2015 36th IEEE Sarnoff Symposium

Radio Resource Dimensioning in a Centralized

Ad-Hoc Maritime MIMO LTE Network
Lina Mroueh∗ , Achraf Kessab†∗§ , Philippe Martins§ , Serge Hethuin† and Emmanuelle Vivier∗
∗
Institut Supérieur d’Electronique de Paris, 75006 Paris
†
Thales Communications and Security, 92230 Gennevilliers
§ Télécom ParisTech, 75013 Paris

{lmroueh, evivier}@isep.fr, {achraf.kessab, serge.hethuin}@thalesgroup.com and martins@telecom-paristech.fr

Abstract—This paper provides a statistical model for radio re- scenario is assimilated to the uplink of LTE network with
source dimensioning within one cluster of a centralized maritime evolved embedded UEs.
LTE Mobile Ad Hoc Network (MANET). The main objective is Our goal is to determine the required bandwidth to be
to compare, for different antenna configurations and different
transmission modes, the bandwidth required to be allocated to allocated at the CH to handle the traffic of all active CNs.
the Cluster Head (CH) to serve all connected Cluster Nodes (CN). This dimensioning problem is studied for two antenna config-
For this purpose, we consider the average statistical behavior of urations of the CN → CH transmission: (i)- the Single Input
the network and we assume that the active nodes are randomly Single Output (SISO); (ii)- the 2×2 MIMO configuration with
distributed according to a Poisson Point Process. We base our sufficiently separated antennas. The 2×2 MIMO configuration
dimensioning results on the computation of an upper-bound
on the resource outage probability. Numerical results based is compliant with the LTE standard [1] and is studied for two
on Monte Carlo simulations validate the analytical model and transmission modes (TMs): (1)- the first one extracts the full
show the efficiency of the MIMO diversity mode in the radio diversity gain of the MIMO using an Alamouti scheme and (2)-
dimensioning for long range maritime communications. the second multiplexes different streams on the same RB using
a Successive Interference Cancellation (SIC) decoder. Unlike
I. I NTRODUCTION
the cellular network UEs, the considered CNs are evolved
In this paper, we consider a maritime Mobile Ad-hoc transmitters and advanced open loop TMs can be defined.
Network (MANET) where nodes are equipped with Long Term The average statistical behavior of the network is con-
Evolution (LTE) [1] radio devices and have one or two anten- sidered: active nodes arriving to a given area are randomly
nas. These nodes correspond to ships and form a navy fleet distributed around the CH according to a random Poisson
that is led by a shipmaster. In the MANET terminology [2] Point Process. The randomness of the wireless channel is
(and references therein), this navy fleet is commonly known as considered as a mark of the Poisson position and results
cluster, the ship-staff as Cluster Node (CN) and the shipmaster from stochastic geometry using marked Poisson Point Process
as Cluster Head (CH). We assume that the maritime MANET will be invoked. These tools were investigated in [3]–[5] to
is centralized meaning that communications are made through compute an upper-bound on the resource outage probability in
the shipmaster that will be in charge of the radio resource a cellular network considering random Point Poisson Process
management in its own fleet. marked by the random shadowing. The main contribution of
One of the main challenge of Radio Resource Management this paper is to study the impact of the MIMO configuration
(RRM) at the CH is to optimize the allocated bandwidth in and the above mentioned TMs (SISO, transmit diversity and
such a way to serve all CNs while minimizing the probability spatial multiplexing) in the radio dimensioning of a maritime
of the event that arises when the CH has no more radio ad hoc LTE network. For each TM, we derive an upper-bound
resources to attribute to active CNs. In the rest of this paper, on the resource outage probability using the Bennett concen-
this event will be denoted by radio resource outage event. tration inequality. Based on this upper-bound, we determine
The radio dimensioning problem depends critically on the the number of RBs required by the network in such a way that
network parameters including the network load, the antenna the occurrence of the radio resource outage does not exceed a
configuration, and the required Quality of Service (QoS). The given threshold. The average spectral efficiency in the ad hoc
radio resources considered in this paper correspond to the network is then deduced.
LTE radio Resource Blocks (RBs) that consist in a 0.5 ms The rest of the paper is organized as follows. We start
slot duration within a bandwidth of W = 180 KHz. The first, by defining in Section II, the system model and the
total bandwidth is divided into several RBs. Each RB can be maritime path loss model. Then, we review in Section III some
allocated to a single CN depending on its availability and the mathematical background from stochastic geometry required
quality of the CN → CH radio link. In the considered military for further derivation. The mean number of required RBs
centralized network, all the communications are assumed to and the resource outage probability are derived in Section IV
transit via the CH. We consider here the scenario in which for different MIMO configurations and numerical results are
several CNs communicate simultaneously with the CH. This presented in Section V. Finally, SectionVI concludes the paper.

978-1-4673-8042-3/15/$31.00 ©2015 IEEE 128

 2015 36th IEEE Sarnoff Symposium

II. S YSTEM MODEL C. Network Parameters

A. Assumptions The cluster nodes characteristics and the propagation pa-
We consider a circular cluster C of radius R = 20 km rameters are summarized in Table I.
belonging to a centralized MANET with limited number of TABLE I
clusters, in which all communications transit via the CH S YSTEM PARAMETERS
situated without loss of generality at the center of this circular
Nodes Parameters
cluster. We assume that the nodes are equipped with 1 or 2
Carrier frequency 800 MHz
transmit and receive antennas with omnidirectional gains. Let
CN Tx power per RB Pt = 23 dBm;
Φnodes be the pattern of active nodes in the cluster and x CN and CH antenna gain G (dBi) = 0 (omni)
be the position of a given CN in the cluster. This network is Tx and Rx height ht = hr = 12 m
studied under the following assumptions: Antenna configuration nt = 1, 2 and nr = 1, 2
- Assumption 1: The surface density of inter-arrival time Propagation Parameters
in min−1 km−2 is defined by ρ. Free Space Path Loss KFS = 10−9 , γFS = 2
- Assumption 2: For all nodes, the mean service time in Maritime Path Loss Km = 10−8.1 , γm = 3.9
min is exponentially distributed with mean ν −1 . Rayleigh parameter σf2 = 1
- Assumption 3: The active nodes are randomly distributed Threshold SNR βmin = −3.28 dB
according to an homogeneous Poisson Point Process
(PPP) of intensity Λ(dx) = λdx with λ = ρν −1 Tx/km2 .
- Assumption 4: The channel gain depends on three III. R ADIO R ESOURCE O UTAGE P ROBABILITY:
DEFINITION AND STATISTICAL TOOLS
components: the path loss, the Rayleigh random fading
attenuation and the log-normal shadowing attenuation. In this section, we start by defining the radio resource outage
We neglect the shadowing effect in a maritime context. probability and we review from [3] the use of the Bennett
The fading coefficients between transmitting antennas are concentration inequality in the computation of an upper-bound
assumed to be Rayleigh distributed due to NLOS trans- on the resource outage probability. We finally show how to
mission induced by long range communication distances. make use of this upper-bound to find the adequate bandwidth
- Assumption 5: The MANET is noise limited: Due to be allocated at the CH and the average total achieved
to the centralized cluster communication, there is no- spectral efficiency in the network.
intra cluster interference. Indeed, the CHs coordinate the A. Radio resource outage probability
choice of the frequency reuse pattern to guarantee that the
Let x the position of a CN in the circular cluster of radius R
inter-cluster interference induced by the remote clusters
and y be the random fading coefficient of the path between the
is negligible compared to the thermal noise power.
CN and the CH. The set of active CNs in the ad hoc network
- Assumption 6: The CN→CH communication is estab-
is then represented by a Marked Point Poisson Process,
lished if the SNR is higher than a threshold βmin .
Φ̃ = {X̃ ∈ C × R+ : X̃ = (x, y)}
B. Maritime propagation path loss model
The adopted maritime propagation path loss is based on the with intensity Λ(dx) ⊗ py (y)dy = λ p(y) dx dy. For a single
ITU recommendation [6] and is derived for a frequency carrier class of traffic with target nodes rate C0 , let NRB (x, y) denote
fc = 800 MHz and for antenna transmit and receive heights the random number of required RBs by a given node situated
ht = hr = 12 m. Using the interpolated path loss from [6], at position x ∈ C and subject to a random fading y, such that
the linear path-loss PL(d) = Kd−γ is,
 
C0
NRB (x, y) = (1)
W × CShannon (x, y)
PL(d) = PLFS 1{d≤dc,0 } + PLm 1{dc,0 ≤d≤R}
where W = 180 kHz is the bandwidth of one RB and
where PLFS = KFS d−γFS is the free space path loss, dc,0 is CShannon (x, y) is the Shannon spectral efficiency that depends
the critical distance, defined in [7] in function of the transmit on the antenna configuration. The distribution of the fading
and receiver heights ht and hr , and the wavelength λ as, depends also on the considered antenna configuration and will
dc,0 = 4hλt hr = 1.54 km. and PLm (d) = Km d−γm is the be further detailed in Section IV. According to Assumption 6,
maritime path loss propagation given by, N (x, y) ≤ Nmax with:
 
PLm (d) = Km,0 d−γm,0 1{dc,0 ≤d≤dc,1 } Nmax =
C0
. (2)
+ Km,1 d−γm,1 1{dc,1 ≤d≤dc,2 } . W log2 (1 + βmin)
The number of RBs required by a CN situated at a position x
with dc,1 = 5 km and dc,2 = R = 20 km. Finally, the atten-
and subject to a fading y is then,
uation factors are KFS = 0.89 × 10−9 , Km,0 = 1.28 × 10−9 , X 
Km,1 = 6 × 10−9 and the distance exponents are such that NRB (x, y) = min j 1{j−1<NRB (x,y)≤j} , Nmax . (3)
γFS = 2, γm,0 = 2.9 and γm,1 = 3.9. j∈N∗

129
 2015 36th IEEE Sarnoff Symposium

The total number of required RBs by the active CNs in Φ̃ is, with Z
NRB,t (Φ̃) =
X
NRB (X̃). (4) Aj = Prob{NRB (x, y) ≤ j}dx (9)
C
X̃∈Φ̃
being the area of the region in which nodes require at most j
The radio resource outage event occurs when the total number RBs. In a similar way,
of RBs required by all active CNs exceeds the number of
available RBs at the CH Nnet , i.e. X−1
Nmax

n o vN = λ j 2 (Aj − Aj−1 ) + λNmax

2
(πR2 − ANmax −1 ).
Pout (Nnet ) = Prob NRB,t (Φ̃) > Nnet . (5) j=1
(10)
Our main objective is to find the optimal number of required Note that this result can be easily generalized to the case of
RBs Nnet at the CH to guarantee that the outage probability multiple class of nodes requiring different values of target rate.
does not exceed a given threshold. The Bennett concentration inequality applies also in this case
and the values of mN and vN are determined by summing the
B. Outage upper-bound and concentration inequality
values relative to each CN class.
The Bennett’s concentration inequality used in [4] is an
efficient statistical tool to find an upper-bound on the sum C. Dimensioning and average total spectral efficiency
of independent randomly distributed bounded variables. This For an outage threshold pth , the required number of RBs in
is the case so far of the radio resource outage probability the network Nnet that guarantees that the outage probability is
defined in (5) where the total number of required RBs is less than the outage threshold Pout (Nnet ) ≤ pth is computed
nothing but the sum of independent random bounded variables using the upper-bound in (6) as Nnet = α mN where
that depends on the CN position and the random fading. In  2

the following theorem, we recall from [8] this concentration vN −1 Nmax
α=1+ g − log(pth ) . (11)
inequality. mN Nmax vN
Theorem 1 (Bennett’s concentration inequality, [8]): Let According to the value of Nnet , the required bandwidth B is
Xi be a bounded independent
PN random variable PXi ≤ a with chosen to fit in with the closest normalized LTE bandwidth
N 2
an average mx = i=1 E(X i ) and let v x , i=1 E(Xi ). (1.4, 3, 5, 10, 15 or 20 MHz) or with the aggregated carrier
For α ≥ 1, the PNcomplementary cumulative distribution bandwidth. The total average achieved rate in the network is,
function of X = i=1 Xi is such that hX
   C̄t = E C0 1{NRB (X̃)≤Nmax }
vx (α − 1)mx a
Prob{X > αmx } ≤ exp − 2 g (6) X̃∈Φ̃
a vx X i
+ Cj 1{(j−1)<NRB (X̃)≤j}
with g(t) = (1 + t) log(1 + t) − t.
j≥Nmax +1
The upper-bound Psup (Nnet ) on the resource outage probabil-
ity Pout (Nnet ) in (5) for Nnet = αmN is computed using the with Cj being the achieved rate over Nmax RBs, provided
concentration inequality as that C0 requires j RBs to be achieved and hence Cj satisfies
   j Cj = C0 Nmax . Applying the property of the marked PPP
vN (α − 1)mN Nmax
Psup (Nnet ) = exp − 2 g in (7) gives
Nmax vN
X
(Aj − Aj−1 ) Cj (12)
hP i
where mN = C̄t = λANmax C0 + λ
E x̃∈Φ̃ NRB (x, y) and vN =
hP i j≥Nmax +1
2
E x̃∈Φ̃ NRB (x, y) . For a marked point process, the where Aj is the area of the region in which CNs require at
expectation of a given function f (X̃) is, most j RBs to achieve C0 as defined in (9). This total rate
hX i Z is transmitted within the bandwidth B and the corresponding
E f (X̃) = f (x, y) p(y) dy λdx (7) average total spectral efficiency of the network in (b/s/Hz) is
C×R+
X̃∈Φ̃ obtained as S = C̄t /B.
Applying this property to the first moment,
IV. R ADIO R ESOURCE O UTAGE P ROBABILITY WITH
X−1
Nmax Z D IFFERENT A NTENNA C ONFIGURATIONS
mN = j Prob{j − 1 < NRB (x, y) ≤ j}λdx
C
In this section, we derive an upper-bound on the resource
j=1
Z outage probability for different antennas configurations. We
+ Nmax Prob{NRB (x, y) > Nmax − 1}λdx. (8) consider first the SISO case. Then we consider the 2 × 2
C MIMO configuration studied for two transmission modes: the
Consequently, first transmission extracts the full diversity of the 2 × 2 MIMO
X−1
Nmax scheme; the second one multiplexes two different streams on
mN = λ j(Aj − Aj−1 ) + λNmax (πR2 − ANmax −1 ) the same radio resource. These antennas configurations are
j=1 compliant with the LTE standard [1].

130
 2015 36th IEEE Sarnoff Symposium

A. SISO communication where

For a SISO communication, the spectral efficiency in (1) is   2  
1 1 γFS 2
computed as, ξFS = Γ ,
  γFS aFS γFS
Pt K  γ 2  
CSISO (x, y) = log2 1 + y 1 1 m,i 2
η|x|γ ξm,i = Γ .
γm,i am,i γm,i
where η is the thermal noise; y is the random fading with
cumulative distribution function, The values of mN and vN for the SISO scheme can be
− 12 y
therefore computed using (8) and (10) and consequently the
σ
F (y) = 1 − e f upper-bound can be derived from (6).
and σf2 is the fading parameter. B. MIMO communication with diversity gain
C0
Let us define β˜j = PtηK β1j and βj = 2 jW − 1. The area of In this case, the CH and the CNs have 2 transmit and receive
the region Aj in (9) in which nodes require at most j RBs is, antennas each. For a 2 × 2 MIMO communication, the LTE
Z n |x|γ o devices extract the full diversity using an Alamouti code [9].
SISO
Aj = Prob y ≥ dx In this case, symbols are decoded over an equivalent fading y
C β̃j that is distributed according to a chi-squared random variable
− 21 |x|γ
Z
= e σf β̃j dx with a CDF,
C
!
1 1
Using the propagation maritime path loss, F (y) = 2 Γinc y, 4 . (14)
Z Z σf σf2
−aFS |x|γFS γm
ASISO
j = e dx + e−am |x| dx Using the Alamouti code for the 2 × 2 MIMO transmission,
Cdc,0 C/Cdc,0
the spectral efficiency in (1) is,
where aFS = σ12 β̃ 1 = σ2 Pηt KFS βj ; am,i = σ12 β̃ 1 =  
f j,FS f f j,m,i d Pt K
η
β and C = {x ∈ C : |x| ≤ d } and C/C dc,0 =
CMIMO (x, y) = log2 1 + y .
σf2 Pt Km,i j dc,0 c,0 2η|x|γ
{x ∈ C and x ∈ / Cdc,0 }. C0
Using polar coordinates (r, θ), the expression of ASISO j is Let β̃j,d = P2η
tK 1
βj and βj = 2
jW − 1. Using (9), the area of

Z 2π Z dc,0 the region in which nodes require at most j RBs is:

γFS
ASISO
j = re−aFS r drdθ Z n |x|γ o
0 0 AMIMO,d
j = Prob y ≥ dx
1 Z 2π Z dc,i+1 C β̃j,d
X γm,i
re−am,i r
!
+ drdθ. Z
|x|γ
i=0 0 dc,i = (1 − F )dx
C β̃j,d
Let uFS = aFS rγFS and um,i = am,i rγm,i , then !
1 |x|γ
Z
γFS 1
 2 Z = (1 − 2 Γinc , 4 )dx.
" #
1 γFS aFS dc,0 γFS

SISO 1 2
−1 −u
σf σf2 β̃j,d
Aj = 2π uFS e du C
γFS aFS 0
"  γ 2 Z am,i dγm,i # Similar to the SISO case, this area can be decomposed with
1  2
X 1 1 m,i c,i+1
γm,i −1 −u respect to the path loss model as,
+ 2π um,i e du .
i=0
γm,i am,i γm,i
am,i dc,i !
1 |x|γFS
Z
MIMO,d 1
Using the definitions of the gamma and the incomplete gamma Aj = (1 − 2 Γinc , 4 )dx
Cdc,0 σf σf2 β̃j,d,FS
functions: !
Z ∞
1 |x|γm
Z
1
Γ(q) = e−t tq−1 dt, + (1 − 2 Γinc , 4 )dx.
0 C/Cdc,0 σf σf2 β̃j,d,m
Z x
1
Γinc (x, q) = e−t tq−1 dt, Finally, the area AMIMO,d is
Γ(q) 0 j

the area Aj can be written as, dc,0

!!
1 rγFS
Z
1
AMIMO,d
j = 2π r 1 − 2 Γinc ,4 dr
σf2 β̃j,d,FS
  
SISO γFS 2 0 σf
Aj = 2π ξFS Γinc aFS dc,0 ,
γFS 1 Z dc,i+1
!!
X 1 1 rγm,i
1  
2
 + 2π r 1− Γinc ,4 dr
γm,i σf2 σf2 β̃j,d,m,i
X
+ 2π ξm,i Γinc am,i dc,i+1 , i=0 dc,i
i=0
γm,i (15)
 
γm,i 2 and it can be numerically computed using the Riemann ap-
− ξm,i Γinc am,i dc,i , (13)
γm,i proximation sum. Once again, the values of mN and vN for

131
 2015 36th IEEE Sarnoff Symposium

this MIMO scheme can be deduced by replacing AMIMO,d

j in NRB = {50, 65, 75, 90 and 100}. These numbers of RBs
(8) and (10) and consequently the upper-bound can be derived correspond in LTE to respective values of bandwidth and
from (6). aggregated bandwidth (denoted in the following by [CC, CA]
where CC is the carrier component and CA is the aggregated
C. MIMO communication with multiplexing gain one) such that B = {10, [10, 3], 15, [15, 3] and 20} MHz.
In this case, the 2 × 2 MIMO antenna configuration is used Using Monte Carlo simulation with 106 iterations1 , we com-
to multiplex on each resource element of the radio RB two data pute for the different antenna configurations and transmission
streams. The total rate C0 is sub-divided into two streams that modes, the mean number of RBs required by this network and
are spatially multiplexed over the two transmitting antennas. the resource outage probability Pout for the above mentioned
The channel state information is only available at the receiver bandwidth values. The log-ratio test
of the CN and not at the CH transmitter. The CN uses a  
Psup
Successive Interference Cancellation (SIC) decoder to cancel ∆ = log10 (17)
the inter-antenna interference. The first stream is first decoded Pout
by nulling out the second stream. The second stream is then compares the analytical and the numerical results.
decoded by subtracting the contribution of the first stream from For the SISO antenna configuration, the numerical and
the observed signal. Two degrees of freedom are left to decode analytical mean number of RBs is mN,SISO = 47 and
the second stream compared to only one degree for the first the numerical and analytical value of vN,SISO = 76. The
stream. The equivalent fading y1 on which the first stream analytical and numerical outage probabilities corresponding
is decoded is an exponential random variable. Whereas, the to the different values of RBs are given in Table II. It can be
second one y2 is a chi-squared random variable with 4 degrees noticed that the resource outage probability decreases when
of freedom. The required number of RBs by each stream s is more RBs are available at the CH.
then,
  TABLE II
(s) C0 /2 S INGLE CLASS WITH SISO SCHEME
NRB (x, ys ) = , 1≤s≤2
W Cs (x, ys )
B (MHz) 10 [10,3] 15 [15,3] 20
The required number of RBs by both stream is dictated by the NRB 50 65 75 90 100
worst fading and α 1.06 1.39 1.6 2 2.12

(s) (s)
 Psup 0.94 0.22 0.04 0.001 1 · 10−4
NRB (x, y) = min NRB (x, y1 ), NRB (x, y2 ) Pout 0.3 0.02 0.0016 1.7 · 10−5 0
∆ 0.5 1.04 1.39 1.9 ×
where y = min(y1 , y2 ) = y1 and is exponentially distributed.
After some straightforward derivations similar to the SISO
C0
case and by replacing βj by βj,m = 2 2jW − 1 and β̃j by Table III contains the resource outage probability 2 × 2
β̃j,m = P2η
tK 1 MIMO configuration with diversity gain. In this case, the
βj,m , we obtain:
mean number of RBs computed analytically and numerically
 
2
 are mdN,MIMO = 38 and vN,MIMOd
= 38. It can be noticed
AMIMO,m
j
γFS
= 2π ξFS Γinc aFS dc , that the resource outage probability is considerably reduced
γFS
1    using this diversity transmission mode compared to the SISO
X γm,i 2 configuration.
+ 2π ξm,i Γinc am,i dc,i+1 ,
i=0
γm,i
  TABLE III
γm,i 2 S INGLE CLASS WITH 2 × 2 MIMO SCHEME AND DIVERSITY GAIN
− ξm,i Γinc am dc,i , . (16)
γm,i
B (MHz) 10 [10,3] 15 [15,3] 20
Based on this, the values of mN and vN in (8) and (10) and NRB 50 65 75 90 100
consequently the upper-bound in (6) are deduced. α 1.32 1.71 1.97 2.36 2.63
Psup 0.29 0.009 0.0004 2.5·10−6 5.3·10−8
V. N UMERICAL RESULTS
Pout 0.0235 1.7·10−5 0 0 0
The system parameters are defined in Table I of Section II. ∆ 1.09 2.72 × × ×
Using the expression of the area Aj in (13), (15) and (16), we
find the values of mN and vN in (8) and (10) for a target rate
Table IV contains the resource outage probability 2 ×
of C0 = 600 kb/s and a network load defined by parameters
2 MIMO configuration with multiplexing gain. The mean
of inter-arrival surface density ρ = 10−3 min−1 km−2 and
number of RBs computed analytically and numerically is
mean service time ν −1 = 30 min. Then, we deduce the
mm m
N,MIMO = 43 and vN,MIMO = 64. As it can be seen, there
outage probability upper-bound in (6) that is denoted by Psup .
is a slight improvement in terms of resource outage probability
Using the threshold SNR value of −3.28 dB, the number
of RBs is limited to 6 RBs per node. The resource outage 1 The limited number of iterations gives a zero-outage probability for values
probability is computed for different values of total RBs less than 10−6 .

132
 2015 36th IEEE Sarnoff Symposium

compared to SISO case. However, the diversity transmission tively 15 MHz with 75 RBs for the MIMO transmit diversity,
outperforms in term of outage probability the multiplexing and finally an aggregated [15, 3] bandwidth with 75 + 15 = 90
one. Actually, the number of RBs with spatial multiplexing RBs for the MIMO multiplexing case. As it can be observed
is dictated by the stream that is decoded on the worst fading from Table VI, for the same target outage threshold, the MIMO
which is weaker than the chi-squared random variable in the diversity optimizes the required number of RBs at the CH.
Alamouti case. The average spectral efficiency that can be achieved with a
TABLE IV network resource outage threshold of at most 10−3 is obtained
S INGLE CLASS WITH 2 × 2 MIMO SCHEME AND MULTIPLEXING GAIN from (12). For the SISO case, the total rate is approximatively
22.4 Mb/s transmitted within a bandwidth of 20 MHz which
B (MHz) 10 [10,3] 15 [15,3] 20 corresponds to a spectral efficiency of 1.1 b/s/Hz. For the
NRB 50 65 75 90 100 MIMO case with transmit diversity, the total rate is 22.8 Mb/s
α 1.16 1.51 1.74 2.09 2.32 transmitted in a 15 MHz bandwidth and the spectral efficiency
Psup 0.73 0.088 0.011 2.3·10−4 1.19·10−5 is 1.52 b/s/Hz. Finally, for the MIMO multiplexing the total
Pout 0.17 5.5·10−3 2.5·10−4 1·10−6 0 rate is 22.5 Mb/s corresponding to a spectral efficiency of 1.25
∆ 0.63 1.2 1.63 2.3 × b/s/Hz.

We evaluate for the case of multi-class services in which VI. C ONCLUSION

nodes target three different rates C1 = 600 kb/s, C2 = 800
kb/s and C3 = 1 Mb/s. We consider the same surface density In this paper, the dimensioning problem in one cluster
of inter-arrival time ρ1 = ρ2 = 1 · 10−3 min−1 km−2 and of a centralized MANET was studied. We assumed that
ρ3 = 3 · 10−3 and a mean service time ν1−1 = 10 min, all communications transit via the Cluster Head node. We
ν2−1 = 10 min and ν3−1 = 2.5 min. The analytical and compared for different antenna configurations and different
numerical average number of required RBs for this scenario transmission modes, the radio resource outage probability
is 49 and the numerical and analytical value of vN is 90. that occurs when the number of required resources by the
The resource outage probabilities are evaluated in Table V for Cluster Nodes exceeds the one available at the Cluster Head.
the case of SISO system. Compared to the single class case, Using this statistical model, we showed that for the long
the network is more loaded than the single class SISO case range maritime communication, the 2×2 MIMO configuration
which explains the higher resource outage probability that is with a full diversity transmission mode improves not only
obtained. the resource outage probability, but also the average network
spectral efficiency compared to the SISO case and MIMO
TABLE V
M ULTI CLASS SISO SCHEME spatial multiplexing modes.

B (MHz) 10 [10,3] 15 [15,3] 20 R EFERENCES

NRB 50 65 75 90 100
α 1.02 1.33 1.54 1.84 2.05 [1] 3GPP TS 36.213 Evolved Universal Terrestrial Radio Access (E-UTRA),
R.11, 2013
Psup 0.99 0.33 0.08 0.004 4·10−4 [2] Yu, J.Y.; Chong, P.H.J., ”A survey of clustering schemes for mobile ad
Pout 0.4 0.05 0.06 0.0001 6·10−6 hoc networks,” Communications Surveys & Tutorials, IEEE , vol.7, no.1,
∆ 0.38 0.84 1.13 1.66 1.89 pp.32,48, First Qtr. 2005
[3] Decreusefond L., Ferraz E. and Martins P., ”Upper Bound of Loss
probability for the dimensioning of OFDMA systems with multi class
We finally provide in Table VI for the above mentioned randomly located users”, WiOpt, workshop SPASWIN 2009, Seoul, South
network context with a single class of QoS (C0 = 600 kb/s) Korea, June 2009.
[4] Decreusefond, L., Ferraz, E., Martins, P., Vu, T.T., “Robust methods
the adequate number of RBs Nnet that guarantees that the re- for LTE and WiMAX dimensioning,” 6th International Conference on
source outage probability Pout occurs in less than pth = 0.1%. Performance Evaluation Methodologies and Tools, pp.74,82, 9-12 Oct.
For each antenna configuration, the value of Nnet = αmN is 2012
[5] Kaddour, F.Z.; Martins, P.; Decreusefond, L, Vivier, E.; Mroueh,
chosen above the mean value by a scaling factor α computed L.,“Outage probability upper bound for multi-class QoS in the uplink of
using (11). LTE networks,” Global Communications Conference (GLOBECOM), 2013
IEEE , vol., no., pp.4129,4134, 9-13 Dec. 2013
TABLE VI [6] ITU-R P.1546-2, ”Method for point-to-area predictions for terrestrial
D IMENSIONING WITH pTH = 0.1% services in the frequency range 30 MHz to 3 000 MHz”.
[7] Kun Yang, Ekman, T., Roste, T., Bekkadal, F., ”A quasi-deterministic path
Mode SISO MIMOd MIMOm loss propagation model for the open sea environment,” 14th International
Symposium on Wireless Personal Multimedia Communications (WPMC),
α 1.94 1.9 1.98 pp.1,5, 3-7 Oct. 2011
Nnet 92 73 85 [8] Bennett, G. (1962). ”Probability Inequalities for the Sum of Independent
B (MHz) 20 15 [15,3] Random Variables”. Journal of the American Statistical Association 57
(297): 3345.
Pout 3·10−6 0 1·10−6
[9] Alamouti, S., ”A simple transmit diversity technique for wireless commu-
Based on this Nnet number, the corresponding LTE band- nications”, IEEE Journal on Selected Areas in Communications, , vol.16,
no.8, pp.1451,1458, Oct 1998
width is 20 MHz containing 100 RBs for the SISO, respec-

133