Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

Comparison of Models

for WCDMA Downlink Capacity Assessment


Based on a MORANS Reference Scenario
1
Andreas Eisenbl atter
2
, Hans-Florian Geerdes
2
, Antonella Munna

, Roberto Verdone

atesio GmbH, Berlin, Germany; eisenblaetter@atesio.de

Zuse Institute Berlin (ZIB), Germany; {eisenblaetter,geerdes}@zib.de

IEIIT-BO/CNR, DEIS, University of Bologna, Italy; {amunna,rverdone}@deis.unibo.it


AbstractThird generation wireless telecommunication net-
works based on WCDMA technology are being deployed across
the world. Since the downlink is likely to be the limiting direction,
it is crucial for network engineers to assess the downlink capacity
of WCDMA radio cells. In this paper, we revisit a semi-analytical
capacity evaluation model involving snapshot simulation. We fur-
ther develop an alternative approach for assessing cell capacity,
which is a generalization of recent analytical dimension reduction
techniques for cell load computation. The second approach works
under average load rather than snapshots, which enables a
quick approximation of the simulation results. We investigate
the relationship between the two approaches. We demonstrate
how the MORANS (MObile Radio Access Reference Scenarios)
reference datasets can be used to compare different approaches
on a common basis. Based on a MORANS real-world scenario,
we compare the capacity of different cells under varying soft
handover parameters. The results show how cells capacities
vary under realistic data. As the approximative method is quite
accurate, we can conclude that no snapshot simulation is needed
for capacity analysis in our setting.
I. INTRODUCTION
Radio networks based on WCDMA technology are currently
being deployed by telecommunication operators across the
world. For dimensioning these radio networks, a capacity
estimation is crucial. In WCDMA, all signals are transmitted
on the same frequency band, so interference is inevitable,
and radio networks are typically interference-limited. Capacity
analysis is more involved than for traditional radio systems
since the amount of interference depends strongly on the
mobiles location. WCDMA technology allows for data rates
that are much higher than with traditional radio technology.
These data rates will support services that are especially
demanding in the downlink direction. The downlink capacity
is thus expected to become the bottleneck. We therefore focus
on analyzing the downlink capacity of radio cells.
Soft handover (SHO), the capability of a mobile device to
be connected to several base stations (BSs) at a time, is a
novel feature of WCDMA radio technology. This mechanism
can be applied if a mobile device receives several radio signals
from different antennas at a comparable strength. The set of
1
This work is a product of the authors participation in COST 273 and the
MORANS initiative (http://www.cost273.org/morans).
2
Supported by the DFG Research Center MATHEON Mathematics for key
technologies in Berlin, Germany.
BSs that the mobile is connected to is called its active set
(AS). Several radio resource management parameters play a
role here, most noteably the maximum allowed active set size
and the AS window, the range in which the signal strengths
received from the BSs in the active set may vary. If no SHO
is used, the mobile is normally connected to the base station it
receives the strongest signal from, this is called site selection
diversity transmission (SSDT). We compare the capacities for
SSDT and SHO mode and also investigate the effect of the
SHO parameters on a cells capacity.
Monte-Carlo simulation using random realizations of static
user distributions (snapshots) is a well-known approach for
this kind of analysis. While it is generally accepted as a fairly
accurate means for capacity prediction, it is computationally
expensive since experiments have to be repeated until the
outcome is stochastically reliable. We will present a method
that uses snapshot simulation along with an analytical one that
does not require simulation and compare capacity results from
both models. The comparison will be done based on a real-
world scenario, which is a result achieved by the MORANS
initiative within COST273 [1].
In the remainder of this section we introduce our system
model and the soft handover scheme we use, in II our
denition of capacity and the two models for capacity are
presented and compared. We provide computational results for
both models in III and draw conclusions in IV.
A. System Model
The scenario contains I cells. The capacity as dened below
is evaluated for several cells. We pick only cells in the center
of the scenario to avoid border effects. For ease of notation, the
cell in question is always denoted with the index 1. Each cell
transmits at maximum power p
max
; a portion < 1 of this
power is allocated to the trafc on dedicated channels. The
rest is for broadcast and shared channels. Shadow fading is
neglected, fast fading is assumed to be averaged out by perfect
fast power control due to its short correlation length. This
implies that the Carrier-to-Interference-plus-Noise-Ratio (CIR)
perceived at mobile m (MS
m
) for the signal on the link to cell
1
is assumed to exactly meet a specic threshold value denoted
by
m
. The value
m
depends on the service and on the soft
handover state (see below). The attenuation between cell
i
and
MS
m
is denoted by
im
. Furthermore, for each mobile MS
m
,
an orthogonality factor
m
, a noise power
m
at its receiver,
and a (service-specic) transmission activity factor
m
is to
be considered. We write m cell
i
if mobile m is connected
to cell
i
.
B. Soft Handover
We sketch the SHO scheme used in this paper. For more
details the reader is referred to [2]. We denote by n
max
the
maximum number of cells that can be in the AS and by n
the number of currently active cells. SSDT corresponds to
n
max
= 1. Users connected to cell
1
are divided into classes
[3] based on the received power and the AS window :
Class A. Users only connected to cell
1
(n = 1); denoted
M
A
. (In SSDT there are only Class A users.)
Class B. Users with cell
1
as best server and active set size
n > 1; denoted M
B
.
Class C. Users in soft handover with cell
1
in their active
set, but not as best server; denoted M
C
.
Part of the input data for each mobile is a CIR target value.
This value relates to the mobile being connected to only one
cell without SHO, we denote it by
(NHO)
m
for now. If mobile
MS
m
is in soft handover (n > 1), its CIR target reduces
due to micro-diversity effects provided by Maximum Ratio
Combining to a value
(SHO)
m
<
(NHO)
m
. We account for this
by using a diversity factor < 1:

(SHO)
m
=
(NHO)
m
We further assume (cf. [4], [5]) that each link to a cell in the
AS provides the same contribution
(LNK)
m
to
(SHO)
m
, that is,

(SHO)
m
= n
(LNK)
m
. The CIR target for (perfect) fast power
control of cell
1
for Class B and C users is thus

m
=
(LNK)
m
=
(NHO)
m
/n. (1)
For Class A users, we simply have
m
=
(NHO)
m
.
II. ASSESSING DOWNLINK CAPACITY
A. Capacity Denition
According to the classication of users introduced in I-B,
we dene the capacity of cell
1
for a snapshot s as:
M

s
= M
A
+ M
B
By this denition, each served user is counted exactly once.
The average capacity is a mean over S uncorrelated snapshots:
M

= (1/S)

S
s=1
M

s
(2)
B. Model A
We briey recall the semi-analytical model from [3]. Let

1m
denote the portion of total transmission power at cell
1
de-
voted to MS
m
. The following system constraint must hold [4]:
1

mcell1

1m
(3)
The CIR received by MS
m
is

m
=

1m

1m
p
max
(1
1m
)
m

1m
p
max
+

I
i=2

im
p
max
+
. (4)
By transformation, we obtain a closed expression for
1m
:

1m
=

m
+
I

i=2

im

1m
+

1m
p
max
(
m
+ 1/
m
)
(5)
When simulating a snapshot, we evaluate the cells capacity
dened in (II-A) by adding users to cell
1
as long as the
fundamental inequality (3) holds.
C. Model B
Under the assumptions for Model A, the vector p of average
transmit powers satises a linear equation system involving an
I I-dimensional coupling matrix C (cf. [6], [7], [8]):
p = C p + p
()
+ p
(x)
, (6)
where
C
ii
:=

mcelli

m
l
m
, C
ij
:=

mcelli
jm
im
l
m
,
p
()
i
:=

mcelli
m
im
l
m
, l
m
:=
mm
1+mmm
,
(7)
and p
(x)
i
is the power emitted on broadcast channels by cell
i
.
Assuming p
i
= p
max
i, we derive an alternative version of
system constraint (3) by considering only the rst cell in (6):
p
max
C
11
p
max
+

j>1
C
1j
p
max
+ p
()
1
+ p
(x)
1
(8)
In a Monte-Carlo scheme similar to Section II-B, we can
evaluate the capacity of the system by adding mobiles to
cell cell
1
, thereby increasing the values C
11
, C
1j
, and p
()
1
,
until the right-hand side of (8) exceeds p
max
for the rst time.
The linear equation system (6) can be formed for average
user load instead of users in a snapshot (cf. [8]). Let S denote
the set of services and T
s
the spatial average user density
function for each s S. A CIR target
s
and user activity
s
are dened per service. The average coupling matrix

C is
obtained by integrating
1
over all points p in the cells area
and weighting the elements with user density and service load
functions:

C
11
:=
_
pcell1

p
l
s
(p)dp ,

C
1j
:=
_
pcell1

jp

1p
l
s
(p) ,
p
()
1
:=
_
pcell1

1p
l
s
(p) , l
s
(p) :=

sS
T
s
(p)

s

s
1 +
p

s
,
The continuous equivalent of adding users to the cell is to
simultaneously increase the average user load in the cell until
the equivalent of (8) is met with equality. For this purpose, we
use a trafc scaling factor . The average maximum amount
of trafc admissible for cell
1
is calculated from
p
max
=
_
p
max

C
1j
+ p
()
1
_
+ p
(x)
1
. (9)
The value of the resulting scaling factor corresponds to the
fraction of average users in classes A and B that can be
served by cell
1
. The total number of average users in cell
1
is

N :=
_
pcell1

sS
T
s
(p) .
1
Usually, data is provided in pixel format. The integrals in the notation
actually become sums over pixels.
Accordingly, the average number of users served is

M

:=


N. Note that it is particularly easy to calculate as no Monte-
Carlo simulation is involved.
D. Comparison of the Models
The fundamental difference between Model A and the
snapshot version of Model B (8) is the notion of intra-cell in-
terference from power-controlled channels for CIR calculation.
A sketch of the situation can be found in Fig.1. In Model A,
the amount of intra-cell interference is always calculated on
the basis of the xed output power p
max
. The power spent
by the base station on the link for which the CIR is to be
calculated is subtracted from p
max
. The result is the intra-cell
interference. The overall signal strengths considered for CIR
calculation thus sum up exactly to p
max
. In Model B, on the
other hand, intra-cell interference is calculated as the sum of
the powers for all other active radio links, weighed with the
respective activity factor
m
. As can be seen in the gure,
this leads to a higher intra-cell interference. The difference is,
however, less pronounced in practically relevant settings since
the granularity of users is much higher.
1/3
2/3
1
CIR calculation
Average output power
1m
signal
interference
11
P
m
m1m
= (1 11)
Model B
P
m=1
m1m
Model A
(1 1)
intra-cell
Fig. 1. Schematic comparison of evaluation models. Three users with a
relative power consumption of 2/3 and an activity factor of 0.5 are assumed
More interference is assumed in Model B, so the capacity
evaluation is more pessimistic. On the other hand, by using the
scaling factor in the average-based version (9) of Model B,
we assume that the fundamental system constraint (3) is
always met with equality (which is not the case in Monte-
Carlo simulations as there is a certain granularity of users).
This pushes Model Bs results into a more optimistic direction
again. It is, however, not completely clear how the averaging
over the cells area inuences the results of Model B compared
to Monte-Carlo simulation. Our computational experiments
below indicate that the approximation is quite accurate.
In the remainder of this section we will give the analytical
analysis corresponding to the preceding discussion. When
calculating the CIR for a link to mobile m in Model A, the
power (1
1m
) p
max
not spent on the link is accounted as
intra-cell interference and appearswith due consideration of
attenuation and orthogonalityin the denominator of (4).
In Model B, on the other hand, intra-cell interference is
calculated as the sum of all third-party link powers weighed
with the respective activity factor. The average output power
is xed at p
max
. In the deduction of the linear equation
system (6) (cf. [9]) the intra-cell interference for a single
mobile m is calculated as
p
(x)
i
+

ncelli, n=m

n
p
in
, (10)
where p
in
denotes the power spent by base station i on the
link to mobile n. So for calculating the CIR for a transmission
towards mobile m, all remaining dedicated links are assumed
to interfere with their average transmission power. Using the
notation of Model A, that is, := 1 p
(x)
/p
max
and
1n
:=
p
in
/( p
max
), the intra-cell interference term (10) becomes

ncell1
n=m

1n
p
max
= (1
m

1m
) p
max
.
Hence, for analysis of the transmission to mobile m the cells
total output power (link power plus intra-cell interference) is
assumed to be higher than the average power, namely

>1
..
(

ncell1
n=m

1n
+
1m
) p
max
.
This concept is inspired by the fact that for practical network
planning, the average transmission power is usually limited
to a certain fraction (for example 70 %) of the equipments
technical maximum transmit power. The headroom is used for
equalizing fading in fast power control and to allow for a
graceful degradation of service in a congestion situation.
The consequences of this different concept can be best
observed when transforming (8) into a version similar to (3):
1
p
(x)
1
p
max

mcell1

m
_

m
+

j>1
jm
1m
+
m
pmax1m
_

m
+ 1/
m
Since 1 p
(x)
1
/p
max
= , the transformed version of (8) reads
1

mcell1

m
_

m
+

j>1
jm
1m
+
m
pmax1m
_
(
m

m
+ 1/
m
)
. (11)
This is almost the same as (3), except for the activity factor

m
that appears in the denominator. However, since is
usually in the range of -15 to -10dB, the term 1/
m
clearly
dominates the denominator of (11) and (3), the difference is
thus negligible for practical purposes.
III. COMPUTATIONAL RESULTS
A. The MORANS Turin Scenario
Our computational results are based on the realistic Turin
scenario developed within the COST 273 MORANS activity [1].
The MORANS (MObile Radio Access Network reference
Scenarios) initiative is undertaken within the Radio Network
Aspects Working Group (WG3) of the COST 273 Action. Its
goal is to increase comparability among results of different ap-
proaches to evaluate radio network planning and radio resource
TABLE I
SERVICE INFORMATION
Service DL bit rate DL Activity m User Mix
[Kbps] Factor [dB]
Voice 12.2 0.500 -17.48 38.1 %
Video Call 64.0 1.000 -14.66 8.1 %
Data 32.0 1.000 -11.43 2.6 %
WWW
a
64.0 0.774 -15.08 50.1 %
a
A packet call data source model is assumed for this service. The activity
factor reects pauses between single packets within a packet call. In addition,
users have a 10 % probability of being in a packet call at any given time. The
remaining 90 % of the users in a snapshot are assumed to be in reading time.
Their connection is idle. They consume no radio resources.
management strategies. To this end, reference scenarios for
radio network evaluation and planning are provided. Besides
simple, synthetic scenarios, two real-world-based scenarios
(Turin and Vienna) are available. Their denition and use is
more involved than in the synthetic case, but they enable tests
of radio network algorithms under more realistic conditions.
This is the rst publication that realizes the MORANS
initiatives goal of comparing results obtained with different
approaches. We have used the Turin scenario. The scenario
includes an area of 17.85 15.35 km
2
. Geographic data
includes a digital elevation model and vector les describing
railways and motorways. Path loss predictions based on the
COST 231-Hata model are used in the current version which
do not use this information, so we consider a at scenario.
Trafc characterization, in terms of service information (see
Table I) and usage on 4 different services in both uplink and
downlink are given. Link level simulation tables and target
block error rates have been used to calculate CIR targets. The
user distribution is not homogeneous, it is sketched for the
service voice in Fig.2(a); the distributions of users of other
services are equivalent but scaled according to the service
mix. For the results of our Monte-Carlo simulation (Model A),
800 independent user snapshots have been used, an example
snapshot is shown in Fig.2(a).
We evaluate a reference radio access network included in
the MORANS Turin scenario. A total of 34 sites are deployed,
32 of which are composed of 3 cells and 2 of 4 cells, according
to Fig. 2. In addition, base station conguration parameters, as
the antenna type, mechanical and electrical tilt, azimuth, height
are given, together with the horizontal and vertical radiation
pattern. The transmit powers of BSs are p
max
= 10 W; a
fraction of = 0.8 is allocated to trafc channels. The SHO
diversity factor (cf. I-B) is = 0.71.
B. Capacity Analysis for Selected Cells
After evaluating all cells in the scenario, we have picked
four cells with results of different characteristics for discus-
sion. Their locations are indicated in Fig. 2(b). We have
analyzed the average capacity (Model A) and its approxima-
tion (Model B) as dened above for different values of the
maximum active set size n
max
and the SHO window .
1) Comparison of Models: As can be seen from Figs. 3
6, the analytical approximation of Model B comes very close
(a) User density for service
voice and example snapshot
(b) Radio network with cell ar-
eas and evaluated cells
Fig. 2. MORANS Turin scenario
to the results of Monte-Carlo-Simulation of Model A. On a
qualitative level, the charts show the same relations between
the different parameter sets (relative position of different
graphs). Quantitatively, the results are very similar as well,
with a maximum relative approximation error of about 1 %.
This essentially means that in our evaluation model there is
no need for costly snapshot simulations.
2) SSDT capacity Results: When analyzing the SSDT re-
sults, it is obvious that the results differ noticeably between
cells. In our examples, values range from about 47 users (BS
2 2) to 59 users (BS 25 1). This was to be expected in a
setting with non-homogeneous trafc and irregular cell layout.
The deviations of cell capacities from the mean can in all
cases be explained by analyzing the specic local situation.
The two main levers on cell capacity are a) the interference
situationrelative strength of the serving signal over the
interfering signals, reected in the sum

I
i=2
im
1m
in (5)and
b) the trafc distribution in the cell relative to the interference
distribution. The more trafc in a cell is placed in areas with
favorable interference situation (areas where the mentioned
sum is small), the higher the capacity. However, some of the
differences could to a certain degree be leveled by considering
shadow fading (cf. the remarks on the effect of SHO).
3) Inuence of Soft Handover: It is striking how the
capacity behavior of the selected cells differs when taking into
account SHO and varying parameters. This diversity applies
for all cells in the scenario. In general, it can be observed that
the capacity of all cells decreases under SHO if the parameter
is increased too much, all example charts presented here
show a decrease from = 3 to = 4. This trend was also
observed for the other cells with very few exceptions. It can
be explained by arguing that with increasing , cells become
members of the active set of users far away from them and
the diversity gain is outweighed by increasing interference.
Beyond this general trendand besides a general benet
from diversity as specied in (1), cells can be roughly
divided into ones that clearly benet from SHO (SHO graphs
lie above the SSDT graph) and ones that sacrice capacity.
An example for the rst case is BS 2 2, for the second
BS 25 1. These two types of cells can often be observed to
46
48
50
52
54
56
1 2 3 4
U
s
e
r
s

s
e
r
v
e
d
(dB)
SSDT
n
max
= 2
n
max
= 3
n
max
= 4
n
max
= 5
(a) Model A
46
48
50
52
54
56
1 2 3 4
U
s
e
r
s

s
e
r
v
e
d
(dB)
SSDT
n
max
= 2
n
max
= 3
n
max
= 4
n
max
= 5
(b) Model B
Fig. 3. Capacity results for BS 2 2
44
46
48
50
52
54
56
58
60
1 2 3 4
U
s
e
r
s

s
e
r
v
e
d
(dB)
SSDT
n
max
= 2
n
max
= 3
n
max
= 4
n
max
= 5
(a) Model A
44
46
48
50
52
54
56
58
60
1 2 3 4
U
s
e
r
s

s
e
r
v
e
d
(dB)
SSDT
n
max
= 2
n
max
= 3
n
max
= 4
n
max
= 5
(b) Model B
Fig. 4. Capacity results for BS 25 1
occur in pairs, as is the case for our examples. The reason
in this specic case is that the second cell includes an area
that it dominates very narrowly. The interference situation is
bad since there are many signals with an almost equivalent
strength. Once SHO is used, these areas are shared among
neighboring cells, leading to a certain leveling of capacity
among the neighbors. This phenomenon can be clearly ob-
served when comparing Figs. 3 and 4. However, it is to be
expected that this is to some extent an artefact of not taking
into account shadowing, which would also lead to a certain
leveling in the SSDT case.
Another example for a clear relationship in SHO behavior
are the last two neighboring cells selected for discussion,
BSs 5 3 and 33 2: while the graph for the rst cell has a
steep increase from = 1 to = 2, the same graph for the
rst cell decreases articulately. Obviously the rst cell gains
capacity at the expense of the second. However, for larger
active set sizes, other neighboring cells come into play and
the relationships cannot be explained as easily anymore.
IV. CONCLUSION
In this paper, we have revisited a semi-analytical model for
evaluating the downlink capacity of WCDMA radio cells using
snapshot simulation. We have then introduced an alternative,
analytical model that provides an efcient estimate for cell
capacity without using simulations. We discussed the relation
between the two models analytically. We have used one of
the MORANS real-world scenarios for computational results.
By comparing the results for different models on a common
data scenario, we have realized the goal of the MORANS
54
56
58
60
62
1 2 3 4
U
s
e
r
s

s
e
r
v
e
d
(dB)
SSDT
n
max
= 2
n
max
= 3
n
max
= 4
n
max
= 5
(a) Model A
54
56
58
60
62
1 2 3 4
U
s
e
r
s

s
e
r
v
e
d
(dB)
SSDT
n
max
= 2
n
max
= 3
n
max
= 4
n
max
= 5
(b) Model B
Fig. 5. Capacity results for BS 5 3
46
48
50
52
54
56
1 2 3 4
U
s
e
r
s

s
e
r
v
e
d
(dB)
SSDT
n
max
= 2
n
max
= 3
n
max
= 4
n
max
= 5
(a) Model A
46
48
50
52
54
56
1 2 3 4
U
s
e
r
s

s
e
r
v
e
d
(dB)
SSDT
n
max
= 2
n
max
= 3
n
max
= 4
n
max
= 5
(b) Model B
Fig. 6. Capacity results for BS 33 2
initiative, namely to increase comparability of results in the
area of radio network analysis and optimization. The data set
includes non-homogeneous trafc and an irregular network
layout. This paper is thus one of the few scientic publications
on WCDMA radio network evaluation that includes compar-
atively realistic data. We furthermore show how realistic data
inuences capacity results for different cells. As the second
models results are strikingly close to the rst ones, our results
show that no snapshot simulation is necessary to calculate the
average capacity in our evaluation model.
REFERENCES
[1] A. Munna, S. Ruiz, R. Verdone, MORANS - MObile Radio Access
Network Reference Scenarios, In Proc. of WPMC04, Abano Terme,
Italy, September 2004.
[2] 3GPP TR 25.922 v5.0.0(2002-03), Technical Spec. Group Radio Access
Network; Radio resource management strategies (Release 5).
[3] C. Balzanelli, A. Munna and R. Verdone, WCDMA Downlink Capacity
- Part 1. In Proc. of IEEE PIMRC, Beijing, China, September 2003.
[4] A. De Hoz and C. Cordier, W-CDMA downlink performance analysis.
In Proc. of IEEE VTC, Vol.2, Sept 1999, pp. 968972.
[5] Q. Zhang, UMTS Air Interface Voice/Data Capacity - Part 2: Forward
Link Analysis. In Proc. of IEEE VTC, Vol.4, May 2001, pp. 27302734.
[6] L. Mendo and J. M. Hernando. On dimension reduction for the power
control problem. IEEE Trans. on Communications, 49(2), Feb. 2001.
[7] U. T urke, R. Perera, E. Lamers, T. Winter, and C. G org. An advanced
approach for QoS analysis in UMTS radio network planning. In Proc. of
the 18th ITC, pages 91100. VDE, 2003.
[8] A. Eisenbl atter and H.-F. Geerdes. A novel view on cell coverage and
coupling for UMTS radio network evaluation and design. In Proc. of
INOC05, Lisbon, Portugal, Mar. 2005. ENOG.
[9] A. Eisenbl atter, H.-F. Geerdes, T. Koch, A. Martin, and R. Wess aly.
UMTS radio network evaluation and optimization beyond snapshots.
Technical Report ZR-04-15, ZIB, Berlin, Germany, 2004.

You might also like