Determination of the duty cycle of WLAN for realistic radio frequency electromagnetic field exposure assessment.

TL;DR: In this paper, duty cycles of WLAN using Wi-Fi technology are determined for exposure assessment on large scale at 179 locations for different environments and activities (file transfer, video streaming, audio, surfing on the internet, etc.).

Abstract: Wireless Local Area Networks (WLANs) are commonly deployed in various environments. The WLAN data packets are not transmitted continuously but often worst-case exposure of WLAN is assessed, assuming 100% activity and leading to huge overestimations. Actual duty cycles of WLAN are thus of importance for time-averaging of exposure when checking compliance with international guidelines on limiting adverse health effects. In this paper, duty cycles of WLAN using Wi-Fi technology are determined for exposure assessment on large scale at 179 locations for different environments and activities (file transfer, video streaming, audio, surfing on the internet, etc.). The median duty cycle equals 1.4% and the 95th percentile is 10.4% (standard deviation SD = 6.4%). Largest duty cycles are observed in urban and industrial environments. For actual applications, the theoretical upper limit for the WLAN duty cycle is 69.8% and 94.7% for maximum and minimum physical data rate, respectively. For lower data rates, higher duty cycles will occur. Although counterintuitive at first sight, poor WLAN connections result in higher possible exposures. File transfer at maximum data rate results in median duty cycles of 47.6% (SD = 16%), while it results in median values of 91.5% (SD = 18%) at minimum data rate. Surfing and audio streaming are less intensively using the wireless medium and therefore have median duty cycles lower than 3.2% (SD = 0.5–7.5%). For a specific example, overestimations up to a factor 8 for electric fields occur, when considering 100% activity compared to realistic duty cycles.

Summary

Introduction

• Elsevier Editorial System(tm) for Progress in Biophysics and Molecular Biology Manuscript Draft Manuscript Number: PBMB-D-12-00066R1 Title: DETERMINATION OF THE DUTY CYCLE OF WLAN FOR REALISTIC RADIO FREQUENCY ELECTROMAGNTIC FIELD EXPOSURE ASSESSMENT.
• Duty cycles of WLAN using Wi-Fi technology are determined for exposure assessment on large scale at 179 locations for different environments and activities (file transfer, video streaming, audio, surfing on the internet, etc.).
• Largest duty cycles are observed in urban and industrial environments.
• For lower data rates, higher duty cycles will occur.

REFERENCE PBMB-D-12-00066 DETERMINATION OF THE DUTY CYCLE OF WLAN FOR REALISTIC

• The authors have included the document “Response to reviewers” at the end of the revised manuscript.
• This is a detailed summary of the changes made in preparing the revised manuscript.
• The authors have put an asterisk before their answers and changes.

REFERENCE PBMB-D-12-00066 DETERMINATION OF THE DUTY CYCLE OF WLAN FOR REALISTIC RADIO FREQUENCY ELECTROMAGNTIC FIELD EXPOSURE ASSESSMENT

• This document contains the detailed summary of the changes made in preparing the revised manuscript, also known as Dear reviewer 1.
• The authors prefer this realistic worst-case approach as one cannot know how long an activity takes: surfing, skype, watching a movie will take longer than 6 minutes, watching/streaming a trailer might be shorter.
• Also for different parallel applications, the same reasoning can be used and the resulting duty cycle will be the sum of the duty cycles of the individual applications with as maximal values the upper limits of Table 3.
• In general, poor connections have low data rates resulting in larger duty cycles and thus an increased exposure at all distance from the access point.

FREQUENCY ELECTROMAGNTIC FIELD

• W. Joseph is a Post-Doctoral Fellow of the FWO-V (Research Foundation–Flanders), also known as Funding sources and acknowledgment.
• The research leading to these results has received funding from the European Union’s Seventh Framework Programme (FP7/2007-2013) project SEAWIND under grant agreement no 244149.
• The research is also partly funded by the Fund for Scientific Research – Flanders (FWO-V, Belgium) project G.0325.11N.
• With thanks to the 1 st author’s mother Lud Depraetere + , who was helpful until the end….

2.1 METHOD TO ASSESS WLAN DUTY CYCLES IN-SITU

• In this way the maximum field level during the measurement time is determined.
• Thirdly, the duty cycle of the active channels is determined.
• These settings are listed in Table 1 and validated in Verloock et al. (2010).
• Finally, the total averaged field is determined from the duty cycle and the max-hold field strength as follows: )/(max mVEDE holdtot avg tot (2) with D the duty cycle, avgtotE the total average (over 6 min, 30 min) electric-field strength due to WLAN, and holdtotE max the max-hold electric field strength (assuming continuously present).

2.2.1 Equipment

• Using a Wi-Fi-packet analyzer, the active Wi-Fi channels are determined.
• The analyzer consists of the software tool Airmagnet (Airmagnet 2011) together with a laptop and a Wi-Fi card (type Proxim ORiNOCO 11 a/b/g Client Combocard gold).
• They were originally defined in IEEE Std 802.11b-1999 and IEEE Std 802.11g-2003, respectively, but both standard documents are currently obsoleted by the revised standard document IEEE Std 802.11-2012, which still includes these physical air interfaces.
• The SA-measurement setup of the narrowband measurements consists of tri-axial Isotropic Antennas (type Rohde and Schwarz TS-EMF, dynamic range of 1 mV/m – 100 V/m for the frequency range of 80 MHz – 3 GHz) in combination with a spectrum analyser (type Rohde and Schwarz FSL6, frequency range of 9 kHz – 6 GHz).
• The measurement uncertainty for the electric field is ± 3 dB for the considered setup (CENELEC 2008).

2.2.2 Configurations

• The WLAN duty cycle is measured with the SA using the procedure of above at a total of 179 locations in two countries, namely, Belgium and the Netherlands.
• At all these locations, the duty cycle could be assessed as WLAN was significantly present (in total 344 locations are considered and at 179 WLAN was measured).
• The considered environments are the following: rural, residential, urban, suburban, office, and industrial environments (Joseph et al. 2012, Verloock et al. 2010).
• Both indoor and outdoor locations are considered.
• Table 2 summarizes the environments and the number of locations per environment where WLAN was measured.

2.3.1 Theory and method

• One can also assess exposure during typical activities or using different applications (VoIP, file transfer, video streaming, audio, surfing on the internet, etc.).
• For their lab assessment, the authors use a 802.11a network instead of a 802.11g network as the former is deployed at 5 GHz instead of 2.4 GHz for the latter.
• To transmit data from a client to an access point (AP), there is a waiting time DIFS (Distributed Inter-Frame Space) and a random backoff time B (to avoid that multiple users would access the wireless medium simultaneously).
• Highest theoretical duty cycles are thus obtained for 6 Mbps (lowest modulation and data rate for 802.11a and thus the worst-quality connection) and equal to 94.7%.

2.3.2 Procedure to estimate exposure using duty cycle

• Firstly, one performs a measurement of the total Wi-Fi exposure using max-hold setting of the SA (huge overestimation: 100% duty cycle), which is a cumulative value if more than one client is present.
• Secondly, one can select the duty cycles from Table 4 for the application used.
• The second reason is that the client data transfers will also throttle back when using the TCP transport layer (which is still mostly used for reliable data transfers).
• So their results are realistic worst-case values.

3.1 GENERAL RESULTS AND RESULTS PER ENVIRONMENT

• Table 2 lists the overall duty cycle measured during the large measurement campaign performed in Belgium and the Netherlands (“All environments”).
• It is clear that the worst-case approaches assuming continuous WLAN exposure (thus D = 100 %) result in large overestimations.
• The mean value of D in Khalid et al. (2011) is higher because networks in schools are considered during lessons thus during activity (overall median of 1.4% versus 4.8% in Khalid et al. (2011)).
• In their study, duty cycles in actual circumstances are measured without knowledge of activity in the different environments.

3.2 DUTY CYCLE FOR DIFFERENT WLAN APPLICATIONS

• Figure 3 shows the duty cycle at 54 Mbps (channel occupation in %) versus time for three applications namely VoIP, video streaming, and file transfer.
• For the YouTube video streaming, the video file is buffered during the first 50 seconds (also around 60%) and after this period with high duty cycle, the channel occupation reduces to 0.1% because all data has been received and only basic control information is still being sent.
• File transfer causes the highest duty cycles up to 66 % (54 Mbps) and 94% (6 Mbps), which approaches the theoretical maximal duty cycle of 69.83 % (54 Mbps) and 94.7% (6 Mbps) (see Section 2.3) and shows an excellent agreement between theoretical calculations and measurements.
• Thus during intensive applications much higher duty cycles can occur and exposures can increase.

3.3 APPLICATION: SIMULATION OF FIELD STRENGTH WITH REALISTIC DUTY CYCLES

• To investigate the impact of the resulting field strength with realistic duty cycles, the authors run simulations (Finite-Difference Time-Domain FDTD, SEMCAD-X, Speag, Switzerland) of a DLink DI-624 AirPlusXtremeG access point with an Equivalent Isotropically Radiated Power (EIRP) of 20 dBm.
• If the authors consider realistic high duty cycles in office environments and use the value of 6.1 % (p95 in office environments, Table 2), they obtain 1.36 V/m (45 times below ICNIRP).
• Institute of Electrical and Electronics Engineers IEEE 802.11a, 1999.
• Exposure to radio frequency electromagnetic fields from wireless computer networks: Duty factors of Wi-Fi devices operating in schools.

Belgium and the Netherlands.

• Wi-Fi occupation during time for different activities at 54 Mbps (VoIP, video streaming, and file transfer), also known as Figure 3.
• Electric field of an access point at 2.4 GHz for different duty cycles (p50, p90 office, video streaming), also known as Figure 4.

Full text

Elsevier Editorial System(tm) for Progress in Biophysics and Molecular Biology
Manuscript Draft
Manuscript Number: PBMB-D-12-00066R1
Title: DETERMINATION OF THE DUTY CYCLE OF WLAN FOR REALISTIC RADIO FREQUENCY
ELECTROMAGNTIC FIELD EXPOSURE ASSESSMENT
Article Type: Review Article
Keywords: WLAN; Wi-Fi; radio frequency; duty cycle; exposure; electromagnetic
Corresponding Author: prof Wout Joseph, Ph.D.
Corresponding Author's Institution: Ghent University/IBBT
First Author: Wout Joseph, Ph.D.
Order of Authors: Wout Joseph, Ph.D.; Daan Pareit; Gunter Vermeeren; Dries Naudts; Leen Verloock;
Luc Martens; Ingrid Moerman
Abstract: Wireless Local Area Networks (WLANs) are commonly deployed in various environments.
The WLAN data packets are not transmitted continuously but often worst-case exposure of WLAN is
assessed, assuming 100 % activity and leading to huge overestimations. Actual duty cycles of WLAN
are thus of importance for time-averaging of exposure when checking compliance with international
guidelines on limiting adverse health effects. In this paper, duty cycles of WLAN using Wi-Fi technology
are determined for exposure assessment on large scale at 179 locations for different environments and
activities (file transfer, video streaming, audio, surfing on the internet, etc.). The median duty cycle
equals 1.4 % and the 95th percentile is 10.4 % (standard deviation SD = 6.4%). Largest duty cycles are
observed in urban and industrial environments. For actual applications, the theoretical upper limit for
the WLAN duty cycle is 69.8% and 94.7% for maximum and minimum physical data rate, respectively.
For lower data rates, higher duty cycles will occur. Although counterintuitive at first sight, poor WLAN
connections result in higher possible exposures. File transfer at maximum data rate results in median
duty cycles of 47.6% (SD = 16%), while it results in median values of 91.5% (SD = 18%) at minimum
data rate. Surfing and audio streaming are less intensively using the wireless medium and therefore
have median duty cycles lower than 3.2% (SD = 0.5-7.5%). For a specific example, overestimations up
to a factor 8 for electric fields occur, when considering 100 % activity compared to realistic duty cycles.
Response to Reviewers: Summary of the changes
October 5, 2012
Wout Joseph
Dept. of Information Technology
Ghent University/IBBT
Gaston Crommenlaan 8 box 201
B-9050 Ghent
BELGIUM
REFERENCE PBMB-D-12-00066 DETERMINATION OF THE DUTY CYCLE OF WLAN FOR REALISTIC
RADIO FREQUENCY ELECTROMAGNTIC FIELD EXPOSURE ASSESSMENT

Dear Editor:
We have included the document “Response to reviewers” at the end of the revised manuscript. This is a
detailed summary of the changes made in preparing the revised manuscript. We have put an asterisk
before our answers and changes.
Yours sincerely,
Prof. Wout Joseph, Ph. D.
Dr. Ir. Daan Pareit, Ph. D.
Ir. Dries Naudts
Leen Verloock
Ir. Günter Vermeeren
Prof. Luc Martens, Ph. D.
Prof. Ingrid Moerman, Ph. D.
Letter for reviewer 1
October 5, 2012
Wout Joseph
Dept. of Information Technology
Ghent University/IBBT
Gaston Crommenlaan 8 box 201
B-9050 Ghent
BELGIUM
REFERENCE PBMB-D-12-00066 DETERMINATION OF THE DUTY CYCLE OF WLAN FOR REALISTIC
RADIO FREQUENCY ELECTROMAGNTIC FIELD EXPOSURE ASSESSMENT
Dear reviewer 1:
This document contains the detailed summary of the changes made in preparing the revised
manuscript. We hope the changes we made according to your suggestions (explanation unit of time,
multiple clients, far vs. near exposure vs. duty cycle, etc.) will satisfy you. We have put an asterisk
before our answers and changes.
We thank you for spending your time reviewing our paper.
Yours sincerely,
Prof. Wout Joseph, Ph. D.
Dr. Ir. Daan Pareit, Ph. D.
Ir. Dries Naudts
Leen Verloock

Ir. Günter Vermeeren
Prof. Luc Martens, Ph. D.
Prof. Ingrid Moerman, Ph. D.
Reviewer 1 Comments:
*We have put an asterisk (*) before every response.
Reviewers' comments:
Reviewer #1: This is a well written and interesting paper, reporting methods to calculate the duty cycle
for WLAN transmitters under different real time scenarios. The data reported will be very useful in
realistic assessment of the exposure of people to WLAN sources.
*Thank you for appreciating our paper.
Here are some comments:
1- The main comment is related to the unit of time (relevant time frame) for duty cycle calculations. In
Khalid et al 2011, the unit of time was considered as the duration of a typical lesson in a school ( about
30min). However the authors here considered the duration of a file transfer for example ( or any other
activity), as the unit of time. This needs to be elaborated further as whether the time to transfer a file
(which is very small) is enough to calculate the average exposure (in light of the ICNIRP's requirement
of 6 min averaging)?
Response to Reviewer comment No. 1
* We found the article of Khalid et al very interesting and informative, but have a slightly different –and
complementary– approach to the latter article. Khalid et al present elaborated exposure results for
classroom scenarios in UK schools. In our article, one of the purposes is to find an upper limit for the
duty cycle, based on the type of traffic that is used, which can then be applied to predict maximum duty
cycles in different scenarios, depending on the application type and usage. We used indeed deliberately
the duration of the considered activity as unit of time for the assessment of the duty cycles of actual
applications (2 to about 6 minutes). For comparison with ICNIRP limits, we advise to use e.g., the mean
duty cycles of Table 4 to have a realistic worst-case value. The reason is the following. If we would
average a single 2-minute-activity over a 6 minute time frame (ICNIRP), we would obtain lower duty
cycles, because no traffic is sent (except for e.g. some sporadic beacon control frames) during the
remaining 4 minutes, which would decrease the obtained average value. But suppose that this 2-
minute-activity is repeated several times for more than 6 minutes (e.g. successively watching different
YouTube videos), we would obtain the average value we consider within this article. This is a “realistic
worst-case” approach (as mentioned in the paper) assuming that the (successive) activities will take
longer than 6 minutes (surfing on internet, skype, streaming at home, watching a movie). We prefer
this realistic worst-case approach as one cannot know how long an activity takes: surfing, skype,
watching a movie will take longer than 6 minutes, watching/streaming a trailer might be shorter.
From our results one could try to determine “actual (on average)” duty cycles by multiplying the duty
cycles we provide with usage patterns i.e., multiplying with the ratio of the average duration of an
activity and 6 minutes (if the duration is longer than 6 minutes the ratio should be equal to 1). E.g., for
a 2h-movie multiply by 1 as this takes longer than six minutes but watching a YouTube trailer of 2
minutes would be multiplying with a ratio of 2/6=1/3. But as one cannot know how long a time frame
takes, we prefer this realistic worst-case approach, which can thus be an overestimation sometimes.
Behaviour of children is even more difficult to determine but can then also to be characterized

(sometimes shorter than 6 min watching, but sometimes trailers of child programs of 10-14 minutes
e.g., Dobus, Big & Small, Bumba (Belgium), Zandkasteel (Belgium, the Netherlands), etc.).
Your remark made us aware of the fact that there might be some confusion about the unit of time that
is used. This has now been explained (we consider realistic worst-case) in Section 2.3 (when discussing
the durations) and the usage patterns are mentioned as future research in the conclusions. The
following text is added in Section 3.2:
“The unit of time in this paper is thus the duration of an activity. For comparison with ICNIRP limits,
we advise to use the mean duty cycles of Table 4 to have a realistic worst-case value. This is a “realistic
worst-case” approach assuming that the activities will take longer than 6 minutes, which is the ICNRIP
time averaging requirement (surfing on internet, skype, streaming at home, watching a movie,
successive activities). We prefer this realistic worst-case approach as one cannot know how long an
activity in reality lasts: surfing, skype, watching a movie will take longer than 6 minutes but
watching/streaming a trailer might be shorter.”
*In the conclusions we added the following:
“Finally, we provided realistic worst-case duty cycles for the various activities. By accounting for usage
patterns for these activities, actual averages could be obtained. The usage patterns might however be
difficult to acquire in some cases (such as child behavior).”
2- The authors mention that one can use the duty cycles reported in this paper for realistic exposure
estimation when activities of people are known well, for example exposure of children during video
streaming in a class. Given that the authors only calculate the duty cycle during the actual file transfer (
unit of time here), what happens if 30 children do this activity all together? how would the authors
estimate the cumulative exposure when the averaging time is not sufficiently long?
Response to Reviewer comment No. 2
*For the effect of multiple users: in Section 3.2 we mention that we assume no other clients are
present. When there are multiple clients that have separate data transfers (thus no multicasting of the
video stream), these different data streams will have to be sent over the wireless network. As you
know, Wi-Fi clients cannot send simultaneously but will send intermittently (after contenting with
each other for wireless medium access). Thus, the duty cycle will gradually increase with multiple
clients until the maximum duty cycle (as indicated in the article) is reached. However, in this case there
are at least two factors which will cause the duty cycle not to increase linearly and which cause the
maximum achieved duty cycle for a group of clients to be lower compared to the case when only one
client would be heavily using the wireless medium at its maximum. The first one is that a single client
results in a minimum back-off time B, but for multiple clients, a higher back-off time will be present.
This results in more idle periods and thus in lower duty cycles. The second reason is that the client
data transfers will also throttle back when using the TCP transport layer (which is still mostly used for
reliable data transfers). This way, we can again state that the presented maximum duty cycles for a
single user are a realistic worst-case value, which is even more pessimistic for a higher amount of
clients.
For the duration of the averaging time: For shorter durations than 6 minutes the usage patterns from
the remark above can again be used.
For the cumulative exposure: one performs a measurement of the total Wi-Fi exposure using max-hold
(huge overestimation), which is a cumulative value from the different clients (children using the
laptop). Next, by multiplying with the duty cycles that are presented in this article, one thus obtains a
realistic worst-case value prediction.
Most scenarios will have nowadays one single concurrent user at a hotspot, for use at house etc.
(information of mobile operators, networks are currently designed in this way). The duty cycles we
provided can then be applied. However there are exceptions like the class room example with children
using a wireless application and in future increasing use may change this statement. Therefore the

following procedure to assess (cumulative exposure) is proposed, this is also now included in Section
2.3:
<here equation (3) from paper>
Firstly, one performs a measurement of the total Wi-Fi exposure using max-hold setting of the SA (huge
overestimation: 100% duty cycle), which is a cumulative value if more than one client is present.
Secondly, one can select the duty cycles from Table 4 for the application used. Finally, by multiplying
with the duty cycles, one thus obtains a realistic worst-case value prediction for single user and single
application.
If multiple clients are present, one can estimate the resulting duty cycle as the duty cycle of one client
times number of clients, with as theoretical maximum the values of Table 3:
(3)
Where Dmultiple clients, D, n, and Dmax represent the duty cycle with n clients, D the duty cycle for a
single client, n the number of clients and Dmax the upper limits of Table 3. However, the upper limits of
Table 3 will be an overestimation due to the following reasons. The first one is that a single client
results in a minimum back-off time B, but for multiple clients, a higher back-off time will be present.
This results in more idle periods and thus in lower duty cycles. The second reason is that the client
data transfers will also throttle back when using the TCP transport layer (which is still mostly used for
reliable data transfers). This way, we can again state that the presented maximum duty cycles are a
realistic worst-case value, which is even more pessimistic for a higher amount of clients. Calculations
show that for maximal occupation the duty cycles from Table 3 namely 70-94% (CW = 15, single user)
reduce to 5.5-31.3% (CW = 1023, a lot of users) for 54 – 6 Mbps, respectively. So our results are
realistic worst-case values. We advise not to apply the numbers for maximal occupation, as these can
be an underestimation when multiple users are present but maximal occupation is not reached.
Also for different parallel applications, the same reasoning can be used and the resulting duty cycle will
be the sum of the duty cycles of the individual applications with as maximal values the upper limits of
Table 3.
*We added this now in the text in Section 2.3, in the new subsection 2.3.2 (Procedure to estimate
exposure using duty cycle).
3- The authors rightly argue that poor WLAN connections result in higher possible exposure. They also
suggest that since positions with bad connections are located far from the access point, it results in
lower received powers and lower exposure due to higher distance. My question to the authors is what
will be the exposure of a person that stands next to the access point? surely he/she would not be
exposed as low as someone in a longer distance. I think the authors need to explain this slightly further
in the text.
Response to Reviewer comment No. 3
*The sentence “Note however that positions with bad connections are located far from the access
point, resulting also in lower received powers and lower exposures (higher distances)” might confuse
the reader. The sentence is written from the view point of the user of a PC or laptop who is at a large
distance of the access point and having a poor wireless connection. This user will be exposed to low
levels of EMF. Of course, when looking at exposure, also persons who are not using the network will be
exposed. These persons can be at any distance from the access point. In general, poor connections have
low data rates resulting in larger duty cycles and thus an increased exposure at all distance from the
access point. In conclusion, the influence of the duty cycle on the exposure is independent of the
distance.
To avoid confusion, we have removed the mentioned sentence from the text and modified the text as
follows:

Frequently asked questions

Q1. What are the contributions in this paper?

In this paper, duty cycles of WLAN using Wi-Fi technology are determined for exposure assessment on large scale at 179 locations for different environments and activities ( file transfer, video streaming, audio, surfing on the internet, etc. ). 

Q2. How many samples are acquired each second?

Samples are acquired each second over at least120 s (or more, up to 350 s) until the video or audio fragment is finished or the file is transferred. 

Q3. What is the reason why the authors suggest that poor connections result in higher exposure?

They also suggest that since positions with bad connections are located far from the access point, it results in lower received powers and lower exposure due to higher distance. 

Q4. What are the environments that are considered?

The considered environments are the following: rural,residential, urban, suburban, office, and industrial environments (Joseph et al. 2012, Verloock etal. 2010). 

Q5. What is the sa-measurement setup of the narrowband measurements?

The SA-measurement setup of the narrowband measurements consists of tri-axial IsotropicAntennas (type Rohde and Schwarz TS-EMF, dynamic range of 1 mV/m – 100 V/m for thefrequency range of 80 MHz – 3 GHz) in combination with a spectrum analyser (type Rohde andSchwarz FSL6, frequency range of 9 kHz – 6 GHz). 

Q6. What is the way to assess the WLAN exposure?

The duty cycles per environment andfor the various applications can be used for practical exposure assessment where currently hugeoverestimations are made by assuming continuous WLAN exposure. 

Q7. How long did Khalid et al use the time to assess the duty cycles of actual?

The authors used indeed deliberately the duration of the considered activity as unit of time for the assessment of the duty cycles of actual applications (2 to about 6 minutes). 

Q8. How long did Khalid et al use for the assessment of the duty cycles of actual?

The authors used indeed deliberately the duration of the considered activity as unit of time for the assessment of the duty cycles of actual applications (2 to about 6 minutes).