Content uploaded by Svetlana Boudko
Author content
All content in this area was uploaded by Svetlana Boudko on Apr 10, 2014
Content may be subject to copyright.
Team Decision Approach for Decentralized
Network Selection of Mobile Clients
Svetlana Boudko and Wolfgang Leister
Norsk Regnesentral, Oslo, Norway
Email: {svetlana.boudko, wolfgang.leister}@nr.no
Stein Gjessing
University of Oslo, Norway
Email: steing@ifi.uio.no
Abstract—We consider a network selection problem for a group
of mobile clients that operate in a heterogeneous wireless access
network environment and are equipped with multiple access
network interfaces. The involved networks cooperate in order
to improve their own and the mobile clients’ performance. We
formulate the problem as a team decision problem. In this
formulation, several decision variables are involved, and these
decisions are made by several decision makers with access to
different information but contributing to a common goal. The
novelty of the proposed approach is that the network selection is
done in a decentralized manner with only limited information
available to decision makers. We present two decentralized
algorithms to this problem that we compare and evaluate in
the OMNet++ simulation environment.
Index Terms—Wireless networking, mobile network selection,
heterogeneous networks, decentralized algorithms.
I. INTRODUCTION
Internet video streaming is taking a significant portion of the
Internet services and the amount of these services delivered
over wireless access technologies is expected to increase
exponentially during the next years. According to the Cisco
visual networking index [1], global mobile data traffic grew
2.6-fold in 2010, nearly tripling for the third year in a row.
Mobile data traffic is expected to increase 26-fold between
2010 and 2015. Mobile data traffic will grow at a compound
annual growth rate (CAGR) of 92 % from 2010 to 2015,
reaching 6.3 exabytes per month by 2015. The number of
mobile devices is expected to pass 7.1 billion by 2015, which
is approximately equal to the estimated world’s population in
2015 (7.2 billion). Traffic of video related services, where the
quality is particular sensitive to network conditions is expected
to account for 52.8 % of total mobile traffic.
This constantly increasing demand for mobile bandwidth
implies that significant improvements in how resources are
allocated in mobile networks and in how the data is delivered
to mobile users are needed to avoid degradation in service
quality and, possibly, congestion collapse. Another issue to
take into account is that tens of thousands of mobile devices
can operate simultaneously inside one limited area that has
overlapping coverage of different mobile networks, such as
UMTS, WLAN, WiMAX, and LTE. Joint consideration of
allocating some users to different networks can improve the
performance of the whole system. By intelligently redistribut-
ing mobile devices among the available wireless connections,
the networks can accommodate more users, improve the users’
QoS and increase the networks’ revenue.
To tackle this challenge, the IEEE 802.21 Media Inde-
pendent Handover (MIH) standard [2] advises mobile nodes
and networks to perform handover decisions collaboratively,
in an environment with multi-interface, multi-technology user
equipment and multiple network points of attachment. How-
ever, the network selection problem is challenging with multi-
ple decision criteria, such as user preferences, user movements,
operations performed, battery limitations, mobile device types,
network load and service provider cost. This problem becomes
even more challenging due to inaccurate and insufficient infor-
mation for decision making, the dynamic nature and inability
to collect all necessary information due to computational
limitations of the devices and large dimensions of the data
involved.
The work proposed in this paper includes the following two
contributions. 1) We propose a distributed solution that allows
network selection in a decentralized manner with only limited
information shared among the decision makers. 2) By focusing
our research on a group of mobile users, we study the impact
of multiple users on decisions of the system.
The benefit of our solution is reduced signaling overhead
and consequently reduced response delays when the network
reconfiguration is needed to improve the performance of
the system. Due to the applied decentralized approach, the
scalability of the system is improved compared to networks
that need to obtain global knowledge of the system state.
We focus on decision support for mobile network selection
and vertical handoff management when a mobile device is
equipped with multiple access interfaces.
The remainder of the paper is organized as follows. After
presenting a representative scenario in Section II, we give an
overview of approaches discussed in related work (Section III).
We present the problem as a team-decision problem in Sec-
tion IV, and outline suitable algorithms. The simulation set-up
and the results of simulations are presented and analyzed in
Section V, before concluding in Section VI.
II. SCENARIO
We consider a network selection scenario for a group of
users in a hotspot area like a crowded city center, a public
transportation node or an exhibition site where a coverage of
several base stations or access points from different networks
is possible. We assume a substantial overlap in coverage of
Figure 1. Network topology built upon multiple mobile networks with
heterogeneous technologies to serve a group of clients.
these stations. A representative scenario of such networking
is illustrated in Figure 1. These user terminals are capable of
connecting to several access networks, and vertical handoffs
between different networks are technically possible. The termi-
nals periodically receive beacon signals from base stations or
access points of the available access networks that are typically
broadcast once per second.
Users located in the same cell of a mobile network can
experience degradation in quality due to shortage of avail-
able bandwidth. Though admission control mechanisms are
designed to ensure the quality of wireless connections and to
prevent network congestion, there is still a possibility that a
user is admitted to the network while requiring low bandwidth,
e.g., for web browsing, will require more resources for video
streaming shortly after. We consider also a situation when
a base station may act in a proactive way and monitor the
available resources in adjacent cells. The users that are going
to move to a cell that, at the moment, is not able to admit new
users can be notified to perform a vertical handoff to another
available network. Since users may have different preferences
and request different types of service their utility functions are
built upon different criteria.
To provide better load balancing between the networks, and
to avoid disturbing ping-pong effects, joint coordination, and
information exchange between the users and the base stations
is essential; both the clients and the networks can benefit
from cooperative handoffs. However, due to strict bandwidth
and power limitations of mobile networks, and also due to
scalability issues, a complete information exchange between
mobile users and networks is not feasible. Decentralized
network selection is therefore essential.
III. REL ATED W OR K
In a wireless heterogeneous environment, resource alloca-
tion and the “always best connected” concept are important
challenges that need to be solved. In this section, we focus
on handoff management and network selection studies that
address these challenges.
A. Handoff Management in Mobile Networks
Prediction-based techniques have been suggested in several
studies [3–6] aiming to reduce handoff delays.
To represent the movement behavior of a mobile user,
Paramvir et al. [3] propose a two-level user mobility model
consisting of a local level and a global level. A hierarchical
location prediction algorithm is proposed based on an approx-
imate pattern-matching algorithm implemented in the global
level and Kalman filtering techniques implemented in the local
level.
Akyildiz and Wang [4] propose a mobility model that
uses historical records and stochastic behavior of mobile
users to predict their future position. The model is built
upon a framework of user mobility profiles (UMP). In the
proposed prediction algorithms, many factors are taken into
consideration including velocity and direction of mobile users,
historical records, stochastic model of cell residence time and
path characteristics. The authors claim that these algorithms
predict more accurately than previous schemes. However,
the complexity of the algorithms make them impractical for
mobile applications.
In two studies, Tseng et al. [5] and Choi et al. [6] propose
using cross-layer information to perform layer-3 handoff in
parallel with or prior to the layer-2 handoff. However, these
schemes can lead to false alarms and cause unnecessary MIP
registrations. Ray et al. [7] conclude that deciding upon the
ideal choice and timing of cross-layer triggers in order to
reduce layer-3 latency is still an open problem.
Vertical handoff is the handoff between the networks of
different wireless technologies and has been addressed in
several studies [8–13]. While horizontal handoffs are typically
triggered when the received signal strength (RSS) of the
serving access router drops below a certain threshold the
vertical handoff can be initiated due to other reasons such
as user preferences or network conditions including coverage,
bandwidth, cost and power consumption. The decision process
is therefore more complex for vertical handoffs than for
horizontal ones.
While some authors only use RSS as an input parameter
for the handoff decision process [4, 14], others combine the
use of RSS with bandwidth information [15–17]. Using cost
functions has been proposed earlier [8–10]. Nasser et al. [18]
propose a cost function that depends of the cost of service,
security, power consumption, network conditions and network
performance. However, in their evaluation, all weights except
the bandwidth weight are set to zero. This renders their cost
function to a function of one parameter: bandwidth.
Algorithms based on fuzzy logic or artificial neural net-
works in combination with multiple criteria [11–13] suffer
from high handover delay because of their complexity and the
training process. Unfortunately, the authors of these algorithms
did not provide throughput results.
B. Admission control and network selection in wireless net-
works
Ormond and Murphy [19] propose a network selection
strategy that explores a number of possible utility functions.
The solution is user-centric, and an interplay between different
users and networks is not considered. Ormond and Murphy
conclude that the impact of multiple users operating in the
same region needs to be further examined.
Tragos et al. [20] propose a generic admission control
algorithm that allows network selection for 4G heterogeneous
wireless networks. The algorithm aims to provide maximum
utilization of the network, prevent overloading situations and
ensure best QoS. However, implementation of the algorithm
requires the presence of a centralized entity.
In our analysis, we recognize that several problems are
not yet addressed, or where the currently available solutions
need to be improved. Decentralized algorithms that rely on
information only partly shared between the decision-makers
need to be implemented and evaluated in multi-user scenarios.
These considerations motivate us to look at distributed and
computationally efficient methods of network selection in
heterogeneous mobile environments.
IV. PROBLEM FORMULATION
Decentralized network selection can be formulated as a team
decision problem [21, 22] where several decision variables
are involved. These decisions are made by different decision
makers that have access to different information but participate
in a common goal.
Team decision theory is concerned with determining the op-
timal decisions, given a set of information for each of several
decision makers, that work together to achieve a payoff. In
team decision problems, these sets of information are different
though often correlated for different decision makers. These
optimal decisions can be either person-by-person optimal or
team optimal. In person-by-person optimal cases, each person
makes the decisions that optimize the individual’s payoff, but
not necessarily the team payoff. These cases are optimal for
a particular team member, given that the decision functions
for other members are fixed. In team optimal cases, the group
payoff is optimized. Team optimality is a stronger condition,
and is thus harder to achieve. Taking into account that person-
by-person optimal strategies may result in unfair distribution
of the resources, we focus our research on team optimal
strategies.
A. System Model
Taking into consideration our understanding about prefer-
ences of mobile nodes and the networks, we are now ready to
formalize our observations into a system model. We consider a
set of networks N= 1,2, ..., n and a set of mobile terminals
M= 1,2, ..., m. For each terminal mjand network nithe
following is defined. Streaming bitrate requirements of mobile
nodes are denoted by rj;rssi,j is the received signal strength
in network ifor terminal mjwhile power consumption and
cost of service in network nifor terminal mjare denoted
by pi,j and ci,j , respectively. The total available bandwidth
of network niis denoted by bi. For each terminal mj, we
define a user preference profile that is described by a tuple
containing Thp
i,j ,Thc
i,j , and Thrss
i,j . These denote thresholds,
or user preferences, for respectively power consumption, cost
of service and received signal strength. We define a time period
τi,j during which terminal mjis served by network nibefore
performing a handoff and moving to the next cell of this
network.
For each mobile terminal mjand each network niwe define
the function xi,j which mimics the decision taken by a mobile
terminal mjto switch to or to stay in mobile network ni.
xi,j =(1,if mjhas roamed to or stays in ni
0,otherwise (1)
We define the common goal of the set of networks and
users of these networks as maximization of consumed band-
width over a period of time, minimization of the number of
handoffs and reduction of signalling between the networks
and terminals. To achieve this goal, the participating networks
and the terminals need to cooperate while trying to maximize
their own performance. To facilitate a decentralized approach,
we define two components, the network component and the
mobile node component. We formulate the problem and solve
it separately for these two components.
1) Network Component: For each network ni, we define
a utility function Uias a sum of consumed bandwidth of all
users of this network over the time the user is connected to it,
as defined in Eq. 2.
∀{i}:Ui=X
j
xi,j ·rj·τi,j (2)
In this sense, the networks benefit if they select the users
that not only request a higher bandwidth but also intend to
stay in the network for longer periods of time, which also
can eliminate the ping-pong effect when the user needs to
change the network again recently after the handoff. Basing
our decision on research done in [23–25], we assume that a
mobile network is capable of predicting the residence time
of a mobile node inside the network based on mobile node’s
velocity, movement patterns and the local area. We realize that
this problem is an ongoing research work. For the purpose of
this paper, we assume that the prediction can be performed
with acceptable precision.
The common goal is the maximization of the expected value
of the sum of network utility functions.
max X
i
E[Ui](3)
The utility function is constrained by the network resources.
As any network of the system has a limited knowledge about
the resources and decisions of the rest of the system, xi,j is
given as its expected value.
∀{i}:X
j
E[xi,j ]·rj≤bi(4)
At Mobile Node mj:
if Network Selection is triggered
search for available networks
for each available network ni
if (pi,j <Thp
i,j ,ci,j <Thc
i,j ,rssi,j <Th rss
i,j ) then
add the network to list of candidate networks
for each network niin list of candidate networks
send requests to candidate network ni
wait for response from candidate networks
upon reception of response from candidate network ni
if (admitted value == true) then
add response to response list
if response list == null
stay in the current network
else
for all responses in response list
choose the network with the highest ti,j
Figure 2. Distributed Algorithm for Network Selection, Mobile Node
Component
At Candidate Network ni:
wait for admission requests from mobile nodes
upon reception of admission requests from mobile node mj
using available knowledge solve Eq. 2 with constraints Eq. 4
return admission message containing
admitted value = true/false and expected time
Figure 3. Distributed Algorithm for Mobile Node Admission, Network
Component
2) Mobile Node Component: For each user we define a
utility function fjas a function of power consumption, cost of
service, available bandwidth and received signal strength. We
realize that the set of parameters that define user preferences
can be larger than the one mentioned above and can also differ
from user to user. We also realize that users might employ
different utility functions but for this work we limit us to this
definition.
max fj(pi,j , ci,j , rj,rssi,j)(5)
Eq. 5 poses a multiparameter optimization problem that can
be solved by introducing weights and normalization. Another
solution is to relax the optimization problem by reducing it to
a one variable optimization Eq. 6. Consequently, we introduce
a set of constraints in Eq. 7, where some of parameters z∈
{p, c, rss}, namely power consumption, cost of service, and
received signal strength, are limited by their thresholds Thz
i,j
defined by the user’s preferences.
max X
i
xi,j ·rj·τi,j (6)
pi,j ≤Thp
i,j , ci,j ≤Thc
i,j ,rssi,j ≤Thrss
i,j (7)
B. Algorithm
The system model defined in Section IV-A is used in the
decentralized algorithm for network selection outlined below.
We build the algorithm based on the following a) maxi-
mization of the total consumed bandwidth by distributing the
users between the networks taking into account the networks’
available bandwidth, and b) minimization of the number of
performed handoffs between the networks.
There are two events that may trigger the execution of
the network selection algorithm: 1) based on monitoring of
available resources in its cells, and the prediction of users’
location information, the network informs mobile nodes that
are about to move to a congested cell to switch to another
available network instead; 2) mobile terminal mjexperiences
degradation of network performance detected by increased
packet loss or delay on the mobile terminal.
When the network selection is triggered, the effected ter-
minal runs the selection algorithm as shown in Figure 2.
As a consequence, the network receives calls from mobile
nodes it runs the algorithm shown in Figure 3. To calculate
the expected values of the utility function defined by Eq. 2,
this algorithm takes as input the knowledge available for this
network. Depending on how the knowledge of the system is
shared among the networks, we differ between two versions
of the algorithm: Algorithm Aand Algorithm B.
1. Algorithm A: Each mobile node mj, while sending a
request to the network ni, informs the network about the
requests sent to other networks. Based on this information,
each network calculates the probability of mobile node mj
to choose this network if accepted.
2. Algorithm B: Each network nishares its information with
exactly one more network nk. The network nidoes not
accept a mobile node mjif the node is accepted by the
network nkand τi,j < τk,j .
C. Algorithm Evaluation
To evaluate the algorithms, we define upper and lower
bounds to their operation. The upper bound is achieved by
applying a centralized solution with fully shared knowledge
of the conditions in all evaluated networks and is further
referred as global knowledge reference. This reference can also
be viewed as a modification of the algorithms [19] and [20]
discussed in Section III-B that are now extended to a multi-
user scenario. Its utility function is defined in Eq. 8. The utility
function is constrained by resource limitations of networks as
described in Eq. 4 and preferences of mobiles nodes described
in Eq. 7. This problem belongs to the class of integer linear
programming problems, which is known to be NP-complete,
thus problematic for real-time tractable implementation and
in most cases can be used only as a reference for evaluating
algorithms.
U=X
iX
j
xi,j ·rj·τi,j (8)
The lower bound corresponds to a situation when all networks
base their decisions only on local knowledge (Eq. 2) and is
further referred as local knowledge reference.
V. SIMULATIONS
The performance and functionality of the algorithms have
been evaluated through multiple simulation runs. We have
implemented both versions of our algorithm in the OMNet++
environment [26]. In Algorithm A, the network gets the
(a) Minimum value (b) Average values (c) Maximum value
Figure 4. Decision errors from the simulation using one iteration of the algorithm run after network selection is triggered. The results for 100, 200, 300
mobile nodes are based on 1000 simulation runs for each group of nodes.
information about how many other networks are requested by
the same mobile node. As no other information is available,
we assume that the probability of being assigned to any
of the networks is equal for all participating networks. In
Algorithm B, each network shares its information with exactly
one more network. In our testing scenario, these networks
do not overlap. We compare the algorithms with the global
knowledge reference and the local knowledge reference.
A. Simulation Setup
For the sake of simplicity, we simulate a scenario with four
wireless networks, which covers quite well the scope of the
evaluation. In this scenario, a group of users from one network
is about to move from one cell of the network to another cell
of the same network that experiences a shortage of available
bandwidth. In consequence, the cell that the users move to is
not able to accommodate all these users.
For our evaluation, we run tests with 100, 200, and 300 users
moving to this congested cell. Further, we divide the users into
four categories in terms of requested bandwidth. To define
these categories we use service class characteristics defined
by Tragos et al. [20] as follows: a) at 64 kbps, for simple
telephony and messaging b) at 512 kbps, for web browsing
c) at 1024 kbps, for interactive media and d) at 2000 kbps,
for video streaming, each category having approximately the
same number of users.
Note that none of the networks have enough resources to
accommodate all users alone. All four networks must be used
in order to meet the requirements of all users. We run also
tests when total bandwidth of all networks is not sufficient
to accommodate all users. The tests are done for network
conditions that result in 5%, 10%, 15%, 20%, 25%, 30%
dropped calls if the global knowledge reference is applied.
The time τi,j for the user jto stay in the network nibefore
performing a horizontal handoff, or a cell residence time, is
randomly distributed in the range [1,100] time units.
B. Simulation Results
We evaluate how mobile nodes are distributed among the
networks after one iteration of the algorithm run. We calculate
the number of decision errors as a number of users whose
connection ends up in dropped calls due to wrong network
allocation. These errors are the results of wrong assignments
to networks that do not have sufficient bandwidth to accom-
modate the assigned users. For each group of users (100, 200,
300 users), we repeat the experiment 1000 times with different
sets of τi,j .
For all tests done, the top and bottom 5 % of the results
are excluded from the evaluation. The results are averaged
over these simulation runs and are depicted for minimum
value results in Figure 4(a), for average value results in
Figure 4(b), for maximum value results in Figure 4(c), for
cumulative distribution function in Figure 5. The values 0 in
the results for Algorithms Aand Bin Figure 4(a) indicate
that all users are assigned to the networks correctly. The
global knowledge reference is 0 for all experiments meaning
that in the centralized solution, all users were assigned to
the networks without any dropped calls. The results with
dropped calls in the global knowledge reference are depicted
in Figure 6. The figure shows the results for 200 mobile nodes.
The results for 100 and 300 mobile nodes are very similar to
the results for 200 nodes and therefore are not included in the
paper.
The tests show that both Algorithms Aand Bcan distribute
the users between the networks significantly better than the
local knowledge reference. Algorithm Aperforms better than
Algorithm Bfor all user groups through all tested values for
dropped calls in the optimum solution.
It shows that sharing partial information in Algorithm B
makes little use of extra information from just one network.
However, Algorithm Brequires significantly more information
to exchange between the networks than Algorithm A. It also
requires more sophisticated mechanisms and protocols to be
implemented in the networks, including security considera-
tions and synchronization of the information flow.
We also evaluate the dynamic scenario. For these tests, the
algorithms are run until all clients are assigned to the networks
with sufficient bandwidth, also considering the arriving calls.
The arrival rate of new calls is modeled with a Poisson stream.
The graphs depicted in Figure 7 show the averaged results for
100, 200, and 300 users over 1000 test runs. The x-axis shows
Figure 5. Cumulative distribution function for percentage of decision errors
using one iteration of the algorithm run after network selection is triggered.
The results are based on 1000 simulation runs for a system consisting of 200
mobile nodes.
the number of iterations of the algorithm. The y-axis shows the
percentage of decision errors. Clearly, both Algorithms Aand
Bconverge faster than the local knowledge reference. There
is very little difference between Algorithms Aand Beven
though Algorithm Brelies on more information.
C. Signaling Overhead
We estimate signaling overhead Sofor the algorithms and
the references. As signaling required to trigger network selec-
tion is the same for the references and the algorithms these
messages are excluded from the estimation. For the global
knowledge reference all nnetworks in consideration need to
exchange the information about musers that get triggered
network selection, and the overhead is estimated as follows
(Eq. 9).
So=n·(n−1) ·m(9)
To make estimations for Algorithms Aand Band the local
reference defined respectively by Eq. 10, Eq. 11, Eq. 12, we
use the results of the dynamic scenario that are depicted in
Figure 7.
So= 0.18 ·m·(n−1) (10)
So= 0.32 ·m·(n−1) (11)
So= 1.06 ·m·(n−1) (12)
Clearly, Algorithm Aprovides a significant reduction of the
signaling overhead.
VI. CONCLUSION
In this paper, we propose a network selection solution for
a group of mobile nodes. Compared to the work specified
in Section III, our main achievement is decentralization of
the network selection formulated as a team decision problem
and consideration of the impact of multiple users. An efficient
decentralized network selection solution is important for future
mobile networks, since it improves utilization of the network
resources and QoS of users and reduces signalling overheads.
Figure 6. Dropped calls from the simulation using one algorithm run with
total available bandwidth less than total required bandwidth.The results are
based on 1000 simulation runs for a system consisting of 200 mobile nodes.
The x-axis shows the percentage of dropped calls for the optimum (global
knowledge reference). The y-axis shows the percentage of dropped calls.
Figure 7. Percent of users received dropped calls, dynamic scenario.The
x-axis shows the number of iterations of the algorithm. The y-axis shows the
percentage of decision errors. The results are based on 1000 test runs.
We study how the solution results depend on the information
available for the decision making. The simulation results of
our algorithms show that blocked calls can be reduced with
approximately 60-50 % compared to the local knowledge
reference. The test results do not differ much for 100, 200 and
300 users, and we expect that these results can be extended to
the general case.
Both Algorithm Aand Algorithm Bdeliver similar results
in terms of number of blocked calls. The implementation
of Algorithm Brequires development of mechanisms for
synchronizing information about the network conditions and
careful security considerations when information from one net-
work is available to other networks. Operation of Algorithm B
requires significantly higher signalling between the networks
and the users. We therefore conclude that Algorithm Ais to
be preferred over Algorithm B.
As a step further, we intend to extend our work to different
simulation scenarios to consider an interplay of several mobile
networks with different pricing strategies. We also intend to
consider scenarios where users from different networks are
part of the same multicast groups and regrouping them can
reduce network load.
VII. ACKNOWLEDGMENT
The work described in this paper has been conducted as a
part of the ADIMUS (Adaptive Internet Multimedia Stream-
ing) project which is funded by the NORDUnet-3 programme.
The authors wish to thank Trenton Schulz and Ragnar Hauge
for discussions during the preparation of this paper.
REFERENCES
[1] Cisco, “Cisco visual networking index: Global mobile
data traffic forecast update, 2010-2015,” 2011.
[2] IEEE, “Media Independent Handover,” IEEE Standard
802.21, 2009.
[3] T. L. Paramvir, T. Liu, P. Bahl, S. Member, and I. Chlam-
tac, “Mobility modeling, location tracking, and trajectory
prediction in wireless ATM networks,” IEEE J. Sel. Areas
Commun., vol. 16, pp. 922–936, 1998.
[4] I. F. Akyildiz and W. Wang, “The predictive user mobility
profile framework for wireless multimedia networks,”
IEEE/ACM Trans. Netw, vol. 12, pp. 1021–1035, 2004.
[5] C.-C. Tseng, L.-H. Yen, H.-H. Chang, and K.-C. Hsu,
“Topology-aided cross-layer fast handoff designs for
IEEE 802.11/mobile IP environments,” Communications
Magazine, IEEE, vol. 43, no. 12, pp. 156–163, Dec. 2005.
[6] Y.-H. Choi, J. Park, Y.-U. Chung, and H. Lee, “Cross-
layer handover optimization using linear regression
model,” in ICOIN 2008. Int’l Conf. on Information
Networking, Jan. 2008, pp. 1–4.
[7] S. Ray, K. Pawlikowski, and H. Sirisena, “Handover in
mobile wimax networks: The state of art and research is-
sues,” IEEE Communications Surveys Tutorials, vol. 12,
no. 3, pp. 376–399, 2010.
[8] Q. Song and A. Jamalipour, “A network selection mech-
anism for next generation networks,” in ICC 2005. IEEE
Int’l Conf. on Communications, vol. 2, May 2005, pp.
1418–1422.
[9] A. Hasswa, N. Nasser, and H. Hassanein, “Tramcar: A
context-aware cross-layer architecture for next generation
heterogeneous wireless networks,” in ICC 2006. IEEE
Int’l Conf. on Communications, vol. 1, Jun. 2006, pp.
240–245.
[10] F. Zhu and J. McNair, “Optimizations for vertical handoff
decision algorithms,” in IEEE Wireless Communications
and Networking Conference, WCNC2004, vol. 2, Mar.
2004, pp. 867– 872.
[11] P. M. L. Chan, Y. F. Hu, and R. E. Sheriff, “Implemen-
tation of fuzzy multiple objective decision making algo-
rithm in a heterogeneous mobile environment,” in IEEE
Wireless Communications and Networking Conference,
WCNC2002, vol. 1, Mar. 2002, pp. 332–336.
[12] L. Xia, L. Jiang, and C. He, “A novel fuzzy logic vertical
handoff algorithm with aid of differential prediction and
pre-decision method,” in ICC’07. IEEE Int’l Conf. on
Communications, Jun. 2007, pp. 5665–5670.
[13] N. Nasser, S. Guizani, and E. Al-Masri, “Middleware
vertical handoff manager: A neural network-based solu-
tion,” in ICC ’07. IEEE Int’l Conf. on Communications,
Jun. 2007, pp. 5671–5676.
[14] A. H. Zahran, B. Liang, and A. Saleh, “Signal threshold
adaptation for vertical handoff in heterogeneous wireless
networks,” Mob. Netw. Appl., vol. 11, pp. 625–640, Aug.
2006.
[15] C. W. Lee, L. M. Chen, M. C. Chen, and Y. S. Sun,
“A framework of handoffs in wireless overlay networks
based on mobile IPv6,” IEEE J. Sel. Areas Commun.,
vol. 23, no. 11, pp. 2118–2128, Nov. 2005.
[16] K. Yang, I. Gondal, B. Qiu, and L. Dooley, “Combined
SINR based vertical handoff algorithm for next genera-
tion heterogeneous wireless networks,” in IEEE Global
Telecommunications Conference, GLOBECOM ’07, Nov.
2007, pp. 4483–4487.
[17] C. Chi, X. Cai, R. Hao, and F. Liu, “Modeling and anal-
ysis of handover algorithms,” in IEEE Global Telecom-
munications Conference, GLOBECOM ’07, Nov. 2007,
pp. 4473–4477.
[18] N. Nasser, A. Hasswa, and H. Hassanein, “Handoffs in
fourth generation heterogeneous networks,” IEEE Com-
munications Magazine, vol. 44, no. 10, pp. 96–103, Oct.
2006.
[19] O. Ormond and J. Murphy, “Utility-based intelligent
network selection,” in In Proceedings of the IEEE In-
ternational Conference on Communications, 2006.
[20] E. Tragos, G. Tsiropoulos, G. Karetsos, and S. Kyriaza-
kos, “Admission control for QoS support in heteroge-
neous 4G wireless networks,” Network, IEEE, vol. 22,
no. 3, pp. 30 –37, 2008.
[21] R. Radner, “Team decision problems,” Ann. Math.
Statist., vol. 33, no. 3, 1962.
[22] Y.-C. Ho, “Team decision theory and information struc-
tures,” Proceedings of the IEEE, vol. 68, no. 6, pp. 644
– 654, june 1980.
[23] B. Liang and Z. J. Haas, “Predictive distance-based mo-
bility management for multidimensional PCS networks,”
IEEE/ACM Trans. Netw., vol. 11, no. 5, pp. 718–732,
Oct. 2003.
[24] I. Akyildiz, J. S. Ho, and Y.-B. Lin, “Movement-based
location update and selective paging for PCS networks,”
IEEE/ACM Trans. Netw, vol. 4, no. 4, 1996.
[25] G. Yavas, D. Katsaros, O. Ulusoy, and Y. Manolopou-
los, “A data mining approach for location prediction in
mobile environments,” Data Knowl. Eng., vol. 54, pp.
121–146, August 2005.
[26] G. Pongor, “Omnet: Objective modular network testbed,”
in MASCOTS ’93: Proc. Int’l Workshop on Modeling,
Analysis, and Simulation on Computer and Telecommuni-
cation Systems. Society for Computer Simulation, 1993,
pp. 323–326.