Content uploaded by Ramon Agustí
Author content
All content in this area was uploaded by Ramon Agustí
Content may be subject to copyright.
MANAG
EM
ENT
OF
N
EXT-G
EN
ERATI
ON
~
WIRELESS
NETWORKS
AND
SERVICES
~
Provisioning Multimedi Wireless
Networks for Better
*QoS:
RRM
Strategies for
3G
W-CDMA
Oriol Sallent, Jordi Perez-Romero, Ramon Agusti, and Ferran Casadevall
Universitat Politecnica de Catalunya
ABSTRACT
Third-generation mobile communication sys-
tems will bring a wide range of new services with
different quality of service requirements and
will
open the ability to exploit radio resource man-
agement functions to guarantee
a
certain target
QoS,
to maintain the planned coverage area and
to
offer a high-capacity while using the radio
resources
in
an efficient way. RRM functions
impact the overall system efficiency and the
operator infrastructure cost,
so
they
will
defini-
tively play an important role
in
a mature 3G sce-
nario. In order
to
provide some insight into
RRM strategies implementation, a range of rep-
resentative case studies with several innovative
algorithms arc presented and supported by simu-
lation results
in
a realistic UMTS Terrestrial
Radio Access Network scenario as devised in the
3GGP standardization forum.
In
particular, a
decentralized uplink transmission ratc selection
algorithm in the short term, a congestion control
mechanism to cope with overload situations, and
downlink scheduling for layered streaming video
packets are proposed.
INTRODUCTION
The evolution of the end user needs towards
multimedia applications has pushed the wireless
community to conceive the so-called third-gener-
ation (3G) systems (e.g., Universal Mobile
Telecommunications System, UMTS,
or
cdma2000), where wideband code-division multi-
ple access (W-CDMA)
is
the predominant tech-
nology. W-CDMA access networks provide an
inherent flexibility to handle the provision of
future 3G mobile multimedia services with dif-
ferent quality
of
service
(00s)
guarantees and
the ability
to
optimize the spectrum efficiency
in
the air interface by means of efficient radio
resource management (RUM) algorithms.
The problem faced by a n'etwork operator is
to offer
a
system where the number of users
is
maximized for a given set of
QoS
requirements.
In this problem
two
aspects can be clearly distin-
guished: network planning (i.e., the design of the
fixed network infrastructure in terms of number
of
cell sites, cell site location, number and archi-
tecture of concentration nodes, etc.) and radio
resource allocation (i.e., for
a
given network
deployment, the way radio resources are dynami-
cally managed in order to meet the instanta-
neuus demands of users moving around the
network).
In the framework
of
2G time-division multi-
ple access (TDMA)-based mobile systems (e.g.,
Global System for Mobile Communications,
GSM, or IS-54), the main problem is in network
planning. The perceived subjective
QoS
for voice
service is mainly controlled through suitable fre-
quency assignment among cell sites in order to
provide a sufficient carrier to interference ratio
(Cll).
On the other hand,
in
2G
the call blocking
probability is the other fundamental
QoS
param-
eter and is controlled through providing in a first
step enough frequencies to a given cell site and
in a second step by adding new sites. For a given
2G network configuration there
is
an almost
constant value for maximum capacity. Addition-
ally, radio resource allocation in the short term
(e.g., on the order of tenthslhundreds
of
mil-
liscconds) has little
to
do in
a
scenario where the
supported service (e.g., voice) requires a channel
with constant quality and tight delay constraints.
In thc framework of 3G mobile systems the
situation
is
significantly different. First, in W-
CDMA-based systems there is not a constant
value for the maximum available capacity, since
it
is tightly coupled to the amount of interfer-
,ence in the air interface. Second, the multiser-
vice scenario drops the stringent delay
requirement for some services and, consequent-
ly, opens thc ability to exploit RRM functions to
100
0163-6804103/$17.00
0
2003
IEEE IEEE
Communications
Magazine
February
2003
guarantee a certain target
QoS,
to maintain thc
planned coverage area and offcr high capacity
while using the radio resources efficiently. It is
worth mentioning that RRM functions can be
implemented in many diffcrcnt ways, having an
expected impact
on
overall system efficiency and
operator infrastructure cost,
so
RRM strategies
will
definitively play an important rolc in
a
mature
3G
scenario. Additionally, RRM strate-
gies are not subject to standardization,
so
they
can he a differentiation issue among manufac-
turcrs and opcrators.
The problem
of
addressing the QoS provision-
ing for wireless multimedia traffic through suit-
able RRM strategies has gained great momentum
during the last years within the framework
of
W-
CDMA schemes. Also, admission control
schemes including handoff procedures with dif-
ferent levels of adaptive resource reservation
mechanisms
[l,
21 as well as short-term protocols
to schedule transmissions according to their hit
error rate (BER) requirements 131 have been
widely proposcd. Nevertheless, few studics in the
open literature
[4]
regarding these RRM mecha-
nisms approach realistic scenarios aligncd with
the actual 3G standards spccifications.
In
the above context, the objectives of this
article arc twofold: first, to contribute to the
assessment of some
QoS
concepts as well
as
to
RRM strategies'devised accordingly in the
framework of a realistic UMTS radio segment
scenario
as
standardized in.the
3G
Partnership
Project (3GPP); and sccond, to provide some
guidelines on how the generic RRM functions
can he implemented in the form of specific and
innovative algorithms. The algorithms intro-
duced
in
this article address uplink rate selec-
tion, congestion control, and downlink packet
schcduling
for
laycred streaming video. For this
purpose,
a
rangc of representative case studies
carried out in the framework
of
the IST
ARROWS European Collaborative project
[5]
are presented and supported by simulation
results for algorithm evaluation. Thus, the rest
of
the article
is
organized
as
follows. We define the
QoS
problem
in
the radio interface. We describe
'
the 3GPP approach for RRM and
also
how the
Radio Resource Control (RRC) protocol
is
defined to execute RRM algorithm decisions.
We propose several new RRM algorithms and
accompany them with results ohtaincd hy means
of a precise UMTS Tcrrestrial Radio Access frc-
quency-division duplex (UTRA-FDD) system
and link-level simulator. Finally, we summarize
the main conclusions reached with this work.
QOS
AND THE
bDl0
INTERFACE
To cope with a certain
QoS
a bearer service with
clcarly defined characteristics and functionalities
must he sct up from the source
to
the destina-
tion
of
the service, maybe including not only the
UMTS Tcrrestrial Radio Access Network
(UTRAN, plus core network) but also external
nctworks
161.
Within the UMTS bearer service,
the role
of
the radio bearer service
is
to cover
all
aspects of the radio interface transport over the
UTRAN; consequently, RRM strategies
will
be
responsible for assuring the defined
QoS
in
this
segment.
The radio interface of the UTRAN
is
laycred
into three protocol layers: the physical layer
(Ll), thc data link layer
(LZ),
and the network
layer (L3). Additionally, L2
is
split into
two
suh-
layers, radio link control (RLC) and medium
access control (MAC). On the other hand, the
RLC and
L3
protocols are partitioned in two
planes, user and control. In the control plane,
L3
is partitioned
into
sublayers where only thc Iow-
est sublayer, denoted radio resource control
(RRC), terminates in the UTRAN
[6].
Connections between RRC and MAC
as
well
as
RRC and
L1
provide local interlayer control
services, and allow the RRC to control the con-
figuration of the lower layers. In the MAC layer,
logical channels are mapped
to
transport chan-
nels. A transport channel defines the way in
which traffic from logical channels
is
processed
and sent to the physical layer. The smallest enti-
ty of traffic that can be transmitted through a
transport channel
is
a transport block (TB).
Oncc
in
a
certain period
of
time, called a trans-
mission time interval (TTI),
a
given number of
TBs will be delivered to the physical layer in
order
to
introduce some coding characteristics,
interleaving, and rate matching to the radio
frame. The sct
of
specific attributes are referred
as the transport format (TF) of the considered
transport channel. Note that the number
of
TBs
transmitted in a
TTI
indicates that diffcrcut hit
rates are associated with diffcrcnt
TFs.
As the
user equipment (UE) may have more than one
transport channel simultaneously, thc TF comhi-
nation (TFC) refers to the sclected combination
of TFs. The list
of
allowed TFCs to be used
is
referred to
as
the transport format combination
sct (TFCS)
[h].
THE
RRM
FRAMEWORK
IN
UTRA
RRM strategies have
to
he applied to both uplink
and downlink
in
a
consistent way. In the uplink
direction, centralized solutions (i.e., RRM algo-
rithms located at the radio nctwork controller,
RNC) may provide better performance com-
pared to a distributed solution (i.e., RRM algo-
rithms located at thc UE) because much morc
RRM relevant information related to all users
involved in the process may be available at the
RNC. On the contrary, exccuting decisions madc
by centralized RRM algorithms would be much
more costly
in
terms of control signaling becausc
in this case the UE must he informed about how
to operate. Consequcntly, strategies face thc pcr-
formancelcomplexity trade-off that usually finds a
good solution
in
an intermediate state whcrc
both centralized and decentralizcd components
are present. The 3GPP approach for the uplink
could be included
in
this category, since
it
can be
divided into
two
parts.
Centralized component (located at the RNC):
-
Admission control. It is employed
to
decidc
whether to acccpt or reject a new connec-
tion depending
on
thc interference (or
load)
it
adds to existing connections. With
acceptance, a TFCS
is
decided so that
thc
maximum allowed hit rate is determined.
*
Congestion control.
It
faces situations in
which the system has reachcd
a
congestion
status and therefore the
QoS
guarantccs
-
To
cope with a
certain
00s
a
bearer service
with clearly
defined charac-
teristics and
functionalities
must be set up
from the source
to the
destination
of
the
service, 'maybe
including not only
the
UMTS
network
(UTRAN
plus Core
Network) but also
external
networks.
IEEE
Communications
Magazine
-
February
2003
101
.
~
~.
..
.~
-
....
~
~.
-
..
~
.-.-
.
i
Service Web
(UL)’
Streaming basic
(DL)
Streaming enhancement
(DL)
’
Transport
channel
DCH
DCH
DSCH
+.__I__-_..
.---.-I
..
_I
..............
1
TB
sizes.
bits
336
(320 payload, 336 bits
(320
payload.
336
bits
(320
payload.
i
TFS
TFO.
bits
0
x
336
~-
0
x
336
0
x 336 ___..~
I
I
i
~
.,
I
-1
-~__-
i-
..
.........
.
16 MAURLC
header)
16 MAURLC
header)
...
.
16 MAURLC
header)
..........
..
.
I
j
--_._~l__-l___
Zx336(16kb/s,SF=
.
....__._____I_
128)
1
....-
.....
1
h,bits
1
X
336
(16‘5b/sL
SF
=
64) 4
x
336 (32 kb/s,
SF
=
64) 1
5
336
(8
kb/s,
SF
=
256)
i
TF2.
bits
_----...-I
2 x
336
(32 kbls.
-I___
SF
=
32)
....
-
,
TF3,
bits
3
.....
.
.
4
x 336j32
...
kb/s,
5F
=
64)
x
336
(48
kbls.
5F
=
16)
-:-
.......
. .~
8
x
336
..
(64
kbls.
SF
=
32)
..
,
PT?:@bi?--
-4
x
336
(64
kb!.
SF
=
16)
-
I__~
..
..
___~_
...
16
x
336
(128
. kbls,
SF
=
16)
i
TF5,
bits ~
............
..
--
-
..........
.-
-
.............
I
__
,
r
-
i
!.
-
.
~--i
El
Table
1
.-Transport
fonnatsforthe
considered
MRS.
....--__I.-
I
-
...
~
,
-.
-
-
i
UE-MAC strategy
!
SCr24 0.54 5.0
I
MR 0.12 11.3
0
Table
2.
Average
delay
poformance
for
UE-MAC
algorithms.
Average packet delay
(5)
Rate
per
page jitter (kb/s)
............
.__I_~._
.
.
.~-
~
.
..
..
are at risk due to the evolution of system
dynamics (mobility aspects, increase in
interference, etc.). An appropriate selection
of
TFCS can he seen
as
a
congestion con-
trol procedure (limiting the maximum hit
rate for example when there is an increase
in the measured interfcrence).
Decentralized
part
(located at
UE-MAC):
This algorithm autonomously decides a
TF
for
each
TT1,
and thus opcrates on a “short” term
in order to take full advantage
of
the timc vaq-
ing conditions. In the uplink, the
TF
is selected
by the MAC layer
of
each terminal within thc
TFCS assigned by the network. In this sense, the
scheduling is carricd out autonomously at the
user level, allowing significant signaling savings.
In the downlink,
a
totally centralized opera-
tion arises naturally, and the
TF
selection can be
performed by considering the information
of
all
the users. An additional function, code manage,
ment, must
also
be considered. Code manage-
ment is devoted to manage the orthogonal
variable spreading factor (OVSF) code tree used
tn allocate physical channel orthogonality among
different users.
Moreovcr, handover procedures have a strong
impact
on
the overall
RRM,
so
it is mandatory
to develop
RRM
strategies that take this influ-
ence into account. Handover management
is
in
charge
of
allowing the continuity of the call in
progress when the mobile moves from one cell
to another and still guaranteeing its
QoS.
Decisions made
by
RRM algorithms are exe-
cuted through radio hearer control procedures
(a subset of RRC procedures) such as
[6]:
Radio bearer setup. This procedure is used to
set up a new radio bearer and specify a suitable
TFCS after performing admission control.
Physical channel reconfiguration. The physical
channel reconfiguration in
UL
can be used to
change the transmission frequency band (hard
handover), maximum allowed transmission power,
minimum W-CDMA spreading factor, and
so
on.
In the downlink, this procedure can also he used
to
dynamically change downlink OVSF codes
allocated
to
different bearers, taking into account
hit rate requirements
of
each connection.
Transport channel reconfiguration. This pro-
cedure is used to changc thc different parame-
ters of a certain transport channel. Particularly,
it can be used to limit the allowed TFCS depend-
ing on the system status (e.g., when congestion
arises).
RRM
ALGORITHM
CASE
STUDIES
In this section sevcral case studies involving dif-
ferent RRM strategies are discussed in order to
provide some insight
on
how these algorithms
and strategies can be implemented in a real SyS-
tcm. Thc algorithms are evaluated and support-
ed by simulation results, and some representative
services are included. In particular, the consid-
ered services and their corresponding transmis-
sion characteristics are:
Web browsing (interactive class). The radio
access hearer considered has a maximum bit rate
of
64
kbls in the uplink and an associated
3.4
khls
signaling radio bearer (Table
1).
The interactivc
traffic model considers the generation of activity
periods (i.e., pages for Web browsing), where
several information packets are generated, and a
certain thinking time between them, reflecting
service interactivity. The specific parameters are:
average thinking time between pages:
30
s,
aver-
age
number of packet arrivals per page:
25;
num-
her of bytcs per packet: average 366 bytes,
maximum
6000
bytes (truncdtcd Pareto distribu-
tion), time betwccn packet arrivals: average
0.125
s,
exponential distribution.
Layered downlink streaming video (streaming
Class). This service has
two
different quality layers:
basic and enhancement. The radio access bcarcr
considered for, the basic layer uses a dedicated
channel (DCH) and has two possihlc TFs:
TFO
(no’ transmission)
or
TFl (allowing the transmis-
sion of four transport blocks). For the enhance-
ment laycr the radio access bearer uses a downlink
shared channel (DSCH) and contains
6
TFs,
defined in Table
1,
which arc selected depending
nn
how the scheduling algorithm hehavcs.
102
IEEE
Communications
Magazine
* February
2003
The simulation modcl includes seven cells
with radii
0.5
km. Physical layer performance is
ohtained from link-level simulations carried
out
to feed the system level simulator presented
hcre with block error rate (BLER) statistics. The
mohility model and propagation models are
defined in
[6].
CASE
STUDY
1:
UPLINK UE-MAC ALGORITHM
This algorithm autonomously decides the trans-
mission rate on a frame-by-frame basis, and thus
operates on a “short” term in ordcr to take full
advantage of the time-varying conditions. This
decision is taken on a decentralized way by the
MAC layer of each UE among thc possibilities
contained within TFCS. Notice that the inclusion
of no transmission is allowed as
a
particular case
of
TF.
In this case
two
approachcs are explored
for interactive-like services (e.g.,
Web
browsing).
Service Credit (SCr) algorithm: This
algo-
rithm aim$ to offer a negotiated average hit rate.
The SCr of a connection accounts
for
the differ-
ence between thc obtained hit rate and the
expected hit rate for this connection. Essentially,
if
SCr
<
0,
the connection has obtained
a
higher
hit rate than expected, and if SCr
>
0,
the con-
nection has obtained
a
lower hit rate than
expected.
In
cach
TTI,
the SCr for a connection
should he updated as follows:
.
SCr(n)
=
SCr(n
-
I)
+
(Guaranteed-rate/ TB-size)
-
Transmitte!_TB(n
-
1)
where
SCr(n),
measured in number of transport
blocks per
TI,
is the service credit for
‘RI
=
n,
SCr(n
-
1)
is the service credit in the prcvious
TTI, Guaranteed-rate is the number of hits per
?TI
that would be transmitted at the guaranteed
hit rate, TB-size is the number of hits
of
the TB
for the considered RAB, and Transmitted-TB(n
-
1)
is
the number
of
successfully transmitted
TB
in the previous TTI. At the beginning of the
connection: SCr(0)
=
0.
Maximum Rate
(MR)
algorithm: This
algo-
rithm selects the
TF
that allows the highcst
transmission hit rate according tu the amount of
bits in the buffer waiting for transmission,
Resuk-
One important measurcment to
understand the behavior of the different
UE-
MAC strategics is the transport format distribu-
tion used. Referring
to
Table
I,
UE-MAC has
the freedom to choose
among
TFO
(when the
buffer is empty or SCr
<
0),
TFI,
TF2, TF3,
and TF4. For SCr24 (Fig.
1,
SCr24 stands for a
service credit strategy with a guaranteed rate of
24
khis) it can be observed that most
of
the
timc
TFI
and TF2 are used because thc UE
buffer queues several packets and tends to
transmit the information at 24
kh/s.
In turns,
in^
the periods when the UE huffcr is empty thc
UE
gains service credits, and when a new pack-
et arrives the transmission ratc is increased
over the guaranteed one (i.e.,
TF3
and TF4 are
used). For the
MR
strategy, since it chooses the
TF
according to the buffer occupancy and tries
to transmit the information as fast as possible,
most of the time TF4
is
used (Fig. 2). Addition-
ally, Table 2 shows the average delay perfor-
mance for both MR and SCr strategies.
It
can
.
SCr24
0354
....
.
...
TFl
TF2 TF3 TF4
Figure
1.
TF.disfributionforSCr.
MR
0.9
0.8
.........
~
......
~.~
..................
~.~
........
~.,~~~~~
.................
~~
.....
~~~
.....
~
......
~~
.....
TFl
TF2
TF3
TF4
Figure
2.
TF
distribution
for
MR.
be seen that MR provides lower delay because
it tends tu maximize the transmission rate
according to huffcr occupancy. On the contray,
since SCr does not take buffer occupancy into
account, it provides better control of the trans-
mission rate, reflccted in a
IOW
rate pcr page
jitter. Notice that for the same user and service,
depending on the spccific UE-MAC algorithm
used, the load or equivalently, the.intcrference
this user will cause the system will hc different,
which should be taken into account in the
admission control phase. To raise this effect,
consider common uplink admission control
based
on
statistical cell load control,
so
a call
request is
only
accepted if thc resulting ccll
load
q
is below a certain threshold
qmax
obtained from radio network planning
[7]:
’
--
where
SFi
is
the ith uscr spreading factor,
(EbiNo);
stands for the ith user requirement,
vi
is
the ith user activity factor,
K
thc number of
IEEE
Communications
Magazine
-
February
ZOO3
103
-~
__
_I~
-
--
.
’
650
!__
”
l~
IL
.-,.I
.Table
3.Resulis
forqcD
=0.8,
qCR
=
0.7, ATcD
=
10,
ATCK
=
10.
i
Packet
delay percentiles Algorithm
1
Algorithm
2
1
during congestion
periods
..~_
50%
<
0.12
I
<0.16s
75%
c
0.84
s
<
1.12s
!
95%
<
2.94
5
<
6.62
5
_-.LI._~
_Ix
..I._.-
.
.
~
-__.._.I_~~^
-
~
..
,
.___
,
.-
~
.~
-.
~
..
.
~&
6Table
4.
ResulfsforqcD
=0.8,
qCK
=
0.7,
ATc.
=
10,
AT~R
=
loand
700
users.
users in the cell, and
f
the intercell
to
intracell
interference factor estimation.
It can he seen from Eq.
I
that for load esti-
mation purposes a certain
SFi
needs to be con-
sidercd. Consequently,
it
can be concludcd that
the UE-MAC algorithm impacts the admission
control process, which should take this fact
into account for accurate cell load estimation.
For further details on other UE-MAC algo-
rithms and statisticlil performance measure-
ments
sec
151.
CASE
STUDY
2:
UPLINK
CONGESTION
CONTROL
A
congestion control mcchanism including the
following parts is presented for interactive ser-
vices:
1)
Congestion detection: The criterion intro-
duced
in
order to decide when
the
network is in
congestion and trigger the congestion resolution
algorithm is whcn thc load factor increases over
a
ccrtain threshold,
qc~,
during a certain amount
of time,
ATcD
(i.e.,
if
q
t
qco
in, e.g.,
YO
per-
cent of the frames within
AT~D).
2)
Congestion resolution: When congestion
is
assumed in the network, some actions must be
taken
in
order to maintain network stability. The
congestion resolution algorithm executes
a
set of
ruics to lead the system out ofcongcstion status.
Three steps are identified:
a.
Prioritization: Ordering the different users
from lower to higher priority (i.e., from those
that expect a lower grade of service
to
those with
more stringent
QoS
requirements)
in
a prioriti-
zation table.
b.
Load reduction: Two main actions can be
taken:
‘
No
new connections are accepted while in
congestion.
*
Reduce the TFCS (i.e., limit the maximum
transmission rate) for a certain number of
users already accepted in the network,
beginning from the top
of
the prioritization
table.
Algorithm
1:
The user is not allowed
to
transmit anymore while
in
a congestion
period (e.g., the TFCS is limited
to
TFO
by
sending from the RNC to the UE the
L3
RRC protocol message Transport Channel.
Reconfiguration).
Algorithm
2:
The TFCS is limited to TF2,
so
users arc not,allowed to transmit at more
than 32 kbis, whereas
in
normal conditions
the maximum rate is
64
kbis.
c. Load check After the actions taken in h),
one would recheck the conditions that triggcred
the congestion status. If congestion persists,
one
would go back
to
b) for the following group
of
users in the prioritization table.
It
could be con-
sidered that the overload situation has been
overcome if, for a certain amount of time
ATCR
the load factor
is
below a givcn threshold,
~CR
(i.e.,
if
q
s
qC,(
in,e.g.,
90
percent of the frames
within
ATcR).
3)
Congestion recovery: A congestion recov-
ery algorithm is needed
in
order to restore to
the different mobiles the transmission capabili-
ties they had before the congestion was trig-
gered. it is worth mentioning that such an
algorithm is crucial because depending on how
the recovery
is
carried out the system could
fall
again in congestion. A “time scheduling” algo-
rithm (user by user restoring approach) is con-
sidered. That is,
a
specific user is again allowed
to
transmit at
a
maximum rate (i.e., a Trans-
port Channel Reconfiguration message indicat-
ing that TFCS includes up to TF4 is sent).
Once this user has emptied the buffer, another
user is allowed to rccover the maximum rate,
and
so
on.
Results
-
Comparisons for the two presented
load reduction algorithms are summarized in
Tables 3 and
4.
The performance figures are
the,admission probability (i.e., the probability
that a user request is accepted
into
the sys-
tem’), the percentage of time the network is
congested, and the delay distribution
of
the
transmitted packcts during the congestion peri-
od. It can be observed that the ”softer” load
reduction actions taken by algorithm
2
lead
to
more time in congestion and consequently a
reduction in admission probability (notice that
the first action in congestion is to reject all
connection requests).
it
seems that the firmer
actions takcn by algorithm
I
result in shorter
congestion periods. Additionally, one
of
the
expected impacts of congestion is
a
delay
degradation due tn the transmission rate capa-
bilities limitation. It can be observed in Table
4
that algorithm
1
provides a nicer delay distri-
bution than algorithm
2,
specially for the 95th
percentile.
CASE
STUDY
3:
,
DOWNLINK
PACKET
SCHEDULING
FOR
LAYERED
STREAMING
VIDEO
In
this case study we focus
on
streaming video
scrvicc, which is onc of the expected intcrcsts
in
3G
systems. Quality rcquirements dcal with the
achievcd hit rate, percentage
of
lost
packets, and
delay jitter (rather than the end-to-end delay). It
is considered that streaming service allows an
initial setup delay that gives room for some
104
IEEE
Communications Magazine
Fchrualy
2003
packet transmissions before the video is repro-
duced. These packets can be stored in the mobile
terminal buffer. Then, with proper buffer dimen-
sioning, the user can be unawarc of possible
packet retransmissions because the stored buffer
allows for continuous packet flow reproduction.
Thus, this property gives some more room for
scheduling the streaming service since packet
retransmissions may play
a
iole.
In order to differentiate quality levels, we
assume for this service
a
two-layered video appli-
cation that is characterized by two different
flows: a basic layer, with the minimum require-
'ments for
a
suitable visualization, and
an
enhanccment layer-that contains additional
information to improve the quality of the
received images. We
will
assume that the hasic
laycr will he transmitted through the DCH,
while
the enhancement layer will be transmitted
only
if
there is capacityin the DSCHs. It
is
assumed that the possible retransmissions
of
the
basic layer can he carried out
in
the DSCH
together
with
the enhancement laycr, and having
higher priority than the latter.
The proposed packet scheduling strategy allo-
cates resources to the different flows that make
use
of
the DSCH channel. It operates
on
a
frame-by-frame basis (i.e., every
10
ms)
as
fol-
lows:
Prioritization: The first step consists of order-
ing the different users' requests in the DSCH
depending on the following priority criteria:
*
The higher the number of basic layer
TBs
to
he retransmitted, the higher the priority.
*
For the same number
of
basic layer TBs,
the priority
is
established according to the
service credit concept, as explained earlier.
The higher the service credit of the
enhancement layer. the higher the priority.
Resource allocation: Once requests are
ordered, the next step consists of deciding
whether or not they are accepted for transmis-
sion in the DSCH channel, and which is the
accepted
TF.
A transmission
is
granted only if
the estimated load factor is below
a
certain
threshold
$,
and the estimated transmitted power
level is below
a
fraction
6
of the maximum trans-
mitted power. Othenvise, the TF is reduced by
one, or equivalently, the transmission bit rate
is
reduced and the conditions rechecked.
If
this~is
not possible, the request should wait for the next
frame.
oi
I
0.8
0.85
0.9
0.95
1
I
Results
-
One of the most relevant parameters
in the design
of
the packet scheduling algorithm
relies
on
the threshold
$.
Figure
3
presents the
average bit rate obtained during
a
streaming ses-
sion for the enhancement layer for different
$
values and number of users in the system.
S
=
1
has been assumed. The hasic layer is not pre-
sented since its achieved bit rate is almost always
32
kbis (i.e., the basic layer
is
guaranteed). It can
he concluded that the selection
0
=
0.95
pro-.
vides the hest behavior, since the enhancement
layer gets the highest possible hit rate for the
different load levels.
In
turns, Fig.
4
plots the
delay jitteiachieved for different
$
and 120
users. Again
$I
=
0.95
is revealed
to
he
a
suitable
value because it provides the lowest possible
delay variation.
1
35,000
,
I
0.8
0.85
0.9
0.95
1
Z
30,000
1
B
E"
20,000
-
225,000
4
r
.a
I
15.000
-
01
I
Ft
I
I
Figure
3.
Achieved
bit
ratr.
CONCLUSIONS
In this article the key role RRM strategies
will
play in
a
mature
3G
W-CDMA based system
such as UTRA-FDD has been strengthened.
On
one hand,
3G
will bring a wide range
of
new ser-
vices with different
QoS
requirements and will
open the ability
to
exploit RRM functions. On
the other hand, W-CDMA performance
is
tight-
ly coupled to the amount of interference in the
air interface and eventually depends
on
many
strongly interrelated system parameters that
need to be suitably managed through RRM
strategies in order to achieve high efficiency.
This article has also proposed several specific
RRM algorithms in order to provide some
insight to the reader on how the generic RRM
functions can he implemented.
In
particular, the
presented case studies have led us
to
conclude
the following:
'Results for UE-MAC strategies show that
SCr tends to use
a
more stabilized dynamic
transmission rate around the agreed value, while
IEEE
Communications
Magazine
-
February
2003
105
W-CDMA
performance
is
tightly coupled
to
the amount
of
interference in
the air interface
and eventually
depends on
many strongly
interrelated system
parame
need
suitably
through RRM
for MR most of the time the highest possible
transmission rate is used. This different behavior
of the UE-MAC algorithms means that, for the
same user and service, depending on the specific
UE-MAC algorithm used the load or, equiva-
lently, interference this user will cause the sys-
tem will be different. Thus-, the UE-MAC
algorithm impacts on the admission control pro-
cess, which should take this fact into account for
accurate cell load estimation.
=When managing congestion situations, it
seems more suitable to adopt strict actions, such
as
not allowing delay-tolerant services to trans-
mit during the congestion, than softer policies,
such as reducing to some extent the transmission
rate. With the strict policy the time the network
is in congestion
is
reduced and the delay degra-
dation is nicer than in the softer case.
.Layered streaming video download allows
for retransmissions and variable quality.
By
sup-
porting this service through
a
dedicated channel
for the basic layer and
a
shared channel for the
enhancement layer and retransmissions, high
efficiency can be achieved with a proper schedul-
ing algorithm, and
OVSF
code is saved with the
use
of
the DSCH.
ACKNOWLEDGMENTS
This work is part of the ARROWS project, par-
tially funded by the European Commission under
the IST framework (IST 2000-25133) and the
Spanish Research Council under grant TIC2000-
2813-CE. The authors wish
to
express their grati-
tude to the other members of the consortium
(Teleconi Italia Labs, Telefbnica I+D, Universi-
ty
of Limerick, and Inesc Porto) for valuable dis-
cussions, with special thanks to Claudio Guerrini
and Paolo Goria from TILab. The help provided
by
J.
Sanchez on some of the simulation work is
also acknowledged.
REFERENCES
[l]
S.
K.
Das et al., "A Call Admission and Control Scheme
for
QoS
Provisioning in Next Generation Wireless Net-
works," Wireless Networks
6,
2000, pp. 17-30,
[2]
Z.
Lui and M. El Zarki, "SIR Based Cal1,Admission Con-
trol for DS-CDMA Cellular Systems,"
/€€€
JSAC,
vol. 12,
1994.
[3]
I.
F.
Akyldiz, D. A. Levine, and
I.
Joe, "A Slotted CDMA
Protocol with
BER
Scheduling for Wireless Multimedia
Networks,"
IEEEIACM
Trans. Net., vol. 7. no. 2, Apr.
1999, pp. 146-58.
[4] W. Rave et al., "Evaluation of Load Control Strategies in
an UTWFDD Network," Proc.
VTC
'01.
pp. 2710-14.
[5] http://www.ist-arrows.upc.es
[6] http://www.3gpp.org
[7]
H.
Holma and A. Toskala, Eds.. W-CDMA
for
UMTS,
Wiley, 2000.
BIOGRAPHIES
ORIOL
SALLENT
[M'98] (oriol@tsc.upc.es) received Engineer
and Doctor Engineer degrees in telecommunication from
the Universitat Politecnica de Catalunya (UPC), Spain, in
1994 and 1997, respectively. He joined the Escola Tecnica
Superior d'Enginyeria de Telecomunicacio de Barcelona,
where he became assistant professor in 1994 and associate
professor in 1998. His research interests are in the field of
mobile communication systems, especially packet radio
techniques and radio resource management for spread-
spectrum systems. He received the Doctorate Award from
the Telecommunication Engineer Association
of
Spain in
1997 for his Ph.D. dissertation on multiple access protocols
for CDMA-based systems.- .
JORDl
PEREZ-ROMERO
[S'981 (jorperez@tsc.upc.es) obtained
an Engineer degree in telecommunications from the Escola
Tecnica Superior d'Enginyeria 'de Telecomunicacio of UPC in
1997. He joined the Radio Communic.ations Group in the
Department of Signal Theory, and Communications at UPC
where he obtained a Ph.D. degree in April 2001. He
is
cur-
rently.assistant professor in the field cif radio communica-
tions. He has been involved in different European projects,
and his main research areas are the packet transmission
mechanisms and the radio resource management strategies
for CDMA mobile communications networks.
FERNANDO
J. CASADEVALL [M'871 (ferranc@tsc.upc.es) received
Engineer
of
Telecommunication and Dr.Eng. degrees from
UPC in 1977 and 1983, respectively. In 1978 he joined
UPC, where he was an associate professor from 1983 to
1991. He
is
currently a full professor in the Signal Theory
and Communications Department. After graduation he was
concerned with equalization techniques for digital fiber
optic systems. He has also worked in the field of digital
communications with particular emphasis on digital radio
and its performance under multipath propagation condi-
tions. In the last 10 years, he has mainly been concerned
with the performance analysis and development of digital
mobile radio systems. In particular, his research interests
include cellular and personal communication systems, mul-
tipath transceiver design (including software radio tech-
niques), mobility, and radio resource management. Over
the last 10 years he participated in more than 20 research
projects funded by both public and private organizations.
In particular, he participated in the research project CODIT
and ATDMA in RACEll and RAINBOW in.ACTS, both funded
by the European Commission. Currently in the context of
the 5th European Framework Program he participates in
the IST projects WINEGLAS and ARROWS, as project man-
ager of the latter. He has published more than 50 technical
papers in international conferences and magazines. From
October 1992 to January 1996 was responsible for the
Information Technology Area in the National Agency for
Evaluation and Forecasting (Spanish National Research
Council).
RAMON
AGUSTI [M'781 (ramon@tsc.upc.es) received an Engi-
neer of Telecommunications degree -from the Universidad
Politecnica de Madrid, Spain,
in
1973, and
a
Ph.D. degree
from UPC in 1978. In 1973 he joined the. Escola Tecnica
Superior d'Enginyers de Telecomunicacio de Barcelona,
Spain, where he became
a
full professor in 1987. After
graduation he worked in the field
of
digital communica-
tions with particular emphasis on transmission and devel-
opment aspects in fixed digital radio, both radio relay and
mobile communications. In this framework he spent six
months at the Polithecnic (of Turin, Italy, 1976-1977, and
one month per year at the CNET laboratories, Lannion,
France, from 1980 to 1984. For the last 15 years he has
mainly been concerned with the performance analysis and
development of planning tools and equipment for mobile
communication systems, and he has published about 100
papers in these areas. He participated in the European pro-
gram COST 231 (1989-1996), Evolution of Land Mobile
Radio, and in COST 259 (Wireless Flexible Personalized
Communications) as Spanish delegate. He also participated
in RACE and ACTS European research programs in the past
(ATDMA, MICC, and RAINBOW projects) and in IST at the
present (WINE GLASS and ARROWS Iprojects) as well as in
many private and public funded projects. In this time he
has also been an advisor to Spanish and Catalonian gov-
ernmental agencies (DGTel, CICYT, ANEP, and CIRIT) on
issues concerning mobile communications.
He
received the
Catalonia Engineer of the Year prize in 1998. He
is
part of
the editorial board of several scientific international jour-
nals. At present his research interest lies in the area of
mobile communications with special emphasis on CDMA
systems, packet radio networks
,
radio resource allocation,
and
QoS.
106
IEEE
Communications
Magazine
February
2003
