Figure 5. The |S11| is -2.4 dB at 3.5 GHz during the presented simulations. This implies that if the

antenna is fed a constant input power, which is the case during the measurements and the

simulations, then the output power reduces since more power is reflected in the antenna.

RF Exposure of Heterogeneous Phantoms

Figure 9 shows sagittal cross sections of the VFM at the location of the maximum SAR. The

figure shows that the SAR is localized to the area in front of the emitting antenna. Therefore, the

localized, peak SAR10g is evaluated for the six different configurations shown in Figure. 4 and

compared to the ICNIRP basic restrictions on the peak SAR10g for the general public, which are 2

W/kg for the head and trunk and 4 W/kg in the limbs.

[Figure 9]

Table 1 lists the peak SAR10g values obtained from the FDTD simulations, renormalized to an

input power of 1 mW. All values are far below the ICNIRP basic restrictions on peak SAR 10g.

Corresponding to these low SAR10g values, we find, using Equation 1, relatively high allowed

input powers (>0.29 W, see Table 1) in comparison to the 1 mW input power, which is applied in

the standard transmission protocol. The |S11| of the studied antenna is located between -19 dB and

-14 dB at 3.5 GHz, depending on the studied configuration.

[Table 1]

DISCUSSION

In this study, the RF exposure near an attocell is studied using both compliance measurements

and numerical simulations. In order to execute numerical simulations, the studied attocell and the

antenna it contains, see Figuress. 1 and 2, have to be modeled in the simulation software under

consideration. The antenna model is based on the real antenna’s dimensions and dielectric

parameters. The antenna’s feed was modeled as a voltage source, whereas this is an SM-A

connector in reality, see Figure. 2. Figure 5 shows that this model has yielded a good agreement

in the S11 between the FDTD simulations and the measurements of the antenna’s S11 in free-space.

An additional validation was executed using MoM, which was in good agreement with the FDTD

simulation and the measurement in terms of the first peak of resonance of the antenna (around 3.4

GHz), but showed a shift of the second resonance peak towards lower frequencies. Such a shift

was also observed for the FDTD simulations that include dielectric materials around the antenna

(wood and acrylic glass). The same effect has already been observed by Aminzadeh et al. [2016],

where dipole antennas were modeled for operation next to a homogeneous phantom. Aminzadeh

et al. [2016] found that resonance peak of the dipole shifted towards lower frequencies next to

dielectric materials and therefore, a smaller dipole, corresponding to a higher frequency of

resonance in free space, is needed to achieve a desired resonance next to the human body. In the

case of the attocell, a redesign of the antenna is not necessary since the |S11| remains <-10 dB in

the target frequency band between 3.25 and 3.75 GHz.

The electric field strengths show a good agreement in shape (Fig. 6) and decay as a function of

distance from the antenna (Fig. 7) between FDTD simulations and measurements. The ICNIRP

reference level (61 V/m) for the general public cannot be exceeded above the acrylic glass, when

applying an input power of 1 mW. However, the numerical simulations show that values higher

than the reference level for the general public exist near the antenna’s surface. In normal

operating conditions, the antennas cannot be approached by the general public, due to the layer of

acrylic glass.

Compliance measurements are also executed with a standardized phantom according to the IEC

62209-2 standard [IEC, 2010]. FDTD simulations were executed as well, using the same set up.

The measurements and simulations show a similar SAR pattern in the homogeneous phantom.

This SAR compliance assessment is complemented by FDTD simulations using a heterogeneous

phantom, which show relatively low peak SAR10g values in comparison to the ICNIRP basic

restrictions, see Table 1. Consequently, relatively high allowed output powers are determined in

comparison to the 1 mW, which is applied in real operation of the attocells. The highest SAR10g is

obtained for the configuration where the VFM’s head is facing the emitting attocell. We attribute

this to a local enhancement of the SAR in the nose, which touched the acrylic glass. This local

enhancement of SAR at frequencies > 900 MHz, including 3.5 GHz, in extremities of the same

heterogeneous phantom has already been demonstrated in Vermeeren et al. [2010].

Exposure from multiple attocells simultaneously has not been studied yet, although Lannoo et al.

[2015] envisage this as the future of attocell communication, in particular during handovers. This

exposure should be studied during future developments of the attocells. However, under the

current assumptions of hard handovers, the single-cell exposure, which is studied in this

manuscript, provides a valid exposure assessment.

RF compliance of antennas that operate at 3.5 GHz has been studied before, mainly for antennas

used for WIMAX or Long Term Evolution (LTE) purposes. Table 2 presents an overview of

SAR values near 3.5 GHz antennas obtained in previous studies.

[Table 2]

The peak SAR10g values found using our FDTD simulations are of the same order of magnitude,

but lower than those obtained for directional antennas placed close to the human body. For an

input power of 1 W, we find peak SAR10g values between 1.2 - 2.8 W/kg, while the values found

in literature (see Table 2) are somewhere between 0.8 - 7 W/kg for the same input power

(excluding Vermeeren et al. [2010]). We attribute these somewhat lower values to the higher

separation distance considered in our configuration (62 mm versus 1-4 mm in Klemm et al.

[2005], 0 mm in Ma et al. [2010], and 10 mm in Ikeuchi et al. [2011, 2012], 0 mm in Ma et al.

[2010], and 1-4 mm in Klemm et al. [2005]) and the fact that we have some dielectric materials

with non-zero conductivity in between the antenna and the phantom, which introduces a loss of

power due to reflections and absorption. The values presented in Vermeeren et al. [2010] are an

order of magnitude higher than what is found for the attocell, even for much larger separation

distances (> 30 cm). This can be attributed to the fact that the gain (17 dBi) of the studied BSA in

Vermeeren et al. [2010] is 13.5 dB higher than the gain (3.5 dBi, determined using FDTD) of the

antenna contained in the attocell.

It is impossible that the attocell induces a SARwb higher than 0.08 W/kg, due to its relatively low

input power (1 mW) of the attocells, as previously mentioned. Our results show that the peak

SAR10g values are also far below the ICNIRP basic restrictions at this input power. This indicates

that RF devices that emit at very low power might not always require a full compliance

assessment. One should take potential separation distances from the antenna, the antenna’s gain,

and the emitted frequency into account in making such considerations. However, the potential of

a simplified compliance assessment for low-power RF devices could be investigated in a future

study.

CONCLUSIONS

This study investigated the exposure to RF EM fields near an attocell operating at 3.5 GHz with

an input power of 1 mW, both by means of compliance measurements and numerical simulations.

A three-dimensional positioning system is used to perform Erms field strength measurements near

a real configuration of an attocell. These measurements are then validated by FDTD simulations

of the same configuration. Both measurements and simulations show compliance with the

ICNIRP reference levels at the used input power, since a maximum value of 5.9 V/m was

measured above the attocell. A compliance assessment of the antenna deployed in the attocell is

executed according to the IEC 62209-2 standard, which resulted in peak SAR10g values of

9.7 mW/kg obtained by measurements and 7.2 mW/kg obtained in numerical simulations. These

are far below (> a factor 200) the basic restriction for the general public issued by ICNIRP. The

compliance assessment is then complimented by FDTD simulations on a realistic, heterogeneous

model in six real-life situations. The simulations result in peak SAR10g values that are smaller

than 2.8 mW/kg. This is more than a factor of 290 smaller than the relevant basic restrictions for

the general public at 3.5 GHz. We conclude that attocells are an interesting solution to provide

high-bandwidth coverage, while maintaining a low exposure to RF EM fields for the users.

ACKNOWLEDGEMENTS

ARNO THIELENS IS A POST-DOCTORAL FELLOW OF FLANDERS INNOVATION AND

ENTREPRENEURSHIP UNDER AGREEMENT NO 150752. THIS WORK HAS RECEIVED FUNDING FROM

THE IMINDS INTERNET OF THINGS PROJECT.

REFERENCES

The 5G Infrastructure Public Private Partnership (5GPP). 2014. 5G Vision: the next generation of

communication networks and services. White paper. Available from: https://5g-ppp.eu/wp-

content/uploads/2015/02/5G-Vision-Brochure-v1.pdf [Last accessed 20 September 2016].

Aerts S, Plets D, Verloock L, Martens L, and Joseph W. 2014. Assessment and Comparison of

Total RF-EMF Exposure in Femtocell and Macrocell Base Station Scenarios. Rad. Prot. Dosi.

162(3): 236 – 243.

Aerts S, Plets D, Thielens A, Martens L, and Joseph W. 2015. Impact of a Small Cell on the RF-

EMF Exposure in a Train. Int. Jour. of Env. Res. and Pub. Health 12(3): 2639-2652.

Aminzadeh R, Thielens A, Bamba A, Koné L, Gaillot DP, Lienard M, Martens L, Joseph W.

2016. On-Body Calibration and Measurements Using Personal Radio-Frequency Exposimeters in

Indoor Diffuse and Specular Environments. Bioelectromagnetics 37(5): 298-309.

Bakker JF, Paulides MM, Christ A, Kuster N, van Rhoon GC. 2010. Assessment of induced SAR

in children exposed to electromagnetic plane waves between 10 MHz and 5.6 GHz. Phys Med

Biol 55:3115-3130.

Bakker JF, Paulides MM, Christ A, Kuster N, van Rhoon GC. 2011. Corrigendum for assessment

of induced SAR in children exposed to electromagnetic plane waves between 10 MHz and 5.6

GHz. Phys Med Biol 56:2883.

Boursianis, A., Vanias, P,. and Samaras, T. 2012. Measurements for assessing the exposure from

3G femtocells. Rad. Prot. Dosim. 150: 158 – 167.

Caytan O, Agneessens S, Lemey S, Vande Ginste D, Demeester P, Rogier H. 2015. Half- Ultra-

wideband cork substrate-integrated-waveguide cavity-backed slot antenna. Proceedings of the

International Conference on Electromagnetics in Advanced Applications: 1104-1107.

CENELEC. 2010. Basic standard for the calculation and measurement of electromagnetic field

strength and SAR related to human exposure from radio base stations and fixed terminal stations

for wireless telecommunication systems (110 MHz–40 GHz). CENELEC EN 50383:2010.

Brussels, Belgium.

Chowdhury MZ, Jang YM, and Haas ZJ. 2010. Network Evolution and QOS Provisioning for

Integrated Femtocell/Macrocell Networks. IJWMN 2(3): 16 pages, DOI :

10.5121/ijwmn.2010.2301.

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. 2010b. The Virtual Family, development of surface-based anatomical

models of two adults and two children for dosimetric simulations. Phys Med Biol 55: N23-38.

Cooper TG, Mann SM, Khalid M, and Blackwell RP. 2006. Public exposure to radio waves near

GSM microcell and picocell base stations. Journal of Radiological Protection 26: 199 - 211.

Dahlman E, Mildh G, Parkvall S, Peisa J, Sachs J, and Selén Y. 2014. 5G Radio Access. Ericsson

Review 6, Published Online: June 18 2014. Available from:

https://www.ericsson.com/res/thecompany/docs/publications/ericsson_review/2014/er-5g-radio-

access.pdf [Last accessed 20 September 2016].

Ericsson. 2015. 5G Radio Access. White paper.

Gabriel S, Lau RW, Gabriel C. 1996. The dielectric properties of biological tissues: III.

Parametric models for the dielectric spectrum of tissues. Phys Med Biol 41: 2271-2293.

Gupta A, and Kumar RJHA. 2015. A Survey of 5G Network: Architecture and Emerging

Technologies. IEEE Access 3: 1206 – 1232.

IEC. 2010. Human exposure to radio frequency fields from hand-held and body-mounted

wireless communication devices - Human models, instrumentation, and procedures - Part 2:

Procedure to determine the specific absorption rate (SAR) for wireless communication devices

used in close proximity to the human body (frequency range of 30 MHz to 6 GHz). Geneva,

Switzerland: 62209-2:2010.

Ikeuchi R, and Hirata A. 2011. Dipole Antenna Above EBG Substrate for Local SAR

Reduction. IEEE Antennas and Wireless Propagation Letters 10: 904-906.

Ikeuchi R, Chan KH, Hirata A. 2012. SAR and Radiation Characteristics of a Dipole Antenna

Above Different Finite EBG Substrates in the Presence of a Realistic Head Model in the 3.5GHz

Band. Progress In Electromagnetics Research B 44: 53-70.

International Commission on Non-Ionizing Radiation Protection (ICNIRP). 1998. Guidelines for

limiting exposure to time-varying electric, magnetic, and electromagnetic fields (up to 300 GHz).

Health Physics 74: 494-522.

Klemm M, Kovcs IZ, Pedersen GF, Troster G. 2005. Novel Small-Size Directional Antenna for

UWB WBAN/WPAN Applications. IEEE Trans. on A&P 53 (12): 3884-3896.

Kos B, Valic B, Kotnik T, and Gajsek P. 2011. Exposure assessment in front of a multi-band base

station antenna. Bioelectromagnetics, 32(3): 234–242.

Lannoo B, Dixit A, Colle D, Bauwelinck J, Dhoedt B, Jooris B, Moerman I, Pickavet M, Rogier

H, Simoens P, Torfs G, Vande Ginste D, and Demeester P. 2015. Radio-over-Fibre for Ultra-

Small 5G Cells. 17th International Conference on Transparent Optical Networks (ICTON). 5-9

July 2015, Budapest, Hungary.

Lemey S, Declerq F, and Rogier H. 2014. Dual-band substrate integrated waveguide textile

antenna with integrated solar harvester. IEEE Antennas and Wireless Propagation Letters 13:

269-272.

Ma J, Yin YZ, Hu W, and Ren XS. 2010. Compact Planar Hexa-Band Inverted-F Antenna for

Mobile Phone Application. Microwave and Optical Technology Letters 53(6): 1233-1236.

Nokia Networks. 2014. FutureWorks:; looking ahead to 5G. Building a virtual zero latency

gigabit experience. White paper. Available from: http://resources.alcatel-lucent.com/asset/200014

[Last accessed 20 September 2016].

Porcino D, and Hirt W. 2003. Ultra-Wideband Radio Technology: Potential and Challenges

Ahead. IEEE Communications Magazine 41 (7): 66-74.

Samsung. 2015. 5G Vision. White Paper. Published online: Feb 2015. Available from:

http://www.samsung.com/global/business-images/insights/2015/Samsung-5G-Vision-0.pdf [Last

accessed 20 September 2016].

SPEAG. DASY 3 measurement system - Application Note: System Description and Setup.

Zurich, Switzerland.

Thielens A, Vermeeren G, Kurup D, Joseph W, and Martens L. 2013. Compliance Boundaries for

Multiple-Frequency Base Station Antennas in Three Directions. Bioelectromagnetics 34(7): 563-

567.

Thors B, Strydom ML, Meyer FJC, and Kimmo K. 2008. On the Estimation of SAR and

Compliance Distance Related to RF Exposure From Mobile Communication Base Station

Antennas. IEEE Trans Electromagn Compat, 50(4):837–848.

Thors B, Thielens A, Colombi D, Tornevik C, Vermeeren G, Martens L, and Joseph W. 2014.

Radio frequency electromagnetic field compliance assessment of multi-band and MIMO

equipped radio base stations. Bioelectromagnetics 35(4): 296-308.

Torgovnikov GI. 1993. Dielectric Properties of Wood and Wood-Based Materials. Springer Series

in Wood Science. Springer-Verlag.

Vermeeren G, Gosselin MC, Kühn S, Kellerman V, Hadjem A, Gati A, Joseph W, Wiart J, Meyer

F, Kuster N, and L Martens. 2010. The influence of the reflective environment on the absorption

of a human male exposed to representative base station antennas from 300 MHz to 5 GHz. Phys.

Med. Biol. 55: 5541-5555.

Zarikoff, B. and Malone, D. 2013. A comparison of RF exposure in macro- and femtocells. Health

Phys. 105: 39 – 48.

List of captions

Table 1: Peak SAR10g values (in W/kg) in the VFM, normalized to an input power of 0 dBm in

the attocell and allowed input powers (see Eq. 1) for the studied attocell, based on the ICNIRP

basic restrictions on peak SAR10g for the general public.

Table 2: Overview of literature on SAR studies of antennas tuned to 3.5 GHz.

Fig.ure 1: Overview of the studied attocell. (a) The wooden framework which supports the

acrylic floor and the antenna (see Fig. 2) placed in one of the cells. (b) Bottom view of four

antennas mounted underneath 4 attocells with an acrylic glass layer on top. (c) Top view of the

studied attocell configuration with dimensions of the configuration. (d) Computer-Aided Design

model of the attocell. (e) Side view of one attocell with an indication of all relevant dimensions.

Fig.ure 2: Illustration of the used antenna, which is based on a design presented by Lemey et al.

[2014] and Caytan et al. [2015]. (a) Top view (b) Bottom view, and (c) Top view with

dimensions indicated.

Fig.ure 3: Setup used for the SAR measurements using a standardized phantom (IEC62209-2)

filled with tissue-simulating liquid and the electric field strength measurements, Courtesy of

Günter Vermeeren.

Fig.ure 4: Configuration of the FDTD simulations. (a) The VFM in a frontal and side view. (b)

Placement of the attocell underneath the right foot of the VFM. The skin of the foot touches the

layer of acrylic glass. (c) Alignment of the VFM’s head with the attocell. (d) The VFM’s head

placed in front of the attocell, facing the attocell. (e) The VFM’s head placed in front of the

attocell, facing the opposite direction.

Fig.ure 5: Voltage reflection coefficient of the used antenna (S11) determined using

measurements, FDTD, and MoM.

Fig.ure 6: Electric field Strength (Erms) simulations and measurements at 8 mm above the acrylic

glass for the attocell which is fed an input power of 1 mW at 3.5 GHz. The location of the

maximum is indicated by a red dot. (a) Simulation and (b) measurements of E rms for the upper

configuration. (c) Simulation and (d) measurement of Erms for the lower configuration.

Fig.ure 7: Erms as a function of the distance from the antenna’s surface following a line orthogonal

to the antenna’s surface through the geometric center of the antenna. The layer of acrylic glass

was removed for this simulation. The whiskers indicate the expanded (k = 2) measurement

uncertainty of 14 %, while the grey area indicates the expanded (k = 2) simulation uncertainty on

the (homogeneous) FDTD simulations of 12 %.

Fig.ure 8: SAR values at 5 mm from the phantom’s inner shell using (a) FDTD simulations and

(b) measurements. The red dot in (b) indicated the location or the peak SAR.

Fig.ure 9: SAR distributions in the VFM (normalized to the maximum SAR in the phantom for

each simulation) for the six studied exposure scenarios.

Back of the head Front of the head Right Foot

Antenna

orientation

peak SAR10g

(mW/kg)1.86 2.09 2.80 2.28 1.21 1.38

P¿allowed (W) 1.1 0.96 0.71 0.88 0.29 0.29

Table 1

Study Description Results

Klemm et al. [2005]

The design of directional and an

omni-directional UWB slot

antennas is described, including

SAR compliance using a three-

layered body model.

The directional slot antennas

shows peak SAR10g values

ranging from 3.8-7 W/kg at

1 mm from the body for an

input power of 1 W at

3.5 GHz, which reduces to

values from 1.9-2.6 W/kg at

a separation distance of 4

mm.

Vermeeren et al. [2010]

The VFM is placed at distances > 1 m from a BSA emitting 1 W at

3.5 GHz

The peak SAR10g can exceed the INCIRP basic restriction for occupational exposure on

SAR10g up to a factor of 2 (20-40 W/kg) at 3.5 GHz.

Ma et al. [2010] RF compliance of a PIFA antenna for mobile phone applications is

The peak SAR10g at 3500 MHz is 0.155 W/kg for

studied using a SAM phantom in six frequency bands. an input power of 23 dBm.

Ikeuchi et al. [2012]

The SAR10g induced by a dipole antenna tuned to 3.5 GHz is

studied above two ground planes at different locations at a separation

distance of 10 mm from an anatomical human head.

Peak SAR10g values in the head model range from 0.7-1.6 W/kg, when a reflective layer is placed next to the dipole which is fed 1 W at

3.5 GHz.

Ikeuchi et al. [2012]

The SAR10g induced by a dipole antenna tuned to 3.5 GHz is

studied above different substrates with the purpose of reducing the

SAR values.

Peak SAR10g in an anatomical human head model at 10 mm from the antenna range from

1.28-3.04 W/kg for an antenna output power of 1

W.

Ma et al. [2010]

RF compliance of a PIFA antenna for mobile phone applications is studied using a SAM phantom in

six frequency bands.

The peak SAR10g at 3500 MHz is 0.155 W/kg for an input power of 23 dBm.

Klemm et al. [2005]

The design of directional and an omni-directional UWB slot

antennas is described, including SAR compliance using a three-

layered body model.

The directional slot antennas shows peak SAR10g values

ranging from 3.8-7 W/kg at 1 mm from the body for an

input power of 1 W at 3.5 GHz, which reduces to

values from 1.9-2.6 W/kg at a separation distance of 4

mm.Table 2

Fig.gure 1

Fig.ure 2

Fig.ure 3

Fig.ure 4

Fig.ure 5

(a) (b)

(c) (d)

Fig.ure 6

Fig.ure 7

(a) (b)

Fig.ure 8

Fig.ure 9