Sunteți pe pagina 1din 6

Dimensioning X

2
backhaul link in LTE networks
Alexandre Blogowski and Olivier Klopfenstein and Benjamin Renard
Orange Labs
38 rue du general Leclerc, 92130 Issy-les-Moulineaux, France
e-mail: alexandre.blogowski@orange.com
AbstractA central objective in designing Long Term Evolu-
tion networks (LTE) is to provide a better quality of experience
for the end user. One of the implemented enhancements is a
direct connection between neighbouring base stations through
X2 interface. This interface enables mobility management and
the transfer of the users data in case of a handover. The aim
of this paper is to study the impact of user mobility on capacity
planning and to provide an analytical model to dimension the
X2 bandwidth under certain performance objectives.
I. INTRODUCTION
LTE networks are widely considered as one solution for
future high-capacity mobile communications. A higher bit
rate, a wider cell coverage, a lower latency, an enhanced and
seamless mobility management... are some of new features of
LTE networks [2], [9], [11]. In LTE architecture, the space is
partitioned into cells, each cell being attached to a base station.
Each base station (also called eNB) provides air interface to a
user equipment (UE) and receives trafc from the core network
through a so called S
1
interface. When a UE moves from
one cell to another (handover process), its data are transferred
through a X
2
interface that connects two neighbouring eNBs.
As we will only consider the path between the eNB and the
rst intermediate node for S
1
, and the path between two eNBs
for X
2
, we will refer to S
1
and X
2
links (see Fig. 1). From
the operational point of view, LTE network planning would
be easier if a simple relation could be established between the
capacities needed on S
1
and X
2
links.
Fig. 1. S
1
and X
2
interfaces.
Main uses of the X
2
interface include intra-LTE mobility
support (handover process), inter-cell interference coordina-
tion and load management. While load management can be
considered as requiring a constant and negligible bandwidth,
handover support and interference coordination bandwidth
depend on the cell state. As interference coordination is not
likely to be used in homogeneous LTE networks, this paper
only investigates the bandwidth needed for handover support.
For this use, trafc on the X
2
link can be generated by
control or user plane. However, it has been demonstrated in
[14] that control plane trafc for X
2
is negligible compared
to user plane trafc. We thus consider only user plane trafc
in this paper. X
2
trafc can also be divided between outgoing
trafc (UEs leaving the cell) and ingoing trafc (UEs entering
the cell). As in an homogeneous network those trafcs are
equal, we only deal with outgoing trafc. Finally, since X
2
and S
1
links shares the same physical link (see Fig. 1), we
have to dimension a single X
2
link per site.
Let us note that X
2
dimensioning has already been inves-
tigated in [14]. Using Erlang formula, the authors propose
dimensioning rules for X
2
link. However, although this so-
phisticated approach may be closer to real-life network, it may
be unnecessarily accurate at the planning scale, where global
assumptions are made to draw future capacity investments. We
propose here a simpler analytical model which can be used
easily by network planners and in case a relation between S
1
and X
2
links can easily be found. Furthermore, the results we
obtain appear to be close to those of [14].
The dimensioning of the X
2
link is driven by a target quality
of service in case of handover. Hence, one of the critical
ingredients for this analysis is a model for UE mobility. This
aspect of mobile networks has been extensively studied in the
literature. A mapping of various mobility models, used for
wireless systems, is given in [1]. To the best of our knowledge,
most of the existing papers dealing with the handover problem
only involve a single UE. In this context, many precise
analytical models are proposed, such as random waypoint [7],
smooth random mobility [1] or one-moment model [3]; some
other approaches are more geometric [13]. Several simulators
have been proposed, see [5], [13]. Based on a uid mobility
model, [14] is one of the rare papers dealing with a population
of UEs and its impact on network dimensioning. Here, we
describe rst a simple mobility model for a single UE, and
extend it to a population of UEs. This approach is natural and
simplies the X
2
capacity analysis.
The paper is organized as follows. In Section II, a mobility
model and its impact for X
2
dimensioning are described for
a population of UEs. Section III investigates the multi service
aspect and the effects of transport protocols on the proposed
improvements. In Section IV a comparison with an other
model [14] is examined. Finally, numerical tests are provided
in Section V where a numerical analysis of the X
2
/S
1
ratio is
2801
performed. Section VI concludes this paper.
II. DIMENSIONING THE X
2
LINK
A. The case of a single UE
A UE can move from one cell to another during the same
communication. Its probability of leaving a cell depends on
cell size, communication duration and mobility parameters
(speed, direction...). Each network handover process requires
some network resources on the X
2
link to route the commu-
nication to the target eNB.
We consider rst the case of a single UE in a given radio
cell based on the same assumptions as in [3], [5], [6], [7],
[10], to establish an analytical mobility model. The cell is
assumed circular with radius R. Within the latter, the UE
keeps the same constant velocity v and the same direction
(moving in a straight line). Let be the constant duration of
a communication. The case where speed and direction of the
UE are not constant within the cell, is treated in [1], [5], [6],
[7]. During its communication, the UE travels a distance v.
If v > 2R, it leaves the cell with probability 1. If v 2R,
the probability of exit depends on its starting position in the
cell and its direction. Under the assumption that this starting
point is uniformly distributed on the cell, the dotted disk in
Figure 2 represents the horizontally translated initial cell by
distance v. Furthermore, as the circle is invariant under
rotation, the selection of the direction (in straight line) has no
inuence in our further analysis. Hence, the probability that
the UE stays in the cell is the ratio between the intersection
of the two disks in Figure 2 (the shaded area S) and the area
of a disk, that is:
P
stay
=
Area(S)
R
2
(1)
Fig. 2. The shaded area represents remaining UEs in the cell.
A simple calculation shows that, for
v
2R
1:
P
stay
=
1

_
_
2 arccos
_
v
2R
_

v
R

1
_
v
2R
_
2
_
_
(2)
Note again that if v = 2R, then P
stay
= 0, and if v = 0,
then P
stay
= 1. In the remaining, let P
exit
= 1 P
stay
. For
small v/2R, Taylor expansion of P
exit
is:
P
exit
=
2v
R
+ o
_
v
2R
_
(3)
Parameters Description Unity
R Radius of the cell m
v UE velocity m/s
Duration of a communication s
Period during which data are stored in the buffer ms
Q Buffer size and data in transit Mb
Delay constraint ms
D
X
2
X
2
link capacity for a single UE Mb/s
d UEs throughput Mb/s
N Number of UEs
s A time slot
k Simultaneous outputs
P
(k,N)
Probability of k outputs among N %
C
X
2
X
2
link capacity Mb/s
C
S
1
S
1
link trafc Mb/s
B
S
1
Bandwidth used on the S
1
link Mb/s
k

min{k N|P
exit

N1
j=k
P
(j,N1)
p}
p Percentage of congestion %
Size of the reordering buffer of RTP ms
C Number of classes of service
k Handover distribution
Pr(k, n) Probability of having a handover distribution k %
Pc(i, k)
Probability that a class-i user
experiences congestion
%
TABLE I
NOTATION TABLE
Therefore, under the above assumption, P
exit
depends linearly
on v and . Replacing v and with their average values,
the above approximation is exactly the value of the mean
number of handover during a communication, as given in [12]
for the Random WayPoint mobility model. This is perfectly
understandable as if a UE is not likely to do more than one
handover, the mean number of handover is the probability of
leaving the current cell.
We already know that a UE can move from one cell to
another during the same communication with probability P
exit
given by equation (2). This causes, for a time , an interruption
of the radio link between the transmitting eNB and the UE.
During this interruption, the source eNB should transfer to
the target eNB an amount Q of data over the X
2
link. This
amount is composed of data buffered at the source eNB and
ongoing data which continue to arrive via the S
1
link. Q is
proportional to the UEs throughput d on S
1
link and we note
= Q/d. Consequently, to ensure transparency for the UE
during the handover process, we need to transfer, on the X
2
link, an amount Q of data in less than seconds. Hence, for
every UE leaving the cell, we need a capacity of:
D
X2
=
d

(4)
B. Dimensioning X
2
for a population
We assume that all UEs have the same mobility and trafc
characteristics (e.g same v, maximum bit rate...) and the same
communication duration . We also assume constant the
expected maximum number N of UEs in the cell and consider
it as the number of active UEs within the cell at any time. This
assumption is considered as providing conservative results.
When a UE leaves the cell, a capacity of D
X2
must be
reserved on the X
2
link. We focus on the probability that the
UE goes out from the cell at a given moment. To do this,
2
2802
we discretize the duration of the communication in /
segments of size (time slots). For v/2R small enough,
the distribution of exit time of a UE has almost a linear
distribution function, i.e.
P
exit
( t)
2vt
R
, t (5)
We can therefore consider that this distribution is uniform over
the interval [1, ] and that the probability of exit during a time
slot s with duration is:
P
exit
(s) =
2v
R
(6)
The probability that a UE goes out on a given slot s is
P
exit
(s) = 2v/R / = P
exit
/, and the probability
that there are exactly k outputs on this slot s is thus given by:
P
(k,N)
= C
k
N

_
P
exit

_
k

_
1 P
exit

_
Nk
(7)
If P
exit
= 1, we truncate the duration of the communication
to the maximum time spent inside the cell, or
max
=
2R
v
.
Note that the above discrete model can be approximated
by a Poisson distribution with parameter . Indeed, eqn. (7)
follows a binomial law with parameters N and P
exit
/,
but for N large enough and P
exit
/ small enough, we have
= NP
exit
/ (8)
From the network point of view, having exactly k outputs
at the same time (i.e. on the same time slot) corresponds to a
throughput of k D
X2
= k Q/ which must be transferred
on the X
2
link. We denote P
(network)
c
(k) the probability that
the X
2
link needs a capacity of less than k Q/. We have:
P
(network)
c
(k) =
k

i=0
C
i
N

_
P
exit

_
i

_
1 P
exit

_
Ni
(9)
Now, we wish to determine the dimensioning of the X
2
link
in order to guarantee a certain target quality of service (QoS)
for all users. It is during the communication time that the
user can perceive the network congestion (overall slowdown
of the network). In our model, we consider that, in case of
congestion on a slot s, ows are blocked and discarded at the
end of the slot. Excess data are thus lost.
If the network has been dimensioned for k simultaneous
outputs, the user will experience congestion if he leaves the
cell on the same time slot as k (or more) other users. We
denote P
c
(k) the probability that at least k+1 UEs (including
the reference UE) leave the cell on the same time slot. We
have, for k N 1:
P
c
(k) =

s=1
_
_
_
P
exit

N1

j=k
P
(j,N1)
_
_
(10)
= P
exit

N1

j=k
P
(j,N1)
(11)
We have thus characterized the quality perceived by a user
(probability of congestion) for a given dimensioning of the
X
2
link. Conversely, we can deduce the capacity of the X
2
link corresponding to a given target QoS.
We are now focusing on the viewpoint of the user and on
a target quality of service. We will denote by p the maximum
congestion probability on the X
2
link, i.e. the user experiences
the congestion due to its mobility with a probability no greater
than p. The capacity C
X2
of the X
2
link is then
C
X2
= k

D
X2
= k

(12)
where
k

= min
_
_
_
k N

P
exit

N1

j=k
P
(j,N1)
p
_
_
_
(13)
III. MULTI-SERVICE MODEL
A. Extending the single service model
The previous model considers only one type of ows. How-
ever, in IP networks, various type of ows can be found, with
different trafc characteristics (data rate, mean communication
duration...) and needs in term of QoS. Assume now that we
have C classes of service. Each class i has a data rate d
i
, a
mean communication duration
i
, an amount of data Q
i
to
transfer in case of handover and p
i
will denote the maximum
congestion probability. We still consider a xed number N
of active UEs and we also consider a xed number of active
UEs in each class. We denote by n = (n
1
, . . . , n
C
) the static
distribution of UEs among classes.
Let P
i
stay
be the probability that a class-i UE will stay in
its cell during the communication. We have the same equation
than eqn. (2) with
i
instead of . We also have P
i
exit
=
1 P
i
stay
. Denote by P
r
(k, n) the probability of having a
handover distribution k = (k
1
, . . . , k
C
):
P
r
(k, n) =
C

i=1
P
i
(ki,ni)
(14)
where P
i
(ki,ni)
is deduced from eqn. (7). Note that

kHn
P
r
(k, n) = 1 where H
n
= {k|i, k
i
n
i
} denotes
the possible handover distributions.
As in the single service model, we consider the point of
view of a user. If the X
2
link has been dimensioned with a
capacity C
X2
, the user will experience congestion if he leaves
the cell on a slot such that the data rate needed by handover
UEs will be higher than C
X2
. We will denote by P
c
(i, k) the
probability that a class-i user experiences congestion if the X
2
link has been dimensioned to allow a simultaneous handover
distribution k:
P
c
(i, k) = P
i
exit
P
_
_
_
total trafc on X
2
>
C

j=1
k
j

j
d
j

_
_
_
= P
i
exit

l
i
k
P
r
(l, n e
i
) (15)
3
2803
where e
i
is the i
th
unitary vector of N
C
and where

i
k
=
_
_
_
l H
nei

j=1
l
j

j
d
j
+
i
d
i
>
C

j=1
k
j

j
d
j
_
_
_
We dene k

as the distribution that minimizes the capacity


needed on the X
2
link (and thus minimizes

i
k
i

i
d
i
/)
while satisfying the conditions i, P
c
(i, k) p
i
. The needed
capacity on X
2
is then
C
X2
=
C

i=1
d
i

i
(16)
B. Lowering the data rate on the X
2
link
When a congestion occurs, the following slot is likely to
be empty or less used. If we postpone some data from the
congested slot to the following slot (by reducing the UEs
data rate on the X
2
link), it will be possible to avoid this
congestion. Until now, all the UEs data was transferred to the
target eNB before the recovery of the radio link in the target
cell. However, it is highly likely that the UE will have the same
data rate on both radio links (as in both cases the UE is on cell
edge). The UEs data will thus be entirely buffered at the target
eNB and then send to the UE with the same data rate d
i
on the
radio link than via the previous eNB. However, if both actions
(transfer on X
2
and radio links) are done simultaneously, the
user wont experience additional congestion.
As a consequence, if the rst packets transferred on the X
2
link arrive before the recovery of the radio link and if the
data rate on the X
2
is equal or higher than d
i
, the user wont
suffer from congestion. The rst condition is fullled as the
end-to-end delay on the X
2
interface is between 10 and 20
ms while the handover duration ranges from 25 to 50 ms.
The second condition implies that we can lower the needed
UEs data rate on the X
2
link from
i
d
i
/ to d
i
. The needed
capacity becomes
C
X2
=
C

i=1
d
i
k

i
. (17)
More details and improvements will be given in Section V.
Nonetheless with our numerical inputs (i,
i
= 100 ms and
= 25 ms), this enhancement divides the needed X
2
capacity
by 4.
C. Impact of transport protocols
The preceding enhancement has however some drawbacks.
If we consider that the target eNB starts receiving packets
from its S
1
link as soon as the radio link is recovered, this
ow will collide with the X
2
ow if the latter is still active
(which couldnt happen with the original model). The UE will
thus receive packets in the wrong order and we are going to see
the impact of two transport protocols on this medley. A simple
solution would be to reorder packets at the target eNB, but this
would need additional network features. We will thus study the
impact of transport protocols when there is no reordering at
the eNB level.
1) TCP: Transmission Control Protocol (TCP) is the main
transport protocol over the internet and is used for elastic ows
(http, ftp...). As supposed in paragraph III-B, the data rate
on X
2
is higher or equal to the data rate on S
1
and thus
there wont be more than one packet from S
1
between two
X
2
packets, as shown on the following table:
Seq. number 0 1 2 10 3 11 4 ...
Link X
2
X
2
X
2
S
1
X
2
S
1
X
2
ACK(n) sent 0 1 2 2 3 3 4
TCP needs at least three consecutive ACKs with the same
sequence number to start a fast-recovery (TCP Reno) or slow-
start (TCP Tahoe) process. TCP wont thus have any impact
on our enhancement.
2) RTP: Real Time Protocol (RTP) is used for real-time
ows (VoIP, video streaming, real-time gaming...) and can be
used over UDP (User Datagram Protocol). We will suppose
that this protocol goes along with a reordering buffer with size
(in seconds): it is possible to reorder two consecutive packets
if their distance is lower than . The end user will experience
congestion if the collision time between ows from S
1
and
X
2
is higher than ; that is if the overall transmission time on
the X
2
link is higher than +. To experience congestion, a
class-i user must thus have an X
2
data rate D
i
X2
such that
D
i
X2
< d
i

i
+
(18)
If i, +
i
, RTP wont have any impact on our
enhancement. Consider now the case where i| + <
i
.
If we suppose that the eNB has a proportionally fair
scheduler, a class-i UE has a data rate d
i
C
X
2

j
kjdj
and the
condition of eqn. (18) becomes:
C

j=1
k
j
d
j
> C
X2
+

i
(19)
The congestion probability, given by (15), is now:

P
c
(i, C
X2
) = P
i
exit

l
i
C
X
2
P
r
(l, n e
i
) (20)
where:

i
C
X
2
=
_
_
_
k H
nei

j=1
k
j
d
j
+ d
i
> C
X2
+

i
_
_
_
(21)
IV. COMPARISON WITH WIDJAJA AND LA ROCHES
MODEL [14]
In [14], the authors propose a richer stochastic model. Based
on a trafc model using the Kaufman-Roberts formula [8],
they give the number of UEs leaving the cell per unit time.
As in our model, they consider that in case of congestion
ows are blocked. However, they drop all excess ows who
arrive while we accept these ows during a time slot s before
stopping them at the end of this slot.
4
2804
Fig. 3. Blocking probabilities for both models.
Their mobility model is based on the uid ow model,
according to which the UE crossing rate out of an enclosed
region with perimeter length L is vL/ where is the UEs
density. If we adapt their model to a model with a xed number
of active UEs, the offered load for class-i UEs becomes:
a
i
=
2vn
i
R
(22)
We can notice that if we replace P
exit
by the mean number
of handover during a communication 2v
i
/R [12] in eqn.
(8), we obtain the same parameter. While we consider in
our model the probability of leaving the cell, Widjaja and
La Roche consider the mean number of handover during a
communication. Another major difference between stochastic
models is that Widjaja and La Roche use the classical multi-
Erlang model developed by Kaufman and Roberts [8] in
place of a simpler comprehensible model and easy to use by
networks planners. Figure 3 compares blocking probabilities
for both models, with the same mobility model (which is our
mobility model). In this gure, BKR denotes our enhanced
model and we have taken C
X2
= 1280 Kbps. We can observe
that blocking probabilities are really close, and thus the same
X
2
bandwidth is required in both models.
In [14], the authors also propose a model with batched
arrivals: UEs are moving in groups (batches) with random
size, and they observe the impact of such a behaviour on
the X
2
link. A state of the art of the Erlang multi-rate loss
model with batched arrivals can be found in [4]. However, no
model has been developed for batches with different classes of
service, which led Widjaja and La Roche to consider a single
service class for their batch study. A complete study of batches
arrivals needs a consequent development. Another reason for
not studying deeply the batches arrivals lies in the fact that
even if active UEs are travelling in batches, it is unlikely that
they will all leave the cell on the same slot. If we consider a
high speed train (300 km/h) or a commuter train (80 km/h),
a 25 ms slot represents approximately 2 meters for the rst
one, and only 0.5 meter for the second. These distances make
it unlikely for few active UEs from the same train to leave
the cell simultaneously and induce congestion. For instance,
with the classes of service {VoIP, video, data}, if C
X2
= 1024
kbps, we can have k

= {0, 0, 1} or {2, 3, 0} or {32, 0, 0}...


V. NUMERICAL ANALYSIS
For this chapter, we use the enhanced model (paragraph
III-B) and the following numerical inputs: R = 250 m and
v = 12 km/h. The various classes of service are {VoIP,
video, data} with a distribution of {25%, 25%, 50%}, with
corresponding data rates of {32, 320, 1024} Kbps and com-
munications duration of {100, 200, 5} seconds. We also
consider i,
i
= 100 ms, and = 25 ms.
A. Variation of the number of active UEs N
Until now, the number of active UEs has always been
considered steady. In this section, we will consider that N
follows a Poisson distribution. Table II shows the maximum
number of active UEs that we can have for a maximum
congestion probability set to 0.01%. We can observe that those
numbers are very close, which conrms our hypothesis of a
steady number of active UEs.
C
X
2
(Kbps) Nmax original amax with Poisson distribution
1024 1 1
1376 25 24
2080 135 134
TABLE II
IMPACT OF A POISSON DISTRIBUTION OF THE NUMBER OF ACTIVE UES
WITH THE ENHANCED MODEL
B. Enhanced model and impacts of transport protocols
In what follows, numerical inputs will be the same as before.
We also consider = 25 ms (real-life systems have usually a
bigger , this small value is used for showing the impact of
RTP in a worst-case scenario).
Fig. 4. The impact of RTP on congestion probabilities for N = 100.
Fig. 4 shows congestion probabilities in terms of X
2
band-
width for a population N = 100 UEs, for the enhanced
model and with the impact of RTP. The original multi-service
model has the same congestion probabilities than the enhanced
model, but for a X
2
bandwidth multiplied by
i
/ = 4. We
can notice that with a very small , RTP has a non negligible
impact on X
2
dimensioning.
For example, a maximum congestion probability p
i
= 0.5%
for all classes requires a X
2
bandwidth of 1344 Kbps for the
enhanced model and 2650 Kbps if we take into account RTP.
5
2805
C. Variation of the maximum congestion probability
It could be interesting to see the impact of a rise of
parameter p
i
(here considered as the same for all classes)
on the needed X
2
bandwidth. Fig. 5 shows this impact for
the enhanced model (without RTP impact) for N = 100.
We observe that the X
2
bandwidth still vary by steps; this
is consistent with the bandwidth evolution illustrated on Fig.
4 for separate services.
Fig. 5. Impact on the X
2
dimensioning of the variation of the maximum
congestion probability
D. Ratio X
2
/S
1
From an operational point of view, nding a relation be-
tween S
1
and X
2
capacities can be very useful. However,
the presence of congestion on the S
1
link changes the actual
bit rate and thus the needed X
2
bandwidth. In the absence
of congestion (

i
n
i
d
i
C
S1
), each class-i UE receives its
maximum bit rate d
i
. The ratio X
2
capacity over bandwidth
used on the S
1
link, denoted by C
X2
/B
S1
, is then

i
k

i
dii/

i
nidi
.
In case of congestion (

i
n
i
d
i
> C
S1
), assuming a propor-
tional share of the capacity, each UE receives on S
1
a bit rate
of d
i
C
S
1

j
djnj
, but the ratio C
X2
/B
S1
remains the same; it is
thus independent of the S
1
link state. This explains why we
were able to consider a non-congested S
1
link.
Fig. 6. Evolution of the C
X
2
/B
S
1
ratio according to v and B
S
1
Fig. 6 shows the evolution of the C
X2
/B
S1
ratio according
to the UEs speed v and different values of bandwidth used
on the S
1
link B
S1
. We can notice that the ratio is quite small
and the X
2
interface can thus be carried with the S
1
interface
on the same physical link. Moreover, if the value for B
S1
is
given, we can easily nd the needed X
2
bandwidth.
VI. CONCLUSION
In this paper, we provide an analytical model that quanties
the impact of UE mobility on the dimensioning of a LTE
mobile network. Based on QoS requirements, dimensioning
rules are given for the X
2
link, and also for relating directly the
X
2
capacity to the S
1
trafc. The simplicity of the approach
makes it easy to integrate in existing network planning tools.
Indeed, network planners must just have the number of active
UEs in the cell and the dimension of the S
1
link to estimate
the capacity needed on the X
2
link. The comparison with the
richer model proposed in [14] shows a quite small differences
on the results. Furthermore, our discrete model can be approx-
imated by a Poisson distribution, allowing a verication of the
assumptions made previously.
Future work include a study where all handover ows would
be considered as elastic and never be blocked and the impact
of radio interface conditions on handover rates. We also plan
to investigate the dimensioning of the X
2
interface in LTE-
Advanced networks, where inter-cell interference coordination
(ICIC) schemes will be used, thus increasing the load on the
X
2
interface.
REFERENCES
[1] Christian Bettsetter. Mobility modeling in wireless networks: Catego-
rization, smooth movement, and border effects. Mobile Computing and
Communications Review, 5(3):5566, 2001.
[2] Yishi Chen. State of the art of LTE architecture v1. Technical report,
France Telecom Group, June 2007.
[3] Edward Chlebus and Wieslaw Ludwin. Is handoff trafc really
poissonian ? IEEE International Conference on Universal Personal
Communications, pages 348353, 1995.
[4] Ioannis D. Moscholios and Michael D. Logothetis. The erlang multirate
loss model with batched poisson arrival processes under the bandwidth
reservation policy. Computer Communications, 2010.
[5] Roch A. Guerin. Channel occupancy time distribution in a cellular radio
system. IEEE Transactions on Vehicular Technology, VT-35(3):8999,
August 1987.
[6] Daehyoung Hong and Stephen S. Rappaport. Trafc model and
performance analysis for cellular mobile radio telephone systems with
prioritized and nonprioritized handoff procedures. IEEE Transactions
on Vehicular Technology, VT-35(3):7792, August 1986.
[7] Esa Hyytia and Jorma Virtamo. Random waypoint mobility model in
cellular networks. Wireless Network, 13:177188, 2007.
[8] Joseph Kaufman. Blocking in a shared resource environment. IEEE
Transactions on Communications, 1981.
[9] Pierre Lescuyer and Thierry Lucidarme. Evolved Packet System (EPS).
Wiley, 2008.
[10] Alcatel Lucent. LTE S1/X2 dimensioning methods. Technical report,
January 2009.
[11] Ph. Niger St. Del Burgo F. Jounay R. Kortebi B. Michau B. Lemoine
S. Jobert M. Duprez, F. Le Clech. LTE transport architectural framework
and scenarios. Technical report, France Telecom Group, December 2009.
[12] Mahmoud M. Zonoozi and Prem Dassanayake. User mobility modelling
and characterization of mobility patterns. 1997.
[13] Sanjiv Nanda. Teletrafc models for urban and suburban microcells:
Cells size and handoff rates. IEEE Transactions on Vehicular Technol-
ogy, 42(4):673682, November 1993.
[14] Indra Widjaja and Humberto La Roche. Sizing X2 bandwidth for inter-
connected eNodeBs. IEEE Vehicular Technology Conference, September
2009.
6
2806

S-ar putea să vă placă și