Documentos de Académico
Documentos de Profesional
Documentos de Cultura
discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/272515576
CITATIONS
READS
37
3 authors:
Teoman Karadag
Mehmet Yuceer
Inonu University
Inonu University
8 PUBLICATIONS 8 CITATIONS
SEE PROFILE
SEE PROFILE
T. Abbasov
Inonu University
54 PUBLICATIONS 205 CITATIONS
SEE PROFILE
All content following this page was uploaded by Teoman Karadag on 07 June 2016.
The user has requested enhancement of the downloaded file. All in-text references underlined in blue are added to the original document
and are linked to publications on ResearchGate, letting you access and read them immediately.
doi:10.1093/rpd/ncv008
The present study analyses the electric field radiating from the GSM/UMTS base stations located in central Malatya, a densely
populated urban area in Turkey. The authors have conducted both instant and continuous measurements of high-frequency
electromagnetic fields throughout their research by using non-ionising radiation-monitoring networks. Over 15 000 instant and
13 000 000 continuous measurements were taken throughout the process. The authors have found that the normal electric field radiation can increase 25 % during daytime, depending on mobile communication traffic. The authors research work has also
demonstrated the fact that the electric field intensity values can be modelled for each hour, day or week with the results obtained
from continuous measurements. The authors have developed an estimation model based on these values, including mobile communication traffic (Erlang) values obtained from mobile phone base stations and the temperature and humidity values in the environment. The authors believe that their proposed artificial neural network model and multivariable least-squares regression
analysis will help predict the electric field intensity in an environment in advance.
INTRODUCTION
In recent years, the effects of electromagnetic field pollution and electromagnetic radiation exposure on the
living environment have grown increasingly in parallel
with the rise in the number of base stations that
operate using ultra-high-frequency electromagnetic
waves(1). A non-ionising energy by nature, electromagnetic radiation spreads through waves or particle
shapes. Today, it has been made clear that a great
proportion of electromagnetic pollution is caused by
ultra-high-frequency electromagnetic waves emitted by
base stations. For this reason, scientific researches conducted in this field of study tend to focus on the effects
of mobile phones and base stations on the living environment(2 5). As shown by Joseph et al. and Genc
et al., most of the researches on human exposure have
been conducted on a short-term basis up to now(6, 7).
The number of studies with continuous and long-term
measurements is very few in the literature(8, 9). Today,
many researches are carried out to evaluate the effects
of the electromagnetic pollution on the biological
systems.(10 19) Most of these researches are about the
carcinogenic effects of base stations and mobile
phones. Some epidemiological studies have shown the
relationship between certain types of cancer and the
rise in the number of base stations and mobile telephone use. There is, however, a significant body of research to prove otherwise(10 13). In most studies, the
effects of the electromagnetic pollution on the wildlife
# The Author 2015. Published by Oxford University Press. All rights reserved. For Permissions, please email: journals.permissions@oup.com
ET AL.
T. KARADAG
changes have been evaluated to develop a mathematical prediction model. To this end, .13 000 000 continuous measurements have been carried out in the
city centre, and the obtained results were modelled by
statistical methods. This study aims to develop artificial neural network (ANN) and MLRS models with
a feed-forward back-propagation learning algorithm
for the estimation of the electric field in the environment, computing the following parameters: month, day,
time, mean temperature, mean humidity and mobile
phone communication traffic. The results and the measurements were found to be compatible.
Page 2 of 14
a representative data set. In general, the neural networks do not need much of a detailed description
or formulation of the underlying process and thus
appeal to practising engineers who tend to rely on
their own data(29). Recently, neural networks have
been successfully applied to process modelling and
control(30 35). By using the experimental observations
as the input data set to identify the effects of operating parameters on the electromagnetic field intensity
values, an ANN model was created. A three-layered
feed-forward and a back-propagation algorithm with
6 neurones in the first layer, 20 neurones in the interim
layer and 1 neurone in the last layer were chosen.
The network had one input layer, one hidden layer
and one output layer as shown in Figure 1. The first
layer has six logarithmic sigmoid (logsig) neurones,
the second layer has four logarithmic sigmoid neurones and last layer has one linear neurone. In the
course of training, which was based on Levenberg
Marquardt method, the number of hidden layers, the
number of neurones in the hidden layer, training accuracy and number of iterations were determined by using
trial and error method. The input data and the output
data were normalised and de-normalised before and
after the actual application in the network. Thus, the
model was trained for inputoutput behaviour of the
system.
For the development of the neural network model,
the Neural Network Toolbox and MATLAB 7.0 were
used. A MATLAB script was written, loaded the data
file, trained and validated the network and saved the
model architecture.
The transfer function logsig used in this study is
given in the following equation:
yi
1
;
1 ezi
Month,
Day,
Time,
Mean temperature (8C),
Humidity (%),
The number of connected subscriber.
Page 3 of 14
Two Wavecontrol MonitEM continuous electromagnetic field monitoring and measurement stations
were used to monitor the existing electric field level
differences in the city centre every half-hour in a day.
While one of these stations took measurements at
GSM and UMTS frequencies, the other one took
measurements in the frequency range of 100 kHz
8 GHz continuously for 24 h. The stations measured
the instant electric field values every half second in
half-hour periods and calculated the average electric
field value in a half-hour by using Sliding Window
method. It recorded the measured average electric
field values on a mini SD storage card and sent the
average value to the server via an internal GPRS
modem. The two MonitEM stations were placed in
the city centre to measure the electric field simultaneously, and half-hour measurement results were evaluated together. The MonitEM for GSM and UMTS
frequencies was placed in the high-density settlement
zone to take continuous measurements throughout
the study period. The continuous electric field measurement values were observed along with the temperature and humidity values of the environment
parameters and mobile communication traffic values.
The environment temperature and humidity values
were taken from Inonu University Observatory and
Meteorology Station (Malatya, Turkey) during the
measurement period. The mobile communication
traffic values were taken from mobile communication
companies. The obtained results were compared with
the national and international standards and modelled by multivariable least-squares regression
(MLSR) and ANN statistical methods.
ET AL.
T. KARADAG
where yi is the ith response; b0, b1 and bn 2 1 are the regression parameters; xi,1, . . . xi,n 2 1 are the ith predictor variables; 1i is the independent random error
associated with the ith response, typically assumed to
be distributed N(0,s). After being built on the available
data, the model will probably be used for predicting the
values taken by the response variable for new data
points fxg that are not in original data set.
minimise the effects of the high-frequency electromagnetic radiation exposure on human health and
living organisms. The standards for GSM (Global
System for Mobile Communication) base stations
were defined for European countries by ETSI (The
European Telecommunication Standardization Institute, 1997) and for the USA by FCC (The Federal
Communications Commission)(37). The limits defined
by ICNIRP are widely accepted in the international
arena as well. The following standards were used in the
evaluation of the measurements taken in the course of
this study: ICNIRP(25), IEEE Institute of Electrical
and Electronics Engineers(38) and International Electro
technical Commission (IEC)(39).
Information and Communication Technologies
Authority (ICTA) is responsible for determining the
electromagnetic limit values in Turkey. The limit
values for GSM Networks and the standards implemented in Turkey have been defined as sublimit
values in the above-mentioned standards. The limit
values are determined not only for overall environment but also for a single device. In this way, people
in the vicinity of an electromagnetic field emitting
source (e.g. base station) are better protected against
the health hazards of electromagnetic radiation(26).
Page 4 of 14
Figure 2. Two-dimensional impression of measured electric field values in the pilot urban area in the frequency range from
100 kHz to 8 GHz.
Figure 3. The geographical map of the pilot urban area with the electric field values measured in the frequency range of
100 kHz8 GHz.
Page 5 of 14
ET AL.
T. KARADAG
Figure 5. Electric field intensity graphic based on the continuous measurements conducted between 2 and 20 January 2014
in the frequency range of 100 kHz 8 GHz.
electric field value was 2.508 V m21. The graphics, including the continuous measurement results, are given
in Figures 5 and 6.
Given the significant changes in the measured
values, the necessity to develop a mathematic model
for evaluating the radiation of electromagnetic pollution became more evident. To analyse the results
better, the authors drew two graphics with the electric
field intensity values measured each half-hour periods
between 2 and 20 January 2014. The electric field
intensity values in the frequency range of 100 kHz
8 GHz (Broadband) and GSM and UMTS frequencies were measured everyday throughout this period,
and grouped on a half-hour basis. The graphics in
Figure 7 demonstrate the fact that there are parallels
and characteristic similarities between the broadband
measurements in the frequency range of 100 kHz
8 GHz and GSM and UMTS frequencies.
According to these results, 77 % of the electric field
intensity values radiate to the environment at the
range of GSM and UMTS frequencies, the rate of
23 % was measured at the range of 100 kHz 8 GHz
frequency from the radio and television broadcasting and high-frequency telecommunication devices.
Page 6 of 14
Figure 4. Three-dimensional impression of measured electric field values in the pilot urban area in the frequency range from
100 kHz to 8 GHz.
Figure 7. The graphics of the electric field intensity change in half-hour periods between 2 and 20 January 2014 as measured
by the continuous electric field monitoring and measurement stations in the frequency range of 100 kHz 8 GHz and GSM
and UMTS frequencies.
However, the electric field intensity levels were measured as spectrum at the frequencies of 900, 1800 and
2100 MHz was calculated as 65 and 35 % of the electric field level by the TV, FM radio, Wi-Fi, etc. frequencies in city centre in a year ago.
The electromagnetic field intensity in the environment resulted from GSM and UMTS signals emitted
from cell phone base stations. The changes in the values
depend on the number of calls, duration of the calls on
the base stations and the physical parameters of the environment (temperature and humidity). Therefore, the
principal parameters affecting the distribution of electromagnetic pollution should be determined.
In the light of the instant measurements, high electric field values were detected in the residential area
where the study was conducted. The point with the
highest mobile communication traffic was detected in
the pilot area, and a continuous monitoring and
measurement station was set there. The continuous
measurements were carried out in GSM and UMTS
Page 7 of 14
Figure 6. Electric field intensity graphic of GSM and UMTS frequencies based on the continuous measurements conducted
between 2 and 20 January 2014.
ET AL.
T. KARADAG
Figure 9. Graphic of hourly mobile phone communication traffic on the base station in the residential area.
Page 8 of 14
Figure 8. Time-varying graphic of electric field intensity values measured in GSM and UMTS frequencies in the residential
area.
Figure 10. Hourly time-varying chart of electric field intensity values measured for 5 weeks during the study period.
Figure 11. Graphics of hourly temperature and humidity changes during the study period.
q ;
q
R P
N
2 PN
y2
i1 xi x
i1 yi
s
PN
2
i1 yi xi
;
RMSE
N
N
1X
yi xi
100;
MAPE %
N i1 xi
Page 9 of 14
ET AL.
T. KARADAG
Table 1. The electric field measured hourly for 24 h, the number of subscriber attempts using the base station, hourly average
temperature and humidity values in the environment.
Date and time
Temperature 8C
Humidity %
1.695
1.48
1.24
1.185
1.02
1.02
1.055
1.205
1.185
1.305
1.54
1.615
1.73
1.71
1.66
1.64
1.71
1.845
1.87
2.02
2.07
2.04
2
1.935
1260
574
265
218
122
218
855
2467
3556
6092
8354
10 235
10 726
9833
8742
9165
10 053
11 517
12 418
12 713
11 311
7807
5301
3292
16.94167
16.80833
16.475
16.05833
15.95
15.56667
16.05833
17.225
17.875
18.65833
19.33333
20.23333
20.925
21.9
22.11667
22.41667
22.2
20.375
19.30833
18.9
19.03333
18.68333
18.39167
17.95
34833.00
35333.00
36.5
38583.00
40.25
41917.00
41833.00
39167.00
37.75
36.25
33583.00
31667.00
30.75
29417.00
29.25
28.5
28833.00
32.75
34083.00
36
36.25
37083.00
37917.00
39167.00
bi
1.2889
3.565` 1024
20.0118
0.0273
2.99` 1025
20.0043
6.17` 1024
The input and output data were normalised and denormalised before and after the actual application in
the network. Thus, the model was trained for input
output behaviour of the system.
The MLSR analysis was performed on the training
set, which is used to develop the neural networks. The
regression parameters of the MLSR model are given
in Table 2.
The test data set was used to assess the performances of the MLSR equations. The result of MLSR
model is shown in Figure 12.
The ANN model results are shown in Figure 13.
The behaviour of the network for the test data is shown
in Figure 14. As detected from Figure 14, the network
model results are compatible with the observations.
A higher value of the correlation coefficient and
the smaller values of MAPE and RMSE mean a
better performance of the model. The correlation
coefficients calculated for training and testing of the
network were 0.9913 and 0.9849, respectively. The
Page 10 of 14
E [V m21]
Figure 13. ANN training results: predicted electric fields (V m21) versus measured values.
Figure 14. ANN testing results: predicted electric fields (V m21) versus measured values.
Page 11 of 14
Figure 12. MLSR results for testing data: predicted electric fields (V m21) versus measured values.
ET AL.
T. KARADAG
Table 3. Performance indexes achieved using ANN and MLSR during training and validation periods.
Training data
Test data
Model
MAPE (%)
RMSE
MAPE (%)
RMSE
ANN
MLSR
0.9913
0.8213
2.2230
10.2869
0.0451
0.1961
0.9849
0.8073
3.0353
11.2745
0.0636
0.2152
The results obtained from continuous measurements can be shared with the national and international public in order to prevent the health hazards
caused by the electromagnetic pollution.
It is possible to form a statistical model of the electromagnetic pollution changes in the city centre, commercial and residential areas based on the results
obtained from continuous measurements. ANN and
MLSR methods can be used for this purpose.
The proposed ANN model predicts the electric
field in the environment, when the month, day, time,
mean temperature, humidity and the number of connected users are provided. The maximum errors for
the MLSR and ANN models for validation were
found to be 0.7789 and 0.2787, respectively. The estimation capability of ANN is better than MLSR in
predicting electric field of complex and non-linear
systems. These results showed that the ANN model is
useful in electric field predictions.
The optimum control systems can be feasible to
monitor the electromagnetic pollution changes caused
by the base stations in the city centre and residential
areas by using mathematical modelling during any
time of the day with the mobile communication traffic
and physical parameters of the environment. Future
electric field predictions could be performed using
these data items and 2-D, 3-D maps.
FUNDING
This work was supported by the Inonu University
.B.A.P. 2012-28
Research Fund [Project Numbers I.U
and 2013-45].
REFERENCES
1. Sorgucu, U. and Develi, I. Measurements and analysis
of electromagnetic pollution generated by GSM-900
mobile phone networks in Erciyes University Turkey.
Electromagn. Biol. Med. 31, 404 415 (2012).
2. Ulaby, F. T. Fundamentals of Applied Electromagnetics.
Prentice Hall (2001).
3. Hutter, H. P. et al. Subjective symptoms, sleeping problems, and cognitive performance in subjects living near
mobile phone base stations. Occup. Environ. Med. 63,
307 313 (2006).
4. Seyfi, L. Measurement of electromagnetic radiation with
respect to the hours and days of a week at 100 kHz3
Page 12 of 14
5.
6.
7.
8.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
Page 13 of 14
9.
ET AL.
T. KARADAG
35. Yldz, Z., Yuceer, M. and Abbasov, T. Comparison of
modeling approaches for prediction of cleaning efficiency
of the electromagnetic filtration process. Appl. Comput.
Electromagn. Soc. J. 26, 899906 (2011).
36. The Mathworks, Inc. Available on http://www.
mathworks.com (2003).
37. Federal Communications Commission Evaluating compliance with FCC guidelines for human exposure to radiofrequency electromagnetic fields. 199701 ed: OET
Bulletin 65 (1997).
Page 14 of 14