Page 1
$3� �� �<Æ 0A �7H ë�H
Master’s Thesis
ì�ríß� RAN (Radio Access Network)���×þ�
~½ÓZO�_� $í0px ì�r$3�
Performance Analysis of Decentralized RAN (Radio Access
Network) Selection Schemes
�ª� nq �&³ (� C & Yang, Sook Hyun)
��������íß��<Æõ� ���íß��<Æ���/BN
Department of Electrical Engineering and Computer Science
Division of Computer Science
ô�Ç ²DG õ� �<Æ l� Õüt "é¶
Korea Advanced Institute of Science and Technology
2005
Page 2
ì�ríß� RAN (Radio Access Network)���×þ�
~½ÓZO�_� $í0px ì�r$3�
Performance Analysis of Decentralized RAN
(Radio Access Network) Selection Schemes
Page 3
Performance Analysis of Decentralized RAN
(Radio Access Network) Selection Schemes
Advisor : Professor Yeom, Ikjun
by
Yang, Sook Hyun
Department of Electrical Engineering and Computer Science
Division of Computer Science
Korea Advanced Institute of Science and Technology
A thesis submitted to the faculty of the Korea Advanced
Institute of Science and Technology in partial fulfillment of the
requirements for the degree of Master of Engineering in the
Department of Electrical Engineering and Computer Science,
Division of Computer Science
Daejeon, Korea
2004. 12. 28.
Approved by
Professor Yeom, Ikjun
Advisor
Page 4
ì�ríß� RAN (Radio Access Network)���×þ�
~½ÓZO�_� $í0px ì�r$3�
�ª� nq �&³
0A �7Hë�H�Ér ô�DzDGõ��<Æl�Õüt"é¶ $3����<Æ0A�7Hë�Hܼ�Ð �<Æ0A�7Hë�Hd����
0A"é¶�r\�"f d���� :�xõ� �%i�6£§.
2004�� 12�Z4 28{9�
d����0A"é¶�©� %i� e�� ï�r (���)
d����0A"é¶ Ðqt 5px §>= (���)
d����0A"é¶ ô�Ç 8� ú (���)
Page 5
MCS
20033371
�è¡ \8; g@. Yang, Sook Hyun. Performance Analysis of Decentralized RAN (Radio
Access Network) Selection Schemes. (Ûo»ÿ� RAN (Radio Access Network) 9ßÌv
�ç¡(« q� )ç��ûB (Ûo)o�. Department of Electrical Engineering and Computer Science,
Division of Computer Science . 2005. 28p. Advisor Prof. Yeom, Ikjun. Text in
English.
Abstract
With the widespread deployment and popular usage of wireless networks, various wireless
access network technologies are being developed. Taking advantages of these access network
technologies will be able to offer a lot of possibilities for increasing bandwidth, accessing
the Internet, and expanding the service range. For supporting interoperability between
a variety of access network technologies, one of the major challenges is the creation of a
new handoff protocol across heterogeneous networks. Because a mobile host with multiple
network interfaces simultaneously has more than one available access network, a new handoff
protocol should perform to discover RAN (Radio Access Network) and select the optimal
RAN among found RANs. We present four RAN selection schemes to discover and choose
the optimal RAN. These four schemes are simple, easy to implement and generic solutions
for vertical and horizontal handoff. Four selection schemes discover available access networks
utilizing a static/dynamic period or a CAN (Candidate Access Network) without the help of
centralized servers, thereby removing unnecessary power consumption without QoS (Quality
of Service) degradation. After implementing a simulator for the wireless overlay network
composed of a different type of networks, we evaluated four selection schemes to observe the
achieved bandwidth, number of handoffs and total amount of power consumption using our
simulator. Results show that the usage of a dynamic period or a CAN reduces large amount
of power consumption without degrading achieved bandwidth.
i
Page 6
Contents
Abstract i
Contents iii
List of Tables iv
List of Figures v
1 Introduction 1
2 Background and Related works 4
3 RAN selection schemes 7
3.1 How to monitor available access networks . . . . . . . . . . . . . . . . . . . . 7
3.2 How to determine the optimal alternative RAN . . . . . . . . . . . . . . . . . 10
4 Network models 13
5 Performance analysis 17
5.1 Achieved bandwidth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
5.2 Number of handoffs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
5.3 Power consumption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
6 Conclusion 25
Summary (in Korean) 26
References 27
iii
Page 7
List of Tables
3.1 Five selection schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
4.1 Wireless Network’s Coverage, Bandwidth, Power consumption . . . . . . . . . 14
4.2 The composition of access points or base stations . . . . . . . . . . . . . . . . 16
4.3 802.11b network adapter’s On/Off transition power consumption . . . . . . . 16
5.1 Simulation parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
5.2 Percentage of four selection to a continuously active scheme on power con-
sumption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
iv
Page 8
List of Figures
2.1 All IP-based Multinetwork . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
3.1 Single CAN monitoring vs. Multiple RAN monitoring . . . . . . . . . . . . . 10
3.2 Alternative access network selection flow chart . . . . . . . . . . . . . . . . . 11
4.1 Wireless overlay network structure . . . . . . . . . . . . . . . . . . . . . . . . 13
4.2 Simulation network topology . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
5.1 Achieved downlink bandwidth . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
5.2 Number of handoffs (per node) . . . . . . . . . . . . . . . . . . . . . . . . . . 22
5.3 The percentage of vertical handoffs (per node) . . . . . . . . . . . . . . . . . 23
5.4 Power consumption per seconds . . . . . . . . . . . . . . . . . . . . . . . . . . 24
v
Page 9
1. Introduction
With successful deployment and popular usage of wireless networks, a large variety of wireless
access network technologies, including second- (2G) and third- generation (3G) celluar, satel-
lite, WiBro/WiMax (IEEE 802.16), Wi-Fi (IEEE 802.11), and Bluetooth (IEEE 802.15),
have emerged. These multiple wireless networks using the spectrum of different frequencies
were independently designed, implemented, and operated to meet different requirements for
mobility, data rates, services, and so on. With advantages of these independently existing
networks, multiple networks are able to offer a lot of possibilities for increasing bandwidth,
accessing the Internet, and expanding the coverage area. For the use and coordination of
heterogeneous networks, there are some trials to integrate and internetwork between these
different types of networks, ultimately toward the fourth generation (4G) of wireless com-
munications [1, 2, 3, 4, 5].
In an environment of multiple wireless access technologies, one of the major challenges
is a handoff problem. Traditional handoff decision algorithms for homogenous networks
have several limitations [6]. Traditional handoff decision algorithms only use signal strength
as a handoff metric, so that traditional algorithms do not provide a sufficient criteria to
compare heterogeneous networks which have different characteristics, such as communication
charge, bandwidth, coverage, and power-efficiency. Furthermore, traditional handoff do not
allow user selection of RAN (Radio Access Network), even if one of heterogeneous networks’
features is that the user can select RAN among available networks based on user’s utilization
policy and the network status of RANs (e.g., available bandwidth, latency, bit error rate,
etc).
A new handoff problem in the 4G network can be divided into two small problems: access
discovery and access selection. The access discovery is that a mobile host or the network
discovers the available access networks and monitors access networks’ status in that area.
The access discovery has three methods [6]: centralized, distributed, and a combination both.
A centralized access discovery is that the network collects the available access networks in
that area and announces the collected RANs’ information to a mobile host. On the contrary,
a distributed method means that a mobile host itself searches for the available RANs. In
this distributed method, a mobile host’s power management is essential because multiple
network adapters should be turned on for monitoring different types of access networks.
The access selection means to select the optimal access network among available access
1
Page 10
networks, which is found after access discovery. The optimality of access network depends
on the selection parameters, such as user’s preferences, communication charge, available
bandwidth, power dissipation of network adapters, etc. The access selection can occur before
handoff-trigger(Pre-selection) or after handoff-trigger (Post-selection). These two selection
methods have trade-off between the chosen access network’s optimality and handoff delay.
Pre-selection does not increase handoff delay, but the selected access network cannot be
necessarily optimal. On the contrary, although post-selection chooses the optimal access
network, post-selection increases handoff delay.
There are several works focusing on a handoff problem in the 4G network. In [3],
they proposed energy-efficient and centralized access discovery mechanism with a access-
information-collecting agent in a network. WISE (Wise Interface SElection) [8] also sug-
gested a centralized server which monitors available access networks instead of a mobile host.
WISE selected independently access networks used for downlink and uplink. On the other
hand, distributed handoff schemes are suggested in [9, 10]. PPM (Power and Performance
Management) [9] tried to decrease power consumption by choosing maximum power-saving
network interface. To support seamless roaming between 802.11b and GPRS or CDMA1X
network, a mobile host decides whether a 802.11b network is available and proper [10]. In
[6, 7], a policy-based algorithm with various kinds of metrics and a cost function are utilized
for vertical handoff.
The goal of this thesis is to evaluate performance of four selection schemes which are
simple, easy to implement and applicable to all networks without the help of centralized
servers. Because these schemes choose an optimal alternative RAN before a handoff is trig-
gered, these schemes enable simultaneous binding with two access networks so that handoff
delay or loss rate can be decrease. Four selection schemes are divided on basis of the fol-
lowing methods: static/dynamic period, and a CAN (Candidate Access Network). After we
constructed network models, we implemented a new simulator for a wireless overlay network
environment which is composed of a different type of wireless networks. A mobile host with
all types of network interfaces moves according to scenario based on the random waypoint
algorithm. We observed number of handoff occurrences, achieved downlink bandwidth, and
total amount of power consumption at two simulation topologies which have sufficient or
insufficient network resources.
In our results, all four selection schemes achieved similar downlink bandwidth to a con-
tinuously active scheme’s downlink bandwidth, even if four RAN selection schemes do not
always turn on all network interfaces. Dynamic or a CAN gives large amount of energy-
saving. However, when network resources are sufficient, a CAN method triggers too many
2
Page 11
handoffs. A CAN method does not always guarantee small amount of power consumption.
In that, when network resources are not enough, static with a CAN consumes more power
than dynamic without a CAN.
The rest of this thesis is organized as follows: In Chapter 2, we present an overview
of related works with the concept of 4G network and handoff schemes for the 4G network.
Chapter 3 elaborates on four RAN selection schemes. Then, we describe network models
for simulating our five RAN selection schemes in Chapter 4. Performance analysis of five
RAN selection schemes using simulation is presented in Chapter 5, and the final Chapter
concludes this thesis.
3
Page 12
2. Background and Related works
The fourth-generation (4G) wireless networks mean heterogenous networks which support
multiple broadband wireless access technologies and seamless mobility across these networks
[11]. For supporting interoperability between heterogeneous networks, the 4G network will
use IP technologies as a common platform at the core network as shown in Figure 2.1 [11, 12].
A multiservice user terminal equipped with multiple network interfaces is being developed
Figure 2.1: All IP-based Multinetwork
for communicating through various networks. In the 4G network, the following challenges
exist [4].
• Access discovery: find available access networks
• Access selection: decide over which access network to connect at any point in time
• AAA support: verify the user identity (authentication) and the service to which the
user is entitled (authorization), and collect data to bill the user for the service (ac-
counting)
• Mobility management: support session continuity between access networks
• Profile handling: handle the user profile containing user’s personal preferences for
choice of access, application adaptation and VPN solutions
4
Page 13
• Content adaptation: application’s adaptation to current conditions
Among above challenges, our thesis tried to solve access discovery and access selection. For
this reason, we present an overview of related works on only access discovery and access
selection in this Chapter.
There was an attempt to perform access discovery without scanning a wide range of
frequencies [3]. This work depended on a centralized server in a network. A centralized
server monitors and collects available access networks, while managing a mobile host’s po-
sition (known by GPS), a mobile host’s IP address and so on. A BAS (Basic Access Signal)
protocol and a BAN (Basic Access Network) are proposed for a mobile host to communicate
with a centralized server. A RAN is selected as a BAN if a RAN has a broad coverage area,
preferably larger than that of the other RANs, and supports a reliable communication for
signaling transmission in which a high data rate is not necessary. With a BAS protocol and
a BAN network, a mobile host and a network exchange control information for performing
a handoff. This work enables amount of power consumption to decrease because only single
network interface for a BAN is activated while all other unused interfaces are off. How-
ever, this work has a problem that to know a mobile host’s position using GPS has several
limitations, and this solution requires additional network infrastructure.
WISE (Wise Interface SElection) [8] constructed a VDC (virtual domain controller),
which is a centralized component and periodically checks the state of BSs or APs. A mobile
host monitors its input and output queue length to check that the current access network is
valid. When an input or an output queue length falls below or exceeds over the predefined
threshold, a mobile host selects an alternative access network based on access information
which is provided from VDC. Because this method independently selects an alternative
access network which is energy efficient according to downlink (receive) or uplink (trans-
mit), downlink and uplink can be different access networks. For example, if CDMA1X’s
interface power dissipation at the receive mode is lower than 802.11b’s power dissipation
while 802.11b’s power dissipation at the transmit mode is lower than CDMA1X’s power
dissipation, a CDMA1X network interface is used as a downlink interface and 802.11b net-
work is determined as an uplink network. Simultaneously using two network interfaces is
not necessarily reducing amount of power consumption in some cases because two network
interfaces are not fully-utilized without idle time, even if a method selects two minimum
power-consuming network interfaces.
In [9, 10], distributed access discovery methods are proposed. A PPM (Power and Per-
formance Management) [9] determines what a network interface is the most suitable for
the application’s need. When an application starts or a currently used network does not
5
Page 14
guarantee application’s data consumption rate, a PPM turns on network interfaces and
selects network interfaces whose average throughput are greater than application’s data
consumption rate. Among selected networks, a PPM chooses maximum power-saving net-
work interface by measuring average power consumption due to the communication and the
usage of RAM. While a PPM stares at the change of application’s data consumption rate
or a network’s available bandwidth, a PPM decides whether a mobile host performs handoff
or not. However, because a PPM finds an alternative RAN after a handoff is triggered, a
PPM algorithm may increase handoff delay. Application’s data consumption rate used for
a PPM is hard to know in practice.
In [10], there was a trial to support a proactive method for roaming between 802.11b and
GPRS/CDMA1X network. Assuming that GPRS/CDMA1X networks are always reachable,
they focused on sensing availability of 802.11b. While periodically listening to and collecting
NAV (network allocation vector) which is the main scheme used in 802.11b, a mobile host
induces 802.11b network’s available bandwidth and access delay from collected NAVs. With
signal strength and available bandwidth/access delay, a mobile host decides whether a mobile
host roams from GPRS/CDMA1X to 802.11b network. Although a mobile host can estimate
available bandwidth and access delay without a network’s help, this work has a limitation
which is constrained to a handoff between GPRS/CDMA1X and 802.11b network.
In [2], this work suggested interface selection algorithm based on signal strength’s dif-
ferences between the current network and alternative network chosen with the priority of
interfaces. On the contrary, the other work in [6] stated that a traditional handoff decision
scheme, which only uses signal strength as a selection parameter, is insufficient for the 4G
network composed of various access networks. They indicated that the following new pa-
rameters should be considered as handoff parameters, not only signal strength: service type,
communication charge, system performance (channel propagation characteristics, bit error
rate, battery power), mobile node conditions (velocity, moving pattern, moving histories,
and location), and user preferences.
6
Page 15
3. RAN selection schemes
This chapter gives descriptions of four RAN selection schemes, simple, generic and easy
to implement. Because these four schemes do not target on a specific network, these four
schemes can be used for both of a horizontal and vertical handoff. These selection schemes
are based on a distributed access discovery method without centralized servers. A mobile
host itself monitors available access networks and selects the optimal alternative access
network with self-collecting RAN information. These schemes are divided on the basis of
access discovery method, such as static/dynamic turning-on period, with/without a CAN
(Candidate Access Network). Four schemes define that the optimal RAN is a RAN which
enables a mobile host to communicate with minimum amount of power dissipation while
guaranteeing QoS requirement. For deciding the optimal RAN which is an alternative to a
currently used RAN, four schemes commonly consider dynamic as well as static selection
parameters and select an alternative RAN according to the flow chart in Section 3.2. By
choosing the optimal RAN before handoff-trigger (pre-selection), selection schemes do not
cause handoff delay to increase. Pre-selection enables simultaneous binding which a mobile
host connects with an alternative and a currently used RAN, so that four schemes are
able to reduce loss from a link outage during a handoff. These four schemes only deal
with the downlink traffic because most of traffic-type are download, and four schemes can
be applicable to the uplink without change of method. Here, by QoS requirement, we
mean target downlink bandwidth. We assume that BSs or APs in a network monitor their
available bandwidth and periodically broadcast monitored available bandwidth by attaching
to advertisement messages.
3.1 How to monitor available access networks
To discover available access networks, a mobile host equipped with multiple network inter-
faces should scan and monitor various frequencies through turning on all network interfaces.
For this reason, the large amount of energy is consumed to perform access discovery. To
reduce network interfaces’ active time is essential for a battery-powered mobile host. We
focus on two methods which manipulate the monitoring period: static turning-on, dynamic
turning-on.
First, we include additional scheme for evaluating four schemes, a continuously active
7
Page 16
scheme. A continuously active scheme means that a mobile host turns on all interfaces
and dose not turn off until the battery-lifetime of a network interface ends. A mobile host
continuously stares at the whole status change of all access networks, so a mobile host is
able to select the exactly best radio access network and achieve downlink bandwidth which
is the closest to QoS requirement. Even though this selection scheme wastes large amount
of energy because of not turning-off network interfaces, this scheme is useful for comparing
with other four selection schemes.
Four selection schemes periodically turns on network interfaces to decrease unnecessarily
wasted energy. While a mobile host periodically turns on network interfaces, a mobile host
monitors and collects reachability/signal strength/available bandwidth of access networks.
At these four schemes, we assume that collected access information during monitoring time
is valid until next monitoring time comes. Although network status (reachability, signal
strength or available bandwidth) changes while a network interface is off, these periodic
schemes do not turn on network interfaces for updating new access information until next
monitoring comes. Based on this assumption, a mobile host selects the nearly best or the
optimal RAN among unexpired access information. If a mobile host still has unexpired
access information on a specific network type at a point of time, a mobile host does not turn
on a corresponding network interface for finding new access information.
Four schemes as shown in Table 3.1 use different methods which manipulate the mon-
itoring period such as static or dynamic, and have different number of monitored access
networks such as all or single.
Table 3.1: Five selection schemes
Selection scheme Monitoring Number of
period type monitored networks
1 Continuously active do not turn off All
2 Static without a CAN static All
3 Dynamic without a CAN dynamic All
4 Static with a CAN static Single (CAN)
5 Dynamic with a CAN dynamic Single (CAN)
Static monitoring period means that the monitoring period is fixed without considering
network status change. On the contrary, dynamic monitoring period changes along with
a mobile node’s velocity. These two monitoring period types basically manipulate each
8
Page 17
network interface’s monitoring period to be proportional to the corresponding network’s
coverage. We reflect network’s service coverage because network status changes slowly if
an access network’s coverage is large. A selection scheme with dynamic period exploits a
mobile host’s velocity, which is able to be extracted from signal strength variation of the
currently used access network. As signal strength variation is bigger, a mobile host adjusts
the monitoring period shorter. This is based on the fact that signal strength variation is
relatively large implies that a mobile host is moving fast, and vice versa. Differently from
the usage of GPS (Global Positioning System), this method using signal strength variation
does not have to turn on an additional network interface for measurement of a mobile host’s
velocity.
static period ∝ {network′s coverage}
dynamic period ∝ {network′s coverage
∆signal strength}
Periodic schemes are divided by number of access networks which a mobile host moni-
tors for selecting the optimal RAN. Number of monitored access networks can be single or
multiple. Here by, we mean that to monitor multiple RANs is for a mobile host to sense
all of network interfaces. When a mobile host monitors only single RAN, we call single
monitored RAN as a CAN. A CAN is the optimal and replacing a currently used RAN. A
mobile host pre-selects a CAN against a future handoff and periodically checks validity of
a CAN without turning on the rest of network interfaces. When using a CAN, a CAN’s
monitoring period is configured by above two period type, in that static or dynamic. A
mobile host confirms validity of a CAN by observing whether a CAN’s available bandwidth
is equal or greater than a mobile host’s downlink bandwidth requirement. If a CAN is not
valid, a mobile host chooses a new CAN among unexpired access information. Except the
case when a CAN is unreachable, a mobile host performs a handoff to a CAN, without
confirming whether a CAN currently guarantees downlink bandwidth requirement unless a
CAN’s monitoring time comes. If a CAN cannot be reachable, a mobile host does not use
the accumulated information and turns on all network interfaces. Then, a mobile host newly
updates its access information and finds a new CAN.
Selecting a CAN reduces larger amount of wasted power than monitoring multiple RANs
because a mobile host almost turns on single network interface for a CAN’s validity check
as shown Figure 3.1. In a Figure 3.1, a mobile host has three network adapters #1, #2 and
#3. Red arrows above the time-line means a corresponding network interface’s turning-on
when a mobile host uses a CAN method. Red arrows below the time-line is a corresponding
network interface’s turning-on while a mobile host monitors all of network interfaces. At the
9
Page 18
Figure 3.1: Single CAN monitoring vs. Multiple RAN monitoring
start time, a mobile host selects network adapter #1 as a CAN. Then, a mobile host only
turns on a CAN for checking validity of a CAN. When a CAN is not valid, a mobile host
tries to find a new CAN. As shown in Figure 3.1.(a), a mobile host turns on network adapter
#2 because access information on network #2 is expired. However, a mobile host does not
turn on network adapter #3 as given in Figure 3.1.(b). This is because access information
of network #3 is still valid.
A CAN can cause an oscillation between a currently used RAN and a RAN selected as a
CAN. To decrease an oscillation, we use a stability period which is a waiting period before
a handoff. If a mobile host performs a handoff to the used RAN, a mobile host waits for
a stability period. After that, if a currently used RAN still does not guarantee bandwidth
requirement, a mobile host retries to perform a handoff.
3.2 How to determine the optimal alternative RAN
A mobile host takes static and dynamic selection parameters for determining the optimal
RAN. Static parameters are made up of access network’s maximum coverage, each network
interface’s power dissipation at the receive/transmit/idle mode and user’s target downlink
10
Page 19
bandwidth. BSs or APs’ available bandwidth and change of signal strength are used as
dynamic parameters.
With above multiple parameters, a mobile host chooses the optimal alternative access
among available access networks according to the flow chart in Figure 3.2. There are a set N
Figure 3.2: Alternative access network selection flow chart
which is composed of access network identifier ni and a set M whose a element is a mobile
host identifier mj that communicates through ni ∈ N . While a mobile host mj performs
access discovery, a mobile host stores a set Navail,mj composed of available access network
identifiers. First, a mobile host makes up a new set with elements that guarantee a mobile
host’s downlink bandwidth requirement (mj,QoS) among the set Navail,mj . At this time, a
mobile host checks that the value obtained from signal strength reflection function f with
advertised available bandwidth (nk,ABW ) is equal or greater than mj,QoS . A signal strength
reflection function f causes an effect that nk,ABW decreases gradually as a distance from
the center of AP or BS is longer. Unless there exists an access network which guarantees
bandwidth requirement, a mobile host chooses an access network having maximum value
11
Page 20
with function f as the optimal alternative.
Second, among the set composed of access networks which guarantee downlink bandwidth
requirement, a mobile host finds access networks whose signal strength is increasing during
the monitoring time. Among these signal-strength-increasing networks, a mobile host finally
determines the optimal access network with minimum power dissipation. However, if there
is no access network whose signal strength is increasing, a mobile host finds minimum power
dissipation access network in the QoS guarantee set and decides minimum power-dissipation
access network as the optimal access network. Minimum power dissipation network means
that a network has minimum value in a cost function g. A cost function g is as follows:
g(nerec, neidle
, nk,BW , mj,QoS) =mj,QoS
nk,BW× nerec
+nk,BW −mj,QoS
nk,BW× neidle
, where nerec , and neidleare respectively power consumption at the receive and idle mode,
and nk,BW is network nk’s bandwidth. In this cost function g, we do not include power
consumption at the transmit mode because we only consider downlink traffic. Because
a network interface becomes idle mode after receiving data, we calculated the portion of
receive and idle time at the unit time, using the ratio of bandwidth requirement to network’s
bandwidth.
12
Page 21
4. Network models
We consider a network model overlaid with heterogeneous wireless networks as shown in
Figure 4.1 [5]. In such an overlay structure, a mobile host is able to simultaneously have more
than one heterogeneous access network which provides a connection. This model assumes
that a network comprised of high-bandwidth wireless cells covers a relatively small area and a
network which provides much lower bandwidth connection is over a much wider geographic
area. Maximum bandwidth change according to mobile host’s velocity is not applied to
simulator’s network model. Regardless of a mobile host’ velocity, available bandwidth is
only influenced by distance between a mobile host and the center of AP or BS. As a mobile
host gets far from the center of AP or BS, available bandwidth decreases.
Figure 4.1: Wireless overlay network structure
Each AP or BS in a network keep traces of available bandwidth, and periodically broad-
casts advertisement message attached its available bandwidth. However, available band-
width of AP or BS can be guaranteed at the very close area form AP or BS, because signal
strength of AP or BS decreases gradually as getting away from AP or BS. In simulation,
we assume that signal strength is proportionally to 1r , where r is distance from the center
of AP or BS. After a mobile host receives advertisement messages, a mobile host computes
13
Page 22
actual available bandwidth with signal strength reflection function f in Section 3.1. Then,
a mobile host stores these re-evaluated available bandwidth values.
Table 4.1: Wireless Network’s Coverage, Bandwidth, Power consumption
Network Type Coverage Bandwidth Power Consumption (W)
(km) (Mbps) Receive Transmit Idle
CDMA2000 2.5 2.4 0.495 2.805 0.082
1x EV-DO
802.16 (1) (5) (0.66) (1.32) (0.2)
802.11a 0.4 54 1.914 1.176 1
802.11b 0.4 11 1.35 2.25 0.75
802.11g 0.4 54 1.384 1.98 0.75
With the above network model, to configure a different type of networks which make
up simulation topology, we refer CDMA 2000 1x EV-DO for WWAN, 802.16 network for
WMAN, and 802.11a/b/g network for WLAN. Table 4.1 [13, 14, 15, 16] shows well-known
values of CDMA 2000 and 802.11a/b/g network. However, 802.16 are going on an attempt to
be standardized and not yet deployed widely. There are no practical values used for 802.16’s
coverage, bandwidth, and power dissipation. 802.16’s network information (coverage, band-
width, and power dissipation) is calibrated to approximately middle of CDMA2000’s and
802.11 networks’ values. Parenthesized values of Table 4.1 means calibrated one. Based on
Table 4.1, two simulation topologies with an area of 25 square kilometers (5km × 5km) is
used as shown in Figure 4.2. Figure 4.2(a) is the case when network resources are not enough
to guarantee QoS, but Figure 4.2(b) means when network resources are enough. Table 4.2
shows the composition of BSs or APs in two simulation network topologies. In Figure 4.2(a),
label 1 depicts insufficient network resource area which is composed of low bandwidth net-
works, CDMA 2000 and 802.16. On the contrary, a high bandwidth 11 ∼ 54Mbps such as
802.11a/b/g is supplied in all areas of Figure 4.2(b).
Mobile hosts moves according to scenario created by the network simulator 2’s node
movement generator [17], setdest. The setdest program generates node movement files using
random waypoint algorithm. In this algorithm, nodes select with a uniform distribution a
destination point (waypoint) from the region and a velocity in the interval [min, max]. When
a node arrives at a destination point, a node remains there for a given amount of time before
selecting another destination.
14
Page 23
(a) With insufficient network resources
(b) With sufficient network resources
Figure 4.2: Simulation network topology
Total consumed energy due to on/off transition is mostly decided, not by the amount
of on/off transition power consumption of each network adapter, but by total number of
turning-on network adapters. It is hard to find on/off power dissipation and transition
time of network adapters. For these reasons, we equally assign values in Table 4.3 [9]
to all these network adapter’s on/off power dissipation and on/off transition time. Five
proposed selection schemes in Chapter 3 are only considered downlink traffic, so that we
takes all transferred data as received data. To calculate power consumed for data transfer,
we induced receive time during the unit time with the below formula:
Trec =mj,rin(t)nk,BW
× Tunit
15
Page 24
Table 4.2: The composition of access points or base stations
CDMA1x 802.16 802.11a 802.11b 802.11g Total
Insufficient 4 10 10 31 10 65
network resource
Sufficient 4 20 21 25 26 96
network resource
Table 4.3: 802.11b network adapter’s On/Off transition power consumption
Transition time (ms) Avg. Power (W)
Off state entry 1 1.7
Off state exit 300 2.3
, where mj,rin(t) is downlink bandwidth of a mobile host mj at time t and nk,BW is currently
used network bandwidth from Table 4.1. After a network adapter receives data from a
network during the unit time, a network adapter becomes idle mode for the rest of the unit
time. Idle time is as follows:
Tidle = Tunit − Trec
Tidle =nk,BW −mj,rin(t)
nk,BW× Tunit
Thus, sum of energy which a mobile host consumes to transfer data, mj,e, is calculated by
the following formula:
mj,e =∑
t
(nk,erec × Trec + nk,eidle× Tidle)
where nk,erec , nk,eidleis currently used network’s receive, idle power dissipation from Ta-
ble 4.1.
16
Page 25
5. Performance analysis
We evaluate performance of five selection schemes through a simulation. In particular, we are
interested in examining the achieved downlink bandwidth, number of handoff occurrences,
and power consumption per seconds. For a simulation, we implemented a new simulator
using C/C++ based on network model of Chapter 4.
Table 5.1: Simulation parameters
Total simulation time 3,600 seconds
Area of network 5,000m × 5,000m
Network types CDMA1X, 802.16, 802.11a/b/g
Total number of BSs or APs 65 or 96
Total number of mobile nodes 100
Node pause time 0 seconds
Node max. speed 11 m/s ( .= 40km)
Min. period 5 seconds
Stability period 20 seconds
Simulation parameters are set-up as shown in Table 5.1. In 5km by 5km simulation
area, 100 mobile nodes move as the random waypoint mobility model with a pause time
of 0 and a maximum speed of 11m/s. For excluding the case that a mobile node’s signal
strength is fixed, we set a pause time as 0. A maximum speed is determined on the basis of
average speed of cars in the downtown. All of the mobile hosts have five types of network
interfaces: CDMA 2000 PCMCIA card, 802.16 network adapter, and 802.11a/b/g network
adapter. For Table 3.1’s 2-5 selection schemes, a minimum period is utilized as a minimum
coverage network’s monitoring period. For example, if coverage of a network A which has
the smallest coverage among wireless networks is 100m and certain network B’s coverage is
1km, a network A’s and a network B’s monitoring period are respectively 5 seconds and 50
seconds (5sec × 1km100m ) in static period schemes. Dynamic period selection schemes use these
static period as a minimum period of each network’s monitoring period. Dynamic schemes
increase these period when a mobile node moves slowly. Thus, a network B’s monitoring
period can be longer than 50 seconds in dynamic schemes. A minimum period 5 seconds is
17
Page 26
induced from dividing minimum coverage network (400m) by maximum velocity of a mobile
host (11m/s). A monitoring period is also used as a valid period for access information
which is whether a BS or AP’s signal strength is increasing or decreasing. For this reason,
we consider a radius (400m) instead of a diameter of minimum coverage network (800m). A
CAN’s stability period is configured to 20 seconds. Each result of five selection schemes is
average of 100 mobile nodes’ results.
5.1 Achieved bandwidth
Figure 5.1 shows results of the achieved downlink bandwidth when target bandwidths are
0.1, 0.2, 0.4, 0.8, 1.6 and 3.2Mbps. Five selection schemes gave similar results at two sim-
ulation topologies. This means that four schemes (2 ∼ 5 in Table 3.1) do not degrade
bandwidth, even though four schemes do not continuously turn on all of the network in-
terfaces and monitor all changes of networks. From 0.1 to 0.8Mbps, a target bandwidth is
enough guaranteed. 1.6 and 3.2Mbps are not guaranteed at two simulation network topolo-
gies. At the target bandwidth 1.6 and 3.2Mbps, 84% (1.4Mbps) and 63% (2.0Mbps) of
mobile hosts’ requirements are guaranteed at the insufficient network, but schemes meet
almost 98% (1.55Mbps) and 90% (2.9Mbps) of requirements at the sufficient network.
5.2 Number of handoffs
Figure 5.2 shows five selection schemes’ total number of handoffs which are composed of
horizontal and vertical handoffs. The ratio of number of vertical handoffs to total number
of handoffs is given in Figure 5.3. A mobile host performs a handoff when the currently
used network does not guarantee a target bandwidth. For this reason, a continuously active
scheme, which monitors all available networks and selects the best network, has the smallest
number of handoffs among five schemes as shown in Figure 5.2. On the contrary, four
schemes can cause more handoffs than a continuously active scheme since four schemes
do not necessarily choose the best network. As given in Figure 5.2, schemes with a CAN
occur more handoffs than schemes without a CAN. This is because schemes without a
CAN monitor all of network interfaces while schemes with a CAN only monitor a CAN for
validity check. Furthermore, schemes using a static period showed fewer number of handoffs
than schemes using a dynamic period. Because static monitoring periods are shorter than
dynamic, schemes using a static period are likely to choose more optimal network than those
using a dynamic period.
18
Page 27
Schemes without a CAN perform similar number of handoffs with a continuously active
scheme when target bandwidths are 0.1, 0.2, 0.4Mbps in Figure 5.2. Number of handoffs of
a continuously active scheme and schemes without a CAN scarcely increase until 0.4Mbps.
As known from Figure 5.1, two simulation network topologies have enough resources to
meet bandwidth requirements to 0.4Mbps, so that mobile hosts’s number of handoff do not
increase. However, schemes with a CAN caused number of handoffs 1.7 ∼ 3.5 times more
than a continuously active scheme from 0.1 to 0.4Mbps, although both simulation topologies
have enough resources to guarantee downlink bandwidth. This is because of an oscillation
problem of a CAN.
At a network with insufficient resources, four schemes has similar number of handoffs
(about two times handoffs of a continuously active scheme) when target bandwidths are
1.6 and 3.2Mbps. As a target bandwidth grows, four schemes’ unnecessary handoffs due
to periodically monitoring increase because there are a few of networks to supply a target
bandwidth. In Figure 5.2(b), there are a little differences between number of handoffs on
five schemes. This result means that four schemes’ unnecessary handoffs due to periodically
monitoring decreases. Because there are lots of networks which are able to guarantee target
bandwidths, four schemes as well as a continuously active scheme easily find the optimal
network.
5.3 Power consumption
Table 5.2 gives comparison between a continuously active and the other four selection
schemes. At two kinds of simulation topologies, four selection schemes only consume from
5 to 40% of a continuously active scheme’s power consumption. By using four selection
schemes, we are able to reduce power consumption maximum 95% comparing with the case
when all network interfaces turn on.
Table 5.2: Percentage of four selection to a continuously active scheme on power consumption
Static with Dynamic with Static w/o Dynamic w/o
candidate candidate candidate candidate
With insufficient 4.7 ∼ 4.5 ∼ 20.4 ∼ 9.0 ∼resource network 29.6% 22.7% 32.9% 23.5%
With sufficient 5.0 ∼ 4.8 ∼ 20.4 ∼ 9.0 ∼resource network 33.3% 30.7% 41.2% 32.8%
19
Page 28
In a Figure 5.4, four selection schemes’ results on power consumption per seconds are
shown. Each bar gets longer as target bandwidths increase because a mobile host consumed
more power when amount of transferred data rises. A static scheme without a CAN or a
dynamic scheme with a CAN respectively gave the largest or the smallest power consumption
among selection schemes. From 0.8 to 3.2Mbps, sufficient resource network case wasted
more power than insufficient resource case because of increase of achieved bandwidth at the
sufficient resource network. Schemes using a static period consumed more energy than those
using a dynamic period. This is because static schemes turns on network interfaces more
frequently than dynamic schemes. Using a CAN reduced power-wastage, but schemes with
a CAN did not always consume smaller amount of energy than schemes without a CAN.
As shown in a Figure 5.4(a)’s 1.6 and 3.2Mbps, if there are enough network resources to
guarantee target bandwidth, a static scheme with a CAN exhausted more power than a
dynamic scheme without a CAN.
At the insufficient resource network, a dynamic scheme without a CAN removed about
20% to 55% of power consumption of a static scheme without a CAN. A dynamic scheme
with a CAN reduced a percentage of 2 to 23 of power consumption of a static scheme with
a CAN. The effect of a dynamic period’s power-saving at schemes with a CAN was smaller
than at schemes without a CAN because a dynamic period’s effect for decreasing number
of turning-on is only constrained to a CAN in schemes with a CAN. A static scheme with
a CAN consumed from 22% to 90% of a static scheme without a CAN. In the case of a
dynamic scheme, using a CAN only wasted from 50% to 95% of a dynamic scheme without
a CAN’s power consumption. At the sufficient network resources, difference between a
static and a dynamic scheme with a CAN decreased with compared at a network with
insufficient network resources. Because there are enough bandwidth-guaranteeing networks
at the sufficient resource network, mobile hosts using a static scheme need not turn on
network interfaces for finding a new CAN.
20
Page 29
(a) With insufficient network resources
(b) With sufficient network resources
Figure 5.1: Achieved downlink bandwidth
21
Page 30
(a) With insufficient network resources
(b) With sufficient network resources
Figure 5.2: Number of handoffs (per node)
22
Page 31
(a) With insufficient network resources
(b) With sufficient network resources
Figure 5.3: The percentage of vertical handoffs (per node)
23
Page 32
(a) With insufficient network resources
(b) With sufficient network resources
Figure 5.4: Power consumption per seconds
24
Page 33
6. Conclusion
In this thesis, we have presented performance analysis of four RAN selection schemes which
are simple, easy to implement and generic solutions for a vertical as well as a horizontal
handoff. Four RAN selection schemes are divided on the basis of static/dynamic period,
with/without a CAN. For removing wasted power owing to unused network interfaces, these
four schemes periodically turn on network interfaces according to each network interfaces’
static or dynamic monitoring period and use a CAN which is a pre-selected alternative RAN
for a future handoff. Schemes with a CAN only turn on single network interface remaining
holding other network interfaces off for monitoring validity of a CAN. We implemented a new
simulator for the wireless overlay network environment. Using our simulator, we simulated
four selection schemes to observe the achieved bandwidth, number of handoffs, and total
power consumption. Our results show that a dynamic period or a CAN is able to reduce
large amount of wasted power without degrading downlink bandwidth, although number of
handoffs increase when a network does not have enough resources to guarantee bandwidth
requirements.
25
Page 34
³À »ÈÐ %K�
(Ûo»ÿ� RAN (Radio Access Network) 9ßÌv �ç¡(« q� )ç��ûB(Ûo)o�
Áº��� W1àÔ0>ß¼�� V,�o� �Ð/åL÷&�¦ s�6 x�� ú ¢ô�Ç /åL7£x ����"f ���ª�ô�Ç +þAI�_� Áº��� W1àÔ0>ß¼
l�Õüt� >hµ1Ï÷&�¦ e����. s��Qô�Ç Áº��� W1àÔ0>ß¼ l�Õüt[þt_� �©�&h�[þt�¦ s�6 xô�Ç����� ��6 x½+É Ãº e��
��H�½�×¼0A×¼Û¼\�¦7£x��r�~�úe��ܼ 9���'��Å\�_�]X�5Åq�¦6 xs� �>�½+É÷�rëß���m���,Áº���W1
àÔ0>ß¼�� ]j/BN½+É Ãº e����H "fq�Û¼ t�%i��¦ V,��n= ú e����H 1px_� ú§�Ér ��0px$í�¦ ]j/BNK�×�¦ ú e��
�¦ �.���sכ �7Hë�H\�"f��H s�l�7áxçß�_� W1àÔ0>ß¼[þt�¦ �©� ñ ���1lx �l� 0A �#� �>rF� ���H ÅÒ¹כô�Ç
ë�H]j[þt ×�æ, s�l�7áxçß�_� W1àÔ0>ß¼ ��s�\�"f_� �½�×¼�áÔ\�¦ t�"é¶ �l� 0Aô�Ç ë�H]j\�¦ Û�¦�¦�� �
%i���. �½�×¼�áÔ�� µ1ÏÒqt �%i��¦ M: �½�×¼�áÔ½+É @/�©� W1àÔ0>ß¼\�¦ ¹1Ô�¦ ���×þ� ���H ~½ÓZO�ܼ�Ð"f
çß�éß� ��¦ ½�&³ �l�/'ĺ 9°ú �Ér7áxÀÓ_�W1àÔ0>ß¼��s�\�"f_��½�×¼�áÔ÷�rëß���m���"f�ÐÓüt
o�&h�ܼ�Ð ���Ér W1àÔ0>ß¼ ��s�\�"f_� �½�×¼�áÔ�� ��0pxô�Ç W1 ��t� ~½ÓZO��¦ ]jr�ô�Ç��. s��Qô�Ç
W1 ��t� ~½ÓZO�[þt�Ér &ñ&h�/1lx&h���� ÅÒl� x9� @/� W1àÔ0>ß¼\�¦ s�6 x�<Êܼ�Ð+� ×�æ�©� "f!Q_� �¹¡§
\O�s� ��6 x ��0pxô�Ç W1àÔ0>ß¼\�¦ ¹1Ô��H��. ¢ô�Ç "fq�Û¼ |9�\� e��#Q �<HK� �Ðt� ·ú§�¦ Ô�¦�9¹כ�ô�Ç ���
§4� �è�\�¦ \O��±p��. #��Q 7áxÀÓ_� Áº��� W1àÔ0>ß¼�� ���5g4R e����H W1àÔ0>ß¼ 8��â\�"f_� r�ÓýtYU
s�����¦ 0A �#� Dh�Ðî�r r�ÓýtYUs�'�\�¦ ½�&³ô�Ç Êê, ½�&³�)a r�ÓýtYUs�'�\�¦ s�6 x �#� ]jr��)a W1
��t� RAN (Radio Access Network) ���×þ� ~½ÓZO�_� �½�×¼0A×¼Û¼, �½�×¼�áÔ µ1ÏÒqtS��úü< �è��)a
���§4�_� |¾Ó�¦ 8£¤&ñ�<Êܼ�Ð+� ���×þ�~½ÓZO�_� $í0px�¦ q��§ �%i���. r�ÓýtYUs���� ���õ�, 1lx&h���� ÅÒl�
�� @/� W1àÔ0>ß¼\�¦ ��6 x�<Êܼ�Ð+� �½�×¼0A×¼Û¼_� y���è \O�s� ú§�Ér |¾Ó_� ���§4��¦ ]X����½+É Ãº e��
%3���.
26
Page 35
References
[1] A. K. Salkintzis, MOTOROLA ”Internetworking Techniques and Architectures for
WLAN/3G Integration toward 4G Mobile Data Networks,” IEEE Wireless Commu-
nications, vol. 11, no. 3, pp. 50-61, June 2004.
[2] M. Buddhikot, G. Chandranmenon, S. Han, Y. W. Lee, S. Miller, and L. Salgarelli, ”In-
tegration of 802.11 and Third Generation Wireless Data Networks,” IEEE INFOCOM
2003, vol. 1, pp. 503-512, April 2003
[3] M. Inoue, K. Mahnud, H. Murakami, and M.Hasegawa, ”Novel Out-Of-Band Signal-
ing for Seamless Internetworking Between Heterogeneous Networks,” IEEE Wireless
Communications, vol. 11, no. 2, pp. 56-63, April 2004.
[4] E. Gustafsson, and A. Jonsson, ”Always Best Connected,” IEEE Wireless Communi-
cations, vol. 10, no. 1, pp. 49-55, Feb. 2003.
[5] E. A. Brewer et al, ”A network architecture for heterogeneous mobile computing,” IEEE
Wireless Communications, vol. 5, no. 5, pp. 8-24, Oct. 1998.
[6] J. McNair, and Fang Zhu, ”Migration toward 4G wireless communications - Vertical
handoffs in fourth-generation multinetwork environments,” IEEE Wireless Communi-
cations, vol. 11, no. 3, pp. 8-15, June 2004.
[7] H. Wang, R. Katz, and J. Giese, ”Policy-enabled handoffs across heterogeneous wireless
networks,” IEEE Workshops on Mobile Computing Systems and Applications, pp. 51 -
60, Feb. 1999.
[8] Minji Nam, Nakjung Choi, Yongho Seok, and Yanghee Choi, ”WISE: Energy-Efficient
Interface Selection on Vertical Handoff between 3G Networks and WLANs,” IEEE inter-
national symposium on personal, indoor and mobile radio communications, September
2004.
[9] W. Quadeer, T. Simunic, J. Ankcorn, V. Krishnan, and G. Micheli, ”Heterogeneous
Wireless Network Management,” PACS, Dec. 2003.
27
Page 36
[10] C. Guo, Z. Guo, Q. Zhang, and W. Zhu, ”A Seamless and Proactive End-to-End Mo-
bility Solution for Roaming Across Heterogeneous Wireless Networks,” IEEE Journal
on Selected Areas in Communications, vol. 22, no. 5, pp. 834-848, June 2004.
[11] Gang Wu, M. Mizuno, P.J.M. Havinga, ”MIRAI Architecture for Heterogeneous Net-
work,” IEEE Communications Magazine, vol. 40, no. 2, pp. 126 - 134, Feb. 2002.
[12] R. Berezdivin, R. Breinig, and R. Topp, ”Next-generation Wireless Communications
Concepts and Technologies,” IEEE Communications Magazine, vol. 40, no. 3, pp. 108
- 116, March 2002.
[13] CDMA2000 1x EV-DO, ”CDMA2000 1x EV-DO’s PCMCIA CARD,” [Online docu-
ment], [cited 2004 Dec 12], Available HTTP: http://www.gtran.co.kr
[14] Cisco Aironet Wireless LAN Clinet Adapters, ”Cisco Aironet 5GHz 54Mbps Wireless
LAN Client Adapter,” [Online document], [cited 2004 Dec 12], Available HTTP:
http://www.cisco.com/application/pdf/en/us/guest/products/ps4451/c1650/ ccmi-
gration 09186a00800c61ea.pdf
[15] Cisco Aironet Wireless LAN Clinet Adapters, ”Cisco Aironet 350 Series
Client Adapters,” [Online document], [cited 2004 Dec 12], Available HTTP:
http://www.cisco.com/application/pdf/en/us/guest/products/ps448/c1650
[16] ORiNOCO Client Products, ”ORiNOCO 11b/g PC Card,”
[Online document], [cited 2004 Dec 12], Available HTTP:
http://www.proxim.com/learn/library/datasheets/11bgpccard.pdf
[17] The Network Simulator ”The Network Simulator - NS-2,” [Online document], [cited
2004 Dec 12], Available HTTP: http://www.isi.edu/nsnam/ns/
28
Page 37
ÇÔ �� �+ ;³�
s� �7Hë�H�¦ ¢-a$í �l���t� ú§�Ér ì�r[þta� úú§�Ér �¹¡§�¦ ~ÃÎ��¤_þvm���. ���$� $3��� õ�&ñ�¦ @/���\�
"f Áº��y� ��}9� ú e���2�¤ °ú¡��� l��r\�¦ ëß�[þt#Q ÅÒ��� %i�e��ï�r �§Ãº_��õ� �$í ü� �§Ãº_��a� ���
d��ܼ�Ð y����×¼wn�m���. $3��� õ�&ñ1lxîß� =åSe�� \O���H �'ad��õ� |9�Õþ�ܼ�Ð s�=åJ#QÅÒr��¦ Âúª�Ér r�çß�
s�%3�t�ëß� �-Áº��� ú§�Ér ��ØÔgË>�¦ ÅÒ��� %i�e��ï�r �§Ãº_��a� ��r� ô�Ç��� y����×¼wn�m���. ¢ô�Ç ]j
�7Hë�H�¦ d����K�ÅÒr��¦ �'ad���¦ l�Ö�¦#�ÅÒ��� Ðqt5px§>= �§Ãº_��õ� ô�Ç8�ú �§Ãº_��a�� y����_� ú�@pu
�¦ ×¼wn�m���.
]j�� s�XO�>� Áº��y� �7Hë�H�¦ ��}9� ú e���2�¤ �ü<ï�r $��B (��ÉÓ'� W1àÔ0>ß¼ êÁ� d��½[þta�
�y����_� ú�@pu�¦��� ��¦z�·_þvm���.Òqty�� ���H~½ÓZO�õ����Òqt�¦¶ú�������HX<e��#Q��4R��½+É#�
Ä»\�¦ L:{��>� K�ï�r 7áx8����, z�+«>\� �¹¡§�¦ ÅÒ�¦ ú§�Ér �����¦ K�ï�r [jô ¥���, 5pxî�r���m�, %ò
ô ¥, ���½©, Õªo��¦ �©�8�, Ö�©6£§�¦ �8K�ï�r o=�Érõ� _�\P�\�>� y����×¼wn�m���. ÷�rëß� ��m��� �½Ó�©� ��
7ᤠ°ú �Ér ì�r0Al� 5Åq\�"f "f�Ð\�¦ ���ø@ � 9 a�¦����¦ ¼#�ô�Çô�Ç ��6£§Ü¼�Ð �7Hë�H�¦ ¢-a$ír�~� ú e���
2�¤ �ü<ï�r ���H $��B êÁ� d��½[þt\�>� ��r� ô�Ç��� ���d��ܼ�Ð y����×¼wn�m���.
¢ô�Ç, $3��� õ�&ñ1lxîß� a%~�Ér 1lxl��Ð"f �H j˵s� ÷&#QÅÒ%3�~�� �©��B ���m�ü< �©�ç�H\�>�� ���d��
ܼ�Рy����׼wn�m���.
��t�}��ܼ�Ð �½Ó�©� ���\�"f s�=åJ#Q ÅÒ�¦ j˵[þt M: �H j˵s� ÷&#QÅÒ%3�~�� 1lx�&³���, ÂÒ�_��a�
� ���d��ܼ�Рy����׼wn�m���.
Page 38
l� �]� "�
T� �mr : �ª� nq �&³
Ú�| [Í i�J G�B : 1980�� 6�Z4 16{9�
_ô> Ú�| U� : "fÖ�¦r� y©�"f½ �o/BG 81lx
(è< +8� U� : "fÖ�¦r� ���í½ ��o 21lx
®£ �Ð : "fÖ�¦r� ���í½ ��o 21lx ĺ$í����àÔ 121lx 1002 ñ
E-mail ®£ �Ð : [email protected]
Áþ� �]�
1996. 3. – 1999. 2. "î ü�ü@²DG#Q�¦1px�<Æ�§ áÔ|½ÓÛ¼#Qõ�
1999. 3. – 2003. 2. ���[j@/�<Æ�§ (��ÉÓ'�íß�\O� ���/BN (B.S.)
2003. 3. – 2005. 2. ô�DzDGõ��<Æl�Õüt"é¶ ���íß��<Æõ� (M.S.)