applied

sciences
Article
Analysis of the Actual Power and EMF Exposure from
Base Stations in a Commercial 5G Network
Davide Colombi *, Paramananda Joshi, Bo Xu , Fatemeh Ghasemifard , Vignesh Narasaraju
and Christer Törnevik
Ericsson Research, 164 83 Stockholm, Sweden; paramananda.joshi@ericsson.com (P.J.);
bo.xu@ericsson.com (B.X.); fatemeh.ghasemifard@ericsson.com (F.G.); vignesh.narasaraju@ericsson.com (V.N.);
christer.tornevik@ericsson.com (C.T.)
* Correspondence: davide.colombi@ericsson.com; Tel.: +46-76-760-20-89




Received: 10 July 2020; Accepted: 29 July 2020; Published: 30 July 2020


Abstract: In this work, monitoring of the transmit power for several base stations operating in a live
5G network (Telstra, Australia) was conducted with the purpose of analyzing the radio frequency
(RF) electromagnetic field (EMF) exposure levels. The base stations made use of state-of-the-art
massive MIMO antennas utilizing beamforming in order to optimize the signal strength at the user’s
device. In order to characterize the actual EMF exposure from 5G base stations, knowledge of the
amount of power dynamically allocated to each beam is therefore of importance. Experimental
data on the spatial distribution of the base stations’ transmit power were gathered directly from
the network by extracting information on the radio and baseband operations. Out of more than
13 million samples collected over 24 h, the maximum time-averaged power per beam direction was
found to be well-below the theoretical maximum and lower than what was predicted by the existing
statistical models. The results show that assuming constant peak power transmission in a fixed beam
direction leads to an unrealistic EMF exposure assessment. This work provides insights relevant for
the standardization of EMF compliance assessment methodologies applicable for 5G base stations.

Keywords: mobile communication; EMF exposure; 5G; base stations; RF EMF compliance; massive
MIMO; antenna arrays; beamforming

1. Introduction
The first 5G NR (new radio) commercial networks were launched in 2019, and the number of
5G subscriptions globally reached 13 million by the end of 2019 [1]. Like any other radio access
technology, 5G base stations (BSs) have to comply with radiofrequency (RF) electromagnetic field (EMF)
exposure limits, such as those specified by the International Commission on Non-Ionizing Radiation
Protection [2]. EMF measurement surveys of 5G BSs have been conducted in some countries [3,4],
and typical exposure levels in areas accessible by the general public were found to be thousands of
times below the limit values and similar to those of other existing mobile technologies (2G, 3G, and 4G).
5G makes use of beamforming and beam steering to adapt the antenna radiation pattern to the
time-varying traffic and the radio propagation conditions. This allows it to efficiently direct the energy
where needed, providing enhanced performance to the users. To be able to steer the beams over a range
of angles, BS products make use of array antennas characterized by many elements, generally known
as massive MIMO antennas. Compared with conventional non-beamformed systems, massive MIMO
BSs are characterized by a larger variability of transmitted signals in time and space. In order to take
this into account, specific measurement techniques to assess the exposure levels in proximity of 5G
massive MIMO BSs have been developed [5–7].

Appl. Sci. 2020, 10, 5280; doi:10.3390/app10155280 www.mdpi.com/journal/applsci

Appl. Sci. 2020, 10, 5280 2 of 10

The effects of beamforming and traffic variation on the EMF exposure levels have been studied
by means of numerical simulations and statistical models [8,9]. Based on conservative assumptions
on the data traffic and on the user distributions, these studies estimate the BS maximum power level
contributing to EMF exposure in any beam direction. Since the beams are steered in order to provide
service to the connected users, the time-averaged power in each beam direction was found to be
well below the instantaneous peak (or theoretical maximum). According to [8], the 95th percentile
of the transmitted power per beam would range between 7% and 22% of the theoretical maximum.
These results have been verified by measurements conducted on massive MIMO BSs [10,11] of 4G
LTE-Advanced networks. While a substantial amount of research available to date already shows that
massive MIMO BSs do not constantly operate at the maximum power, experimental data specific for
5G live networks are yet not available.
In this study, a large amount of experimental data of relevance for EMF exposure was collected
from a 5G NR commercial network. The dynamic of the spatial distribution of the transmit power
serving 5G users was continuously monitored for 25 massive MIMO BSs over 24 h. Data were extracted
from network counters and events, which directly provide information on the BS baseband operations.
In addition to the time-averaged transmit power, information on the beams scheduled by the BSs when
serving the connected users is obtained and processed. By this knowledge, the spatial distribution of
the transmit power was determined. The method is developed in Section 2 and results on the actual
power are given in Section 3. A discussion, including implications of the results on EMF exposure
from massive MIMO BSs is provided in Section 4 and some conclusions are drawn in Section 5.
This work is of interest to support the ongoing revision of IEC 62232 [5], which provides methods
to conduct EMF compliance assessment of massive MIMO BSs by taking the effects of beamforming
into consideration. The availability of 5G experimental data can be relevant to regulators when defining
guiding principles on the usage of the actual maximum power to avoid unrealistic estimates of the
EMF exposure levels.

2. Method
Measurement data were collected from 25 NR BSs of the Telstra commercial network in Australia.
Telstra was among the first operators to launch 5G, and in January 2020, more than 100,000 5G-enabled
mobile devices were connected to its network [12]. The massive MIMO BSs selected for this study
were located in dense urban areas, typically characterized by higher traffic than rural areas. The BSs
operated within the NR band 78 (3300–3800 MHz) with a channel bandwidth varying among the sites
and ranging from 40 to 80 MHz. The time division duplexing (TDD) downlink duty cycle, which is
defined as the fraction of the time available for downlink transmission to the total transmission time,
was about 75%.
Ericsson Network Manager (ENM) was used to access information on the 5G BSs operation for
24 h over a weekday at the end of January 2020. By means of gathering data directly through the ENM,
it was possible to simultaneously collect information and statistics for many sites over a relatively
long period. Moreover, it allowed us to analyze the spatial distribution of the BS transmit power
in a three-dimensional space within the scan range of the antenna rather than having to conduct
measurements at a fixed location.
The transmit power per direction or more specifically the equivalent isotropic radiated power
(EIRP) in azimuth and elevation (ϕ, θ) was determined as a function of time by continuously logging
two metrics:

• The total (cell-wide) time-averaged transmit power, Pavg ;

• The time-averaged gain, Gavg (ϕ, θ).

While the recently updated exposure guidelines, ICNIRP [2], have extended the averaging time
applicable to whole-body exposure to thirty minutes, six minutes was applied in this study according
to the previous version of the ICNIRP guidelines [13]. This choice was made in order to allow a direct
Appl. Sci. 2020, 10, x FOR PEER REVIEW 3 of 10

to the
Appl. Sci.previous version of the ICNIRP guidelines [13]. This choice was made in order to allow a direct
2020, 10, 5280 3 of 10
comparison with what was predicted by the statistical models, e.g., [8], published when [2] was not
yet available. Moreover, it might take some time until the updated ICNIRP guidelines are adopted in
comparison with whatSince
national regulations. was the
predicted by the statistical
time-averaged power over models,
thirtye.g., [8], published
minutes is equal orwhen [2] was
less than notis
what
yet available. Moreover, it might take some time until the updated ICNIRP guidelines are
obtained by averaging over six minutes, the results presented in this paper are also conservative withadopted in
national regulations.
respect to Since the
the latest ICNIRP time-averaged power over thirty minutes is equal or less than what is
guidelines.
obtained by averaging over six minutes, the results presented in this paper are also conservative with
respect to the latest
2.1. Six-Minute TotalICNIRP guidelines.
(Cell-Wide) Time-Averaged Power (𝑃𝑎𝑣𝑔 )
𝑃avg was assessed
2.1. Six-Minute in the baseband
Total (Cell-Wide) based
Time-Averaged on the
Power (Pavgamount
) of physical resource blocks (PRBs)
scheduled by the BS for downlink transmission. By summing up the power over all transmission time
Pavg was assessed in the baseband based on the amount of physical resource blocks (PRBs)
intervals, a sample of the six-minute time-averaged transmit power was generated every 600 ms and
scheduled by the BS for downlink transmission. By summing up the power over all transmission time
the distribution of the time-averaged power was then reported through the ENM every fifteen
intervals, a sample of the six-minute time-averaged transmit power was generated every 600 ms and
minutes (a typical reporting interval for network counters). The reported values of the time-averaged
the distribution of the time-averaged power was then reported through the ENM every fifteen minutes
power were organized into bins characterized by a width of 2.5% with respect to the maximum
(a typical reporting interval for network counters). The reported values of the time-averaged power
available power (i.e., corresponding to 100% of the PRB usage). Therefore, a total of 1500 six-minute
were organized into bins characterized by a width of 2.5% with respect to the maximum available power
time-averaged transmit power samples collected over 15 min were distributed over a total of 40 bins
(i.e., corresponding to 100% of the PRB usage). Therefore, a total of 1500 six-minute time-averaged
(ranging from 0% to 100%). It is worth noticing that each sample was determined by applying a
transmit power samples collected over 15 min were distributed over a total of 40 bins (ranging from
rolling average and the reported distribution is therefore representative of any six-minute interval
0% to 100%). It is worth noticing that each sample was determined by applying a rolling average and
within fifteen minutes. Figure 1 provides an illustrative example of the data as collected through the
the reported distribution is therefore representative of any six-minute interval within fifteen minutes.
network counter.
Figure 1 provides an illustrative example of the data as collected through the network counter.

Figure1.1.Example
Figure Example of
ofaasix-minute
six-minutetime-averaged
time-averagedtransmit
transmitpower
powerdistribution
distribution (normalized)
(normalized) for for aaBS
BS
overaafifteen-minute
over fifteen-minuteinterval.
interval. Each
Each barbar shows
shows the
the number
number ofof samples
samples falling
falling into
into aaspecific
specificrange
rangeof of
values(100%
values (100%would
wouldcorrespond
correspond to tothe
theBSBSoperating
operatingat atthe
themaximum
maximumavailable
availablepower
powercontinuously
continuously
for
for66minutes).
minutes). The
The overlaying
overlaying text
text represents
representsthetheraw
rawdata
dataasascollected
collectedfrom
fromthetheENM,
ENM,i.e.,
i.e.,indicating
indicating
that966
that 966samples
samplesofofP𝑃avg belonged to
avg belonged to the
the 0–2.5%
0–2.5% interval,
interval, 30
30 to
to 2.5–5%,
2.5–5%, 32
32 to
to 5–7.5%...304
5–7.5%...304 to to 22.5–25%.
22.5–25%.
No
Nosamples
samplesininthe
theexample
examplewerewereobtained
obtainedabove
abovethe
the22.5–25%
22.5–25%interval.
interval.

2.2.
2.2. Six-Minute
Six-Minute Time-Averaged
Time-AveragedGain
Gain(G
(𝐺avg )
𝑎𝑣𝑔 )
Pavg provides a measure of the total cell-wide time-averaged transmit power, but it does not
𝑃avg provides a measure of the total cell-wide time-averaged transmit power, but it does not
provide information on its spatial distribution, which for a massive MIMO BS will vary over time
provide information on its spatial distribution, which for a massive MIMO BS will vary over time
depending on the beam selected for transmission. This was addressed by monitoring the BS antenna
depending on the beam selected for transmission. This was addressed by monitoring the BS antenna
radiation patterns during operation. While 3GPP allows for different beamforming schemes [14],
radiation patterns during operation. While 3GPP allows for different beamforming schemes [14], the
the software running on the selected BSs at the time of the investigation supported single-user MIMO
software running on the selected BSs at the time of the investigation supported single-user MIMO
codebook-based beamforming with 8 CSI-RS ports. In this configuration, user equipment (UE) makes
codebook-based beamforming with 8 CSI-RS ports. In this configuration, user equipment (UE) makes
use of channel state information (CSI) reference signals (RS) transmitted from the BS to measure
use of channel state information (CSI) reference signals (RS) transmitted from the BS to measure the
the radio channel quality. Based on this, the UE sends the CSI report, which contains a precoder
radio channel quality. Based on this, the UE sends the CSI report, which contains a precoder matrix
matrix indicator (PMI), a rank indicator (RI), and some other indicators, to the BS. Then, the BS selects
indicator (PMI), a rank indicator (RI), and some other indicators, to the BS. Then, the BS selects the
the corresponding beam(s) for downlink transmission of traffic data using the reported PMI and RI.
corresponding beam(s) for downlink transmission of traffic data using the reported PMI and RI. The
The PMI and RI can be seen as a reference into a codebook of antenna precoding matrix (known
Appl. Sci. 2020, 10, x FOR PEER REVIEW 4 of 10

Appl.
PMISci. RI 10,
2020,
and can5280
be seen as a reference into a codebook of antenna precoding matrix (known by4 bothof 10

the BS and the UE) corresponding to a certain beam shape. The dynamic of the antenna radiation
patterns was therefore studied by logging PMI and RI reported periodically to the BS by the UEs. As
by both the BS and the UE) corresponding to a certain beam shape. The dynamic of the antenna
long as a UE is ‘active’ (i.e., receiving data from the BS), the BS will request PMI and RI reporting
radiation patterns was therefore studied by logging PMI and RI reported periodically to the BS by
from it. Therefore, the time interval between consecutive reports collected in this study varied
the UEs. As long as a UE is ‘active’ (i.e., receiving data from the BS), the BS will request PMI and
depending on the number of connected UEs with non-empty downlink transmit buffer (with a
RI reporting from it. Therefore, the time interval between consecutive reports collected in this study
median value of 70 ms). For this study, a total number of about 13 million PMI and RI samples were
varied depending on the number of connected UEs with non-empty downlink transmit buffer (with a
collected in the 24 h.
median value of 70 ms). For this study, a total number of about 13 million PMI and RI samples were
In Figure 2, antenna radiation patterns are shown as a function of the azimuth angle for a small
collected in the 24 h.
subset of the available beam configurations (four out of sixteen patterns available for rank 1
In Figure 2, antenna radiation patterns are shown as a function of the azimuth angle for
transmission are displayed; transmission up to rank 4 was supported by the BSs [14]). For the 8 CSI-
a small subset of the available beam configurations (four out of sixteen patterns available for rank
RS port codebook-based beamforming that was configured on the selected sites, only beam steering
1 transmission are displayed; transmission up to rank 4 was supported by the BSs [14]). For the
in the azimuth plane was supported (no vertical beam steering). A discussion about the effects of
8 CSI-RS port codebook-based beamforming that was configured on the selected sites, only beam
different beamforming implementations on the evaluation of the actual maximum exposure is
steering in the azimuth plane was supported (no vertical beam steering). A discussion about the
provided in Section 4.
effects of different beamforming implementations on the evaluation of the actual maximum exposure
The time-averaged gain, 𝐺avg (𝜑, 𝜃), was obtained by averaging the antenna radiation patterns
is provided in Section 4.
corresponding to all PMI and RI reported by all ‘active’ UEs to the BS within every six-minute interval
The time-averaged gain, Gavg (ϕ, θ), was obtained by averaging the antenna radiation patterns
(a moving average was applied) over the 24 h. 𝐺avg was calculated at any time when a new CSI report
corresponding to all PMI and RI reported by all ‘active’ UEs to the BS within every six-minute interval
was
(a received.
moving Those
average wastime intervals,
applied) over for
the which
24 h. Gno PMI and RI were reported, meaning that no data
avg was calculated at any time when a new CSI report
traffic was served by the BS or that the data logging
was received. Those time intervals, for which no PMI had failed,
and werereported,
RI were excludedmeaning
from the that
analysis.
no data
traffic was served by the BS or that the data logging had failed, were excluded from the analysis.

(a) (b)
Figure2.2.(a)
Figure (a)Antenna
Antenna radiation
radiation patterns
patterns for four
for four beambeam configurations
configurations (in different
(in different colors)
colors) in in the
the azimuth
azimuth
plane (θ = 90◦ ).(𝜃The
plane = 90°). The
values in values in plot
the polar the polar plot are normalized
are normalized to the peaktogain
the peak
of thegain of the beam
broadside broadside
and
beam andinpresented
presented in a logarithmic
a logarithmic scale ainsubset
scale in dB. Only dB. Only a subset
of the of BS
available thebeams
available BS beams
is shown. is shown.
(b) Reference
(b) Reference
system systemtowith
with respect respect to the antenna.
the antenna.

2.3. Six-Minute Time-Averaged EIRP (EIRPact )

2.3. Six-Minute Time-Averaged EIRP (𝐸𝐼𝑅𝑃𝑎𝑐𝑡 )
By combining knowledge of the time-averaged transmit power, Pavg , with the average gain,
By combining knowledge of the time-averaged transmit power, 𝑃avg , with the average gain,
Gavg (ϕ, θ), the six-minute time-averaged actual EIRP was determined according to:
𝐺avg (𝜑, 𝜃), the six-minute time-averaged actual EIRP was determined according to:
EIRPact (ϕ, θ, t) = Gavg (ϕ, θ, t ) × Pavg (t) (1)
𝐸𝐼𝑅𝑃act (𝜑, 𝜃, 𝑡) = 𝐺avg (𝜑, 𝜃, 𝑡)× 𝑃avg (𝑡) (1)
EIRP
𝐸𝐼𝑅𝑃act provides a measure of the spatial distribution of the time-averaged transmit power, and it
act provides a measure of the spatial distribution of the time-averaged transmit power, and
is therefore directly related
it is therefore directly to EMF
related exposure.
to EMF Equation
exposure. (1) implicitly
Equation assumes
(1) implicitly the time-averaged
assumes power
the time-averaged
Ppower
avg to be equally
𝑃avg distributed
to be equally among the
distributed beam
among theconfigurations reported
beam configurations within the
reported withinsamethesix-minute
same six-
interval. In addition, while P provides a measure of the total transmit power, including
minute interval. In addition, while 𝑃avg provides a measure of the total transmit power, including
avg the broadcast
channel, Gavg ischannel,
the broadcast determined
𝐺avgbased on the antenna
is determined basedpattern
on the only usedpattern
antenna for dataonly
traffic.
used Since
for the
dataantenna
traffic.
gain for the broadcast signals is several dB lower than the peak antenna gain for the traffic beams,
Appl. Sci. 2020, 10, x FOR PEER REVIEW 5 of 10

Appl. Sci. 2020, 10, 5280 5 of 10

Since the antenna gain for the broadcast signals is several dB lower than the peak antenna gain for
the traffic beams, 𝐸𝐼𝑅𝑃act provides a conservative estimate of the time-averaged EIRP. In any case,
the
EIRPbroadcast signal in 5G NR only counts for a very small amount of the BS maximum transmit
act provides a conservative estimate of the time-averaged EIRP. In any case, the broadcast signal in
power
5G NR (the
onlyresource
counts forelements
a very small allocated
amount foroftransmission
the BS maximum of the synchronization
transmit signal block
power (the resource elementsare
typically lower than 0.1% of the total available [6]).
allocated for transmission of the synchronization signal block are typically lower than 0.1% of the total
[6]). of 𝑃avg at time 𝑡 was approximated by the largest six-minute averaged power sample
The value
available
reported
Thein the corresponding
value of Pavg at time fifteen-minute
t was approximated intervalby as
theextracted from the network.
largest six-minute averagedWith powerreference
sample
to the fifteen-minute interval in Figure 1, the
reported in the corresponding fifteen-minute interval asavg selected 𝑃 extracted from the network. With referenceof
would be 23.75% (i.e., the center value to
the
thelargest bin, for which
fifteen-minute interval at in
least one 1,
Figure sample was observed).
the selected Pavg wouldThisbeis23.75%
a conservative
(i.e., the assumption,
center value sinceof the
for many
largest instances
bin, for which of at𝑡 least
within onethe fifteen
sample wasminutes, the six-minute
observed). time-averaged
This is a conservative transmitsince
assumption, powerfor
could
manybe lower (e.g.,
instances in Figure
of t within the1fifteen
most of the samples
minutes, belong totime-averaged
the six-minute the bin corresponding
transmit to a power
power could of be
0
to 2.5%).
lower As shown
(e.g., in Figurein 1Section
most of 3, the
thissamples
assumption might
belong lead
to the bintocorresponding
a large overestimate of 𝐸𝐼𝑅𝑃
to a power of 0actto. 2.5%).
As shown in Section 3, this assumption might lead to a large overestimate of EIRPact .
3. Results
3. Results
𝑃avg as a function of time is shown for the 25 BSs in Figure 3. As discussed in Section 2, within
Pavg as a function
each fifteen-minute of time
interval, is shown
this for the 25 BSstaken
was conservatively in Figure
as the3. largest
As discussed
sampleinofSection 2, within
the six-minute
each fifteen-minute
time-averaged powerinterval,
reportedthisamongwas all
conservatively
samples in the taken as the largest
distribution. sample of purposes,
For illustration the six-minute
this
istime-averaged
to avoid largepower reported
overlapping of among all samples
the markers. Only in the distribution.
samples correspondingFor illustration purposes,
to a 𝑃avg larger than this
2.5%is
to avoid
are large
depicted the figure. of the markers. Only samples corresponding to a Pavg larger than 2.5% are
inoverlapping
depicted in the figure.
The maximum 𝑃avg observed for all BSs ranged between 22.5–25% (i.e., 𝑃avg = 23.75%), but for
The maximum
most of the cases, 𝑃avg Pavg observed
was found to forbe
allwithin
BSs ranged
0% and between
2.5% of22.5–25% (i.e., Pavg
the maximum = 23.75%),
(i.e., but for
𝑃avg = 1.25%).
most ofthe
During thenight
cases, Pavgmidnight
(from was found to be within
to about 6 a.m.),0% andwas
no BS 2.5% of the
found maximum
to operate at power avg = 1.25%).
(i.e., Plevels above
During the night
2.5% of the maximum. (from midnight to about 6 a.m.), no BS was found to operate at power levels above
2.5% of the maximum.

Figure3.3. 𝑃Pavg
Figure avg (as
(asaapercentage
percentageofofthe
theavailable
availablemaximum)
maximum)for for24
24hh(starting
(startingand
andending
endingatat8.30
8.30Am.).
Am.).
Eachcolor
Each coloridentifies
identifiesaa5G
5GNRNRBS.
BS.For
Forillustration
illustrationpurposes,
purposes,only
onlysamples reporting𝑃Pavg
samplesreporting avg larger
larger than
than
2.5%are
2.5% areplotted.
plotted.No Nodots
dotsare
areshown
shownbetween
betweenmidnight
midnightup
upto
toabout
about66Am.
Am.indicating
indicatingthat
thatall
allBSs
BSswere
were
operatingatataacell-wide
operating cell-widepower
powerlevel
levelbelow
below2.5%
2.5%of
ofthe
themaximum.
maximum.

Anexample
An exampleof
avg
(ϕ,𝜃)
of 𝐺Gavg(𝜑, θ)observed
observedfor forone
oneBS BSatatan
anarbitrary
arbitrarytime
timeisisshown
shownin inFigure
Figure44(in(inthe
the
azimuth plane). The envelope of the antenna radiation patterns for all available beams
azimuth plane). The envelope of the antenna radiation patterns for all available beams is also plotted is also plotted
in the
in the same
same figure
figuretogether
togetherwith
withthe theradiation pattern
radiation of the
pattern beam
of the pointing
beam towards
pointing boresight.
towards In order
boresight. In
to limit the number of computations in the analysis, the G angular domain was
order to limit the number of computations in the analysis, the 𝐺avg angular domain was limited to
avg limited to the region
of interest −90 ◦ ≤ ϕ ≤ 90◦ , i.e., in front of the antenna. G in Figure𝐺4 is well below the peak envelope
the region of interest −90° ≤ 𝜑 ≤ 90°, i.e., in front of the avgantenna. avg in Figure 4 is well below the
gain (e.g., 9 dB in the boresight direction) and it is quite uniform in
peak envelope gain (e.g., 9 dB in the boresight direction) and it is quite uniform inthe azimuth plane, indicating
the azimuth that
plane,
the time-averaged
indicating that the transmit power transmit
time-averaged is evenlypower
distributed over distributed
is evenly the scan rangeoverof the
the scan
antenna.
range of the
antenna.
Appl. Sci. 2020, 10, 5280 6 of 10
Appl. Sci. 2020, 10, x FOR PEER REVIEW 6 of 10
Appl. Sci. 2020, 10, x FOR PEER REVIEW 6 of 10

Figure 4. Example of Gavg (dB, in red) in the azimuth plane (θ = 90◦ ) for one BS within an arbitrary
Figure 4. Example of 𝐺avg (dB, in red) in the azimuth plane (𝜃 = 90°) for one BS within an arbitrary
six-minute
Figureinterval and normalized
4. Example
six-minute 𝐺avgnormalized
intervalofand
to to
(dB, in red) the
in thepeak
the antenna
azimuth gain.
plane gain.
peak antenna (𝜃 The
= 90°)
The for
green
one
green
dashed
BS within
dashed
line
lineanshows
shows
arbitrary
the
the
envelope of the
six-minute antenna
interval gain
and for all
normalized theto available
the peak beams.
antenna In blue
gain. The is the
green gain
dashedfor the
line beam
shows
envelope of the antenna gain for all the available beams. In blue is the gain for the beam pointing pointing
the
towardsenvelope
boresight.
towards of the antenna gain for all the available beams. In blue is the gain for the beam pointing
boresight.
towards boresight.
EIRPact𝐸𝐼𝑅𝑃 was calculated
wasactcalculated as a as a function
function ofoftime
timeaccording
according totoEquation
Equation (1) (1)
for all
forBSs.
all The
BSs.envelope
The envelope
𝐸𝐼𝑅𝑃
pattern of wasactcalculated
𝐸𝐼𝑅𝑃 (max 𝐸𝐼𝑅𝑃asacta(𝜑,
function
𝜃, 𝑡)) of time according toreporting
Equationthe
(1)highest
for all BSs. The envelope
𝑡 act (ϕ, θ, t)) for the base station reporting the highest EIRP is plotted in
act
pattern of EIRPact (maxEIRP for the base station EIRP is plotted in
pattern of 𝐸𝐼𝑅𝑃act
t (max 𝐸𝐼𝑅𝑃act (𝜑, 𝜃, 𝑡)) for the base station reporting the highest EIRP is plotted in
Figure 5Figure 5 (in red). In the
(in red). In the same𝑡 same figure, the envelope of the theoretical maximum EIRP (i.e., obtained
figure, the envelope of the theoretical maximum EIRP (i.e., obtained
Figure
assuming5 (inconstant
red). Inpeak
the same
power figure, the envelope
transmission of the
for each theoretical
beam) maximum
is also shown (greenEIRP (i.e., line).
dotted obtained
The
assuming constantnormalized
assuming peak power
values areconstant
transmission
peaktopower
the peaktransmissionfor each
EIRP for the forBS. beam)
each beam)
Figure
is also shown
is also
5 shows shown
that
(green
𝐸𝐼𝑅𝑃(green dotted
dotted line). The values
line). The
act is only a fraction of
are normalized
values to the peak
are normalized
the theoretical maximum EIRP
to thefor
and the
peak BS.
EIRP8.8
at least Figure
fordB 5 shows
thelower.
BS. Figure that EIRP
5 shows
For the otherthat
act is only
BSs,𝐸𝐼𝑅𝑃 a fraction
act is only𝐸𝐼𝑅𝑃
the recorded of the
a fraction theoretical
of
act were
maximum the and at least
theoretical
considerably 8.8 dBThe
maximum
smaller. lower.
and at
overallForresults
leastthe other
8.8 dB BSs,
lower.
for all 25 the
For theincludingEIRP
BSs,recorded
other BSs, theact were
the 95threcorded 𝐸𝐼𝑅𝑃
considerably
percentile act 99thsmaller.
and were
considerably
percentile
The overall results smaller.
of 𝐸𝐼𝑅𝑃
for all TheBSs,
act with
25 overall
respect toresults for
theall
the maximum
including 25
95th BSs, including
theoretical
percentile EIRP,
and the
are 95thpercentile
percentile
summarized
99th andEIRP
in Table
of 1.99thact with
percentile of 𝐸𝐼𝑅𝑃act with respect to the maximum theoretical EIRP, are summarized in Table 1.
respect to the maximum theoretical EIRP, are summarized in Table 1.

Figure 5.Figure 5. Envelope of 𝐸𝐼𝑅𝑃(dB),(dB),

i.e.,i.e., max 𝐸𝐼𝑅𝑃act (𝜑, 𝜃, 𝑡)) in the azimuth plane (in red) for the BS
Envelope of EIRP act act max 𝑡EIRPact (ϕ, θ, t)) in the azimuth plane (in red) for the BS
Figure 5. Envelope of 𝐸𝐼𝑅𝑃act (dB), i.e., tmax 𝐸𝐼𝑅𝑃act (𝜑, 𝜃, 𝑡)) in the azimuth plane (in red) for the BS
having the largest observed 𝐸𝐼𝑅𝑃.act . The values are normalized to the peak EIRP. The green dashed
having the largest observed EIRP 𝑡
act The values are normalized to the peak EIRP. The green dashed
having the largest
line shows observed
the envelope 𝐸𝐼𝑅𝑃
of the act . The values
(normalized) areassuming
EIRP normalized to the peak
constant peakEIRP.
powerThe green dashed
transmission for
line shows
line
the envelope
shows
of the (normalized) EIRP assuming constant peak power transmission for
every beam.the envelope of the (normalized) EIRP assuming constant peak power transmission for
every beam.
every beam.
Table 1. Maximum, 99th percentile and 95th percentile of 𝐸𝐼𝑅𝑃 act with respect to the envelope of the
Maximum,
Table 1.Table 99th 99th
1. Maximum,
percentile and
percentile
95th
andwithin
95th percentile
EIRP
percentileofof𝐸𝐼𝑅𝑃 act with
withfor
respect
respect
to the envelope of the
theoretical maximum for all 25 BSs 24 h. For each BS
act and eachtotime
the envelope
t, 𝐸𝐼𝑅𝑃actof was
the
theoretical maximum
theoretical for allfor
maximum 25 BSs within
all 25 24 h. 24
BSs within Forh.each
For BS and
each BSforandeach
for time t, EIRP
each time act was
t, 𝐸𝐼𝑅𝑃 considered
act was
in the angular direction providing the smallest difference with respect to the theoretical maximum.

Max 99th Percentile 95th Percentile

−8.8 dB 1 −18 dB −20.2 dB
1 −15 dB if Pavg was estimated from the mean of the six-minute time-averaged power distribution rather than
extrapolated from the upper bound of the reported values.
 Max 99th Percentile 95th Percentile
−8.8 dB 1 −18 dB −20.2 dB
1 −15 dB if 𝑃avg was estimated from the mean of the six-minute time-averaged power distribution
rather than extrapolated from the upper bound of the reported values.
Appl. Sci. 2020, 10, 5280 7 of 10

The values in Figure 5 and Table 1 provide a conservative estimate of 𝐸𝐼𝑅𝑃act . In fact, as
described in Section
The values 2, 𝑃avg
in Figure used
5 and in 1Equation
Table provide a(1)conservative
was overestimated
estimate from the reported
of EIRP distribution
act . In fact, as described
of the six-minute time-averaged power by selecting the sample with the highest
in Section 2, Pavg used in Equation (1) was overestimated from the reported distribution transmit power
of thein
the fifteen-minute
six-minute interval even
time-averaged though
power most of the
by selecting the six-minute
sample with time-averaged samples reported
the highest transmit power inwerethe
much lower (see Figure 1). By using the mean of the distribution rather than
fifteen-minute interval even though most of the six-minute time-averaged samples reported the maximum value for
were
𝑃avg , the
much lowerresulting 𝐸𝐼𝑅𝑃
(see Figure 1). would
actBy be at
using the least
mean of 15
the dB (rather than
distribution 8.8than
rather dB)the
below the theoretical
maximum value for
maximum.
Pavg , the resulting EIRPact would be at least 15 dB (rather than 8.8 dB) below the theoretical maximum.

4. Discussion
4. Discussion
Measurement data
Measurement data ofof the
the directional
directional time-averaged
time-averaged transmittransmit power power inin aa commercial
commercial 5G 5G network
network
show that
show that the
the actual
actual values
values are
are well
well below
below thethe (theoretical)
(theoretical) maximum
maximum obtained obtained assuming
assuming peak peak power
power
transmission for each beam. Telstra was among the first operators worldwide
transmission for each beam. Telstra was among the first operators worldwide launching 5G, and the launching 5G, and the
amount of traffic, which is carried, is therefore expected to be representative
amount of traffic, which is carried, is therefore expected to be representative of the state-of-the-art 5G of the state-of-the-art 5G
networks. Nevertheless,
networks. Nevertheless, 5G 5G NRNR isis comparably
comparably new new with
with respect
respect to to other
other existing
existing mobile
mobile technologies,
technologies,
such as LTE and WCDMA, and the number of devices supporting
such as LTE and WCDMA, and the number of devices supporting 5G is expected to grow 5G is expected to grow inin the
the
future. Since P𝑃avg
future. Since avg isis directly
directly related
related to to the
the amount
amount of of downlink
downlink traffic,
traffic, thethe total
total (cell-wide)
(cell-wide) six-minute
six-minute
time-averaged transmit
transmit power
power isisexpected
expectedtotoincrease.
increase.AtAtthe thesame
same time,
time,more
more users
userswill result
will in an
result in
even larger spread of the energy over the antenna scan range, which will
an even larger spread of the energy over the antenna scan range, which will contribute to a reduction contribute to a reduction of
𝐺avg
of with
Gavg respect
with respect totothe peak
the peak envelope
envelopegain. Therefore,𝐸𝐼𝑅𝑃
gain.Therefore, EIRP actis
act islikely
likelynot
not to
to increase
increase substantially
when increasing the
when increasing the number number of users. This is shown in Figure 6, in which
is shown in Figure 6, in which EIRPact 𝐸𝐼𝑅𝑃 for the investigated
act for
BSs is
BSs is given
givenasasaafunction
functionofofthe thescheduled
scheduled users
users in in a six-minute
a six-minute interval.
interval. TheTheblue blue
solidsolid
lineline shows
shows the
the mean
mean valuevalue
amongamong all 𝐸𝐼𝑅𝑃
all EIRP samples
samples
act act corresponding
corresponding to the to the
same same
numbernumberof of
served served
UEs. UEs.
The The
red
red dotted
dotted line shows
line shows the maximum
the maximum EIRPact 𝐸𝐼𝑅𝑃
among among
act these these samples.
samples. The valueThe value
of EIRP 𝐸𝐼𝑅𝑃act is
ofnormalized
act is
normalized
with respect with
to therespect to the EIRP
EIRP envelope envelope
of the of the
theoretical theoretical
maximum, and maximum,
for each time andt,for each
it is time t, it in
determined is
determined
the in the angular
angular direction (ϕ, θ),direction
for which(𝜑, the𝜃), for of
ratio which
EIRPtheact ( ϕ, θ,
ratio t of
) to 𝐸𝐼𝑅𝑃
the (𝜑,
maximum
act 𝜃, 𝑡) to
EIRPthe maximum
envelope is
EIRP
the envelope is the largest.
largest.

Figure 𝐸𝐼𝑅𝑃act
Figure 6. EIRP act as
asaafunction
functionof
ofthe
thenumber
numberof ofserved
servedusers
userswithin
withinthe
the six-minute
six-minute intervals.
intervals. The red
dashed
dashedline
lineshows
showsthethe
maximum valuevalue
maximum among the EIRP
among theact 𝐸𝐼𝑅𝑃
samples, while the while
act samples, blue solid
the line corresponds
blue solid line
to the mean. EIRP
corresponds is normalized
to theactmean. tonormalized
𝐸𝐼𝑅𝑃act is the envelopeto of
thethe theoretical
envelope maximum
of the EIRP.
theoretical maximum EIRP.

In order to correctly interpret Figure 6, it should be observed that Pavg (for each BS) in every
fifteen-minute interval was conservatively approximated with the maximum value from the distribution
of the reported six-minute time-averaged transmit power. Therefore, every six-minute averaging-period
within fifteen minutes was assumed to be characterized by the same Pavg (disregarding of the number
of users). This likely leads to a relatively large overestimate of Pavg and therefore EIRPact when the
number of users is low, and it shows that the values reported in Table 1 might therefore be overestimated.
In addition, as 99% of the collected Pavg samples belonged to the bin corresponding to the lowest
transmit power utilization (0–2.5% of the maximum), variations within this range of utilizations were
Appl. Sci. 2020, 10, 5280 8 of 10

not detected. As a result, while Pavg as used in Equation (1) is substantially constant, the spatial spread
of energy (represented by Gavg ) will increase with increasing the number of users leading to a reduction
of EIRPact .
5G NR massive MIMO BSs can also support a larger number of beams compared with the 8 CSI-RS
ports codebook-based beamforming software, which was running on the investigated sites (e.g., based
on 16 CSI-RS or 32 CSI-RS [14]). In addition to narrower beams, such configurations might also support
steering of energy in elevation and not just within the azimuth plane. Moreover, if reciprocity-based
beamforming (i.e., non-codebook) is supported, sounding reference signals transmitted by the UEs will
be used by the BS to estimate the radio channel, allowing energy to be transmitted in several different
paths rather than in specific directions. By means of reciprocity-based beamforming, BSs might
also simultaneously send different layers in separate beams to different users using the same time
and frequency resource (multi-user MIMO). For such solutions, an even larger spread of the BS
transmit power over different propagation paths is to be expected. The experimental data obtained
for 8 CSI-RS ports codebook-based beamforming are therefore expected to provide a conservative
estimate of the actual transmit power of relevance for EMF assessments compared with other (existing
or upcoming) implementations.
Although the spreading of energy is deemed to increase when traffic increases, the results in
Section 3 show that EIRPact is also well below the theoretical maximum for a limited number of active
users. Out of about 13 million samples, including instances for which only one user was active within
a six-minute interval, not a single case was found when any of the BSs would constantly schedule
the same beam with the maximum power. Mobile networks are in fact deployed to provide services
to a myriad of UEs, with enough margins to address occasional traffic peaks without reaching traffic
congestion. When only a few users are served, the amount of traffic, and therefore the BS time-averaged
transmit power, will account for a small fraction of the available maximum. Moreover, even the same
UE might be served by different beams over time as a consequence of the changes in the propagation
conditions or due to simultaneous transmission of parallel data streams (when two or more layers are
transmitted to the same UE, the allocated power will be simultaneously spread among different beams).
Network measurements provide an efficient tool to gather comprehensive information on the
actual EMF levels of live network for several BSs and long periods. This is even more relevant for BS
implementing beamforming, as the gathering of data directly from the network manager allows one to
study the distribution of the transmitted signal in the spatial domain rather than at a fixed location.
At the same time, measurements conducted in-situ are complementary and equally important, as they
provide a direct measure of the typical EMF exposure levels in areas accessible by the general public.
Although results from in-situ measurement campaigns addressing EMF exposure for massive MIMO
are already available (e.g., [10,11]), including 5G (e.g., [3,4,6]), additional data are desirable while 5G
services become widely spread.
The maximum EIRPact reported in Section 3, obtained from an overestimate of Pavg was found to
be 8.8 dB below the theoretical maximum. This value would drop to −15 dB if Pavg was estimated from
the mean of the six-minute time-averaged transmit power distribution reported for every fifteen-minute
interval. EIRPact was determined with respect to the peak instantaneous EIRP including the effect of
the TDD downlink duty cycle (TDC). This allows for a direct comparison with other studies available
in the literature for which the TDC factor was also included when evaluating the actual power of
massive MIMO BSs.
The experimental values of EIRPact observed in this work are below the maximum actual power
levels previously estimated by statistical and numerical models (e.g., [8,9]). According to a technical
report from the International Electrotechnical Commission [15], the actual maximum transmit power
predicted for a massive MIMO site with horizontal beam scanning would only correspond to about 25%
(−6 dB) of the theoretical maximum (including the effect of 75% TDD downlink duty cycle). Such values
are consistent with measurements previously conducted on massive MIMO BSs of commercial 4G
networks [10,11]. In [10], the actual maximum exposure was found to be about 8.5 dB lower than
 levels previously estimated by statistical and numerical models (e.g., [8,9]). According to a technical
report from the International Electrotechnical Commission [15], the actual maximum transmit power
predicted for a massive MIMO site with horizontal beam scanning would only correspond to about
25% (−6 dB) of the theoretical maximum (including the effect of 75% TDD downlink duty cycle). Such
values
Appl. Sci. are
2020,consistent
10, 5280 with measurements previously conducted on massive MIMO BSs of commercial 9 of 10
4G networks [10,11]. In [10], the actual maximum exposure was found to be about 8.5 dB lower than
the theoretical maximum by using network counters like those utilized in this paper. In [11], field
the theoretical maximum
measurements conductedby in using network
front of counters
a massive MIMO like those
LTE utilized
antenna, ledintothis paper. values
exposure In [11],(when
field
measurements conducted in front of a massive MIMO LTE antenna, led to
scaled for the maximum BS utilization) ranging from 10 dB to 7 dB below the theoretical maximum. exposure values (when
scaledInfor the maximum
Figure BS utilization)
7, a comparison of the BS ranging from 10
compliance dB to 7 dB
boundary (CB)below the theoretical
or exclusion maximum.
zone delimiting the
area outside of which the EMF exposure is below the ICNIRP limits, is given for the delimiting
In Figure 7, a comparison of the BS compliance boundary (CB) or exclusion zone theoretical
the area outside
maximum EIRPof which
and theactual
for the EMF exposure
maximumisEIRP. belowThe theCBICNIRP limits, is based
is determined given for the envelope
on the theoretical of
maximum
all available EIRP
beams andfor
forthe
the8actual
CSI-RSmaximum EIRP. The CB
ports beamforming andisfor
determined
a maximum based on the envelope
instantaneous transmitof
all available
power of 200beams
W. For forthe
theactual
8 CSI-RS ports beamforming
maximum, the CB assessmentand forisadone
maximum instantaneous
by scaling transmit
down the power by
power of 200 W. For the actual maximum, the CB assessment is done by scaling down
8.8 dB (i.e., for a power level of about 26.4 W). The figure shows that the front compliance distance is the power by
8.8 dB (i.e.,
more than for a power
halved whenlevel of about 26.4
considering the W). The
effect of figure shows that
beamforming andthe frontvariation
traffic complianceon distance
the EMF
isexposure
more than halved when considering the effect of beamforming and traffic
levels (for a maximum 𝐸𝐼𝑅𝑃act of −15 dB the compliance distance in the front would variation on the EMF be
exposure levels (for a maximum EIRP
only 2.5 m, i.e., more than 5 times smaller). act of −15 dB the compliance distance in the front would be only
2.5 m, i.e., more than 5 times smaller).

Figure 7. Compliance boundary (CB) or exclusion zone for the BS determined according to EIRPact
Figure 7. Compliance boundary (CB) or exclusion zone for the BS determined according to 𝐸𝐼𝑅𝑃act
(red solid line) and the theoretical maximum EIRP (green dashed line).
(red solid line) and the theoretical maximum EIRP (green dashed line).
5. Conclusions
5. Conclusions
Experimental data on the actual EIRP levels of the massive MIMO BSs (based on single-user MIMO
Experimental
codebook-based data on thewith
beamforming actual EIRP ports)
8 CSI-RS levels operating
of the massive MIMO
in a live BSs (based
5G network on single-user
(Telstra, Australia)
MIMO codebook-based beamforming with 8 CSI-RS ports) operating in a live 5G
were obtained by monitoring the dynamics of the antenna radiation pattern and the time-averaged network (Telstra,
Australia)
transmit were About
power. obtained by monitoring
13 million samplestheof dynamics
the actual of thewere
EIRP antenna radiation
collected and pattern
processed andforthe
25time-
BSs
averaged transmit power. About 13 million samples of the actual EIRP were collected
for 24 h. The results show that assuming peak power transmission in a fixed beam for a long-time and processed
for 25 BSs
interval forto
leads 24unrealistic
h. The results show levels.
exposure that assuming peak power
This translates into atransmission
compliance in a fixedthat
distance beam is for
lessa
long-time
than interval
half than whatleads to unrealistic
is obtained for the exposure
theoreticallevels. This translates
maximum EIRP. Theseinto a compliance
findings distancewith
are consistent that
is less than half than what is obtained for the theoretical maximum EIRP.
the assessments made by the International Electrotechnical Commission in [15] and provide usefulThese findings are
consistent with the assessments made by the International Electrotechnical
insights for the development of RF EMF exposure assessment procedures of 5G BSs. Commission in [15] and
provide useful insights for the development of RF EMF exposure assessment procedures of 5G BSs.
Author Contributions: D.C. and C.T. conceived and designed the experiments; D.C. and P.J. analyzed the data;
Author
D.C., B.X.Contributions: D.C. with
and P.J. contributed and C.T. conceived
analysis tools.and
All designed the experiments;
authors contributed D.C. andthe
with writing P.J.paper.
analyzed the data;
All authors
have
D.C.,read
B.X.and
andagreed to the published
P.J. contributed versiontools.
with analysis of theAll
manuscript.
authors contributed with writing the paper. All authors
have readThis
Funding: andresearch
agreed to the published
received version
no external of the manuscript.
funding.
Funding: This research
Acknowledgments: Thereceived no external funding.
authors acknowledge the support received from the Ericsson colleagues in Australia
Troy Stephan, Christian Le Jalle, Daniel Ho and Radek Wrzesinski with collecting network data, and Dimple
Thomas, Denis Shleifman and Babak Tootonchi from Ericsson in Ottawa for their help with the processing tools.
The authors are also grateful to Debbie Wills, Mike Wood and Steve Iskra at Telstra for providing fruitful comments
and suggestions on the manuscript.
Conflicts of Interest: The authors declare no conflict of interest.
Appl. Sci. 2020, 10, 5280 10 of 10

References
1. Ericsson Mobility Report. Available online: https://www.ericsson.com/4acd7e/assets/local/mobility-report/
documents/2019/emr-november-2019.pdf (accessed on 11 May 2020).
2. ICNIRP. Guidelines for limiting exposure to electromagnetic fields (100 kHz to 300 GHz). Health Phys. 2020.
[CrossRef]
3. Ofcom. Electromagnetic Field (EMF) Measurements near 5G Mobile Phone Base Stations. Available online: https:
//www.ofcom.org.uk/__data/assets/pdf_file/0015/190005/emf-test-summary.pdf (accessed on 11 May 2020).
4. Telstra. 5 Surveys of 5G. Available online: http://1u0b5867gsn1ez16a1p2vcj1-wpengine.netdna-ssl.com/wp-
content/uploads/2019/07/5-Surveys-of-5G-flyer-A4.pdf (accessed on 11 May 2020).
5. IEC TC106. IEC 62232: Determination of RF Field Strength, Power Density and SAR in the Vicinity of
Radiocommunication Base Stations for the Purpose of Evaluating Human Exposure (Committee Draft); IEC:
Geneva, Switzerland, 2020.
6. Aerts, S.; Verloock, L.; Van Den Bossche, M.; Colombi, D.; Martens, L.; Törnevik, C.; Joseph, W. In-situ
measurement methodology for the assessment of 5G NR massive MIMO base station exposure at sub-6 GHz
frequencies. IEEE Access 2020, 7, 184658–184667. [CrossRef]
7. Keller, H. On the assessment of human exposure to electromagnetic fields transmitted by 5G NR base stations.
Health Phys. 2019, 117, 541–545. [CrossRef] [PubMed]
8. Thors, B.; Furuskär, A.; Colombi, D.; Törnevik, C. Time-averaged realistic maximum power levels for the
assessment of radio frequency exposure for 5G radio base stations using massive MIMO. IEEE Access 2017, 5,
19711–19719. [CrossRef]
9. Baracca, P.; Weber, A.; Wild, T.; Grangeat, C. A statistical approach for RF exposure compliance boundary
assessment in massive MIMO systems. In Proceedings of the 22nd International ITG Workshop Smart
Antennas (WSA), Bochum, Germany, 14–16 March 2018; pp. 1–6.
10. Colombi, D.; Joshi, P.; Pereira, R.; Thomas, D.; Shleifman, D.; Tootoonchi, B.; Xu, B.; Törnevik, C. Assessment
of actual maximum RF EMF exposure from radio base stations with massive MIMO antennas. In Proceedings
of the 2019 Photonics & Electromagnetics Research Symposium-Spring (PIERS-Spring), Rome, Italy,
17–20 June 2019. [CrossRef]
11. Werner, R.; Knipe, P.; Iskra, S. A comparison between measured and computed assessments of the RF
exposure compliance boundary of an in-situ radio base station massive MIMO antenna. IEEE Access 2019, 7,
170682–170689. [CrossRef]
12. Telstra. What You Need to Know about Our 2020 Half-Year Financial Results. Available online: https:
//exchange.telstra.com.au/what-you-need-to-know-about-our-2020-half-year-financial-results/ (accessed on
13 May 2020).
13. ICNIRP. Guidelines for limiting exposure to time-varying electric magnetic and electromagnetic fields (up
300 GHz). Health Phys. 1998, 74, 494–522.
14. 3GPP. 5G; NR; Physical Layer Procedures for Data (3GPP TS 38.214 version 15.9.0 Release 15). Available online:
https://www.etsi.org/deliver/etsi_ts/138200_138299/138214/15.09.00_60/ts_138214v150900p.pdf (accessed on
14 May 2020).
15. IEC TC106. IEC 62669 Ed.2: Case Studies Supporting IEC 62232—Determination of RF Field Strength, Power
Density and SAR in the Vicinity of Radiocommunication Base Stations for the Purpose of Evaluating Human Exposure;
IEC: Geneva, Switzerland, 2020.

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).