Você está na página 1de 13

This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2016.2541172, IEEE
Transactions on Vehicular Technology

Call Blocking Probability and Effective Throughput


for Call Admission Control of CoMP Joint
Transmission
Seung-Yeon Kim and Choong-Ho Cho

Abstract—Coordinated multi-point transmission (CoMP) with


the joint transmission (JT) scheme is employed in cellular
networks to improve the data rate, the cell-edge throughput PRB a
and/or to increase the system throughput. In CoMP-JT, since the
user equipment (UE) is capable of receiving the desired signal UE 2 PRB b
from an adjacent cell as well as from its serving cell, it can achieve UE 1
an improved signal to interference ratio. However, this scheme Center zone
UE 3 PRB c
UE 4
entails exhaustion of resources in serving base station (BS). In
this paper, we propose a call admission control scheme (CAC) to Edge zone
improve the performance of CoMP-JT based cellular systems.
Target BS Cooperating BS
In this CAC scheme, CoMP-JT technique is applied to UEs
located near the edge of a cell. In order to establish an analytical Desired signal from target BS
model for CoMP-JT based cellular systems, we introduce a model Desired signal from cooperating BS
for traffic analysis using a two-dimensional Markov chain and Interfering signal from cooperating BS
approximate the computation of the power sum of multiple
log-normal random components in a multi-cell environment. Fig. 1: Example of CoMP Joint Transmission technique sce-
Performances of CoMP-JT based cellular systems are evaluated nario; UE 1 uses PRB a, UE 2 uses PRB b, UE 3 uses PRB
with respect to two quality of service (QoS) constraints pertain- c and UE 4 uses PRB c. UE 1 is called the center UE, UE
ing to call-blocking and outage probabilities, respectively from
the network-layer and the physical-layer perspectives. We first
2 is called the JT UE and UE 3 is called the edge UE. The
measure the resource utilization and call blocking probability center UE and the edge UE are not JT served call (we call
for the downlink resources for various offered loads in the cell. them non-JT served call), and the JT UE is JT served call.
Based on that, we obtain the outage probability and effective
throughput of the system. The analytical results are compared
with computer simulations. Finally, we consider the dynamic
point selection (DPS) scheme that is a modified version of the JT diversity technique, a coordinated multi-point (CoMP) trans-
scheme, in which some resources of adjacent cells are muted. mission and reception with joint transmission (JT) scheme has
Index Terms— Coordinated multi-point transmission, Joint trans- been considered [7]. Fig. 1 shows the example of JT scheme
mission, Dynamic point selection, Interference, Call admission
control, Two-dimensional Markov chain. scenario. As shown in this, the same physical resource block
(PRB) b from the target BS and the cooperating BS is assigned
to the user equipment 2 (UE 2) to be jointly served by CoMP-
I. I NTRODUCTION JT system, which can increase the desired signal power for a
user and hence improve spectral efficiency. The resources of
A. Motivation the cooperating BSs, however, can be exhausted when CoMP-
System capacity and spectral efficiency are main focuses JT is highly utilized due to sharing the resources of BSs. Thus,
in the design of cellular networks [1]. Long Term Evo- it might lead to a high call blocking probability, in which a
lution (LTE)-Advanced improves those performances with call blocking probability represents the likelihood that a UE’s
a frequency reuse of one, while the users around a cell call request is blocked due to limited resources.
boundary would suffer from high inter-cell interferences (ICIs) In order to illustrate CoMP-JT in details, suppose that we
[2]. To overcome this problem, radio resource management use it only in an ICI susceptible region of a cell, which we
approaches based on optimized frequency allocation policies, call an edge zone as shown in Fig. 1. Depending on the size
optimal power assignment and control schemes, and smart of the edge zone and traffic loads in it, the availability of the
antenna techniques were studied [3]-[6]. As a cooperation resources for other UEs in the non-edge zone, say a center
zone, can be determined. Thus, in designing CoMP-JT based
Copyright (c) 2015 IEEE. Personal use of this material is permitted. on the size of the edge zone, it is important to characterize a
However, permission to use this material for any other purposes must be
obtained from the IEEE by sending a request to pubs-permissions@ieee.org. tradeoff between call blocking in the center zone and improved
The review of this paper was coordinated by Prof. Wei Song. (Correspond- cell-boundary performance by CoMP-JT. Accordingly, this
ing author : Choong-Ho Cho) paper investigates this tradeoff by proposing a simple cross
S.-Y. Kim and C.-H. Cho are with the Department of Computer and
Information Science, Korea University, Sejong City, South-Korea (e-mail: layer call admission control (CAC) in CoMP-JT in order to
kimsy8011@korea.ac.kr, chcho@korea.ac.kr). maximize the total throughput of the center and edge zones

0018-9545 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2016.2541172, IEEE
Transactions on Vehicular Technology

while keeping some constraints of quality of service (QoS). network layer and of physical layer respectively pertain
to the call blocking probabilities of user calls and the
modulation and coding scheme (MCS) based on the
B. Related Work and Main Contributions
SIR of accepted calls. In other words, the effective
In previous work regarding JT, [8] and [9] showed that a throughput can be obtained from the effective carried
cooperating operation using a diversity technique can achieve load and with consideration to the MCS levels (bits per
improved diversity gain. CoMP-JT could be non-coherent symbol).
or coherent [10]; Non-coherent JT may use techniques like 4) We examine the dynamic point selection (DPS) scheme
single-frequency network or cyclic delay diversity schemes, [11] for the cell-edge UE by differentiating the size
which target diversity gains and also enable increased transmit of the edge zone. With the DPS scheme, by muting a
power to the UE. On the other hand, coherent transmission potentially interfering signal, the sum of the interference
could be based on channel state information (CSI) feedback powers received at a UE can be decreased [12]. The DPS
relative to two or more cooperating BSs, which can be scheme outperforms the JT scheme with regard to the
used to perform MIMO transmissions from the corresponding normalized effective throughput. In contrast, for the cell
antennas. The most researches for CoMP-JT focus on non- edge UEs which are served by cooperating BS, the JT
coherent CoMP-JT due to less stringent synchronization and scheme achieves lower outage probability than the DPS
CSI requirements [10]-[13]. [14] and [15] showed improved scheme.
performance of a non-coherent CoMP-JT system through The remainder of this paper is organized as follows. Section
simulation and field trials. In [16], authors developed a model II introduces the system model with the JT scheme, our
for analyzing non-coherent CoMP-JT cooperation, in which assumptions, and an optimization problem for designing this
the stochastic geometry framework was considered, and char- system. In Section III, we analyze this system to solve our
acterized the signal to interference ratio (SIR) cumulative optimization problem. Additionally, we propose a CAC of the
distribution function (CDF) for a user. [17] characterized the DPS scheme in Section III. In Section IV, we discuss numer-
outage probability of non-coherent CoMP-JT based on the ical studies and give some concluding remarks in Section V.
Alamouti scheme, in which a UE is randomly located. In [18]
and [19], a non-coherent CoMP-JT scheme was applied to II. S YSTEM M ODEL AND P ROBLEM F ORMULATION
heterogeneous cellular networks and the coverage probability
A. Cellular System with Joint Transmission
for this scheme was derived under Rayleigh fading. Addition-
ally, to improve the throughput of the network, [20] and [21] We consider a cellular network model consisting of 19
established the model to decide the position of a user taking hexagonal cells1 , and a BS located at the center of the cell,
CoMP-JT service according to traffic density in homogeneous in which the length of one side of a hexagonal cell is R as
and heterogeneous cellular networks. shown in Fig. 2. The radius of a circular cell qwhose area is

From a modeling point of view, in this paper, we will the same as a hexagonal cell R0 is given by 3 3R /(2π). 2
consider non-coherent CoMP-JT as in [16]-[19]. In contrast From Fig. 2, we can classify the 18 neighboring cells into three
to [16]-[21], we will furthermore consider large-scale fading sets based on the distance from the target cell, in which BS0
as well as a traffic model. The four key contributions in our is considered to be the target cell’s BS. The first set of cells
paper are summarized below. is the six neighboring√cells numbered from 1 to 6 in the first-
1) A CAC algorithm is proposed in which a CoMP-JT tier which are d1 = 3R away from the center of the target
technique is applied to UEs located near the edge of cell. The second set includes the six cells in the second-tier,
a cell. We model a two-dimensional Markov chain for numbered from 7 to 12. These cells are d2 = 2d1 away from
traffic analysis of CoMP-JT based cellular systems, in BS0 . The third set is the set containing the remaining six cells
which we assume that a call inter-arrival time and call in the second-tier cells, numbered from 13 to 18 and adjacent
duration are exponentially distributed. In the analysis, to d1 cells and d2 cells. These are d3 = 3R away from BS0 .
the call blocking probabilities for users such as non-JT For CoMP-JT, we assume that each BS uses non-coherent
service and JT service users, and the resource utilization JT, with which cooperating BSs transmit the same signal with
are obtained. These are used for some design criteria the same physical resource block (PRB) to a UE. We also
such as the size of the edge zone in order to maximize assume that only one neighboring cell nearest to a target UE
the system throughput. in the first tier is selected as the cooperating BS.
2) We derive a CDF of the SIR for a user accounting for
cellular geometry, ICI and log-normal shadowing effects, B. Call Admission Control for Joint Transmission
in which we consider only one BS from the first tier as To introduce a traffic model, we assume that a number of
a cooperating BS. This paper, in contrast to [20] and UEs can initiate multiple calls occupying radio resources. We
[21], assumes that some neighboring BSs statistically also assume that one PRB of a serving BS is allocated to a
generate ICIs depending on the resource utilization of UE either for non-JT or for JT service.
each cell.
1 Usually, the received signal power is modeled as Eq. (2). From this model,
3) The normalized effective throughput with QoS con-
the impact of interference is mitigated by the exponentially decaying nature
straints in both the network and physical layers is of the path-loss. [25] showed that the two- tiers of interferers are considered
defined as the system throughput, in which the QoS of as the minimum in order to accurately estimate interference.

0018-9545 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2016.2541172, IEEE
Transactions on Vehicular Technology

11 18 12

17 6 1
13
Cooperating
d1 cell
d3
!
10 d2 5 0 2 7
Target cell

R0
16 4 3 14

BS
R
9 15 8

Fig. 2: System description for two-tier cells in CoMP-JT based cellular systems.

As mentioned in Section I and shown in Fig. 1, each cell Algorithm 1 : Call Admission Control of JT scheme
is divided into two zones: the center zone of radius ξ and the
edge zone. When a UE is in the center (or edge) zone, we call 1: if A call is generated in the center zone then
it a center (or edge) UE. Especially, we assume that JT service 2: if a PRB is available then
is available for the edge UEs only over a downlink (DL) 3: Accept the call (assigned PRB i which is one of the
transmission to improve cell-edge performance. In addition, remaining PRBs.)
we assume that UEs are uniformly distributed over a cell and 4: else
that the new call arrival process in a cell follows a Poisson 5: Reject the call: center call blocking.
process with a mean request rate of λ calls/second. Then, the 6: end if
center calls, i.e., new calls originated from the center zone, 7: else
follow a Poisson process with mean rate being λc = Aλ, 8: if a PRB is available then
where A is the fraction of the area of the center zone to the 9: Accept the call (assigned PRB i which is one of the

cell coverage, i.e., A = (ξ/R0 )2 = 2πξ 2 /(3 3R2 ). The edge remaining PRBs.)
calls, i.e., new calls originated from the edge zone, follow also 10: if a PRB i is available in a cooperating BS then
a Poisson process with mean rate λe = (1 − A)λ. Finally, 11: Accept the call (JT UE).
the call holding times of either the center or the edge calls 12: else
are assumed to be exponentially distributed with mean 1/µ 13: Reject the JT service: JT call blocking.
seconds2 . 14: end if
When a new call arrives, the UE requests a PRB. If there is 15: else
no PRB available, the call is blocked and is cleared. If the UE 16: Reject the call: edge call blocking.
is in the center zone, we call this event a center call blocking 17: end if
and if the UE is in the edge zone, we call this event an 18: end if
edge call blocking. On the other hand, if there is at least one
available PRB, one of the available PRBs is randomly chosen
and is assigned to the UE. If the accepted call is from an UE 3 which requests JT service is blocked. An accepted UE
edge UE, then the UE further requests JT service from the is one of three types as shown in Fig. 1 :
cooperating BS. That is, depending on whether the assigned
PRB is available or not in the cooperating BS, the JT service • Center UE - The call is initiated by the center UE in
request is successful or blocked (called JT call blocking). For the center zone, it uses only the PRB assigned from the
example, as shown in Fig. 1, when PRB c is assigned to edge target BS as UE 1 in Fig. 1.
UE (UE 3) in target BS, this call requests the same PRB c • Edge UE - The call is initiated by the edge UE in the
to the cooperating BS. If PRB c of a cooperating BS is not edge zone, it uses only the PRB assigned from the target
available (in the example, the cooperating BS serves UE 4), BS as UE 3 in Fig .1.
• JT UE - The call is initiated by the edge UE in the edge
2 This assumption has been widely used for its analytical simplicity [29]. zone, it uses not only the PRB assigned from the target
Some papers showed that internet traffic appears to be well described by BS but also the PRB assigned from its cooperating BS as
Poisson packet arrivals. For voice-over-IP (VoIP) traffic, it can be represented UE 2 in Fig. 1.
by the ON-OFF pattern, with an ON period corresponding to a talk spurt and
an OFF period corresponding to a period of silence. The duration of ON and Thus, center UE and edge UE involve non-JT-served calls, and
OFF periods can approximately be modeled by an exponential distribution
[30, 31]. For video streaming it can be modeled as a self-similar process with the JT UE involves a JT-served call. The CAC algorithm of
Pareto distribution [32]. JT scheme is given as Algorithm 1.

0018-9545 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2016.2541172, IEEE
Transactions on Vehicular Technology

C. Signal to Interference Ratio Model of Joint Transmission TABLE I: Nomenclature for Analysis
Notation Definition
We consider a channel model consisting of a path-loss ξ The normalized radius of the center zone
component and a log-normal shadowing component. Let (r, θ) The fraction of the area of the center zone
A
be the polar coordinate of a UE in the target BS, BS0 . The to the cell coverage
λc , λe The arrival rates of the center and edge zones UEs
joint probability density function (PDF) for the position of a The resource utilizations for center-, edge-,
UE, fp (r, θ) is given as ρc , ρe , ρj
and JT- UE
The cumulative distribution function (CDF)
1 2r Fc (·), Fe (·), Fj (·) of signal to interference ratio for center-, edge-,
fp (r, θ) = · , r ≤ R0 , 0 ≤ θ ≤ 2π. (1) and JT- UE
2π R02 The steady state probability mass function (PMF)
p(ni , nj )
for (ni , nj )
For given (r, θ), we assume that an average power received at c,Pe,Pj The call blocking probabilities of center-, edge-,
PB
a UE from a BSi , Si (r, θ) is B B and JT- UE
c ,Pe ,Pj The outage probabilities for center-, edge-,
Pout out out and JT- UE
Si (r, θ) = Pt gi (r, θ)−α Li , (2) The interference set with a cardinality is n,
Zn in which each element of set indicates
where gi is the distance between the UE and BSi , and α is the the index of neighboring BSs.
path-loss exponent. Further, L represents the effect of shadow- The probability that there is an interference is ρ,
B(Z n , ρ)
and the number of interfering BSs is n.
fading, which is assumed to be log-normally distributed. Note The mean and the variance of
ms , σs2
that the natural logarithm of L is a Gaussian with a mean of the desired signal power
zero and a variance of σ 2 , which is denoted by N (0, σ 2 ). 2 The mean and the variance of
m̂Z n , σ̂Z n
the sum of the interference power for Z n
First, the SIR of non-JT-served calls are obtained. The SIR
of a center UE is denoted by γc and that of an edge UE is
denoted by γe .
consideration to the MCS levels (bits per symbol) chosen
S0 (r, θ) according to γc , γe or γj . T can be expressed as follows:
γc (r, θ, Z n ) (or γe (r, θ, Z n )) = , (3)
IˆZ n (r, θ) 8
X n
T (λ/µ, ξ) = LM (l) λc (1 − PBc )Fc (l) (6)
where the denominator, IˆZ n indicates n ICIs, which results l=1
from simultaneous use of the same PRB by the cells in the + λe (1 − PBe )[PBj Fe (l)
first- and the second-tier. We can write IˆZ n as o
+ (1 − PBj )Fj (l)] /µ/C,
IˆZ n (r, θ) =
X
Si (r, θ), (4)
where C is the number of PRBs, LM (l) is the bit per symbol
i∈Z n
according to the MCS level l as shown in [24, Table I], and
where Z n denotes the interference set, whose cardinality is n, PBc , PBe and PBj are the call blocking probabilities of center-,
in which each element of set indicates the index of neighboring edge- and JT UEs, respectively, for given (λ/µ, ξ). F∗ (·) is
BSs, as shown in Fig 2. For example, when the indexes of the defined, for ∗ = {c, e, j}, as
interfering BSs are from 1 to 6, the elements of the set of F∗ (l) , Pr[l ≤ γ∗ < l + 1] = F∗ (l + 1) − F∗ (l), (7)
6-tuple are from 1 to 6 which is denoted by Z 6 .
Secondly, for the SIR of a JT UE, the portion of the desired where F∗ (·) is the CDF of the SIR according to the types of
signal transmitted from the cooperating BS that arrives with the UEs.
no excessive delay can be viewed as constructive interference By appropriately choosing ξ and λ/µ (Erlangs) subject to a
(i.e., gain), with the remaining portion of the signal behaving set of constraints consist of the call blocking probabilities and
as destructive interference. Thus, the impact of the signal the outage probability, we can maximize the normalized ef-
power from the cooperating BS is related to the propagation fective throughput for CoMP-JT based cellular systems. Then,
delay [16, 23], and we assume that the propagation delay is an optimization problem to maximize T can be formulated as
less than the duration of the guard interval. Thus, we can have follows:
γj as maximize T (λ/µ, ξ), (8)
P
Sk (r, θ) subject to PBc , ≤ηB , PBe ≤ ηB , Pout ≤ ηout ,
γj (r, θ, Z ) = P k∈J
n
, (5) variables λ/µ, ξ,
/ n Si (r, θ)
i∈Z,k∈Z
where Pout denotes the outage probability, ηB is the threshold
where J is the set of serving BSs including the cooperating of the call blocking probability, and ηout is the tolerance of
BS. the outage probability. The call blocking probabilities and the
CDFs of the SIR will be derived in next Section.
D. Problem Formulation III. P ERFORMANCES A NALYSIS
Let us denote T as the normalized effective throughput, Before proceeding with our analysis, we summarize in Table
which is obtained from the effective carried load and with I the notations, which will be used throughout this paper.

0018-9545 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2016.2541172, IEEE
Transactions on Vehicular Technology

nj = C


aj (0, C − 1)

ai (0, C − 1)
nj = C − 1
µ

ai (0, 2)
nj = 2
µ

aj (0, 1) 2µ aj (1, 1) 2µ

ai (0, 1) ai (1, 1)
nj = 1
µ 2µ

aj (0, 0) µ aj (1, 0) µ aj (2, 0) µ aj (C − 1, 0) µ

ai (0, 0) ai (1, 0) ai (C − 1, 0)
nj = 0
µ 2µ Cµ
ni = 0 ni = 1 ni = 2 ni = C − 1 ni = C

Fig. 3: State transition rate diagram of (ni , nj ) for CAC of the JT scheme.

A. Traffic Analysis + p(ni + 1, nj )(ni + 1)µ + p(ni , nj + 1)(nj + 1)µ.


In this section, we will find the network layer QoS such as Note that p(ni , nj ) = 0 unless ni + nj ≤ C. By solving the
the call blocking probability and the resource utilization for balance equations together with the normalization condition
CoMP-JT based cellular systems. As mentioned above, there
C C−n
Xi
are two kinds of UEs using PRBs of the target cell. Thus, we X
p(ni , nj ) = 1, (12)
define that Ni (t) is the number of UEs in the target cell at
ni =0 nj =0
t while Nj (t) is the number of UEs in the neighboring cells
which are using the target cell’s PRBs. Then, the variable set, we can obtain the steady state PMF.
{Ni (t), Nj (t)}, becomes a two-dimensional Markov chain. In Eq. (11), the left hand side states the flow rate out of
Suppose that ak (ni , nj ), (k = i, j), is the state transition state (ni , nj ) and the right hand side states the flow rate
rate of the two-dimensional Markov chain for the arrival into state (ni , nj ). The balance equations can be visualized
process depending on Ni = ni and Nj = nj . The transition by the state transition flow diagram shown in Fig. 3, where
rates due to call arrival for the target cell’s own UEs and the the flows
 into and out of states are depicted. For example,
neighboring UEs in the target cell are defined, respectively, as p(1, 1) ai (1, 1)+aj (1, 1)+µ+µ is the flow out of state (1,1)
follows and p(0, 1)ai (0, 1) + p(1, 0)aj (1, 0) + 2µ · p(2, 1) + 2µ · p(1, 2)

λ, ni + nj ≤ C, is the flow into state (1,1).
ai (ni , nj ) = Once p(ni , nj ) is obtained, we can obtain the following
0, otherwise,
network layer performances.
λe (1 − PBe )Pc , ni + nj ≤ C,

aj (ni , nj ) = (9) 1) Resource utilization: From the steady state PMF, the
0, otherwise, mean occupied number of PRBs used for the target cell’s own
where PBe denotes the edge call blocking probability in the UEs, mNi , and used for neighboring UEs, mNj , in the target
cell, and Pc is the probability that a PRB of the cell which cell are respectively obtained,
is assigned for an edge UE in one of the neighboring cells C C−i
is available for CoMP-JT, which is {C − (ni + nj )}/C. The
X X
mNi = E[Ni ] = i p(i, j), (13)
second part of (9) is based on the assumption that all the cells i=0 j=0
are stochastically identical. C C−j
X X
The steady state probability mass function (PMF) of mN j = E[Nj ] = j p(i, j).
{Ni (t), Nj (t)} is defined as j=0 i=0

p(ni , nj ) , lim Pr Ni (t) = ni , Nj (t) = nj ,


 
t→∞
(10) Here, Ni , lim Ni (t) and Nj , lim Nj (t). And the mean
t→∞ t→∞
occupied number of total PRBs, mNJT , is
which satisfies the following balance equations for ni + nj ≤
C, C
X k
X
  mNJT = k p(i, k − i). (14)
p(ni ,nj ) ai (ni , nj ) + aj (ni , nj ) + ni µ + nj µ (11) k=0 i=0
= p(ni − 1, nj )ai (ni − 1, nj ) It is also given by mNi + mNj . Then, the utilization of PRBs
+ p(ni , nj − 1)aj (ni , nj − 1) used for the target cell’s own UEs, ρi , and the utilization of

0018-9545 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2016.2541172, IEEE
Transactions on Vehicular Technology

PRBs used for neighboring UEs, ρj , and the overall PRB both the numerator and denominator are a normal distribu-
utilization of a JT based system, ρJT , in the target cell are tion, the SIR for a UE is also a normal distribution, with
respectively obtained, N (ms − m̂Z n , σs2 + σ̂Z
2
n ). Therefore, we can rewrite (18) as

mN i mN j mNJT !
ρi = , ρj = , ρJT = , (15) n ln γ − ms + m̂Z n
C C C F (γ|r, θ, Z ) = 1 − Q p
2
, (19)
σs2 + σ̂Z n
where ρJT is also given by ρi + ρj .
√ R∞
For a given PRB utilizations, in the cell, the PRB utilizations 2
where Q(a) = (1/ 2π) a e−t /2 dt [26].
by JT, center- and edge UEs are, respectively, as follows Remark 2: If the impact of small-scale fading such as
λc Rayleigh fading is considered, the power of signal is no more
ρ j = ρ j , ρc = · ρi = A · ρi , (16) a log-normal distribution. The exponential distribution is then
λ
λe to be considered. Note that the interference is composed of
ρe = · ρi − ρj = (1 − A) · ρi − ρj . several exponential variables, whose sum can be derived by
λ
[17].
2) Call blocking probability: The call blocking probabili- Considering the region Ω in serving cell, for Eq. (19), we
ties for center-, edge- and JT UEs are, respectively, obtained can have
as follows Z π/2 Z R0
C
X F (γ|Z n ) = F (γ|r, θ, Z n ) · fp (r, θ)drdθ. (20)
PBc = PBe = p(i, C − i), PBj = ρJT . (17) π/6 0
i=0
Using the total probability law for the interference set Z n ,
Remark 1: When a data call such as a full queue file transfer we can get the CDF of a UE given by
protocol (FTP) is considered, it stays in the resource until
18
downloading a file with exponentially distributed size of mean X
F (γ) = F (γ|Z n ) · B(Z n , ρ), (21)
1/µ. Note that if the resource holding time is not Markovian,
n=1
call blocking probability may be obtained by [27, 28].
where B(Z n , ρ) denotes the probability that there is an in-
terference is ρ, and the number of interfering BSs is n. Note
B. SIR Analysis that we considered two-tier neighboring BSs. Thus, we assume
In this section, we derive the CDF of the SIR for three UE the probability that there are n interfering BSs is a binomial
types which are center-, edge- and JT UEs. The occurrence distribution with parameters 18 and ρ as given by
of interference depends on the resource utilization in a cell  
18 n
and we assume that the distribution of a sum of independent B(Z n , ρ) = ρ (1 − ρ)18−n , n = 1, 2, · · · , 18, (22)
log-normal random variable can be approximated by a log- n
normal random variable. In addition, we consider that a UE is where ρ is the resource utilization of a conventional system.
randomly located in the region Ω of BS0 as shown in Fig. 2, The above procedure for F (γ) will be applied to obtain
which is defined as the area for π/6 ≤ θ ≤ π/2 and r ≤ R0 . Fc (γ), Fe (γ) and Fj (γ).
Note that we assumed only one neighboring cell nearest to a 1) Fc (γ): For a center UE, since a UE is located in center
target UE in the first tier is selected as the cooperating BS. region, i.e. r < ξ and served as conventional system, we can
Therefore, since Ω lies in a sector adjacent to BS1 , the UE write Fc (γ|r, θ, Z n ) as
choose BS1 as a cooperating BS.
!
In a conventional cellular system, when a tagged UE is at ln γ − ln P t r −α
+ m̂ Z n
(r, θ) in the target BS with n interferers in the cells belonging Fc (γ|r, θ, Z n ) = 1 − Q p
2
. (23)
σ 2 + σ̂Z
to the interference set Z n , we can obtain the conditional CDF n

of SIR for a UE as Considering the center region of Ω in the target cell, we can
have
 
S0 (r, θ)
F (γ|r, θ, Z n ) = Pr P < γ r, θ, Z n (18)

i∈Z n Si (r, θ) Fc (γ|Z n ) = Pr[γc (r, θ) < γ|r, θ, Z n ]
" #
Pt r−α L0
n
Z π/2 Z ξ
= Pr < γ r, θ, Z .

IˆZ n (r, θ) = Fc (γ|r, θ, Z n ) · fp (r, θ)drdθ. (24)
π/6 0

Taking the natural logarithm of both sides, in the numerator Using the total probability law for Z n , we can get the CDF
we have ln S0 (r, θ), which is a normal distribution whose of a center UE as
mean ms and variance σs2 are expressed as E[ln S0 (r, θ)] =
18
ln Pt r−α and Var[ln S0 (r, θ)] = σ 2 , respectively. For the X
Fc = Fc (γ|Z n ) · B(Z n , ρJT ), (25)
natural logarithm of the denominator of (18), ln IˆZ n (r, θ), we
n=1
can obtain a normal distribution whose mean and variance
2
are m̂Z n and σ̂Z n , respectively from Appendix A. Since where ρJT is the resource utilization of JT system.

0018-9545 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2016.2541172, IEEE
Transactions on Vehicular Technology

2) Fe (γ): In case of edge UE, since a UE is located in Algorithm 2 : Call Admission Control of DPS scheme
edge region of Ω in the target cell and served as conventional
system, we can have Fe (γ|Z n ) as 1: if A call is generated in the center zone then
Z π/2 Z R0 2: if a PRB is available then
n
Fe (γ|Z ) = n
Pr[γe (r, θ) < γ|r, θ, Z ] · fp (r, θ)drdθ 3: Accept the call (assigned PRB i which is one of the
π/6 ξ remaining PRBs.)
Z π/2 Z R0 ( else
!)
−α 4:
ln γ − ln Pt r + m̂Z n
= 1−Q p
2
5: Reject the call: center call blocking.
π/6 ξ σ 2 + σ̂Z n
6: end if
· fp (r, θ)drdθ. (26) 7: else
8: if a PRB is available then
Note that since edge UE is not JT served call as UE 3 in
9: Accept the call (assigned PRB i with one of the
Fig. 1, it suffers the interference from BS1 . Thus, BS1 is an
remaining PRBs.)
element of Z n . Taking the total probability law for the possible
10: if a PRB i is available in a cooperating BS then
outcomes of Z n , we can get the CDF of edge UE as
11: Accept the call (DPS UE).
X18 12: else
Fe = Fe (γ|Z n ) · B(Z n , ρJT ), (27) 13: Reject the mute service: DPS call blocking.
n=2 14: end if
where 15: else
 
18 n 16: Reject the call: edge call blocking.
B(Z n , ρJT ) = ρJT (1 − ρJT )18−n . (28) 17: end if
n
18: end if
3) Fj (γ): For a JT UE, since a UE is located in edge region
and JT served call, we can have Fj (γ|Z n ) as
Z π/2 Z R0
n
Fj (γ|Z ) = Pr[γj (r, θ) < γ|r, θ, Z n ] · fp (r, θ)drdθ From (25), (27) and (30), when γ < −6.5dB, the outage
c e j
π/6 ξ probabilities of center UE, Pout , edge UE, Pout , JT UE, Pout
Z π/2 Z R0 ( and overall, Pout , can be obtained.
!)
ln γ − m̂s + m̂Z n
= 1−Q
5) Dynamic Point Selection scheme: Additionally, we con-
p
σ̂s2 + σ̂Z2
π/6 ξ n
sider the dynamic point selection (DPS) scheme that is a
· fp (r, θ)drdθ, (29)
modified version of the JT scheme, in which some resources
where m̂s and σ̂s2 are the mean and variance of a normal of the cooperating BS is muted. When a new call arrives, the
distribution for the sum of desired signals from BS0 and BS1 , UE requests a PRB. If there is no PRB available, the call
respectively, which can be obtained from Appendix A. In the is blocked and is cleared. If the UE is in the center zone,
all possible outcomes for Z n , since BS1 is cooperating BS, we call this event a center call blocking and if the UE is in
Z n excepts the index 1 for BS1 . Using the total probability the edge zone, we call this event an edge call blocking. On
law for Z n , we can get the CDF of a JT UE as the other hand, if there is at least one available PRB, one of
17
the available PRBs is randomly chosen and is assigned to the
(30) UE. If the accepted call is from an edge UE, then the UE
X
n n
Fj = Fj (γ|Z ) · B(Z , ρJT ),
n=1
further requests mute service from the cooperating BS. That
is, depending on whether the assigned PRB is available or not
where in the cooperating BS, the mute service request is successful
 
17 n d
n
B(Z , ρJT ) = ρJT (1 − ρJT ) 17−n
. (31) or blocked (called DPS call blocking and denoted by PB ).
n Thus, an accepted UE is one of three types in DPS scheme.
4) Outage Probability: Once Fc (γ), Fe (γ) and Fj (γ) are • Center UE - The call is initiated by the center UE in
obtained 3 , we can get the overall CDF of a UE, F (γ) as the center zone, it uses only the PRB assigned from the
follows target BS.
ρc Fc (γ) + ρe Fe (γ) + ρj Fj (γ) • Edge UE - The call is initiated by the edge UE in the
F (γ) = . (32)
ρc + ρe + ρj edge zone, it uses only the PRB assigned from the target
BS. The request of mute service is not accepted. That is,
3 In our work, we assume that when a user is accepted in BS, one of the
the PRB is used by its cooperating BS.
remaining resources is assigned. Furthermore, we assume that neighboring
BSs statistically generate inter-cell interferences depending on the resource
• DPS UE - The call is initiated by the edge UE in the
utilization of each cell. Since a user, despite of its location, can basically take edge zone, it uses only the PRB assigned from the target
one of the available resources of BS, utilization of JT system is ρJ T , which is BS. The PRB is now muted by its cooperating BS.
a probability that interference occurs. If we assume that the different resource
sets are assigned to users according to arrived region, for example, center Thus, center UE and edge UE are non-DPS served call, and
users select the center resource set and edge users select the edge resource
set, for each user type, different utilization will be given in Eqs. (25), (27), DPS UE refers to DPS-served call. The CAC algorithm of
(30). DPS scheme is given as Algorithm 2.

0018-9545 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2016.2541172, IEEE
Transactions on Vehicular Technology

1 1
ξ=0.3,0.6,0.9 ξ=0.3,0.6,0.9
0.9
0.9
line : analysis blue line : C=50
♦ : simulation 0.8 red dash : C =10
0.8

Call blocking probability


Call blocking probability
blue line : C=50 0.7
0.7
red dash : C =10
0.6 0.6

PBj 0 PBd 10
0

0.5 10 0.5
ξ increase ξ increase ξ increase ξ increase
0.4 0.4 −1
−1 10
10
0.3 0.3

−2 −2
0.2 10 0.2 10
PBc , PBe PBc , PBe
0.1 ηB =2 × 10 −2 0.1 ηB =2 × 10 −2
−3 −3
10 10
0 0.4 0.6 0.8 1 0 0.4 0.6 0.8 1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Offered load : λ/µ/C Offered load : λ/µ/C

(a) JT scheme (b) DPS scheme

Fig. 4: Call blocking probability versus offered load with various ξ (0.3, 0.6, 0.9) and C (10, 50), in which PBc , PBe , PBj and
PBd are the call blocking probabilities of center-, edge-, JT- and DPS- UE, respectively.

IV. N UMERICAL R ESULTS the call blocking probabilities of DPS scheme show the
same performance as those of JT scheme.
From the previous section, we obtained an analytic model
2) Fig. 5(a)-(f) shows the resource utilizations for two
as a function of ξ and the offered load, λ/µ/C. In this section,
schemes. In Fig. 5(a,c,e), we can observe that each
the performance of the analytic model of a JT based cellular
utilization of JT scheme is not balanced. Note that in
system will be analyzed and compared with the simulation
Eq. (16), ρc , ρe , ρj and ρJT depend on ξ. As mentioned
results to verify its accuracy. Additionally, we will evaluate
above, this is because ξ effects the number of UEs
the performance of the DPS scheme.
arriving in each region. The large ξ provides a large
We examine the effects of the values of ξ, C, σdB on
center zone and a higher ρc , while ρJT decreases. For
the performance metrics of the network layer QoS and the
the effect of C, it can be seen that as C increases, ρj
physical layer QoS. Note that when ξ is one, the JT- and DPS
and ρJT increase. The reason is that as C increases, the
based systems become the conventional system, which are
number of PRBs which can be shared in a cooperative
non-JT- and non-DPS systems, respectively. For these figures,
manner, increases. In Fig. 5(b,d,f) which are the cases of
the curves are numerically obtained from the equations given
DPS scheme, ρc , ρe and ρd have the same performance
in the preceding analysis, whereas the symbols indicated the
as those of JT, while ρDP S is less than ρJT . This is
corresponding simulation results.
because some PRBs in DPS scheme are muted from
neighboring BS. We can see that ρDP S becomes less
A. Network layer QoS of the two schemes sensitive to the change of ξ as C is increased. We can
also see that as ξ increases, ρJT and ρDP S are similar.
In Figs. 4−5, we investigate the network layer QoS of JT-
and DPS schemes such as the call blocking probability and
the resource utilization. Mostly, for figures of JT scheme, our B. Physical layer QoS of the two schemes
analysis and simulation agree well with each other. Figs. 6−7 show the outage probability of JT- and DPS
1) In Fig. 4(a)-(b), we present the call blocking probabil- schemes as physical layer QoS in different C = 10, 50.
ities for varying offered loads. From Fig. 4(a), we can Mostly, for figures of JT scheme, our analysis agrees well
see that PBc and PBe are identical as expected from (17), with simulations.
and PBj is greater than PBc and PBe . The reason is that 1) Fig. 6(a)-(d) show the outage probabilities of two
JT UE has to select the same PRB with the PRB number schemes in C = 10, in which red color and blue color
assigned in target BS from the cooperating BS. For the represent the results of the JT and the DPS, respectively.
effect of ξ, as this value increases, PBc , PBe and PBj tend In the results of JT scheme, it is observed that there is a
to decrease. The reason is that, in a JT based system, significant disparity in the outage probability depending
j c e
the additional PRBs in the target BS are being shared on the types of UEs. Pout is lower than Pout and Pout .
e c
by the UE that belong to the cooperating BS and as ξ Under the lower offered load, Pout is higher than Pout
j
increases, the number of calls arriving in the edge zone and Pout . For the effect of ξ on the JT scheme, as ξ
decreases. For the effect of C, as C decreases we can increases we can see that the outage probability of each
see that PBc , PBe and PBj increase. For PBc and PBe , the UE decreases. The reason is that, as expected from the
threshold of the call blocking probability ηB is satisfied result of the resource utilization, an increased center
for offered load (λ/µ/C) 0.55, 0.7 and 0.8 according to zone decreases ρJT and leads to the diminishment of
C =10 and 50, in which ηB = 2 × 10−2 . From Fig. the probability that interference occurs from neighboring
e
4(b), we can see that under the proposed CAC for DPS, cells. In the results of DPS scheme, although Pout suffer

0018-9545 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2016.2541172, IEEE
Transactions on Vehicular Technology

1 1
line : analysis blue line : C = 50
symbol : simulation red dash: C = 10
0.8 blue line : C = 50 0.8
red dash: C = 10

Utilization

Utilization
0.6 0.6

0.4 0.4

0.2 0.2

0 0
0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1
Offered load : λ/µ/C Offered load : λ/µ/C
(a) JT scheme ξ = 0.3 (b) DPS scheme ξ = 0.3

1 1
line : analysis blue line : C = 50
symbol : simulation red dash: C = 10
0.8 blue line : C = 50 0.8
red dash: C = 10
Utilization

Utilization
0.6 0.6

0.4 0.4

0.2 0.2

0 0
0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1
Offered load : λ/µ/C Offered load : λ/µ/C
(c) JT scheme ξ = 0.6 (d) DPS scheme ξ = 0.6

1 1
line : analysis blue line : C = 50
symbol : simulation red dash: C = 10
0.8 blue line : C = 50 0.8
red dash: C = 10
Utilization

Utilization

0.6 0.6

0.4 0.4

0.2 0.2

0 0
0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1
Offered load : λ/µ/C Offered load : λ/µ/C
(e) JT scheme ξ = 0.9 (f) DPS scheme ξ = 0.9

Fig. 5: Resource utilizations versus offered load with various ξ and C, in which ρc , ρe , ρj , ρd and ρDPS are the resource
utilizations for center-, edge-, JT- and DPS- UE, respectively. Additionally ρJT and ρDP S are the overall utilizations of JT
and DPS, respectably. For the simulation results, ρc : ∗, ρe : ♦, ρj : , ρd : +, ρJT : ◦, ρDP S : ⋆.

from a higher outage probability, the differences between result of the resource utilization, the outage probabilities
the outage probabilities are not significant. In other for two schemes increase as C increase. For the effect
words, from the outage probability aspects, there are of ξ on the DPS scheme, as expected from the result for
not many differences depending on the types of UEs. ρDP S , ξ has almost no effect on the outage probabilities
c e
We can see that Pout and Pout of the DPS scheme is of this scheme.
j
less than those of the JT scheme, while Pout and Pout
d
of the JT are less than Pout and Pout of the DPS. From C. Normalized effective throughput of two schemes
this result, we can see that the JT scheme can achieve Fig. 8(a)-(d) show the normalized effective throughput, τ
the increased signal to interference ratio (SIR) of UEs under different levels of offered loads as defined in (6) for
for a given interference. a different σ = 8, 12. In the results, red color and blue color
2) Fig. 7(a)-(d) show the outage probabilities of two represent the performances of the JT and the DPS, respectively.
schemes in C = 50, in which red color and blue color 1) Fig. 8(a)-(b) show τ with respect to Pout . It is notable
represent the results of the JT and the DPS, respectively. that τ cannot be increased without increasing the toler-
This results show the effect of C. As expected from the ance of the outage probability ηout for all users in the

0018-9545 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2016.2541172, IEEE
Transactions on Vehicular Technology

10

1 1
ξ=0.1(o),0.3(+),0.6(♦),0.9(*) ξ=0.1(o),0.3(+),0.6(♦),0.9(*)
0.9 red line : analysis of JT 0.9 red line : analysis of JT
red symbol : simulation of JT red symbol : simulation of JT

Outage probability of center UE

Outage probability of edge UE


0.8 0.8
blue line & symbol : simulation of DPS blue line & symbol : simulation of DPS
0.7 0.7

0.6 0.6

0.5 0.5

0.4 0.4

0.3 0.3

0.2 0.2

0.1 0.1

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Offered load : λ/µ/C Offered load : λ/µ/C

c
(a) Pout e
(b) Pout

1 1
ξ=0.1(o),0.3(+),0.6(♦),0.9(*) ξ=0.1(o),0.3(+),0.6(♦),0.9(*)
Outage probability of JT UE and DPS UE

0.9 red line : analysis of JT 0.9 red line : analysis of JT


red symbol : simulation of JT red symbol : simulation of JT
0.8 0.8
blue line & symbol : simulation of DPS blue line & symbol : simulation of DPS
0.7 0.7

Outage probability
0.6 0.6

0.5 0.5

0.4 0.4

0.3 0.3

0.2 0.2

0.1 0.1

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Offered load : λ/µ/C Offered load : λ/µ/C

j d (d) Pout
(c) Pout , Pout

Fig. 6: Outage probabilities for two schemes versus offered load with various ξ, in which ξ =0.1(o), 0.3(+), 0.6(♦), 0.9(*),
c e j d
C = 10, α = 4 and σ = 12. Pout , Pout , Pout and Pout are the outage probabilities of center-, edge-, JT- and DPS- UE,
respectively. Additionally Pout is the overall outage probability of JT and DPS.

system. From Fig. 8(a), we can see that when ηout <0.12 utilization, outage probability and effective throughput. In an
and ξ is small, JT scheme provides more outstanding analytical model, we first derived the call blocking probability
τ . While for a small ξ, when ηout is large, τ of DPS and the resource utilization for the different types of UEs using
scheme is maximized. For example, for the DPS scheme, a two-dimensional Markov chain analysis. We also derived
when ηout =0.39, τ can be maximized at 0.67. In case the CDF of the SIR by approximating the ICI by the power
of JT scheme, when ηout =0.3, τ can be maximized at sum of multiple log-normal random components. Then, we
0.49. In Fig. 8(b), to show the effect of C, we set C=50. established the effective throughput of the proposed model
It is notable that increasing C reduces τ . For the DPS by applying the call blocking probability and the MCS level
scheme, when ηout =0.29, the maximum τ is 0.4. In case to the CDF of the SIR value. The majority of our analytical
of JT scheme, when ηout =0.31, the maximum τ is 0.50. results show good agreement with the simulations. The results
We can see that when ξ is small, τ of DPS scheme is also show that the performance mainly depends on the center
maximized. zone size and offered load. Finally, we evaluated the DPS
2) Fig. 8(c)-(d) show τ versus Pout for two schemes. To scheme, which is a simple extension of an existing JT scheme.
show the effect of σ, we set σ=8. It is notable that The DPS scheme outperforms the JT scheme with regard to
decreasing σ decreases τ with respect to Pout . For the normalized effective throughput. In contrast, for the cell
example, for σ=8, τ of DPS scheme is to be 0.61 at edge UEs which are served by cooperating BS, the JT scheme
ηout =0.33 and τ of JT scheme is to be 0.40 at ηout =0.19. achieves lower outage probability than the DPS scheme.
For increasing C, the maximum τ of JT scheme is 0.37
at ηout =0.24 and the maximum τ of DPS scheme is 0.46 A PPENDIX A
at ηout =0.26. A PPROXIMATION ON THE SUM OF LOG - NORMAL RANDOM
VARIABLES
V. C ONCLUSIONS 2
To obtain m̂Z n and σ̂Z n , we consider Aysel Safak’s method

In this paper, we proposed a CAC scheme to improve the [25]. In this method, by assuming that the sum of log-normal
performance of CoMP-JT systems and introduced an analytic random variables is a log-normal random variable, a recursive
model to evaluate the performance of this system. We analyzed technique is developed for approximating the first and second
its performance in terms of call blocking probability, resource moments of a sum of log-normal random variables.

0018-9545 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2016.2541172, IEEE
Transactions on Vehicular Technology

11

1 1
ξ=0.1(o),0.3(+),0.6(♦),0.9(*) ξ=0.1(o),0.3(+),0.6(♦),0.9(*)
0.9 red line : analysis of JT 0.9 red line : analysis of JT
red symbol : simulation of JT red symbol : simulation of JT

Outage probability of center UE

Outage probability of edge UE


0.8 0.8
blue line & symbol : simulation of DPS blue line & symbol : simulation of DPS
0.7 0.7

0.6 0.6

0.5 0.5

0.4 0.4

0.3 0.3

0.2 0.2

0.1 0.1

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Offered load : λ/µ/C Offered load : λ/µ/C

c
(a) Pout e
(b) Pout

1 1
ξ=0.1(o),0.3(+),0.6(♦),0.9(*) ξ=0.1(o),0.3(+),0.6(♦),0.9(*)
Outage probability of JT UE and DPS UE

0.9 red line : analysis of JT 0.9 red line : analysis of JT


red symbol : simulation of JT red symbol : simulation of JT
0.8 0.8
blue line & symbol : simulation of DPS blue line & symbol : simulation of DPS
0.7 0.7

Outage probability
0.6 0.6

0.5 0.5

0.4 0.4

0.3 0.3

0.2 0.2

0.1 0.1

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Offered load : λ/µ/C Offered load : λ/µ/C

j d (d) Pout
(c) Pout , Pout

Fig. 7: Outage probabilities for two schemes versus offered load with various ξ, in which ξ =0.1(o), 0.3(+), 0.6(♦), 0.9(*),
c e j d
C = 50, α = 4 and σ = 12. Pout , Pout , Pout and Pout are the outage probabilities of center-, edge-, JT- and DPS- UE,
respectively. Additionally Pout is the overall outage probability of JT and DPS.

For given (r, θ) of a UE, the interference power from where mw and σw are the mean and standard deviation of
interfering BSi can be obtained as w = y2 − y1 , respectively, which are given by

Si (r, θ) = Pt gi (r, θ)−α Li , (33) 2


mw = my2 − my1 , σw = σy22 + σy21 . (36)
where gi (r, θ) is distance between a UE and the BSi which
The function G1 (σ, m) is given by
is obtained from Appendix B. The natural logarithm of an
interference power from BSi is ln Si (r, θ), which is a normal m σ m2
distribution whose mean and variance are ln Pt gi (r, θ)−α and G1 (σ, m) = mΦ + √ e− 2σ2
σ 2π
σ 2 , respectively. ∞
For the given interference set Z n , let yk = ln Si (r, θ),
X h i
+ Ck D(σ, m, k) + D(σ, −m, k) , (37)
and myk and σyk are the mean and standard deviation of yk , k=1
respectively, where i is the k th element in Z n . Based on Aysel
Safak’s method, ln IˆZ n (r, θ) can be written as R x − t2 k+1
where we have Φ(x) = √1
2π −∞
e 2 dt, Ck = (−1)k ,
k2 σ 2
 2

n
e−km+ 2 Φ m−kσ
! !
and D(σ, m, k) = σ . G2 (σ, m) and
ln IˆZ n (r, θ) = ln
X X
ln S (r,θ) y
e i
= ln e k
. (34)
G3 (σ, m) are given by
i∈Z n k=1
m σ m2
For n = 2 the mean and standard deviation of Z 2 = G2 (σ, m) = (m2 + σ 2 )Φ + (m + ln 4) √ e− 2σ2
ln IˆZ 2 (r, θ) are given by σ 2π

X
+2 Ck (m − kσ 2 )D(σ, m, k)
m̂Z 2 = my1 + G1 (σw , mw ),
k=1
2 2 2
σ̂Z 2 = σy − G1 (σw , mw ) + G2 (σw , mw ) ∞
1 X h i
σy2 + Bk−1 D(σ, m, k) + D(σ, −m, k)
− 2 21 · G3 (σw , mw ), (35) k=2
σw (38)

0018-9545 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2016.2541172, IEEE
Transactions on Vehicular Technology

12

0.7 0.7
red = JT ξ = 0.1 (o) red = JT ξ = 0.1 (o)
blue = DPS ξ = 0.3 (+) blue = DPS ξ = 0.3 (+)

Normalized Effective Throughput

Normalized Effective Throughput


0.6 0.6
ξ = 0.6 (♦) ξ = 0.6 (♦)
0.5
ξ = 0.9 (*) 0.5
ξ = 0.9 (*)

0.4 0.4

0.3 0.3

0.2 0.2
τ =0.49 τ =0.67 τ =0.40 τ =0.50
ηout =0.3 ηout =0.39 ηout =0.29 ηout =0.31
0.1 0.1

0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Outage Probability Outage Probability

(a) C = 10, σ = 12 (b) C = 50, σ = 12

0.8 0.8
red = JT ξ = 0.1 (o) red = JT ξ = 0.1 (o)
0.7 blue = DPS ξ = 0.3 (+) 0.7 blue = DPS ξ = 0.3 (+)
Normalized Effective Throughput

Normalized Effective Throughput


ξ = 0.6 (♦) ξ = 0.6 (♦)
0.6 0.6
ξ = 0.9 (*) ξ = 0.9 (*)
0.5 0.5

0.4 0.4

0.3 0.3

0.2 0.2

τ = 0.37 τ = 0.46
0.1 τ = 0.61 0.1
τ = 0.40 ηout = 0.24 ηout = 0.26
ηout = 0.33
ηout =0.19
0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Outage Probability Outage Probability

(c) C = 10, σ = 8 (d) C = 50, σ = 8

Fig. 8: Normalized effective throughput versus outage probability for JT- and DPS scheme, in which ξ=0.1, 0.3, 0.6, 0.9, C=10,
50, α=4, and σ = 8, 12.

2(−1)k+1 Pk 1
with Bk = k+1 j=1 j , and Θ2 (θ) = 2π/3−θ, Θ3 (θ) = π −θ, Θ5 (θ) = π/3+θ, Θ6 (θ) =

θ, and
X h i
2
G3 (σ, m) = σ (−1)k D(σ, m, k) + D(σ, −m, k + 1) .

2π/3 + θ if π/6 ≤ θ ≤ π/3,
Θ4 (θ) = (43)
k=2 4π/3 − θ if π/3 ≤ θ ≤ π/2.
(39)
Similarly, for n ≥ 3 the mean and variance of Z n = For the second tier, we have Θ7 (θ) = Θ6 (θ), Θ8 (θ) = θ −
n−1
ln(eZ + eyn ) are given by π/6, Θ9 (θ) = Θ1 (θ), Θ10 (θ) = π/2 − θ, Θ11 (θ) = Θ2 (θ),
Θ12 (θ) = 5π/6 − θ, Θ13 (θ) = Θ3 (θ), Θ14 (θ) = 7π/6 − θ,
m̂Z n =m̂Z n−1 + G1 (σwn , mwn ),
Θ15 (θ) = Θ4 (θ), Θ16 (θ) = π/2 + θ, Θ17 (θ) = Θ5 (θ), and
2 2 2
σ̂Z n =σ̂Z n−1 − G1 (σwn , mwn ) + G2 (σwn , mwn ) Θ18 (θ) = π/6 + θ. Once Θi (θ) is obtained, we get gi (r, θ) as
σ̂ 2 n−1 p
− 2 Z2 · G3 (σwn , mwn ), (40) gi (r, θ) = d2 + r2 − 2dr cos(Θi (θ)), (44)
σwn
where where d = d1 for i ∈ {1, . . . , 6}, d = d2 for i ∈ {7, . . . , 12}
and d = d3 for i ∈ {13, . . . , 18} from Fig. 2.
wn =yn − Z n−1 , mwn = myn − m̂Z n−1 ,
2
σw n
=σy2n + σ̂Z
2
n−1 . (41) R EFERENCES
[1] C. Xiong, G.-Y. Li, S. Zhang, Y. Chen, and S. Xu, “Energy- and
A PPENDIX B Spectral-Efficiency Tradeoff in Downlink OFDMA Networks,” IEEE
D ISTRIBUTION OF UE Trans. Wirel. Commun., vol. 10, no. 11, pp. 3874-3886, Nov. 2011.
[2] X. You, D. Wang. P. Zhu, and B. Sheng, “Cell edge performance of
To find gi (r, θ), let us denote by Θi (θ) the angle: one cellular mobile systems,” IEEE J. Sel. Areas Commun., vol. 29, no. 6,
line between the UE and the center of BS0 and the other pp. 1139-1150, Jun. 2011.
[3] S.-Y. Kim, S. Ryu, C.-H. Cho, and H.-W. Lee, “Performance analysis of
line between the center of BSi and that of BS0 . When θ is a cellular network using frequency reuse partitioning,” Perform. Eval.,
randomly drawn between π/6 and π/2, we can find Θi (θ) for vol. 70, no. 2, pp. 77-89, Feb. 2013.
the first tier as follows: [4] D. Lopez-Perez, X. Chu, A. Vasilakos, and H. Claussen, “Power
 minimization based resource allocation for interference mitigation in
π/3 − θ if π/6 ≤ θ ≤ π/3, OFDMA femtocell networks,” IEEE J. Sel. Areas Commun., vol. 32,
Θi (θ) = (42)
θ − π/3 if π/3 ≤ θ ≤ π/2, no. 2, pp. 333-344, Feb. 2014.

0018-9545 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2016.2541172, IEEE
Transactions on Vehicular Technology

13

[5] H. Dahrouj and W. Yu, “Coordinated beamforming for the multicell [29] C. Yao, Q. Wu, Y, Xu, and W, Yin, “Sequential Channel Sensing in
multiantenna wireless system,” IEEE Trans. Wirel. Commun., vol. 9, no. Cognitive Small Cell Based on User Traffic,” Commun. lett., vol. 19,
5, pp. 1748-1759, May. 2010. no.4, pp. 637-640. Apr. 2015.
[6] J. Denis, M. Pischella, and D. L. Ruyet, “A Generalized Convergence [30] Q. Bi, Y. Yang, Y. Yuan and Q. Zhang, “Performance and Capacity
Criterion to Achieve Maximum Fairness Among Users in Downlink of Cellular OFDMA Systems with Voice-Over-IP Traffic,” IEEE Trans.
Asysnchronous Networks Using OFDM/FBMC,” IEEE Commun. Lett., Veh. Technol., vol 57, no.6, pp. 3641-3652, Jun. 2008.
vol. 18, no. 11, pp. 2003-2006, Nov. 2014. [31] S. Castellanos-Lopez, F. A. Cruz-Perez, M. E. Rivero-Angeles, and G.
[7] M. Sawahashi, Y. Kishiyama, A. Morimoto, D. Nishkawa, and M. Hernandez-Valdez, “Joint connection level and packet level analysis of
Tanno, “Coordinated multipoint transmission/reception techniques for cognitive radio networks with VoIP traffic,” IEEE J. Sel. Areas Commun.,
LTE-Advanced,” IEEE Trans. Wirel. Commun., vol. 17, no. 3, pp. 26- vol. 32, no. 3, pp. 601-614, Mar. 2014.
34, Jun. 2010. [32] V. Paxson and S. Floyd, “Wide-area traffic: The failure of Poisson
[8] S. M. Alamouti, “A Simple Transmit Diversity Technique for Wireless modeling,” IEEE/ACM Trans. Netw., vol. 3, no. 3, pp. 226.244, Jun.
Communications,” IEEE J. Sel. Areas Commun., vol. 16, no 8, pp. 1451- 1995.
1458, Oct. 1998.
[9] A. Ghosh, J. Zhang, J. G. Andrews, and R. Muhamed, “Fundamental
limits of cooperation,” IEEE Trans. Inf. theory, vol. 59, no. 9, pp. 5213-
5226, Sep. 2013.
[10] D. Lee, B. Clerckx, E. Hardouin, D. Mazzarese, S. Nagata, K. Sayana,
and H. Seo, “Coordinated multipoint transmission and reception in
lte-advanced: deployment scenarios and operational challenges,” IEEE
Commun. Mag., vol. 50, no. 2, pp. 148-155, Feb. 2012. Seung-Yeon KIM received the Ph.D. degree
[11] 3GPP, “Coordinated multi-point operation for LTE physical layer as- in electronics and information engineering in 2012 from
pects,” TR 36.819, Tech. Rep., Sep. 2011.
Korea University, Korea. He is currently an assistant professor
[12] K. M. S. Huq, S. Mumtaz, F. B. Saghezchi, J. Rodriguez, and R. L.
Aguiar, “Energy efficiency of downlink packet scheduling in CoMP,” in the Department of Computer and Information Science.
Trans. Emerging Tel. Technol., vol. 26, no. 2, pp. 131-146, Feb. 2015. He has participated in various research projects involving
[13] H. Lee, S. Kim, and S. Lee, “Combinatorial Orthogonal Beamforming an IMT-Advanced system, cognitive radio, and interference
for Joint Processing and Transmission,” IEEE Trans. Commun., vol. 62,
no. 2, pp. 625-637, Feb. 2014. management. His research interests include queueing theory,
[14] V. Jungnickel, T. Wirth, M. Schellmann, T. Haustein, and W. Zirwas, coding theory, and performance evaluation of communication
“Synchronization of cooperative base stations,” in Proc. IEEE ISWCS, networks.
pp. 329-334, Oct. 2008.
[15] L. Thiele, V. Jungnickel, and T. Haustein, “Interference Management
for Future Cellular OFDMA Systems Using Coordinated Multi-Point
Transmission,” IEICE Trans. Commun., vol.E93-B, no. 12, pp. 3228-
3237, Dec. 2010.
[16] R. Tanbourgi, S. Singh, J. G. Andrews, and F. K. Jondral, “A Tractable
Model for Non-coherent Joint-Transmission Base Station Cooperation,”
IEEE Trans. Wirel. Commun., vol. 13, no. 9, pp. 4959-4973, Sep. 2014.
[17] J.-B. Seo, S.-Y. Kim, and C. M. Leung, “Outage Probability characteriza- Choong-Ho CHO received B.S. and M.S.
tion of CoMP-Joint Transmission with Path-loss and Ralyleigh Fading,”
IEEE Commun. lett., vol. 19, no. 1, pp. 78-81, Jan. 2015. degrees in industrial engineering from Korea University in
[18] G. Nigam, P. Minero, and M. Haenggi, “Coordinated Multipoint Joint 1981 and 1983, respectively. He received M.S. and Ph.D.
Transmission in Heterogeneous Networks,” IEEE Trans. Commun., vol. degrees in computer science from the Institute National des
62, no. 11, pp. 4134-4146, Nov. 2014.
[19] R. Tanbourgi, S. Singh, J. G. Andrews, and F. K. Jondral, “Analysis of Sciences Appliques, Lyon, France, in 1986 and 1989, re-
Non-coherent Joint-Transmission Cooperation in Heterogeneous Cellu- spectively. He is currently a professor at Korea University.
lar Networks,” in Proc. IEEE ICC, pp. 5160-5165. Jun. 2014. His research interests include IT convergence technology,
[20] A. H. Sakr and E. Hossain, “Location-Aware Cross-Tier Coordinated
Multipoint Transmission in Two-Tier Cellular Networks,” IEEE Trans.
Building Energy Management System, 4G/5G mobile/wireless
Commun., vol. 13, no. 11, pp. 6311-6325, Nov. 2014. networks, D2D communication, small cell and SDN.
[21] F. Baccelli and A. Giovanidis, “A Stochastic Geometry Framework
for Analyzing Pairwise-Cooperative Cellular Networks,” IEEE Trans.
Commun., vol. 14, no. 2, pp. 794-808, Feb. 2015.
[22] Y.-S. Yeh, J. C. Wilson, and S. C. Schwartz, “Outage probability in
mobile telephony with directive antennas and macrodiversity,” IEEE
Trans. Veh. Technol., vol. 33, no.3, pp. 123-127, Mar. 1984.
[23] L. Rong, S. E. Elayoubi, and O. B. Haddada, “Performance Evaluation
of Cellular Networks Offering TV Services,” IEEE Trans. Veh. Technol.,
vol. 60, no. 2, pp. 644-655, Feb. 2011.
[24] D. Lopez-Perez, A. Ladanyi, A. Juttner, H. Rivano, and J. Zhang,
“Optimization Method for the Joint Allocation of Modulation Schemes,
Coding Rates, Resource Blocks and Power in Self-Organizing LTE
Networks,” in Proc. IEEE INFOCOM, pp. 111-115, Apr. 2011.
[25] S. C. Schwartz and Y. S. Yeh, “On the Distribution Function and
Moments of Power Sums with Lognormal Components,” Bell Syste.
Tech. J., vol. 61, no. 7, pp. 1441-1462, Apr. 1984.
[26] T. S. Rappaport, Wireless Communications: Principles and Practice.
Upper Saddle River, NJ: Prentice-Hall, 1996.
[27] S-E. Elayoubi, O. B. Haddada, and B. Fourestie, “Performance Eval-
ualtion of Frequency Planning Schemes in OFDMA-based Networks,”
IEEE Trans. Wirel. Commun., vol. 7, no. 5, pp. 1623-1633, May. 2008.
[28] M. K. Karry, “Analytical Evaluation of QoS in the Downlink of OFDMA
Wireless Cellular Networks Serving Streaming and Elastic Traffic,”
IEEE Trans. Wirel. Commun., vol. 9, no. 5, pp. 1799-1807, May. 2010.

0018-9545 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

Você também pode gostar