Performance Analysis of GSM Networks with Intelligent Underlay-Overlay
Khalid Begain
, Gerg˝o Istv´an R´ozsa
, Andr´as Pfening
, Mikl´os Telek
Dept. of Computing, University of Bradford BD7 1DP, Bradford, West Yorkshire, England, UK, kbegain@bradford.ac.uk
Mat´av Hungarian Telecommunications Company Ltd., Budapest, Hungary, Rozsa.Istvan.Gergo@mail.pki.matav.hu
Nokia Telecommunications, Mobile Switching Budapest, Hungary, andras.pfening@ntc.nokia.com
Department of Telecommunications, Technical University of Budapest, H-1521 Budapest, Hungary, telek@hit.bme.hu
Abstract
The paper presents an analytical model for a GSM-based
cellular mobile network that applies the Intelligent Under-
Overlay (IUO) scheme to increase the capacity by increas-
ing the frequency reuse while maintaining the service qual-
ity. The IUO is a multi-layer cell structure that is based on
dividing the frequency band into the super layer and the reg-
ular layer frequency group. The super frequencies (chan-
nels) can be used by mobile stations with good C/I (car-
rier/interferer) ratio, while the regular frequencies can be
used over the whole cell. The use of IUO is expected to pro-
vide up to 40% gain of capacity [1]. In this paper, we study
the effect of various parameters on the performance of the
networks using IUO and provide practical planning support
based on the analytical results. The considered parameters
include network parameters like super area coverage and
mobile user mobility parameters like moving mobile ratio
and average mobile speed.
Keywords: Cellular mobile networks, GSM, IUO, Per-
formance analysis.
1 Introduction
The demand for wireless communication grows rapidly
nowadays. The capacity of cells in the existing digital cel-
lular mobile networks, like GSM, will have to meet the in-
creased demand. The network operators have to face the
problem, how to increase the capacity of an existing net-
work without noticable degradation of quality of service.
Different solutions can be foreseen. The most obvious so-
lution would extend the GSM band or increase the number
of serving channels or frequencies in an area. However the
overall available GSM spectrum is limited and is usually di-
vided between 2 or 3 network operators leaving a spectrum
of not more than 10 MHz for each operator. Another solu-
tion is to deploy more base stations (cell splitting), or to in-
This work has been partially supported by OTKA T 034972.
Correspondence author
troduce hierarchical structures like micro- and picocells [2],
[3]. This approach works well to a certain extent, however
the denser base station grid results in increased interference
that limits the quality and so the capacity in terms of soft
blocking. Furthermore, the additional network elements
like base stations and transmission network cost quite a lot.
The most promising way of capacity enhancement requires
minor investment while allowing more capacity with minor
or no degradation of service quality. Two methods shall be
mentioned here, the frequency hopping (FH, [4]) and the
intelligent underlay-overlay proposed by Nokia Telecom-
munications (IUO, [1]). (Recently Nokia proposed the IFH
technology, which combines IUO and FH. This is out of the
scope of the paper.) Both methods are similar in the sense
that they allow more frequencies to be used in the exist-
ing cells, that is increase the reuse factor, introducing minor
investment cost. As a result of the tighter reuse, the interfer-
ence in the network will be higher. The two methods differ
in how they cope with the increased interference. In FH,
the established speech connection is hopping on a number
of frequencies. A number of those frequencies are ”clean”,
that is do not suffer from serious interference, while the
other frequencies are interfered. Due to the averaging ef-
fect of the different coding principles, the noise brought by
the interfered frequencies is eliminated. The IUO princi-
ple is to make use of the measurements done by the mobile
station, MS. The MS always measures the strongest neigh-
bors of the serving cell in order to know when to make a
handover to a neighbor cell. This measurement data is used
to estimate the C/I conditions of the MS. If the estimated
C/I is good enough, the MS is assigned a (heavily reused)
so-called super frequency, while if the C/I is bad, a clean
regular frequency is assigned to the MS. This way the super
frequencies (channels) can be used by mobiles with good
C/I ratio, while the regular frequencies can be used over the
whole cell. The frequency band is divided into two groups,
a super layer and a regular layer frequencies, and a lower
frequency reuse factor, thus a smaller area of coverage is
assigned to the super layer.
1
The origin of IUO principle was reuse-partitioning. It
was presented in some papers [5, 6, 7]. [5] presented the
DCA (dynamic Channel Assignment) with a greedy algo-
rithm, and investigated 1 dimensional cellular radio system.
Simulation results were reported on IUO in [8, 9, 6]. In [10],
the performance of GSM network implementing IUO in
combination with Frequency hopping was studied by sim-
ulation. Some further improvements on the original IUO
scheme were suggested. The effect of these improvements
was reported in [11]. In this paper, we introduce an analyti-
cal model of GSM network implementing IUO and provide
results on the performance of the network taking into ac-
count many parameters like the coverage factor of the super
layer to the whole cell, the ratio of moving mobile stations,
and the speed at which they are moving. Our paper gives
practically useful view on the above mentioned parameters.
Network designers can directly apply the providedresults to
dimension their network with improved reuse of the spec-
trum in cellular radio systems.
With the applied analysis approach we calculate the net-
work performance parameters by considering a large ho-
mogeneous network composed by a lot of identical cells,
whose handover traffic to the neighboring cells is symmet-
ric. This assumption on the homogeneous network with
symmetric handover traffic allows us to calculate the net-
work performance based on the analysis of a single cell.
The analysis of inhomogeneous network with asymmetric
handover traffic of neighboring cells, that would require a
much more complex analysis approach, is out of the scope
of this paper.
In Section 2, we give a description on the operation of
IUO scheme. Section 3 summarizes the main assumption
on the studied systems and defines the main processes that
affect the performance of the system. The analytical model
will be defined in Section 4 with which the performance
analysis of a cell with IUO is carried out in Section 5. Some
planning issues of the obtained results is summarized in
Section 6 and the paper is concluded in Section 7.
2 Intelligent Underlay-Overlay
2.1 Principle
The IUO is a feature designed to allow a tighter fre-
quencyreuse for some of the available radio frequencies and
therefore achieve a higher network capacity. It implements
a two-layer network structure with a different reuse factor
for each layer. The underlay adds capacity, the overlay pro-
vides coverage. To maintain optimum capacity, the base
station assigns mobile traffic to either layer of the network
according to actual interference levels. The IUO solution
splits the available frequencyspectrum into two bands. One
consisting of frequencies that can only be used when a high
C/I ratio is ensured, the super frequencies, which are usu-
ally used only near the Base Transceiver Station (BTS). The
other band contains frequencies that can be used throughout
the whole cell, the regular frequencies.
Every IUO cell has regular and super Transmitter Re-
ceivers (TRXs). Regular frequencies completely cover the
cell. These frequencies can be reused by conventionalcrite-
ria, using safe hand-over bounds to provide low probabili-
ties for interference. Mobile stations are assigned to regular
frequencies at the boundary where C/I rate is under a spe-
cific level. Figure 1 shows the principle of the layer struc-
ture of IUO.
C/I bad
Super layer
C/I good
Regular layer
Figure 1. IUO principle
Super frequencies provide services in heavy traffic areas
(downtown) of the cell, where C/I rate is good (interference
free area). Using different C/I ratios, the coverage of the
super layer (super frequencies) can be controlled. If IUO is
combined with downlink power control, even better inter-
ference conditions can be maintained resulting in better call
quality.
2.2 Operation of IUO
We assume the call admission operation of GSM net-
work without IUO is known [12]. The IUO cell operation
is described separately for standing (non-moving)and mov-
ing mobile stations (MS). The IUO algorithm does not dis-
tinguish moving or not moving MSs, this separation is done
for a modeling assumption: we assume that a non-moving
mobile’s interference conditions will not change, and that
the mobile will not try any handovers, unlike the moving
mobile station.
For both moving and non-movingmobile stations, the call
request will first be served by a regular frequency, because
the C/I ratio of the call is not yet known. The C/I ratio is
calculated by comparing the downlink signal of the serv-
ing cell with the downlink signal of all neighboring cells,
that use the same super frequencies.
If the C/I value is better than a predefined ”C/I good”
threshold, then the connection will be served by a super
frequency, otherwise it will remain at the regular layer.
If this handover to the super frequency fails, because all
super frequencies are occupied, then the connection will
stay at the regular layer and try to get into the super layer
after a specific time.
2
For the connections of non-moving MS, the call will be
served at the same layer chosen according to the value of
C/I until its normal termination.
For the connections of the moving MS, the value of C/I ra-
tio may change after the setup of the connection. This im-
plies a number of hand-overpossibilities as follows:
1. For a connection going on a super frequency, the fol-
lowing hand-overs exist due to the movement of the
MS:
If the value of C/I ratio falls below the ”C/I good”
threshold, but there is another free super frequency
where the C/I ratio is appropriate, hand-over will be
done, and the call will be kept on the super layer.
If there is no free super frequency, or none of these has
an appropriate C/I ratio, the call gets back to regular
layer, or if all the regular frequencies are occupied, the
call will be lost.
2. For a connection going on a regular frequency, the
following handovers exist due to the movement of the
MS:
If the C/I value improves and reaches the ”C/I good”
and there is a free super frequency, the connectionwill
be moved to the super layer.
An intra-cell hand-over to regular frequency of an-
other TRX in the same cell.
An inter-cell hand-over to regular frequency of an-
other neighboring cell.
3. The case of direct inter-cell hand-overfrom a super fre-
quency to a neighboring cell is not considered in the
model, because we assume that the C/I ratio is deter-
mined by the distance of the MS fromthe (closest) BTS.
Hence, in our model, a super-regular hand-over always
preceeds the inter-cell hand-over if the coverage factor
of super layer is less than 100%. In the applied frame-
work it would also be possible to model direct inter-cell
hand-overfrom super layer based on a probabilistic rule
that describes the probability of direct inter-cell hand-
over from super layer to a neighboring cell as a function
of the coverage factor of super layer, but it is not con-
sidered in this paper.
Figure 2 summarizes all possible handovers in the IUO
cell.
3 Performance analysis
3.1 Modeling assumptions
For the sake of building an analytical model of the sys-
tem, we make the following assumptions:
Regular to Regular
Regular to Super
Super to Super
SUPER
REGULAR
Super to Regular
Intercell hand-over
Incoming intercell
hand-over
Figure 2. Possible handovers in the system
Only one cell is considered in the analysis. All the inter-
actions from the neighboring cells to the studied cell are
taken into account as an aggregate incoming hand-over
request process, which will first be served by regular fre-
quencies.
The studied cell implements IUO scheme with the super
layer coverage factor
of the cell area. Only one super
frequencygroup is assumed, i.e., the same coverage factor
is assumed for all super frequencies.
The Base Transceiver Station of the cell manages
reg-
ular and
super frequencies. Thus, the total number of
frequencies (channels) in the cell is
.
The cell contains moving and non-moving mobile sta-
tions. The ratio of the moving MSs is
.
The moving MSs are assumed to move in a uniformly dis-
tributed random direction and with a constant speed
.
Super and regular frequencies of the same cell are not dis-
tinguished. The hand-over between regular frequencies or
between super frequencies in the same cell are not taken
into account, since it does not modify the number of busy
frequencies of the layers.
The different processes that control the events in the sys-
tem are:
1. the new call arrival process is assumed to be a Poisson
process with rate
. We assume that the arrivals are uni-
formly distributed on the area of the cell and, therefore,
of the calls will have C/I value better than the ”C/I
good” threshold.
2. the aggregate incoming hand-over is assumed to be a
Poisson process with rate
.
3. the time needed for the calculation of the C/I parameter
for a new connection is assumed to have an exponential
distribution with mean of
! #"
.
3
4. the call holding time, i.e., the time for normal call
termination, is assumed to be exponentially distributed
with average duration of
.
5. The hand-over rate from super to regular frequencies
and vice versa resulting from the movement of the mov-
ing MS are calculated taking into account: the speed of
the MS
, the radius of the cell, the coverage factor
of the super layer, and the probability that the move-
ment of the MS will lead to increase or decrease the
value of C/I ratio. According to this, we assume that
the time needed for a moving MS to move from super
to regular area or from regular to super area is an expo-
nentially distributed random variable with mean value
or
, respectively.
6. Similarly, we assume the channel dwell time to be an
exponentially distributed random variable with mean
value
, which is calculated with similar consid-
erations to the previous point.
3.2 The model
With the above listed modeling assumptions the consid-
ered system (a cell) behaves as a Continuous Time Markov
Chain. At any time instant the state of the cell is determined
by the number of active calls of each class of frequencies,
so we define it as the vector:
"!
where
is the number of connections served by regular fre-
quencies and have C/I
#
”C/I good”,
is the number of connections with C/I
$
”C/I good”
that are served by regular frequencies, (there are two rea-
sons for this situation: new calls with C/I
$
”C/I good” are
servedby regularfrequencies for the period of C/I calcula-
tion; and calls with C/I
$
”C/I good” are served by regular
frequencies if all the super frequencies are occupied.)
is the number of connections served by super fre-
quencies.
Let
%
!
and
%&
!
denote the number
of occupied regular and total frequencies in the cell, respec-
tively. It is clear that a permissible state must satisfy the
conditions
%&
!('
and
%
!('
. Let
)
denote
the number of feasible states. Then, we can define
*
to
denote the state space of the system given that the states
are conveniently ordered from
+
,,,
).-
. The result-
ing model is thus homogeneousand irreducible on the finite
state space
*
and therefore the steady state distribution p
=
/
1032
,
4
5+
,,,
)6-
, exists, unique and can be com-
puted through the linear system of equations
798:
<;
and
=>@?ABDCEGF
SUPER (
>H?A
)
=ICEGFJBK>@?A
=IL
=IL
=IL
=MN
?
L
=OHPRQ
REGULAR (
C
>@?A
)
REGULAR (
C
CEGF
)
IF
S
>H?AUTWVX3Y
=
CEGFJBK>@?A
IF
S
>H?A[Z ZWVX Y
=
>H?ABDCEGF
LOSS IF
\
M
\[]
O@NG^
\
S`_
B
O@NG^
Y
S
C
CEHFa
C
>H?A
Z Zcb X3Y
Figure 3. Queuing model of IUO cell
dfe
g
0h[i
"0
, where the matrix Q denotes the infinitesimal
generator of the Markov chain.
The transition rates, the elements of matrix Q, can be
obtained from the analysis of the driving process in the sys-
tem. We follow the style used in the rule definition syntax
of the model specification language, MOSEL [13], which
is based on a ”Which state follows
"!
if...”
logic. MOSEL is a model specification and evaluation lan-
guage developed at the Department of Computer Science
IV, University of Erlangen, Germany. Figure 3 shows the
considered queuing model of the cell and Table 1 provides
the rules that determine the transition rates following this
style The Mosel description of the model is directly read-
able from this set of rules.
Event Condition Successor State Rate
New Call in reg.
j3kl`monqp"rUs tJkuGvw
Ix
tJyz{
x|}~
l
H
nJ
New Call in sup.
j
k
l`monqp"rUs t
kuGv
x
t
yz{
w
Ix|}~
R
Incoming HO
j
k
l`monqp"rUs t
kuGv
w
Ix
t
yz{
x|}~
D
good C/I calls
|R}~
ps t
kuGv
x
t
yz{
Ix|}~
w
t
yzI{DGI
Super to regular
j3kl`monqp"rUs tJkuGvw
Ix
tJyz{
x|}~
|R}~
yzI{@kuGv
j kl`monqp"rUs tJkuGvw
Ix
tJyz{
IxG|R}~
tJyzI{
KyzI{@kuGv
Soft blocking
j
k
l`monrUs t
kuGv
x
t
yz{
x|}~
|R}~
yzI{@kuGv
Regular to super
|R}~
p"1s t
kuGv
Ix
t
yz{
x|}~
w
t
kRuHvU
kuGvyz{
|}~
s tJkuGv
Ix
tJyz{Rw
IxG|R}~
tJkRuHv
kuGvyz{
Call termination
tkuGv
Ix
tJyz{
x|}~
tJkRuHv
K
t
kuGv
x
t
yz{
Ix|}~
t
yzI{
t
kuGv
x
t
yz{
x|}~
|R}~
Outgoing HO
t
kuGv
Ix
t
yz{
x|}~
t
kRuHvU
z
Table 1. Transition rules and rates
Parameter Value
Number of regular frequencies,
rs
28 (4 TRX)
Number of super frequencies,
1s
16 (2 TRX)
Super area coverage factor,
50% (var. of study)
Call holding time,
J
80s (exponential)
Time for C/I calculation,
J
HI
5s (exponential)
Cell radius,
r
3km
Ratio of moving MS,
t
50% (var. of study)
Speed of moving MS,
50km/h (var. of study)
Table 2. Default value of parameters
4 Performance analysis of a cell with IUO
In this section, we present some numerical results using
the defined model. Table 2 summarizes the default values
4
Offered Traffic [Erlang]
Loss probability
0 20 40 60 80 100 120 140
1e−3
1e−2
0.1
1.0
Super coverage factor
0.25
0.5
0.75
0.9
Figure 4. Loss probabilities
versus super area coverage
Offered Traffic [Erlang]
Utilization
0 20 40 60 80 100 120 140
0.4
0.5
0.6
0.7
0.8
0.9
1.0
Super coverage factor
0.25
0.5
0.75
0.9
Figure 5. Utilization for differ-
ent super area coverage
Offered Traffic [Erlang]
Loss probability
0 20 40 60 80 100 120 140
1e−6
1e−5
1e−4
1e−3
1e−2
0.1
1.0
Ratio of moving mobile
0% moving
25% moving
50% moving
75% moving
Figure 6. Loss probabilities
versus moving MS ratio
for the parameters used in the study. The use of different
values will be explicitly mentioned. We present results on
the call loss probability and the average utilization of chan-
nels (i.e., the carried traffic per channels) versus the offered
traffic. Different results are provided by varying different
parameters: the super area coverage factor, the number of
and
(but keep their sum constant,
), the ratio of
moving MS, and the speed of moving MS. In all cases, the
call loss probability is computed by summing the blocking
probability of new calls and incoming handovers and the
probability that a connection going on a super frequency
will fail to be served due to the lack regular frequency
when it’s C/I value drops bellow ”C/I good” threshold. In
the model quality problems can only be due to interference
problems indicated by leaving the super coveragearea. This
interference problem can lead to call drop, referring to soft
blocking.
First, we study the effect of the super area coverage on
the performance of the studied cell. Figure 4 shows the loss
probability versus the offered traffic. The different curves
refer to different super area coverage factor,
. It can be
seen that the higher
results in lower loss probability. On
the other hand, Fig. 5 shows that decreasing the area cov-
ered by the super layer has a negative impact on the utiliza-
tion of the cell. This is in consistence with the conclusions
obtained in [10]. This means that a trade-off or optimization
should be done in the design phase of the network between
the reduced utilization and increased loss probability versus
the capacity increased by the higher reuse factor. It can be
seen that the soft blocking has minor effect on the overall
blocking rate even in case of high load situations.
In the rest of the results, we study the effect of move-
ment on the performance of the IUO cell. Figures 6 and
7 show curves on the call loss probability and utilization.
The different curves refer to different value of the ratio of
moving MSs,
. The results show that the higher ratio
of movement results in higher loss probability which can
be explained with the higher number of handovers resulting
from the movement. The utilization point of view is shown
in Figure 7. When the load is low then the movement helps
in reaching better utilization in the cell, while in higher load
conditions the increased movement has a negative effect on
the utilization of the super frequencies which results in a
reduced overall utilization of the cell. This is the reason
for the crossing of the utilization curves between 20 and 60
Erlang offered traffic.
Figures 8 and 9 examine the effect of the speed of the
moving MS given that
. Three values were
chosen to reflect the movement in the downtown area of
city; walking speed (3km/h), vehicle speed in residential
area (30km/h), and vehicle speed in in-city main streets
(50km/h). The results show that the increase of the speed
implies very slight increment on the overall call loss proba-
bility. It is because the loss probability is composed by two
main factors: the probability of rejecting new or incoming
handover calls (referred to as blocking) and the soft block-
ing. Figure 8 depicts these factors separately. The over-
all loss probability practically coincides with the blocking
probability. According to the expectations the soft blocking
probability is significantly affected by the speed of moving
MSs, but it has negligible effect on the overall loss since the
blocking probability is at least an order of magnitude higher
than the soft blocking probability. The decrease of the soft
blocking probabilities from 80 Erlang offered load is due to
the high blocking probability.
The results shows that the usage of IUO has a negative
effect on the performance of one cell and, therefore, there
should be an optimization for the gain of tighter reuse in
the multiple cell network and the loss in the performance
of each cell. The other conclusion is regarding the effect of
the mobility of the performance of the IUO scheme. The
results show that the IUO performs worse as mobility of the
mobile stations increase.
5 Planning issues
To assist traffic dimensioning considerations additional
computations were performed to evaluate the offered traffic
with associated 2% loss. The input and output parameters
5
