TRANSACTIONS ON EMERGING TELECOMMUNICATIONS TECHNOLOGIES

Trans. Emerging Tel. Tech. (2015)

Published online in Wiley Online Library (wileyonlinelibrary.com). DOI: 10.1002/ett.2957

RESEARCH ARTICLE

On the dimensioning of LTE and

LTE-advanced networks
Marwane Ben Hcine* and Ridha Bouallegue
Innovation of Communicant and Cooperative Mobiles Laboratory, INNOV´COM, Higher School of Communication, Sup´COM,
University of Carthage, Ariana, Tunisia

ABSTRACT
Network design consists of evaluating cell coverage and capacity and may involve many parameters related to environment,
system configuration and quality of service requirements. The key parameter for cost optimised Long Term Evolution
(LTE)-based network dimensioning is to provide a tight approximation of the effective signal-to-interference-plus-noise
ratio (SINR) distribution. This paper provides a novel framework for LTE and LTE-advanced network dimensioning, where
the classical single-carrier SINR performance evaluation is extended to multi-carriers systems operating over frequency
selective channels. Extension is achieved by expressing the link outage probability in terms of the statistics of the effective
SINR. For effective SINR computation, the exponential effective SINR mapping technique is used. Closed-form expression
for the link outage probability is achieved assuming a log skew normal approximation for single-carrier case. Then we
rely on the log-normal approximation to express the exponential effective SINR distribution as a function of the mean and
standard deviation of the SINR of a generic subcarrier. Effective SINR values are evaluated at click speed in each point on
the network. Simulations show that the proposed framework provides results with accuracy within 0.5 dB for any defined
outage probability target. Copyright © 2015 John Wiley & Sons, Ltd.

*Correspondence
M. Ben Hcine, Innovation of Communicant and Cooperative Mobiles Laboratory, Higher School of Communication, University of
Carthage, Ariana, Tunisia.
E-mail: [email protected]

Received 13 January 2015; Revised 16 April 2015; Accepted 26 May 2015

1. INTRODUCTION scheme (MCS)] in order to cater for the coverage and

quality requirement. Network design relies essentially on
LTE, a broadband wireless access technology, is based the signal-to-interference-plus-noise ratio (SINR) distribu-
on orthogonal frequency division multiple access as the tion. On the one hand, radio coverage is indeed defined
multiple access scheme on the downlink transmission. In as the probability that SINR at cell edge is above a given
orthogonal frequency division multiple access, the sys- threshold. On the other hand, capacity and traffic studies in
tem bandwidth is subdivided into subcarriers, which are LTE-based networks rely on the MCS probabilities, which
grouped to form subchannels units called resource blocks can be directly derived from the effective SINR cumulative
(RBs). An RB is, in turn, the elementary radio resource distribution function (CDF). Thus, the key parameter for
that can be allocated to a user. Each RB consists of 12 con- cost optimised LTE-based network planning is to provide a
secutive subcarriers, or 180 kHz, for the duration of one tight approximation of effective SINR distribution for each
slot (0.5 ms). point in the network.
Recently, LTE has gained considerable attention from Planning procedures for multicarrier systems such as
the telecommunication industry because of its capability to LTE/LTE-advanced are usually based on block error
provide faster mobile broadband services. Thus, it opens rate curves as a function of the effective SINR. Block
the door to the ultra-fast 4 G network generation. As error rate curves are obtained after simulating the entire
LTE networks are being deployed, the need arises for transmitter–receiver chain, including multipath channel
cellular operators to obtain efficient dimensioning tools and coder/decoder subsystems (link level simulation). Tra-
for LTE-based networks. The dimensioning exercise is to ditional methodologies for effective SINR distribution
find out the required number of evolved nodes B (eNB) evaluation in LTE networks rely on time-consuming sys-
and to identify the system optimal configuration [system tem level simulations. These methodologies are usually
bandwidth, operating frequency, modulation and coding not appropriate for practical purposes such as network

Copyright © 2015 John Wiley & Sons, Ltd.

 M. Ben Hcine and R. Bouallegue

dimensioning. In this paper, we substitute the resource- 1.2. Contribution of the current work
greedy system level simulations by an analytical frame-
work for effective SINR distribution derivation. First, we The contributions of this paper are summarised as follows.
provide a tight approximation of the SINR CDF on a (i) A simple formula for outage probability computation
generic subcarrier at any distance from the eNB, while in the presence of correlated log-normal interferes is pro-
taking into account the joint effect of path loss, shadowing posed. The proposed formula is based on log skew normal
and fast fading. Then, we extend this approach to multi- approximation to the sum of correlated log-normal RVs.
carriers system. Effective SINR distribution can thus be An analytical method is provided to derive log skew nor-
obtained at a click speed with very good accuracy. The mal distribution parameters. Simulations show that the
exponential effective SINR mapping (EESM) is considered proposed formula provides exact evaluation of outage
on this work. The validity of the EESM method to predict probability in the presence of correlated log-normal
the multi-carriers link error performance, under a variety interferes. (ii) An analytical model that considers shad-
of operating conditions, has been proved in several papers owing and fast fading effects is developed. SINR distri-
[1–3]. It has been shown that the EESM provides accuracy bution for single-carrier case is accurately approximated
within few tenths of a decibel, even when interference is considering real scenario deployment case, which takes
included [3]. cross-correlation matrix into consideration. (iii) An exten-
sion to multi-carriers case is proposed in order to provide
a tight approximation for the effective SINR distribution
1.1. Related work
on LTE networks based on log-normal distribution. Accu-
racy of approximation is within 0.5 dB. The proposed
The main objective of network dimensioning is to guar-
analytical framework allows us to analyse instantaneously
antee a minimum value of interference at cell edge while
the impact of various parameters involved in the dimen-
maintaining an acceptable overall cell throughput. In the
sioning of LTE-based networks. This includes environment
literature, several techniques have been proposed to miti-
parameters (shadowing, fast fading and path-loss expo-
gate interference and increase throughput especially at cell
nent), network and system parameters (cell range, eNBs
edge. In [4], a load balancing technique has been pro-
density and number of allocated RBs) and quality of
posed in order to reduce inter-cell interference. In [5],
service parameters (outage probabilities).
authors propose a demand-based spectrum orthogonalisa-
tion scheme for interference avoidance in LTE-advanced
networks. In [6], an expression of the SINR distribution 1.3. Outline of the present paper
conditionally on the average received powers has been
derived. Authors rely on Monte Carlo simulations at cell This paper is organised as follows. In Section 2, sys-
edge to obtain the symbol error rate. In [7], authors propose tem model and parameters description are given. Inter-
a simple planning procedure based on the evaluation of two ference expressions based on this model are elaborated.
averages of the SINR. Effective SINR is approximated by In Section 3, we study the SINR distribution for single-
a Gaussian random variable (RV). This approach is limited carrier case and give a closed-form expression for outage
by the resources-consuming simulation needed to com- probability, taken into account the shadowing and the fast
pute the involved SINR averages. A statistical link budget fading effects, for a user equipment (UE) located at any dis-
model for LTE networks has been proposed in [8]; how- tance from its serving eNB. Then, we extend our approach
ever, the system throughput performances have not been for multi-carriers case using EESM technique. Assuming
studied. In [9], authors propose a semi-analytical method log-normal approximation to the effective SINR distribu-
for effective SINR modelling based on the generalised tion, a closed-form expression for outage probability in
extreme value distribution. They rely on a time-consuming multi-carriers case is given. In Section 4, we validate our
Monte Carlo simulation for the estimation of distribution approach and compare our analytical results to Monte
parameters. A paper close in its objectives and its demarche Carlo simulation results. An example of dimensioning
to our work is in [10]. Authors assume a log-normal involving a streaming service is given in Section 5: studies
approximation to SINR distribution for a single-carrier of the outage capacity and selection of adequate channel
case while taking into account the shadowing effect. Then quality index (CQI) at cell edge are performed with our
they jointly take into account the shadowing and the fast formulas. At last, Section 6 concludes the paper.
fading effects to derive a simple outage probability for-
mula. They extend their approach to multi-carriers case 2. SYSTEM MODEL
using the capacity effective SINR mapping (CESM). How-
ever, it has been shown in [1] that CESM model does We consider a homogeneous hexagonal network made of N
not provide an acceptable accuracy when modelling effec- eNBs with three sectors per site. Figure 1 shows an exam-
tive SINR in LTE networks. As alternative to CESM, the ple of such a network with the main parameters involved
EESM technique is used in this work. EESM has been in the study: R, the cell radius (500 m), and 3 R, the inter-
validated in [2, 3] as an accurate approach for effective site distance. A reuse factor equal to 1 is assumed: all
SINR modelling. cells use the whole system bandwidth. The UE u is located

Trans. Emerging Tel. Tech. (2015) © 2015 John Wiley & Sons, Ltd.
DOI: 10.1002/ett
 M. Ben Hcine and R. Bouallegue

Figure 1. Network layout [11].

using polar coordinates .r,  /. On the downlink, the power  A.i,u /

S Pi Kri,u .10 10 .Yi,u Xi,u
received by UE u can be described by the three-stage prop- SINR D D
agation model that considers path loss, shadowing effect I C Nth P
3N1

A.j,u /
Pj Krj,u .10 10 .Yj,u Xj,u C Nth
and fast fading. Let Pi be the transmission power of cell i, jD0,j¤i
the power Pi,u received by UE u can thus be written as (4)

 A.i,u / Nth represent the thermal noise power. In urban areas,

Pi,u D Pi Kri,u .10 10 .Yi,u Xi,u (1)
thermal noise can be neglected with respect to inter-cell
The path-loss model is characterised by parameters K and interference. Furthermore, we assume that all eNBs have
 A.i,u / identical transmitting power. As we focus on a single UE,
 > 2. The term Pi Kri,u .10 10 is the mean value of we may drop the indexes i, u and set ri,u D r, Yi,u D
the received power at distance ri from the cell i. Shadow- Y0 , Xi,u D X0 , i,u D  , Yj,u D Yj , Xj,u D Xj and j,u D j .
i,u
ing effect is represented by log-normal RV Yi,u D 10 10 , The SINR perceived by the UE u can be written in the
where i,u is a normal RV, with zero mean and standard following way:
deviation  typically ranging from 3 to 12 dB. Rayleigh
A./
fast fading effects are represented by the RV Xi,u with r .10 10 .Y0 .X0
a probability density function pX .x/ D ex . In the rest SINR D (5)
P  A.j /
3N1
of this paper, we assume that fYi,u giD1:::3N , fXi,u giD1:::3N rj .10 10 .Yj .Xj
are statistically independent. We define inter , intra as the jD1

shadowing correlation factors between sites and between

sectors of the same site. i,u is the horizontal angle between 3. SINR DISTRIBUTION
the UE and the middle of antenna beam of cell i, i AND OUTAGE
PROBABILITY CALCULATION
i,u D   i (2)

The macro BS antenna radiation pattern used in each sector 3.1. Single-carrier case
is defined by
The outage probability is defined as the probability for the
"  2 # SINR  to be lower than a given threshold value ı

A. / D  min 12 , Am where 180 6  6 180
3dB 0 1
(3)
B A./ C
B r .10 10 .Y0 .X0 C
3dB is the 3-dB beam width, which corresponds to 65ı , P. < ı/ D P B
B 3N1 < ı C
C (6)
and Am D 20 dB is the maximum attenuation. The SINR @ P  A. j/ A
rj .10 10 .Yj .Xj
at user u is thus given by jD1

Trans. Emerging Tel. Tech. (2015) © 2015 John Wiley & Sons, Ltd.
DOI: 10.1002/ett
 M. Ben Hcine and R. Bouallegue

˚ 
Assuming that RVs Xj j>1 are equal to their average be approximated by a log skew normal distribution
  SN. ƒ , !ƒ , "ƒ /, where ƒ is defined as the solution the
value E Xj D 1, we may write
following nonlinear equation:
0 1
P  A.j /
3N1 0 1
B r j .10 10 .Yj
C
B jD1 C P
3N1  2  @2 r  A
P. < ı/ D P B
BX 0 < ı A./
C (7)
C
2
e2i ei ei 1 r 1C2
P
P
Bi,j
@ r .10 10 .Y0 A iD1 Bi,j 16i,j63N1
!2 D e 16i,j63N1 0 1 1
P
3N1 2

ei ei =2 2 2@r P A
Introducing the RV, iD1 Bi,j
16i,j63N1
P
3N1

A.j /
(14)
rj .10 10 .Yj
jD1 Such nonlinear equation can be solved using different
Zf D (8)
A./ mathematical utilities (e.g. fsolve in MATLAB). Accord-
r .10 10 .Y0
ing to Benhcine and Bouallegue [12], we can use upper
The outage probability is now expressed as tail slopes match of both distributions to find out an opti-
C1 mal starting guess solution 0 to (14) in order to converge
Z x  rapidly (only few iterations are needed), where
P. < ı/ D P < Zf px .x/dx (9)
ı v2 13
0 u
u X
u4
Zf is a location dependent factor. The numerator is the sum 0 D t Max fB.i, i/g
2 Bi,j A5  1 (15)
i
of log-normal RVs, which can be approximated by a log 16i,j63N1
skew normal RV [12].
For the sake of simplicity, we define ƒ as Optimal location and scale parameters "ƒ , !ƒ are obtained
according to ƒ
3N1
X A.j /
3N1
X
ƒD

rj .10 .Yj D ƒj (10) v
10 u
u 1 C 2ƒ
jD1 jD1 !ƒ D u t P (16)
  Bi,j
 16i,j63N1
so that ƒj is a log-normal RV with mean j D log rj C
a.A.j / and variance j D a. , where a D ln.10/ 10 . !
!
 3N1
X 2
ƒ D .ƒ0 , ƒ1 : : : ƒ3N1 / is a strictly positive random i i2 =2 !ƒ
! "ƒ D ln e e 
vector such that the vector L D .L0 , L1 : : : L3N1 / with 2
iD1
Lj D log.ƒj / has an n-dimensional normal distribution 0 0 11
(17)
with mean vector  D .0 , 2 : : : 3N1 / and covariance B B ƒ CC
!
  ln B B
@ @r P CC
AA
matrix M with M.i, j/ D Cov.Li , Lj /, 0 6 i, j 6 3N  1, ƒ Bi,j
is called an n-dimensional log-normal vector with param- 16i,j63N1
eters E and M. Let R be the shadowing correlation matrix.
ƒ
We have On the one hand, as Zf D ƒ0 , COV.ƒ, ƒ0 / may be
8 expressed as
ˆ
<1 if i D j p p
COV.ƒ, ƒ0 / D Corr.ƒ, ƒ0 /. Var.ƒ0 /. Var.ƒ/
R.i, j/ D intra if cell i, cell j belong to the same eNB q
:̂ 
inter if cell i, cell j belong to different eNBs
2
D Corr.ƒ, ƒ0 / e0  1
(11) q (18)
2
. 2.e! .2ˇ!/  2 2 .ˇ!//
The covariance matrix is computed by Balakrishnan and
Lai [13] (Eq 11.71) 1 2 
. exp " C 0 C ! C 02
( ) 2
r  2 
ei  1 ej  1 C 1
2
M.i, j/ D ln R.i, j/. (12) On the other hand, using the moment generating function
of the bivariate skew normal distribution, defined in [14],
Cov.ƒi , ƒj / may be expressed as COV.ƒ, ƒ0 / may be computed by

 
i Cj C 12 i2 Cj2
COV.ƒ, ƒ0 / D E Œƒƒ0  E Œƒ E Œƒ
Cov.ƒi , ƒj / D e .eM.i,j/  1/ (13)
D 2'.ˇ!/ .exp. !0 /  1/
(19)
Let B D M 1 be the inverse of the covariance matrix. 1 2 
exp " C 0 C ! C 02
According to Benhcine and Bouallegue [12], ƒ can 2

Trans. Emerging Tel. Tech. (2015) © 2015 John Wiley & Sons, Ltd.
DOI: 10.1002/ett
 M. Ben Hcine and R. Bouallegue

where is the correlation factor between ln.ƒ/ and A fast and accurate calculation of Owen T function used
ln.ƒ0 /, which can be expressed as in our simulation is provided in [15].
2 s  3
Note that if we consider only shadowing effect, the
p 
2 !2
e0 1. 2 e ƒ '.2ˇƒ !ƒ /2' 2 .ˇƒ !ƒ / single-carrier outage probability can be written as
6 7
ln 4Corr.ƒ, ƒ0 /. C 15
2'.ˇƒ !ƒ / 2   3
D
ln ı1  "f
0 ! ƒ P. < ı/ D Qf 4 5 (32)
(20) !f

and ˇƒ D q ƒ .
1C2ƒ
3.2. Effective SINR distribution
We adopt the multivariate extension for skew normal distri-
bution defined in [14]. As a quotient of two dependent log The exponential effective SINR [1] is expressed as
skew normal RVs, Zf has a log skew normal distribution
LSN. f , !f , "f / (Appendix), where M  !
1 X SINRn
SINReff D ˇ ln exp  (33)
M ˇ
!ƒ .˛2 C ˛1 /  0 .˛1 C ˛2 / nD1
f D q (21)
2 C .1  2 /.˛ ! C ˛  /2
!ƒ 1 ƒ 2 0 where ˇ is an optimisation factor that depends only on the
q MCS. M is the number of allocated subcarriers.
2  2 !  C 2
!f D !ƒ ƒ 0 (22) We assume that SINReff expressed in (33) can be approx-
0
imated by a log-normal RV. Such assumption seems to
"f D "ƒ  0 (23) be accurate for small number of allocated subcarriers, and
and it remains that way as M increases. The validity of this

! assumption is proved in the following by simulation.
  T ‰ 1 1
!
 ˛1 In order to compute the mean SINReff and the standard
˛ D D q (24)
˛2 ! ! deviation SINReff , we define the RV n as
1 C T ‰ 1

! ! SINRn

‰ D 1  1  T
(25) n D e ˇ (34)
0q 1
 
1  ı12 0 We compute E Œn and E 2n using the classical for-
D@ q A (26)
0 1  ı22 mula for each positive RV X and p 2 N  : E ŒX p D
C1
R
0 1 pxp1 P.X > x/dx
0 0
ıE D @ q ƒ A (27)
1C2ƒ C1
Z Z1
  E Œn D .Rf . , t//et d dt (35)
!
 0
D (28) 0 0
ƒ
  C1
1 h i Z Z1
D (29)
1 2
E n D 2 Rf . , t/et d dt (36)
Thus, the single-carrier SINR CDF for a UE located at a 0 0
"   #
distance r from its serving eNB, taking into account the t
ln  ˇln./ "f
joint effect of shadowing and fast fading can be written, where Rf . , t/ D Qf !f . We define now the
using (9), as
RV  as
C1
Z " #
x
 "f M  
P. < ı/ D Qf
ln ı
ex dx (30)
SINReff
1 X SINRn
De ˇ D exp  (37)
!f M ˇ
0 nD1

where Qf .x/ D 1  .x/ C 2T.x, f / and .x/ is the stan- We have

dard normal CDF.  D E Œn (38)
T.x, / is the Owen T function expressed as
n o 1  h 2i 
2 D E n  E Œn 2 (39)
Z exp  12 x2 .1 C t2 / M
1
T.x, / D dt (31) Using the Taylor expansion, we may compute the first and
2 .1 C t2 /
0 second central moments of SINReff as

Trans. Emerging Tel. Tech. (2015) © 2015 John Wiley & Sons, Ltd.
DOI: 10.1002/ett
 M. Ben Hcine and R. Bouallegue

ˇ Finally, as SINReff is approximated by a log-normal RV,

mSINReff D E Œˇ ln./  ˇ ln. / C 2 (40) we have
22
0 1
m2
ˇ2 ˇ2 B SINReff C
D2SINReff D Var Œˇ ln./  2 C 4 (41) SINReff D ln @ q A (42)
2 44 2 2
DSINReff C mSINReff

Table I. Simulation assumptions [11]. v !

u
u D2SINReff
Parameter Assumption
SINReff D tln C 1 (43)
m2SINReff
Environment Macro cell, urban area
Carrier frequency 2000 MHz
Cellular layout Hexagonal grid, 19 cell sites
The multi-carriers outage probability for a UE at distance
57 sectors with 65ı
r from its serving eNB can be expressed as
sectored beam !
BTS antenna height 30 m ln.ı/  SINReff
P.SINReff < ı/ D (44)
Inter-site distance 1500 m SINReff
Path-loss model 128.1 C 37.6 log10 .r /
C 21 log10 .fc =2/
where .x/ is the standard normal CDF.
log-normal fade shadow 10 dB
The mean and standard deviation of the effective SINR
Shadowing correlation Between cells: 0.5
depend mainly on the location of UE, the optimisation
Between sectors: 1.0
factor ˇ and propagation model parameters  and  .

Figure 2. Single-carrier outage probability for a UE located at different positions with (left) and without fast fading (right).

Figure 3. Multi-carriers outage probability for 4-QAM modulation, CQIindex D 5 and NRB D 5 with (left) and without fast fading (right).

Trans. Emerging Tel. Tech. (2015) © 2015 John Wiley & Sons, Ltd.
DOI: 10.1002/ett
 M. Ben Hcine and R. Bouallegue

4. VALIDATION 5. APPLICATION TO THE

DIMENSIONING OF A
Simulations were run to validate our approximations for STREAMING SERVICE
single-carrier and multi-carriers SINR distributions. Typi-
cal results with two tiers of interfering cells (N D 19) are 5.1. Outage capacity
presented. Simulation assumptions are given in Table I.
To verify our approximations for the single-carrier case, A first simple application of our framework is the deter-
Figure 2 shows the outage probability for a UE located at mination of the outage capacity. Subchannel capacity is
different positions with and without fast fading. It is obvi- defined as
ous that log skew normal assumption provides accurate
Cch D NRB WRB log2 .1 C SINReff / (45)
approximation to the single-carrier SINR distribution. The
difference between analysis and simulation results is less
where WRB D 180 KHz is the RB bandwidth in LTE
than few tenths of decibel for all cases. We note that our
and NRB is the number of allocated RBs.
framework provides exactly the simulation results if we
For example, Figure 5 shows that 98 per cent of the
consider only shadowing effect.
users reach more than 4 Mbps at distance d D 500 m from
Multi-carriers SINR performance evaluation is cru-
the eNB with modulation 16-QAM and CQIindex D 9 if
cial for LTE dimensioning process because scheduling
seven RBs are allocated to the user. Figure 6 shows the out-
and MCS reporting are carried out per RB. We use
the calibrated ˇ values for CQI from [1]. In order to
study the impact of involved parameters (i.e. number of
RBs, CQIindex ) on the accuracy of our approximations,
Figure 3 shows the multi-carriers outage probability for a
UE located at different positions, for 4-Quadrature Ampli-
tude Modulation (QAM) modulation and CQIindex D 5
assuming that five RBs (i.e. 60 subcarriers) are allocated
to UE. Accuracy of results gets better as the distance
to eNB decreases. We have checked our framework with
other CQIs and number of allocated RBs, the accuracy of
results remains the same. The proposed framework pro-
vides exactly the simulation results if we consider only the
shadowing effect.
It is possible to use our analytical framework to compute
several link budget margins such as fast fading margin,
shadowing margin and interference margin. For example,
Figure 4 illustrates the fast fading margin (2 dB) for a
UE located inside the cell .r,  / D .R, 0/ with 4-QAM Figure 5. Outage capacity (2 per cent) as a function of UE
modulation, CQIindex D 3 and NRB D 3. coordinates with NRB D 7 RBs.

Figure 6. Outage capacity (2 per cent) as a function of UE coor-

Figure 4. Fast fading margin for 4-QAM modulation, CQIindex D dinates (a) 4-QAM, CQI-5; (b) 16-QAM, CQI-9; and (c) 64-QAM,
3 and NRB D 3 with (left) and without fast fading (right). CQI-13 with NRB D 7 RBs.

Trans. Emerging Tel. Tech. (2015) © 2015 John Wiley & Sons, Ltd.
DOI: 10.1002/ett
 M. Ben Hcine and R. Bouallegue

edge, based on our formulas. Figure 7 shows the needed

steps to provide a list of couples .CQIindex , NRB / satisfying
operator requirements at cell edge. The choice is based,
then, on the available resources (number of RBs, channel
bandwidth). It is worthy to note that this approach provides
only an assessment for the number of required RBs and not
a strict value.

6. CONCLUSION

In this paper, we propose an analytical framework for

LTE-based networks dimensioning. We adopt the EESM
technique to model the effective SINR. Our analytical
framework provides an easy-to-use formula for outage
probability and effective SINR evaluation, while consider-
ing the joint effect of path loss, shadowing and fast fading.
Assuming a log skew normal approximation to the sum
of log-normal RVs, we rely on the log-normal approxima-
tion to express the exponential effective SINR distribution
as a function of the mean and standard deviation of the
SINR of a single subcarrier. The proposed framework
allows us to obtain performances results instantly, which is
very useful in the dimensioning process because no more
Figure 7. Flow chart for selection of accepted .CQIindex , NRB /. time-consuming system level simulation is required.

age capacity (2 per cent) as a function of UE coordinates

APPENDIX: DIFFERENCE OF TWO
.r,  / for 4-QAM, CQI-5, 16-QAM, CQI-9 and 64-QAM,
CORRELATED SKEW NORMAL RVS
CQI-13 with NRB D 5 RBs. With this kind of curve, the
operator is able to know at each position which capac-
Given two skew normal dependent RVs .X, Y/  SN2 .‰,
ity can be guaranteed given a target outage probability. !

˛ / as defined in [14], that is, the joint distribution of X and
This result is particularly needed for streaming services,
Y is given by
for which the average capacity is not sufficient in the
dimensioning process. q ( "
2 2
1 .x  1 /2
f .x, y/ D 1   exp 
5.2. Resource blocks allocation 21 2 2.1  2 / 12
#)
2 .x  1 / .y  2 / .y  2 /2
In LTE link budget process, required SINR at cell  C ,
edge depends heavily on the selected combination of 1 2 22
.CQIindex , NRB / in order to reach the targeted quality .x  1 / .y  2 /
of service. ˆ ˛1 C ˛2
1 2
Let .D, Pout / be the downlink cell edge throughput
and outage probability at cell edge, defined by operator.
The probability density function (pdf) of the difference
In practical system, The UE sends CQI feedback as an
Z D Y  X is given by
indication of the data rate, which can be supported by
the downlink channel. This allows the eNB to select the C1
Z
appropriate MCS for downlink transmission. The number
of allocated RBs is computed, then, based on the transport 'z .z/ D f .x, z C x/dx
block size table [16]. 1
For the dimensioning purpose, the selected CQI at cell C1
Z ( "
edge is chosen by the network planner. The selection of 1 1 .x  1 /2
D p exp 
MCS is a trade-off between coverage and resource utili- 1 2 1  2 2.1  2 / 12
1
sation. The more robust the selected MCS (e.g 0-QPSK), #)
the lower the allowed required SINR at cell edge. But on 2 .x  1 / .z C x  2 / .z C x  2 /2
the same time, the higher the resource consumption, which  C ,
1 2 22
leaves less resources for the rest of scheduled users.
We propose here an alternative approach to select the .x  1 / .z C x  2 /
ˆ ˛1 C ˛2 dx
optimal CQI and estimate the number of needed RBs at cell 1 2

Trans. Emerging Tel. Tech. (2015) © 2015 John Wiley & Sons, Ltd.
DOI: 10.1002/ett
 M. Ben Hcine and R. Bouallegue

Let t D x 1
1 , for more convenience, we define  D 2 
and
1 , so that
C1
Z
1
C1
Z A2 D p exp
1 1 2 .˛1 2 C ˛2 1 / 1  2
'z .z/ D p exp  1
2 1   2 2.1  2 / (  2 )
1
"     #) 1 ˇ
z   C 1 t z   C 1 t 2  x  .z  /
2
t  2t C 2.1  2 /.˛1 2 C ˛2 1 /2 
2 2  
  x
z   C 1 t erf p dx
ˆ ˛1 t C ˛2 dt 2
2
h i h i  
Let y D px , we have
Let x D ˛1 C ˛2 12 t C ˛2 z
2 , after some algebra, 2
we may write
n o Z
C1
exp  .z/
2
Z
C1 1
2 2 A2 D p p exp
'z .z/ D p 2 .˛1 2 C ˛2 1 / 1 2
 .˛1 2 C ˛2 1 / 1  2 1
1 (  2)
( 2 ) p
1 ˇ
1 ˇ  2 y  .z  / erf .y/dy
exp  x .z  / 2.1  2 /.˛1 2 C ˛2 1 /2
2.1  2 /.˛1 2 C ˛2 1 /2
ˆ.x/dx Using Equation 13 from [17],
where
q C1
Z p !
n o  b
D 12  21 2 C 22 exp .ax C b/2 erf .x/dx D  erf p
a a2 C 1
ˇ D 2 .˛2 C ˛1 /  1 .˛1 C ˛2 / 1

  
Given the fact that ˆ.x/ D 1
1 C erf px , we We may write
2 2
!
may write 1 ˇ .z  /
A2 D p erf p p
( ) 2 2 2 C .1  2 /.˛ 
1 2 C ˛2 1 /2
.z  /2   
'z .z/ D .A1 C A2 / exp  1 ˇ .z  /
2 2 D p 2ˆ p
2
1
2 C .1  2 /.˛ 
1 2 C ˛2 1 /2

where so that

C1
Z p " !#
1 2 ˇ.z  /
A1 D p 'z .z/ D p 1ˆ p
2 .˛1 2 C ˛2 1 / 1  2    2 C .1  2 /.˛1 2 C ˛2 1 /2
1 ( )
(  2)
1 ˇ .z  /2
exp  x .z/ dx exp 
2.1  2 /.˛1 2 C ˛2 1 /2  2 2
C1
Z
1 Let Fz .z/ denote the CDF of the difference Z D Y  X,
D p
2 .˛1 2 C ˛2 1 / 1  2 we have
1
1 p ( )
exp  y2 dy; Zz Zz
2.1   /.˛1 2 C ˛2 1 /2
2 2 .x  /2
Fz .z/ D 'z .x/dx D p exp  dx
ˇ  2 2
y D x  .z  / 1 1
 !
Zz p
C1
Z 2 ˇ.x  /
1 1  p ˆ p
D exp  t2 dt;    2 C .1  2 /.˛1 2 C ˛2 1 /2
2 2 1
1 ( )
1 .x  /2
tD p y exp  dx
.1  2 /.˛1 2 C ˛2 1 / 2 2
1
D p .x/
2 Let y D , we may write

Trans. Emerging Tel. Tech. (2015) © 2015 John Wiley & Sons, Ltd.
DOI: 10.1002/ett
 M. Ben Hcine and R. Bouallegue

.z/ .z/ and scheduling in LTE. IEEE Transactions on Wireless

Z  Z 
1 1 2 Communications 2011; 10(10): 3436–3448.
Fz .z/ D 2 p exp  y2 dy  p
2 2 2 3. Hanzaz Z, Schotten HD. Analysis of effective SINR
1 1
! mapping models for MIMO OFDM in LTE system.
ˇy In The 9th International Wireless Communications and
ˆ p
2
C .1  2 /.˛1 2 C ˛2 1 /2 Mobile Computing Conference, IWCMC’13, Sardinia,
1 z   Italy, 2013; 1509–1515.
exp  y2 dy D 2ˆ
2  4. Li Z, Wang H, Pan Z, Liu N, You X. Heterogenous
!
z ˇ QoS guaranteed load balancing in 3GPP LTE multi-
 Sk ,p cell fractional frequency reuse network. Transactions
  2 C .1  2 /.˛1 2 C ˛2 1 /2
z   z   on Emerging Telecommunications Technologies 2014;
D 2ˆ ˆ 25(12): 1169–1183.
  !
z ˇ 5. Siti M, Maharum M, Fisal N, Lo A, Ghafar ASA,
C 2T ,p Saparudin FA, Katiran N. A demand-based spectrum
  2 C .1  2 /.˛1 2 C ˛2 1 /2
orthogonalisation scheme for interference avoidance in
LTE advanced heterogeneous networks. Transactions on
where T.x, / is the Owen T function expressed as
Emerging Telecommunications Technologies 2014. DOI:
n o 10.1002/ett.2860.
Z exp  1 x2 .1 C t2 /
1 2 6. Seol C, Cheun K. A statistical inter-cell interfer-
T.x, / D dt
2 .1 C t2 / ence model for downlink cellular OFDMA networks
0
under log-normal shadowing and multipath Rayleigh
We then have fading. IEEE Transactions on Communications 2009;
z   57(10): 1011–1013.
Fz .z/ D ˆ
 7. Giuliano R, Mazzenga F. Exponential effective SINR
!
z ˇ approximations for OFDM/OFDMA-based cellular sys-
 2T ,p tem planning. IEEE Transactions on Wireless Communi-
  2 C .1  2 /.˛1 2 C ˛2 1 /2
! cations 2009; 8(9): 4434–4439.
z ˇ 8. Jaber M, Akl N, Dawy Z, Yaacoub E. Statistical link
D SN ,p
  2 C .1  2 /.˛1 2 C ˛2 1 /2 budget analysis approach for LTE cellular network
dimensioning. In European Wireless, Barcelona, Spain,
where 2014; 1–6.
q 9. Maqbool M, Coupechoux M, Godlewski P. A semi-
D 12  21 2 C 22 analytical method to model effective SINR spatial distri-
ˇ D 2 .˛2 C ˛1 /  1 .˛1 C ˛2 / bution in WiMAX networks. In Proceedings of the 32nd
 D 2  1 International Conference on Sarnoff Symposium. IEEE
Press, USA, 2009; 531-535.
10. Kelif JM, Coupechoux M, Godlewski P. On the dimen-
ACKNOWLEDGEMENT
sioning of cellular OFDMA networks. Physical Commu-
This work was supported in part by the Laboratory nications 2012; 5(1): 10–21.
Innov´COM of Higher School of Telecommunication, Uni- 11. 3GPP, TR 36.942. Evolved universal terrestrial
versity of Carthage, Tunisia. radio access (E-UTRA); radio frequency (RF) system
scenarios.
12. Benhcine M, Bouallegue. R. Highly accurate log skew
REFERENCES normal approximation to the sum of correlated log-
normals. In The Proceedings of the 6th International
1. Hanzaz Z, Schotten HD. Performance evaluation of link
to system interface for long term evolution system. Conference on Networks & Communications (NeTCoM
In The 7th International Wireless Communications and 2014), Chennai, India, 2014; 41–52.
Mobile Computing Conference, IWCMC’11, Turkey, 13. Balakrishnan N, Lai CD. Continuous Bivariate Distribu-
2011; 2168–2173. tions. Springer: New York, USA.
2. Donthi SN, Mehta NB. An accurate model for EESM 14. Azzalini A, Dalla Valle A. The multivariate skew-normal
and its application to analysis of CQI feedback schemes distribution. Biometrika 1996; 83: 715–726.

Trans. Emerging Tel. Tech. (2015) © 2015 John Wiley & Sons, Ltd.
DOI: 10.1002/ett
 M. Ben Hcine and R. Bouallegue

15. Patefield M. Fast and accurate calculation of Owen’s t 17. Ng EW, Geller M. A table of integrals of the error
function. Journal of Statistics 2000; 5(5): 1–25. functions. Journal of Research of the National Bureau
16. 3GPP TS 36.213. Evolved universal terrestrial radio of Standards, B. Mathematical Sciences 1969; 73B (1):
access (E-UTRA); physical layer procedures (release 10). 1–20.

Trans. Emerging Tel. Tech. (2015) © 2015 John Wiley & Sons, Ltd.
DOI: 10.1002/ett