Indoor-to-outdoor channel characterization for modeling and prediction of interference in next generation wireless networks

Abstract

This paper presents an indoor-to-outdoor signal propagation model based on exhaustive measurements campaign. The conducted measurements considered an indoor transmitter and an outdoor receiver to analyze the interference in future 5G networks with femtocells overlaying macrocells. To propose a practical model, architectural details of the buildings and the exact position of the transmitter, which are not typically known to the network designers are considered random in the proposed model, which resulted in a stochastic non-site-specific model for application in 5G networks.

Full text

Indoor-to-Outdoor Channel Characterization for
Modeling and Prediction of Interference in Next
Generation Wireless Networks
S. Hamid
1
, A. Al-Dweik
1,2
, M. Mirahmadi
2
, K. Mubarak
1
, and A. Shami
2
1
Khalifa University, Abu Dhabi, UAE, e-mail:{sanaa.hamid, dweik, mubarak}@kustar.ac.ae
2
Western University, London, Ontario, Canada, e-mail:{mmirahma, ashami2}@uwo.ca
Abstract—This paper presents an indoor-to-outdoor signal
propagation model based on exhaustive measurements campaign.
The conducted measurements considered an indoor transmitter
and an outdoor receiver to analyze the interference in future
5G networks with femtocells overlaying macrocells. To propose
a practical model, architectural details of the buildings and the
exact position of the transmitter, which are not typically known
to the network designers are considered random in the proposed
model, which resulted in a stochastic non-site-specific model for
application in 5G networks.
Index Terms—Femtocell, indoor-to-outdoor channel, measure-
ments, propagation, fading, shadowing, building model.
I. INTRODUCTION
Deployment of femtocells over macrocells is considered an
attractive solution for increasing the coverage and providing
high data rates services to indoor users. Femtocells, which are
also known as femto basestations (FBS) in WiMAX or Home
NodeB (HNB) in 3GPP [1] utilize unused spectrum, or share
resources with existing macrocells networks [2]. Accordingly,
deploying femtocells without proper planning can cause severe
interference with macrocells, i.e., cross-tier interference, or
with neighboring femtocells, i.e., co-tier interference.
The optimal goal in femtocells deployments is to increase
the coverage and maintain the required quality-of-service
(QoS) within the building while reducing the interference
and leakage outside the building. Achieving these two con-
flicting goals simultaneously requires accurate modeling for
the indoor-to-outdoor signal propagation, which is affected by
several factors such as the buildings architecture or construc-
tion materials, the surrounding environment, and the complex
effects of shadowing and fading.
The analysis of the propagation in such scenarios could be
numerically evaluated using ray tracing [3], Finite-Difference
Time-Domain (FDTD) [4], or other hybrid methods [5]. These
methods allow for deterministic modeling of the propagation
even in mixed outdoor-indoor environments. However, they
are usually complex, time-consuming, and their results are
site-specific. They also require large amount of details in-
cluding geometric descriptions, material characterization, and
topographical maps of the environment.
In contrast to huge body of knowledge regarding outdoor-
to-indoor and indoor propagation, e.g., [6], [7], very few
considered the indoor-to-outdoor signal propagation. In [8],
the authors developed an empirical indoor-to-outdoor prop-
agation model based on power measurements undertaken at
two different residential areas. A frequency-dependent model
was obtained by considering carrier frequencies of 0.9, 2, 2.5,
and 3.5 GHz. In this model, the outdoor and indoor path losses
were evaluated separately and the examined indoor transmitter
locations were limited to external rooms and adjacent rooms.
In [4], [5], the results of FDTD are compared to measurements
for a similar scenario. A comparison of the ray tracing
results of indoor-to-outdoor against measurements for indoor
transmitter located in different floors is reported in [3]. In an
another approach, the model in [9] describes the propagation
through windows and confirms that most of signal leakage is
caused by locating the transmitter nearby windows. This model
is based on measurements at frequencies of 0.4 to 18 GHz.
The study in [10] also focused on the propagation through
windows or window-like structures and utilized FDTD, ray
tracing, and measurements results in the model development.
The authors in [11]–[13], reported various empirical path
loss models for the indoor-to-outdoor channel while con-
sidering receiver locations in surrounding streets. The work
reported in [14] compares the indoor, indoor-to-outdoor,
and indoor-outdoor-indoor propagation and provides statis-
tical modeling for the delay spread, coherence bandwidth,
and angle-of-arrival through extensive multiple-input multiple-
output (MIMO) channel measurements. Similar measurements
are reported in [15] with the focus on interference charac-
terization in residential areas. Indoor-to-outdoor path loss is
included in the WINNERII channel model [16] and in COST
Action 273 [17]. In [18], indoor-to-outdoor measurements
around buildings at frequencies of 837 MHz and 2.426 GHz
were performed to address the effects of femtocell unit place-
ments and other factors such as floor loss, and the height of
the transmitter inside the room. However, the analysis of the
transmitter locations effects is restricted.
The effects of architectural layouts of buildings are consid-
ered in [19], [20]. An algorithm namely; Building Procedural
Generation (BPG) [21] was developed to allow for realistic
floor plans generation. Based on BPG, a statistical model for
the indoor-to-outdoor signal propagation was obtained by cal-
culating the signal strength for random transmitter placements
inside the building and receiver placements in the outside.
Furthermore, the obtained model allows for optimizing the

locations of femtocell units inside the building.
In this work, and in contrast to the above mentioned
measurements investigations, we present a stochastic non-site-
specific indoor-to-outdoor propagation model to be utilized for
femtocell interference analysis regardless of the details of the
building that host the femtocell or the exact position of it.
The modeling idea stems from COST-231 MultiWall Model
(MWM) [22] and can be viewed as an improvement of the
model that is tailored to be used in femocell scenarios where
the architecture of the building is not available.
The rest of this paper is organized as follow: In section
II, a brief description of the COST-231 is presented. Section
III presents the measurements setup and the equipment used.
Section IV presents the indoor-to-outdoor measurements of
the received signal power and attenuation, and finally the
conclusion is given in Sec. V.
II. I
NDOOR-TO -OUTDOOR CHANNEL MODELS
Several models were developed in the literature to de-
scribe mixed indoor-outdoor signal propagation. The COST-
231 Multi Wall Model (MWM) [22, Eq. 4.7.2], which is a
semi-empirical model, was used in several 3GPP standards [1]
and in [19], [20]. Using this model, the received signal power
(P
r
) in decibels can be expressed as
P
r
= P
t
− 20 log

4πd
λ

− k

k
f
+2
k
f
+1
−0.46

f
α
f
−
W

i=1
k
i
α
i
+ ζ
(1)
where P
t
is the transmitted signal power, d is the transmitter-
receiver distance, λ is the carrier wavelength, k
f
is the number
of penetrated floors and α
f
is the attenuation factor for each
floor, k
i
is the number of penetrated walls of type i and α
i
is
the corresponding attenuation. The attenuation factors α
i
and
α
f
depend on the dimensions of the walls and floors as well
as their constituting materials. The power variation due to the
multipath fading is represented by the random variable ζ.
Typically, femtocells are deployed to serve users in residen-
tial buildings and hence, the FBS will be installed at the street
level. Thus, the second term in (1) can be eliminated. Then,
P
r
= P
t
− 20 log

4πd
λ


 
L
p
−

W
i=1
k
i
α
i
  
L
B
+ζ. (2)
The first term (L
p
) is the free space path loss (FSPL)
which depends on the transmitter-receiver distance, the second
term (L
B
) depends on the walls, doors, windows and other
obstacles in the building. Generally speaking, the network
operator can assume that the FBS is located close to the center
of the building. Therefore, estimating L
p
at any particular
point out of the building is straightforward. On the contrary,
computing L
B
requires the knowledge of W , k
1
, k
2
, ···,
k
W
, which are unknown to the network operator unless the
detailed floor plan is available. Moreover, the fading factor ζ
is random and can be accounted for only statistically. In the
absence of detailed floor plans to the network operator, which
is usually the case, the building parameters become random as
well. In such scenarios the building shadowing L
B
has to be
characterized statistically. Towards this goal, a measurement
campaign has been setup where a massive number of power
measurements was recorded and analyzed as detailed in the
following sections.
III. M
EASUREMENTS SETUP
The measurements campaign was held at a university cam-
pus that consists mainly of offices, laboratories, and class-
rooms. The building consists of two floors and has an area of
about 5,800 m
2
. The floor plan for the ground level, where
the measurements were conducted, is given in Fig. 1. As
depicted in the Figure, the architectural layout of the building
shows that the first layer of the building is composed of outer
rooms with different sizes, each of which has a wide 6m×2m
double glazed window. The 50-cm width external walls of the
outer room are made of enforced concrete. The second layer
of the building is a 2.3 meters wide corridor that separates
the outer rooms from the internal rooms. The corridor walls
are 30-cm width and made of concrete masonry units (CMU)
covered with an additional 1 cm of concrete. The rooms in
the building are separated using two types of walls, the first
type is similar to the corridors’ walls, and the second is made
of wood covered by plasterboards.
7UDQVPLWWHU
5HFHLYHU
2XWHU
5RRP
,QWHUQDO
:DOO
,QWHUQDO
:DOOV
,QWHUQDO
:DOOV
,QWHUQDO
:DOOV
,QWHUQDO
:DOOV
Fig. 1. The map of the campaign site with examples of the transmitter-
receiver pair placements [23].
A. Measurements Equipment
To take measurements, the Agilent N9310A RF signal
generator was used as the transmitter, whereas the Agilent
CXA-N9000A signal analyzer was used as the receiver. The
transmitted signal is an unmodulated 2 GHz carrier. Two Aa-
ronia OmniLOG 90200 omnidirectional broadband antennas
with operational frequency of 0.7 to 2.5 GHz were used. A
30-watt RF power amplifier (PA) from Mini-Circuits, namely
ZHL-30W-252, with operating frequencies in the range of 0.5
to 2.6 GHz, typical gain flatness of ±1.0 dB, and typical gain
of 49 dB at 2 GHz was also utilized. The PA is used to provide
sufficient transmission power when the measurements consider
large number of walls between the transmitter and the receiver.
The antennas were fixed onto a wooden post attached to a
trolley, so that the height of the antenna is 1.5 m above the
floor level. Fig. 2 shows an example of the transmitter setup

placement in the indoor environment (left) and the receiver
setup placement in the outdoor environment (right).
Fig. 2. Indoor transmitter and outdoor receiver setup.
To calibrate the readings for the losses of cables and connec-
tors, antennas gains and efficiencies, and the gain of the PA,
the reference power P
0
measured at distance d
0
in an outdoor
environment is evaluated and used. At a typical distance d
0
=1
m, P
0
was found to be -23.48 dB. All measurements were
collected relatively, where the power budget in (2) can be
rewritten as [24],
P
r
= P
0
− 20 log
10

d
d
0

− L
C
(3)
where L
C
 L
B
−ζ is the composite shadowing/fading factor.
B. Measurements Procedure
In femtocell applications, the FBS can be installed any-
where inside the building, hence its location is fixed at each
installation but random for different installations. Similarly, the
receiver will be located at a random position in the vicinity of
the building, and the measurement will be recorded while the
receiver is fixed as well. To capture the effect of the building
floor plan, random placements of the transmitter-receiver pair
were considered to include various scenarios and different
number of walls between the transmitter and the receiver.
Needless to say that, in each location, the process starts by
scanning the spectrum around 2 GHz to confirm that spectrum
is empty in the neighborhood of the desired frequency.
The next step in the process is to record the values of
P
r,
for a massive number of scenarios sufficient to build
the statistical model of the channel. In every realization, 200
power readings are recorded at a frequency of 1 Hz, and
then averaged to remove fast fading, i.e., any variations in
the channel due to the movement of people or vehicles. It is
worth noting that in all cases, the distance and the number
of penetrated walls were recorded. The number of penetrated
walls is used for initial categorization for the measurements
results. The distance values which varied between 2 m and 40
m are used to evaluate L
P
and hence compute L
C
. The total
number of realizations performed Λ=5, 500, which makes
the total number of samples recorded 1.1 × 10
6
.
IV. M
EASUREMENTS RESULTS
The received power values obtained by indoor-to-outdoor
measurements were used to evaluate the building shadow-
ing/fading factor L
C
, which models the loss caused by the
building’s walls and the multipath fading. However to have an
in-depth understanding of the collected results, we first group
the results based on the number of walls penetrated before the
signal arrives at the receiver. Therefore, the transmitter position
in such cases is conditionally random. Fig. 1 illustrates some
examples of the transmitter-receiver pair random placements.
These placements generate two main cases of interest based
on the number of walls that transmitted signal penetrates:
• Outer room: A room that has a window facing the outdoor
environment where the receiver is located, therefore k
1
=
1 and k
i
=0∀ i>1. The outer rooms in the building
where the measurements were conducted have wide glass
windows that cover a large portion the external wall.
• Inner rooms: Rooms that has one or more internal walls in
addition to the external wall. In this work we considered
internal rooms such that k
2
∈{1, 2, ..., 5}. The mea-
surement procedure applied for the internal rooms case
is similar to the one applied to the outer rooms.
The probability density function (PDF) of L
C
for k
1
=1
and k
2
=[0, 1, 2, 3, 5] were found to be best fitted into
Gaussian PDFs as shown in Fig. 3. The numerical properties
of the PDFs are given in Table I where m
L
C
, σ
2
L
C
, L
C min
,
L
C max
and Δ
L
C
are the mean, variance, minimum, maximum
and range of L
C
in dB, respectively. Note that most of
walls that belong to the k
2
=4case do not have the same
structure as the other internal walls, these walls are made of
plasterboards, and hence their attenuation is between 1 to 3
dB. Consequently, the results for the k
2
=4case are very
close to the k
2
=3case as depicted in Table I.
TABLE I
S
TATISTICAL PROPERTIES OF L
C
IN dB, k
1
=1.
k
2
m
L
C
σ
2
L
C
L
C min
L
C max
Δ
L
C
0 0.60 53 −23 23 46
1 16.3 71.5 −10 42 52
2 24.3 70.8 −1 50 51
3 33.9 72.5 8 60 52
4 35.5 74.1 9 61 52
5 47.6 74.9 21 74 53
−20 0 20 40 60 80
0
0.01
0.02
0.03
0.04
0.05
0.06
L
C
(dB)
Probability
k
2
=0
k
2
=1
k
2
=2
k
2
=3
k
2
=5
Fig. 3. The PDF of attenuation categorized by number of penetrated walls.

The first case of interest to be discussed is the outer
room case. The outer rooms’ walls are composed of ∼ 70%
windows and ∼ 30% concrete enforced with steel bars. Since
the transmit and receive antennas are much higher then the
concrete level, in all cases, a line-of-sight component through
windows exists and the attenuation will be mainly determined
by the glass attenuation, which is typically 3 to 10 dB [22].
The conditional average power at the receiver can be obtained
using (3) where
E {P
r
| [k
1
,k
2
]} 
¯
P
r
| [k
1
,k
2
]
= P
0
− 20 log
10

d
d
0

− L
B
| [k
1
,k
2
]+
¯
ζ
(4)
where E {ζ} 
¯
ζ. By defining L
B
| [k
1
=1,k
2
=0] >0
dB, and noting that all the parameters in (4) are known except
for
¯
ζ, then
¯
ζ =
¯
P
r
| [k
1
,k
2
] −P
0
+20log
10

d
d
0

− . (5)
The first three terms can be obtained from the measurement
results shown in Fig. 3, where
¯
P
r
| [k
1
=1,k
2
=0]−P
0
+
20 log
10

d
d
0

=0.6 and hence
¯
ζ =(0.6 − ) dB. Such result
is quite interesting because it shows that the average power of
the received signal is larger than the power of the line-of-sight
component. Based on Sinusoidal Addition Theorem (SAT)
[25], adding L sinusoids with random phases and amplitudes
h
L
=[h
1
, h
2
, ..., h
L
] gives a sinusoid which has an average
amplitude E {H
L
} > max{h}. The results in [8] and [24]
have demonstrated similar behavior. It is worth noting that
such results may not be obtained for outer rooms with smaller
windows because the wall attenuation is much larger than the
increase in the average power due to the multipath reflections.
And hence, the average power increase and external wall
attenuation will be lumped into a single parameter [8].
As it can be noted from Fig. 3 and Table I, adding one
internal wall (k
2
=1) shifts the mean of the conditional
PDF f
L
C
(l
C
|[k
1
=1, k
2
=1])to 16.4 dB and increases
the variance significantly. The attenuation for the k
2
=2case
is 24.3 dB while the variance has less than 1 dB difference
from the k
2
=1case. As it can be noted from these results,
the first significant attenuation was caused by first internal
wall because the external wall effect is very limited in our
case. The attenuations for k
2
=3, 4 and 5 are 33.9, 35.5 and
47.6 dB, respectively. As it can be noted from the results, the
attenuation increases nonlinearly as a function of the number
of walls, which is expected in multiwall scenarios [26].
The full set of power measurements obtained during this
work is presented in Fig. 4. The Figure also presents the power
attenuation assuming that the FSPL is the only cause for power
loss. The measurements collected for every room are used to
find the value of L
C
that minimizes the mean squared error
(MSE) with respect to (3). The values of L
C
that minimized
the MSE (
˜
L
C
) are −2.67, 12.26, 19.97, 29.2, 31.13 and 42.31
for k
2
=0, 1, ..., 5, respectively. It is interesting to note that
0 10 20 30 40
−120
−110
−100
−90
−80
−70
−60
−50
−40
−30
−20
Distance (m)
P
r
(dBm)
k
2
=0
k
2
=1
k
2
=2
k
2
=3
k
2
=5
FSPL
Fit
Fig. 4. Measured attenuation overlaid by the estimated attenuation by the
proposed model given the number of penetrated walls, k
1
=1.
the fitting results for the k
2
=0case shows that the received
signal power is 2.67 dB higher than the case when only FSPL
is involved, which is due to the existence of the line-of-sight-
signal in addition to a large number of reflected components.
A. Gaussian-Mixture Distribution for Modeling Indoor-to-
Outdoor Attenuation
The overall signal propagation model for the indoor-to-
outdoor channel can be obtained by removing the conditioning
on number of penetrated walls and combining all attenuation
measurements, which gives the results in Fig. 5. The previous
section established that the number of penetrated walls greatly
affects the observed attenuation. In fact, one can theorize that
the peaks in Fig. 5 are associated with number of traversed
walls. The hypothesis can be tested by analyzing the attenua-
tion histogram for specific numbers of traversed walls depict
in Fig. 3. Since the total attenuation is composed of different
path, or in other words, different processes, the Gaussian-
Mixture (GM) distribution is employed to model the effect.
GM models are well-known for their ability to model the
random processes that are composed of multiple underlying
independent random processes.
The histogram in Fig. 5 includes all the measurements.
Therefore, similar to any random effect that caused by several
individual random process, it is best modeled by a mixture dis-
tribution. In this case, a GM distribution is used to statistically
model L
B
. The PDF of a GM distribution is the weighted sum
of the PDF of M Gaussian distributions,
f
L
C
(l
C
)=
M

i=1
ω
i
N (l
C
|μ
i
,σ
i
) (6)
where N (l
C
|μ
i
,σ
i
) represents the PDF of a Gaussian random
variable with mean μ and standard deviation σ, and ω is
the weight of each Gaussian distribution

M
i=1
ω
i
=1 [19].
Expectation Maximization algorithm has been used to find
the proper coefficients. Akaike analysis [27] is employed

−20 0 40 60
0
0.005
0.01
0.015
0.02
0.025
20
$WWHQXDWLRQ

G%

3UREDELOLW\
Fig. 5. Building attenuation histogram and the proposed GM model.
to determine the minimum number of components that can
adequately obtain the GM model. The results show that the
best model have four components. Fig. 5 shows the fitted GM
model and the obtained parameters are listed in Table II.
TABLE II
GM
MODEL PARAMETERS BASED ON THE MEASUREMENTS.
i 1 2 3 4
μ
i
-8.5822 4.5985 21.6105 36.4812
σ
i
3.8245 6.5394 7.7956 8.9461
ω
i
0.0823 0.2418 0.3966 0.2793
V. C ONCLUSION
Exhaustive measurements have been performed to investi-
gate indoor-to-outdoor signal propagation. The attenuation is
divided to deterministic part, namely path loss, and a random
part which is referred to as building shadowing. Then, based
on the measurements, a Gaussian-Mixture model has been
proposed to model the building shadowing. Since the proposed
model considers the effect of the unknown architecture of
the building, it can be used in practical scenarios, where the
detailed plan of the building that hosts the femto basestation
is not known. It is expected that this model can be tuned
for other types of buildings with different architectures and
materials. The results presented here confirm the generalized
model obtained in the previous work reported in [19]. Hence,
the obtained model therein could be used for interference
studies of future femtocells deployments.
A
CKNOWLEDGMENT
The authors would like to acknowledge funding from Khal-
ifa University (KURIF 210007).
R
EFERENCES
[1] H. Thomas, “3G home NodeB study item technical report (release 9),”
3GPP technical specification group radio access network, Tech. Rep. TR
25.820, 2009.
[2] H. Claussen, L. T. W. Ho, and L. G. Samuel, “An overview of the
femtocell concept,” Bell Labs Tech. J., vol. 13, no. 1, pp. 221-245, 2008.
[3] M. Lillywhite and G. Liang, “Prediction of indoor and outdoor signal
strengths from indoor transmitter,” in Proc. IEEE Int. Symp. Antennas
and Propagation Soc., 2005, vol. 1B, pp. 661-664.
[4] A. Valcarce, D. Lopez-Perez, G. de la Roche, and J. Zhang, “Predicting
small-scale fading distributions with finite-difference methods in indoor-
to-outdoor scenarios,” in Proc. IEEE Vehicular Technology Conf. (VTC),
2009, pp. 1-5.
[5] G. de la Roche, A. Valcarce, and J. Zhang, “Hybrid model for indoor-to-
outdoor femtocell radio coverage prediction,” in Proc. IEEE Vehicular
Technology Conf., 2011, pp. 1-5.
[6] H. Nikookar and H. Hashemi, “Statistical modeling of signal ampli-
tude fading of indoor radio propagation channels,” in Proc. Int. Conf.
Universal Personal Communications, 1993, vol. 1, pp. 84-88.
[7] A. De Toledo and A. M. d. Turkmani, “Propagation into and within
buildings at 900, 1800 and 2300 MHz,” in IEEE Vehicular Technology
Conference, pp. 633-636, 1992.
[8] A. Valcarce and J. Zhang, “Empirical indoor-to-outdoor propagation
model for residential areas at 0.9-3.5 GHZ,” IEEE Antennas Wireless
Propag. Lett., vol. 9, pp. 682-685, 2010.
[9] A. Seville, S. Cirstea, and J. F. Taylor, “Effects of propagation between
the indoor and outdoor environment,” in Proc. Int. Conf. Antennas and
Propagation vol. 2, 2003, pp. 586-589.
[10] Z. Zhang, R. Sorensen, Z. Yun, M. Iskander, and J. Harvey, “A
ray-tracing approach for indoor/outdoor propagation through window
structures,” IEEE Trans. Antennas Propag., vol. 50, no. 5, pp. 742-748,
2002.
[11] M. Ho, S. Berber, and K. Sowerby, “Building leakage propagation
measurements for indoor cognitive radio systems,” Electronics Letters,
vol. 48, no. 23, pp. 1508-1510, 2012.
[12] S. Celik, Y. Yoruk, M. Bitirgan, O. Kurnaz, B. Basyigit, S. Helhel,
and S. Ozen, “Indoor to outdoor propagation model improvement for
GSM900/GSM1800/CDMA-2100,” in General Assembly and Scientific
Symp., 2011, pp. 1-4.
[13] Y. Corre, J. Stephan, and Y. Lostanlen, “Indoor-to-outdoor path-loss
models for femtocell predictions,” in Proc. IEEE Int. Symp. Personal
Indoor and Mobile Radio Communications (PIMRC), 2011, pp. 824-
828.
[14] C. Fang, B. Allen, E. Liu, P. Karadimas, J. Zhang, A. Glazunov, and F.
Tufvesson, “Indoor-indoor and indoor-outdoor propagation trial results
at 2.6 GHz,” in Proc. Antennas and Propagation Conf. (LAPC), 2012,
pp. 1-4.
[15] X. Gao, A. Alayon Glazunov, J. Weng, C. Fang, J. Zhang, and F. Tufves-
son, “Channel measurement and characterization of interference between
residential femto-cell systems,” in Proc. European Conf. Antennas and
Propagation (EUCAP), 2011, pp. 3769-3773.
[16] P. Ky
¨
osti, et al., “WINNER II channel models part II radio channel
measurement and analysis results,” Tech. Rep., 2007.
[17] L. M. Correia, Mobile Broadband Multimedia Networks. Techniques,
Models and Tools for 4G. Academic Press, 2006.
[18] D. Rose, T. Jansen, and T. Kurner, “Indoor to outdoor propagation -
Measuring and modeling of femto cells in LTE networks at 800 and
2600 MHz,” in IEEE Global Communication (GLOBECOM) Workshops,
2011, pp. 203-207.
[19] M. Mirahmadi, A. Shami, and A. Al-Dweik, “A building architecture
model for predicting femtocell interference in next-generation networks,”
in Proc. IEEE Int. Conf. Communication (ICC), 2012, pp. 5059-5063.
[20] Mirahmadi, M.; Al-Dweik, A.; Shami, A., ”Interference Modeling and
Performance Evaluation of Heterogeneous Cellular Networks,” Commu-
nications, IEEE Transactions on , vol.62, no.6, pp.2132,2144, June 2014
[21] M. Mirahmadi and A. Shami, “A novel algorithm for real-time pro-
cedural generation of building floor plans,” CoRR, vol. abs/1211.5842,
2012.
[22] “Digital mobile radio towards future generation systems: COST Action
231,” European Cooperation in the field of Scientific and Technical
Research Tech. Rep., 1999.
[23] Sharjah Training College Ground Floor Plan. Etisalat, 2010.
[24] C. Oestges et al., “Experimental Characterization and Modeling of
Outdoor-to-Indoor and Indoor-to-Indoor Distributed Channels,” IEEE
Trans. Veh. Technol., vol. 59, no. 5, pp. 2253-2265, Jun. 2010.
[25] Y. Maghsoodi and A. Al-Dweik, “Exact formulae for error-rate analysis
of synchronous FHSS networks using envelope characteristic functions
of sums of stochastic signals,” IEEE Trans. Veh. Technol., vol. 57, pp.
974-985, Mar. 2008.
[26] A. Durantini, and D. Cassioli, ”A multi-wall path loss model for indoor
UWB propagation,” in Proc. IEEE Vehicular Technology Conf. (VTC),
2005, pp. 30-34.
[27] G. Claeskens, and N. L. Hjort, Model Selection and Model Averaging,
Cambridge: Cambridge University Press, 2008.