The 17th Annual IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC'06)
A SEAMLESS SERVICE CONTINUITY SCHEME FOR ENHANCED NETWORK
PERFORMANCE IN UMTS/WLAN NETWORKS
George Lampropoulos, Nikos Passas, Lazaros Merakos
Department of Informatics and Telecommunications
University of Athens
Panepistimiopolis, Ilisia
15784 – Athens, Greece
{glambr|passas|merakos}@di.uoa.gr
ABSTRACT
The proliferation of wireless access technologies, mostly
wireless LANs, along with the ubiquitous coverage of cellular
networks, has spurred many researchers to consider schemes
for integrated network provision. In overlaying network
environments, flexible resource management and service
continuity during handover play a crucial role in the quality-
of-service seen by the user. This paper describes an integrated
UMTS/WLAN architecture that manages to establish every
connection in a mobile terminal through the most appropriate
access network, based on both network policies and user
preferences. Moreover, connections are handed over
independently between UMTS and WLAN, aiming at
seamless service continuity. A detailed simulation model is
used to measure the performance of the proposed architecture
against existing schemes.
I. INTRODUCTION
In recent years, the exploitation of wireless LANs (WLANs)
as a supplementary technology to cellular networks (mainly
GPRS and its successor UMTS) is gathering momentum,
while several efforts are focusing on the standardisation of
different UMTS/WLAN integrated schemes [1-2]. Mobility
management is an important aspect in such architectures,
where the main target is to integrate WLAN and UMTS in a
way that provides seamless service continuity. A critical part
of handover in multi-network environments is the
consideration of the numerous parameters located in the
terminal and the network, in order to choose the best possible
access network. The plethora of the proposed architectures
can be categorised into three major types of interconnection
[3]: loose, tight and very tight coupling. In loose coupling,
the point of integration is after the interface of GGSN with
the IP network, whereas in tight and very tight coupling, the
inter-connection is made at the core network level and the
access network level of UMTS respectively. Loose coupling
solutions offer simple and cost-effective implementation
prospects, at the expense of larger handover execution time.
In contrast, tight and very tight coupling are considered more
complex but efficient in providing seamless connectivity.
Considerable efforts have been placed on tight architectures
as a means towards seamless service continuity. In [4], an
interworking entity hides the particularities of WLAN and
allows it to connect directly to the SGSN. A large part of the
existing infrastructure is reused in a way that ensures at least
service continuity. Despite that, when the signal is lost in one
network, all the connections of the terminal are handed over
1-4244-0330-8/06/$20.00©2006 IEEE
Alexandros Kaloxylos
Dept. of Telecommunications Science and Technology
University of Peloponnese
St George Park, Karaiskaki Area
22100 – Tripoli, Greece
kaloxyl@uop.gr
to the other network. A more flexible architecture has been
described in [5], where terminals are able to have connections
over UMTS and/or WLAN at the same time. However, no
sophisticated architecture for efficiently handing over the
connections is given. In terms of efficient radio resource
management, [6] proposes a scheme for managing terminals
in heterogeneous networks following a policy-based
perspective. The network manages to distribute the terminals
properly, according to parameters such as user preferences,
network policy, network availability, and cost, but treats all
connections as one group during handover.
In this paper, a tight coupling scheme for handling
connections independently and with policy considerations in
integrated UMTS/WLAN networks is presented. The main
system architecture is described in section II, while the
procedures of connection establishment and handover are
presented in section III. In section IV, the simulation model
along with evaluation results is described. Finally, section V
concludes the paper.
II. SYSTEM ARCHITECTURE
A. General Description
The proposed architecture considers a UMTS system
interconnected with various WLAN hot-spots dispersed in the
area of UMTS coverage. All active terminals have a
signalling connection with the UMTS established at all times.
Their multimode capability enables them to connect to both
networks simultaneously, when WLAN coverage is also
available, and handover their connections from one network
to another. Moreover, new functionality is introduced in order
to take proper handover decisions and differentiate between
disparate connection requirements. The reference
UMTS/WLAN architecture is depicted in Fig. 1, while a more
detailed and generalised description of this architecture can be
found in [7].
As shown in Fig. 1, UMTS and WLAN networks are
interconnected in a tight coupling way. This is feasible
through an RNC Emulator (ERNC) that manages the WLAN
network resources similar to a UMTS RNC. More
specifically, ERNC collects information concerning the
resources of the attached APs (traffic load, serving terminals,
signal strength of the terminals, etc.) and is responsible for the
establishment of radio paths with the User Equipment (UE).
In order to provide sophisticated usage of the radio resources
and advanced handover decision capabilities, three new
functional entities are introduced: the Mobile Terminal
The 17th Annual IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC'06)

Controller (MTC), the Advanced Radio Resource Controller
(ARRC) and the Light Radio Resource Controller (LRRC).
ARRC is focused on the radio resource management of the
supervising network elements. ARRC is an extended version
of RRC. It manages the resources of the underlying UEs and
Node-Bs, while it communicates with LRRC to acquire load
traffic information. LRRC is limited to gathering and
reporting this information to ARRC and does no further
sophisticated action. Therefore, it is referred to as “light”
RRC. Finally, MTC is integrated into the RRC of the UE and
is responsible for radio link monitoring and processing of
exchange required in heterogeneous networks, where users
will have the freedom to dynamically choose between
different access technologies or even operators, will not
favour for centralised processing.
B. Protocol Stacks
The location of the new functional entities and the necessary
modifications in the system protocol stacks are presented in
Fig. 2 and 3 for the control and the data plane respectively.
ARRC RANAP

terminal parameters. The important property of these new GMM/SM

RLC
SCCP

functional entities is that they are incorporated in existing MTC WLM

IP
UMTS
M3UA

SCTP
GMM/SM

network components and protocols and do not burden the RLC WLAN
LLC
MAC IP

L2
RANAP

SCCP
GTP-C GTP-C IP

network infrastructure. This scheme is similar to the UMTS

MAC
WLAN
MAC
L1 L1
M3UA
SCTP
TCP/UDP TCP/UDP

Integrated CRRM concept by 3GPP [8] and is a logical choice L1 L1

RNC
IP IP IP

if minimal alterations in existing UMTS functionality is the UE

L2

L1
L2

L1
L2

L1
L2

L1
WLM LRRC RANAP
main target. The exact placement of these entities in the IP SCCP SGSN GGSN

UMTS and WLAN protocol stacks will be described in the WLAN

LLC
L2 IP M3UA

SCTP

next subsection. WLAN

MAC
L2
IP

L2

L1 L1 L1 L1

ERNC Modified or new protocols

AP
Functional Network
entities Components
MTC ARRC Fig. 2. Protocol stacks in the system (control plane)

In the control plane, RRC in the MT and the RNC are

Node-B RNC SGSN GGSN PDN
UE enhanced with MTC and ARRC functionality respectively. In
the WLAN part, LRRC is placed besides a new protocol
AP ERNC HLR entity named WLAN Management protocol (WLM) that is
responsible for establishing the radio connections over
LRRC WLAN with its counterpart in the UE. RANAP message
semantics have been slightly changed to uniquely indicate
Fig. 1. UMTS/WLAN reference architecture
handover per connection and allow independent handling of
resources.
Concerning the interaction between the new functional APPL

entities for advanced handover decision, the main aim is to IP IP

TCP/UDP

IP

minimise signalling exchange. Therefore, some fast changing APPL

PDCP

RLC
GTP-U

TCP/UDP
GTP-U GTP-U

TCP/UDP TCP/UDP
GTP-U

TCP/UDP

parameters are used close to the point of generation (e.g., the

TCP/UDP
IP IP IP IP
IP
MAC L2 L2 L2 L2 L2 L2

UE for radio signal measurements), while others are IP

NASL

PDCP
PDCP
L1 L1 L1 L1 L1 L1 L1

RNC SGSN GGSN HOST

requested when needed (e.g., WLAN traffic load report WLAN
LLC RLC

towards RNC). WLAN

MAC
UMTS
MAC
IP

The system assumes two major points for radio resource L1

UE
L1 WLAN
LLC
L2
IP GTP-U

management decisions: i) the MTC (in the UE) and ii) the WLAN
MAC L2
TCP/UDP

IP

ARRC (in the RNC). Each time the UE wants to establish a L1 L1

L1
L2

L1
Modified or new protocols

new connection, the locally kept parameters are processed in AP

ERNC Tunnel

order to choose the target network. More specifically, the Fig. 3. Protocol stacks in the system (data plane)
user, the terminal and the service profiles are filtered along
with the network availability measurements and the result is
Routing of downlink packets for each connection is feasible
compiled in an ordered list. In this way, the UE prioritises the
due to network layer routing in the UE and the SGSN, which
candidate access networks and sends this provisional decision
is based on the NSAPI/IMSI pair. Moreover, since each
to RNC for further processing. Upon receipt, the RNC may
connection is associated with one PDP Context, one NSAPI
re-order the list, based on the traffic load of the candidate
and one RAB ID in a one-to-one relationship, the allocation
access networks and the operator’s policy. The first network
of resources for each connection in both access networks can
in the list is the target network and is reported to the UE as
be made in terms of different RABs.
the final decision. In this way, the network has full control of
the radio resources, but the user preferences are taken into The routing of uplink packets can be based on the same
account as well. Also, some processing load is transferred to identification pair as in the downlink direction. However, the
the UEs, unburdening the RNC. This is considered a selection of the proper radio interface takes place in the
significant improvement, since the amount of information Network Access Selection Layer (NASL). NASL is
The 17th Annual IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC'06)

configured appropriately in order to route IP packets over the UE AP Node B RNC SGSN ERNC GGSN

correct interface even after network attachment point changes.

RRC signaling connection establishment

III. CONNECTION ESTABLISHMENT AND HANDOVER Listen signals from

neighboring APs or

PROCEDURES Node-Bs

Activate PDP Context Request

Create PDP Context Request
In this section, the establishment of an outgoing connection
Create PDP Context Response
through WLAN and the handover from UMTS to WLAN
RAB Assignment Request
procedures are presented in more detail. Handover from Measurement Control
WLAN to UMTS follows a similar procedure. The first
Build ordered list
procedure is presented in Fig. 4. After the UE powers on, it of candidate
establishes a signalling connection with the UMTS that access networks
Measurement Report
remains active until power-off. The UE listens to radio signals
from other APs or Node-Bs periodically. When a new ARRC selects
access network
connection has to be established, the UE issues an Activate
Association to WLAN
PDP Context Request message to the SGSN. This message
contains several parameters such as QoS information, the WLAN Association

requested NSAPI identifying the connection and a PDP RAB Assignment Response

address field (if a dynamic PDP address is assigned by the RAB Assignment Request

network, this field will be filled by the APN Server). Upon Resource Reservation

receipt, SGSN checks if the user has a valid subscription and

chooses the proper GGSN for the UE. By sending a Create
PDP Context Request message, SGSN asks from the GGSN
to allocate an IP address for the UE and establish a tunnel
between SGSN and GGSN. When SGSN receives a Create
PDP Context Response, it orders the serving RNC to allocate
resources (RAB Assignment Request). RNC, in turn, asks from
the UE to report its measurements. This triggers a process that
prioritises the accessible APs or Node-Bs according to signal
strength, user, terminal and service profiles and creates an
ordered list that is transmitted to the RNC. This list is
checked to comply with the available network resources and
operator’s policy, and may be re-ordered by the RNC. If the
WLAN network is on top of the list, an ARRC message
indicates to the UE that it should associate with the WLAN.
With this message the UE is informed to configure the NASL
to route outgoing packets over the WLAN interface as soon as
the new path is established.
In the mean time, the RNC answers to the SGSN that it
cannot serve the current connection and SGSN issues a new
RAB Assignment Request towards the ERNC. After the end of
the reservation procedure, a Successful Resource Allocation
Indication message informs the ERNC that resources have
been reserved over WLAN. ERNC sends a RAB Assignment
Response to SGSN to report the successful outcome, while
SGSN sends an Activate PDP Context Accept message to the
UE. With this message, the establishment procedure is
completed and communication may commence.
The handover from UMTS to WLAN is illustrated in Fig. 5.
The UE monitors periodically for conditions that may initiate
a handover (e.g., signal deterioration, user preferences). When
such a condition arises for a specific connection, a
measurement report is sent to the RNC. This message
contains an ordered list of candidate access networks
according to radio signal measurements, user, terminal and
service profiles.
Successful Resource Allocation Indication
RAB Assignment Response
Activate PDP Context Accept
Fig. 4. Establishment of an outgoing connection through
WLAN
When the RNC receives this report, it may re-order the
reported network list based on the available network
resources and the operator’s policy. If the target network on
top of the list is the same with the one that serves the specific
connection, then the request is simply rejected. Here, the
WLAN is assumed to be on top of the list and the serving
network for the specific connection is UMTS. Therefore, a
relocation preparation phase begins and RNC issues a
Relocation Required message towards the SGSN. SGSN
orders ERNC to allocate resources (Relocation Request and
Relocation Request Ack) and then issues a Relocation
Command to the RNC. The RNC informs the UE that it
should associate with the WLAN and that resources should be
reserved (Inter RAT Handover From UTRAN). In addition,
UE is informed to configure the NASL to send all outgoing
packets for this connection over WLAN, when the new path
is established. When WLAN association and resource
reservation procedures end, the UE informs the ERNC that
the resources have been reserved successfully (Inter RAT
Handover From UTRAN Indication) and ERNC reports the
detection of the relocated connection to SGSN (Relocation
Detect), which triggers the data flow over WLAN. Finally,
the completion of handover procedure is reported to SGSN
(Relocation Complete) and a pair of Iu Release messages is
exchanged between RNC and SGSN in order to release old
radio bearer for this connection.
The 17th Annual IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC'06)
UE AP Node B RNC SGSN
Listen signals from
neighboring APs or
Node-Bs
Build ordered list
of candidate
access networks
Measurement Report
ARRC selects
access network
Relocation Required
Relocation Request
Relocation Request Ack
Inter RA T Handover From UTRAN
Relocation Command
WLAN Association
Resource Reservation
Inter RA T Handover From UTRAN Indication
Relocation Detect
Relocation Complete
Iu Release Command
Iu Release Complete
Fig. 5. Connection handover from UMTS to WLAN
IV. SIMULATION MODEL AND RESULTS
To evaluate the performance of the proposed architecture, a
reference system following the same architectural principles
with the emerging “Generic Access to A/Gb interface”
standard (also known as “Unlicensed Mobile Access-UMA”)
[2] was considered. According to [2], the UE can be either in
WLAN or GPRS/UMTS mode, and handover involves the
relocation of all connections over the same network. To make
the comparison more interesting, a load balancing mechanism
was also considered in both the UMA and the proposed
architecture.
More specifically, the simulation model evaluated three
different architectures: i) simple UMA, ii) load-balanced
UMA and iii) the proposed system. In the case of simple
UMA, the UEs initially connect to WLAN (WLAN preferred
mode is assumed) and handover to the alternative network is
based on user preferences only (no network decision
involved). In the load-balanced UMA, UEs are allocated to
the network with the less traffic load. Finally, in the proposed
system, the same decision policy with load-balanced UMA is
followed, but per connection and not per UE.
The simulation model was built on ns-2 [9]. The architecture
modelled two RNCs and one ERNC, all located under the
same SGSN in a tight coupling way. In order to investigate
the performance of each of the three different architectures in
a dynamic multi-network environment, two different areas
were considered: i) a common UMTS/WLAN area and ii) a
UMTS-only area. In the first area, one RNC and one ERNC
formed an integrated network where dual-mode terminals
ERNC were able to communicate with UMTS and/or WLAN and
handover from one to another. In this composite area, it has
been assumed that a WLAN hot-spot was overlaid by a
UMTS cell. The second area, where only UMTS connectivity
was provided by one RNC, served to model UEs that
leave/enter the first area.
During simulation, the terminals initiated and terminated
connections over WLAN or UMTS according to each
architecture (i.e., UMA, load-balanced UMA, proposed
system). Three types of applications were considered, i.e.,
voice calls, video steaming and ftp sessions. For each kind of
application, sessions followed a Poisson distribution with
mean arrival rate of 100 seconds, resulting in approximately 3
new connections for each terminal every 100 seconds. The
duration of these sessions was considered to be governed by
an exponential distribution of 100 seconds mean. Voice calls
were modelled as constant bit rate traffic streams of 64 kbps
(128 kbps bi-directional), while video sessions were modelled
as ON-OFF sources with activity period of 1/3 and mean rate
of 384 kbps. Ftp sessions requested a 250kB file every 40
seconds. Concerning the bandwidth of the system, the UMTS
cell was assumed to provide 2 Mbps and the WLAN hot-spot
11 Mbps.
simple UMA
UMA (Load balancing)
Proposed system (Load balancing)
Handover Blocking Probability
0.1
0.01
1E-3
1E-4
0.3 0.4 0.5 0.6 0.7
System Load
Fig. 6. Handover blocking probability
The simulations included light, medium and heavy traffic
load scenarios (8, 13 and 18 terminals respectively).
According to the application configuration per UE, this
corresponded to approximately 30%, 50% and 70% of the
system capacity. The target was to evaluate the performance
in terms of handover blocking probability, connection
blocking probability and number of handover procedures
executed by the network. For all the scenarios, the simulation
duration was 50000 seconds, which was sufficient to provide
with stable results.
As shown in Fig. 6, the proposed architecture manages to
decrease the handover blocking probability even when a load
balancing mechanism is assumed in UMA. More specifically,
the handover blocking probability in the proposed
The 17th Annual IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC'06)
architecture is actually 100 times better than in the simple
UMA in the medium load scenario and 5 times in the heavy
load scenario. In light load conditions, the proposed system
has zero handover blocking probability. Compared to load-
balanced UMA, the handover blocking probability is reduced
from 1.6% to 0.1% in the medium load scenario and from
8.9% to 5.8% in the heavy load scenario. This improvement is
mainly attributed to the flexibility the handover management
per connection offers.
simple UMA
UMA (Load balancing)
Proposed system (Load balancing)
Connection Blocking Probability
0.1
0.01
0.3 0.4 0.5
System Load
Fig. 7. Connection blocking probability
Similar results were produced for the connection blocking
probability as well (Fig. 7). In simple UMA, the ratio of the
new sessions that get blocked is 10%, 14% and 18% for the
three load scenarios respectively. With load-balanced UMA,
these values are considerably reduced and reach 0.52%, 1%
and 4% for the same scenarios. However, in the proposed
architecture these values are further decreased to 0.5%, 0.6%
and 2% respectively.
Although the proposed system manages to enhance the
overall performance, it is clear that this has an impact on the
number of handover procedures. In fact, it would be expected
that the handovers in the proposed system would outnumber
the ones in the other architectures. However, the proposed
architecture triggers less handover procedures compared to
simple UMA in the medium load scenario. This is because a
large percentage of the handovers requested by the terminal
do not proceed due to the final decision taken in the RNC.
The underlying reason for this is that the system is already
load balanced and no handover is needed. Instead, in simple
UMA, no handover request is blocked at the RNC, since no
mechanism as in the two other architectures exists. For the
load-balanced UMA and the proposed system, it is clear that
the difference is contingent on the number of connections per
terminal. Here, three connections on the average per UE have
been assumed, so the load-balanced UMA is approximately
three times better in terms of handover signalling overhead.
This, however, can be considerably mitigated by grouping
signals destined to the same network elements. This is similar
to the way signalling for different radio bearers are
aggregated in standard RANAP messages. If this feature is
used, then it has been measured that the signalling overhead
in the proposed architecture is only 1.27 times greater than in
the load-balanced UMA.
Another interesting result concerning the number of
handovers in the system is the fact that the percentage of the
handover requests from UEs that do not necessitate handover
according to RNC decision is 69% in the load-balanced
UMA, while it reaches 73% with the proposed architecture.
This verifies the improved effect of a load balancing
mechanism when applied per connection instead of per
terminal.
V. CONCLUSIONS
Emerging WLAN technology and existing cellular network
infrastructure comprise a unique challenge for providing
advanced services to the end users. Seamless service
continuity can be achieved if flexible architectures exist. In
this paper, a UMTS/WLAN integrated architecture has been
presented and evaluated. The scheme provides seamless
service continuity by introducing sophisticated entities that
allow per connection handover and flexible connection
0.6 0.7 establishment. Performance results have shown that this can
improve the network availability, and reduce both handover
and connection blocking rates at the expense of a small
signalling overhead increase.
ACKNOWLEDGEMENT
This work was performed in the context of the project entitled
“PENED 03ED 909” co-funded by the General Secretariat for
Research and Technology of Greece and the European Social
Fund.
REFERENCES
[1] 3GPP TR 22.934 V6.2.0, “Feasibility study on 3GPP system to Wireless
Local Area Network (WLAN) interworking (Release 6),” September 2003.
[2] 3GPP TR 43.901 V6.0.0, “GSM/EDGE Radio Access Network;
Feasibility Study on generic access to A/Gb interface (Release 6),” August
2004.
[3] G. Lampropoulos, A. Kaloxylos, N. Passas, and L. Merakos, “Handover
Management Architectures in Integrated WLAN/Cellular Networks,” IEEE
Communications Surveys and Tutorials, Fourth Quarter 2005, Vol.7 No.4.
[4] A. Salkintzis, C. Fors, R. Pazhyannur, “WLAN – GPRS Integration for
next-generation mobile data networks,” IEEE Wireless Communications,
October 2002.
[5] M. Jaseemuddin, “An Architecture for Integrating UMTS and 802.11
WLAN Networks,” Proceedings of IEEE Symposium on Computers and
Communications (ISCC 2003), Antalya, Turkey, pp. 716-723, 2003.
[6] R. Ferrús, et. al., “Vertical Handover Support in Coordinated
Heterogeneous Radio Access Networks,” IST Mobile Summit 2005, Dresden
(Germany) 19-22 June, 2005.
[7] A. Kaloxylos, G. Lampropoulos, N. Passas and L. Merakos, “A Flexible
Mechanism for Service Continuity in 4G Environments,” Elsevier Computer
Communications Journal, Special Issue on “End-to-end QoS Provision
Advances,” Volume 29, Issue 6, Pages 669-798 (31 March 2006).
[8] 3GPP TR 25.881 V5.0.0, “Improvement of RRM across RNS and
RNS/BSS (Release 5),” December 2001.
[9] The Network Simulator - ns-2, http://www.isi.edu/nsnam/ns/