Content uploaded by Ryo Hamamoto
Author content
All content in this area was uploaded by Ryo Hamamoto on Feb 12, 2018
Content may be subject to copyright.
An Access Point Selection Mechanism Based on
Cooperation of Access Points and Users Movement
Ryo Hamamoto, Chisa Takano, Hiroyasu Obata, and Kenji Ishida
Graduate School of Information Sciences,
Hiroshima City University,
3-4-1 Ozuka-Higashi, Asa-Minami-Ku, Hiroshima, 731-3194 Japan
Email: ryo@net.info., takano@, obata@, ishida@ hiroshima-cu.ac.jp
Tutomu Murase
Cloud System Research Laboratories
NEC Corporation,
Kawasaki-shi, 211-8666 Japan
Email: t-murase@ap.jp.nec.com
Abstract—Public areas, such as train stations and airports,
providing wireless Local Area Network (WLAN) services are
increasing and expanding because of the rapid development of
WLANs based on IEEE 802.11 standard. Moreover, because
of the advances in smartphone tethering technology, portable
access points (APs) such as mobile Wi-Fi routers are being
utilized more frequently. Consequently, there are increasing
circumstances where a user needs to select and connect to one of
the many APs. AP selection significantly determines the quality
of service of the subsequent communication session. Existing
AP selection algorithms consider user movement but not AP
movement. We propose an AP selection method that handles
both types of movement e ectively. Moreover, we show that the
proposed method improves the throughput significantly compared
to the existing method.
Keywords—mobile AP; AP selection; throughput; AP coopera-
tive; moving;
I. I
Mobile communication technology continues to develop
rapidly and has facilitated the use of laptops and smartphones
in public spaces. Wireless local area networks (wireless LANs)
based on the IEEE 802.11 standard have also become popu-
lar [1]. As a result, the number of public areas, such as train
stations and airports, providing wireless LAN services and
access points (APs) is increasing. A more recent trend is the
popularity of portable APs, such as the mobile Wi-Fi router. A
key factor in this trend is recent advancements in smartphone
tethering technology. This technology creates an environment
in which APs are expected to move frequently. This can result
in situations in which multiple APs share the same area. In
such a situation, users must select one of the many APs.
To ensure su cient communication quality, the standard
wireless LAN protocol based on IEEE 802.11 usually selects
an AP with the highest received signal strength indicator
(RSSI) [2]. However, this approach can su er from uneven
AP loading, where a disproportionate number of wireless
terminals attempt to connect to the same AP even though
alternative APs are under low load. This leads to AP overload
and congestion, which greatly reduces communication quality.
Selecting an AP with the largest RSSI value may not provide
a fair bandwidth allocation or e ective bandwidth utilization
in environments such as train stations and hotel lobbies [3].
Moreover, multi-rate wireless LAN environments where the
transmission rate depends on the distance between the AP and
the wireless terminal su er from what has been referred to as
the performance anomaly [4], [5]. This problem decreases the
throughput of the AP because the AP is connected to terminals
whose transmission rates are extremely low.
The e ects of these problems can be reduced by optimizing
AP selection. Some studies [6]–[9] have proposed and eval-
uated many AP selection methods. In particular, [9] proposed
an AP selection method based on user cooperation during
movement. The method of [9] improves the system throughput
(the sum of the throughput of all APs in the system) by coop-
erative user movement. Thus, the proposed method is based on
[9]. Furthermore, various studies [10]–[14] have demonstrated
the e ectiveness of cooperative user mobility. [10] proposed
a tra c control mechanism that considers user collaboration.
Moreover, [11] introduced a method to increase the system
spectral e ciency of cellular orthogonal frequency-division
multiplexing access systems by cooperation of participating
users in wireless cellular networks. In addition, [12]–[14]
have shown the relationship between communication quality
and user route selection. However, previous studies have not
considered AP mobility. It is expected that AP mobility will
become more common in future. Note that we assume a
mobile AP (3G LTE equipped portable Wi-Fi AP) that can
move easily (low cost). We give the following examples of
AP movement: (a) Multiple users move with a mobile Wi-Fi
router to more convenient places in a conference room (b) In
a public space, such as a baseball stadium or concert hall,
the network manager moves an AP installed on the ceiling to
areas of high temporary-population density (c) After a disaster,
an o cial moves an AP originally installed on a balloon or
helicopter to a shelter within the disaster area It is expected that
the throughput can be improved considerably by considering
the movement of both APs and users.
We propose an AP selection method on the basis of
the cooperation of APs and user movement. In addition, we
analyze the system throughput and average AP throughput
obtained by the proposed method.
This study is organized as follows. Section II describes
the performance anomaly and a current AP selection method
that is based on user movement. In Section III, we propose
an AP selection method that considers both user and AP
movement. We evaluate the proposed method in Section IV.
Finally, Section V presents conclusions and suggestions for
future work.
II. R
Here we explain the performance anomaly faced by multi-
rate wireless LANs and review an AP selection method that is
based on user movement.
A. Performance anomaly in multi-rate environment
The performance anomaly degrades the throughput of all
terminals connected to a single AP in a multi-rate environment.
The total throughput of terminals connected to the same
978-3-901882-76-0 @2015 IFIP
926
TABLE I. R (IEEE 802.11g) [9]
Distance dm 5 7 9 20 25 40 50 60
Transmission rate bMbps 54 48 36 24 18 12 9 6
E ective transmission rate bef fMbps 26.1 24.4 20.4 15.3 11.9 8.5 5.8 4.7
Fig. 1. AP selection method based on AP movement
AP becomes approximately the same as that of the terminal
with the lowest transmission rate. This problem is caused by
carrier sense multiple access with collision avoidance. In such
a wireless LAN environment, each terminal can obtain fair
transmission opportunity. In other words, CSMA CA provides
each terminal with equal access to the communication channel.
However, in a multi-rate environment, the time for which a
terminal occupies the channel depends on its transmission rate.
A terminal with a high transmission rate has short channel
occupancy time; therefore, the wait time of other terminals
is reduced. However, as the channel occupancy time of a
low transmission rate terminal is long, the wait time of other
terminals increases. In this situation, the transmission cycle of
terminals with high transmission rates becomes equal to that of
the terminal with the lowest transmission rate. As a result, the
throughputs of high transmission rate terminals are decreased.
In a multi-rate environment, the throughput aof the a-th
AP can be estimated using Eq. (1) [15]. According to Eq. (1),
AP throughput is equal to the harmonic average value of the
transmission rate of terminals connected to the same AP in a
multi-rate environment.
ana
na
i1
(bi a)1
1
(1)
In Eq. (1), naand bi a denote the number of terminals
connected to the a-th AP and the transmission rate of the i-
th terminal connected to the a-th AP, respectively. Hereafter,
we assume that Eq. (1) indicates the AP throughput. The
throughput u a of the u-th terminal connected to the a-th AP
is obtained by dividing the throughput aof the a-th AP by na
(Eq. (2)). This is because all terminals connected to the same
AP have the same values of u a.
u a a(na)1(2)
B. AP selection method based on user movement
Here we briefly explain an AP selection method based
on user movement [9]. We use this method as a baseline
in our comparative evaluations, and we refer to this method
as the existing method in our evaluation (Sec. IV). This
existing AP selection method maximizes system throughput
by cooperating the actions of a new terminal (new user).
denotes the sum of the throughput of all APs in the system.
The desired action is to move a new user to the AP within an
acceptable area (distance) prior to establishing a connection.
A new user has acceptable movement distance dth. Note
that a new user attempts to connect to the AP that can
maximize . To improve user and system throughput, it is
assumed that the new user is willing to move up to acceptable
movement distance dth. In [9], one of the stepped transmission
rates is selected according to the distance (Table I). In this
study, transmission rate is based on [9]. Thus, AP selection is
performed as follows. First, the throughput for all APs in the
system is calculated assuming the new user moves to an AP
within dth. If several APs o er equal maximum throughput, the
AP with the shortest distance moved is selected. According to
the above consideration, the AP selection method based on user
movement is considered a type of optimization problem; i.e.,
identifying AP athat o ers the maximum and minimum
movement distance mfor a new user connection. Note that
the user movement distance is never greater than dth.
III. AP AP
This section describes the proposed AP selection method
based on AP movement. For simplicity, we explain the proce-
dure for selecting a single AP.
A. Overview of AP selection method based on AP movement
First, we overview the proposed AP selection method
(Fig. 1). The proposed method is an extension of a previously
proposed method [9] where movable distances were set for
both users and APs. Here we define the movable distance of a
new user and that of the i-th AP as dth and ei th, respectively. na
denotes the number of users connected to the i-th AP. The new
user and the AP can move freely within the specified distance
in the system area. The new user selects an AP to maximize the
system throughput . Here has the same meaning as above.
The new user moves to a position where system throughput
can be maximized. The AP moves to maximize while
minimizing the reduction in the transmission rates of users
already connected to the AP. If there are several APs that can
maximize , the proposed method selects a combination with
the shortest movable distances for the AP and the user. Note
that the proposed method is the same as the existing method [9]
if e0.
To define the method as an optimization problem, the
objective function and constraints are defined as follows. The
objective function maximizes by connecting a new user
to the AP. Further, depends on the selected AP a, the
distance moved by a new user to the AP m, and the distance
moved by the AP la. The constraint conditions are the movable
distances of the user and the AP. The maximum movable
distance of a new user and the i-th AP are denoted as dth and
ei th, respectively. In addition, the allowable number of users is
typically limited by the AP’s specifications and policy in real
environments; the maximum number of users that can connect
to any one AP is denoted as N.
B. Determination of communication position of APs and users
to maximize system throughput
This subsection describes a method that yields the com-
munication positions of APs and users to maximize system
throughput using Fig. 2. In Fig. 2, three users are already
connected to the AP and a new user attempts to connect to the
AP. We assume that each AP and user can obtain the position
information (coordinates) of each terminal using certain tools
such as GPS. Note that this system assumes movement on a
two-dimensional surface; i.e., users and APs exist on the same
plane.
2015 IFIP/IEEE International Symposium on Integrated Network Management (IM2015): Short Paper
927
Fig. 2. Determination of communication position of APs and new users
First, the new user moves toward the gravity point (G1)
of the plane formed by all existing users connected to the
AP within the movable distance (dth) (Fig. 2 1). If the new
user can move to G1, the user stops at G1. If the new user
cannot move to G1, the user stops at the point that minimizes
the distance between the user and G1. Here the gravity point
of the plane is defined as the point that minimizes the sum
of the squares of the distances from each vertex constituting
the plane. Using this procedure, new and existing users can
improve throughput after AP movement. Note that if there are
no existing users, the new user moves to the position of the
AP. If only one user is connected to the AP, the new user
moves to the position of the existing user. If two users are
connected to the AP, the new user moves to the midpoint of
the line connecting the two existing users. Next, the AP moves
the gravity point (G2) of the plane formed by all existing
connected users. The AP moves up to its movable distance
(eAP th) (Fig. 2 2). If the AP can move to G2, it stops at
G2. If the AP cannot move to G2, it stops at the point that
minimizes the distance between itself and G2. In addition, if
there is no existing user, the AP moves to the position of the
new user. Through movement of the AP and the new user,
all existing users and the new user can minimize the distance
to the AP without exceeding the upper limit of the movable
distance. Thus, users can communicate at higher transmission
rates because the distance between the users and AP is reduced.
As a result, these procedures maximize the system throughput.
Note that the proposed method performs the above procedures
for all APs and only one AP with maximum among all APs
moves.
IV. T
Here we present an evaluation of the proposed method
using a simulator written in the C programming language.
Note that we focus on system throughput and average AP
throughput. We evaluate the following two characteristics.
I Impact of movable distance of APs on system throughput
performance
II Impact of the number of APs on AP throughput perfor-
mance
The first characteristic is an indication of the e ectiveness
of AP movement, which is the key idea of the the proposed
method. The second characteristic indicates the e ectiveness
of AP movement in terms of the throughput of each AP.
Note that AP throughput is obtained using Eq. (1), and
we use the stepped e ective transmission rate bef f as the
transmission rate. In addition, the radio property of the phys-
ical layer changes drastically if the communication environ-
ment changes. In this situation, the Media Access Control
(MAC) characteristics also change because they depend on
the physical layer. Thus, if we assume the physical layer and
MAC, it is expected that the potential e ect of cooperative
Fig. 3. Relationship between system throughput and movable distance of
both the AP and the new user (vs. minimum distance selection method)
Fig. 4. Relationship between system throughput and movable distance of
both the AP and the new user (vs. existing method)
movement will become unclear. Therefore, to achieve our
purpose, we do not consider the physical layer and MAC
in our evaluations. In future, we will evaluate the proposed
method using a network simulator considering more realistic
network environment characteristics, such as control overhead
and interference between APs.
A. Movable distance of APs and throughput performance
First, we evaluate the relationship between the system
throughput and the movable distance of an AP. Note that the
system throughput is the sum of the throughput of all APs in
the network. In a 120m 120m region, there are APs at (30
m, 60 m) and (90 m, 60 m). This system configuration follows
a previous study [9]. In the initial state, only APs are present.
Next, users enter the region randomly. All users connect to the
AP that o ers the highest system throughput. In this situation,
the proposed method improves system throughput over the
minimum distance selection method and that of [9]. In the
minimum distance selection method, the user connects to the
nearest AP without moving. This method is similar to the AP
selection method using RSSI values as the metric, which is
a common approach. Table I shows the transmission rate of
the user, which has been used in previous studies. This study
assumes a saturated UDP flow (i.e., the user always has data
to send). In our evaluations, each simulation ran for 30 trials
and the results are the averages of the 30 trials. The initial
position of the user changed in each simulation.
In the proposed method, both the AP and user move. Thus,
in the proposed method, the distance between the AP and
user is shorter than in the minimum distance selection method
and the method in which only new users move. We have
clarified the value by which the proposed method improves
the throughput over the minimum distance selection method
and the method of [9].
Figure 3 shows the throughput improvement ratio of the
proposed method compared to the minimum distance selection
method. In the proposed method, the movable distance of both
the APs and new users was set to the same value. In Fig. 3,
the number of join users is 1 and 10. In Fig. 3, the horizontal
axes and vertical axes represent the movable distance of the
AP and the new user, respectively. From Fig. 3, if the number
of join users is 1, the system throughput of the proposed
method is 2 8 times greater than that of the minimum distance
selection method when the proposed method can obtain the
highest values. In addition, for 10 users, the proposed method
2015 IFIP/IEEE International Symposium on Integrated Network Management (IM2015): Short Paper
928
Fig. 5. Throughput performance vs. number of APs in the system (movable
distance of AP 10 m, new user 10 m).
can triple system throughput compared to the minimum dis-
tance selection method. However, compared to the minimum
distance selection method, when the number of join users
is high and user movement distance is short, the proposed
method does not improve the throughput significantly even
if the movable distance of the AP increases. This is because
it is di cult to improve throughput by AP movement since
many users are distributed widely across the region. Therefore,
throughput improvement is best achieved by increasing the
movable distance of users rather than that of the AP.
Next, Fig. 4 shows the improvement ratio of system
throughput compared to the existing method. For the proposed
method, the movable distance of both APs and the new user
is changed from 10 m to 60 m. When the number of join
users is 1, the proposed method improves system throughput
up to 2 4 times that of the existing method. If there are 10
users, the system throughput is improved up to 1 4 times.
However, throughput does not improve significantly when the
user’s movable distance is greater than 60 m. This is because
users can move to a position where they can obtain a higher
transmission rate by themselves. Therefore, the proposed sys-
tem can obtain a higher throughput without AP movement.
In summary, the proposed method can improve throughput
drastically compared to conventional methods.
B. Throughput performance vs. number of APs
Here we evaluate average AP throughput when the number
of APs in the system changes. We use a 120 m 120 m area,
as in IV-A. In the initial state, the specified number of APs is
set in the region. Users then enter the region randomly until
the number of users connected to the AP becomes greater than
two. At that point, we calculate the average AP throughput.
As the number of users is greater than 2, all APs have
connected users and experience the performance anomaly. The
transmission rate, tra c type, and the number of trials of the
simulation are the same as discussed in IV-A.
Figure 5 shows the average AP throughput for the existing
method [9] and the proposed method. The movable distance
of users is 10 m for both methods and the movable distance of
the AP is 10 m for the proposed method. This aforementioned
value indicates that the movable distance of the user and the AP
is the shortest. From Fig. 5, the maximum di erence between
the proposed method and the existing method is approximately
8 Mbps when there are 10 APs. Moreover, with four APs, the
proposed method obtains the same average AP throughput as
the existing method with 10 APs (dotted line in Fig. 5).
From the above results, it is clear that the proposed method
can achieve higher average AP throughput with a fewer number
of APs than the existing method.
V. C
This study has proposed an AP selection method that
considers the movement of users and APs and has evaluated
its performance. We evaluated the e ectiveness of portable
APs, such as mobile Wi-Fi routers, to increase the system
throughput in multi-rate wireless LAN environments. The pro-
posed method can improve the system throughput drastically
compared to the existing method, which moves only the new
user. This is because the AP can move to a position where
existing users can obtain higher transmission rates. In addition,
we compared the system throughput in two environments, i.e.,
when the users and APs can occupy only particular points
and when they can occupy any points in the service area.
The results indicate that the di erence in destination points
has little impact on throughput. Future work will include the
following evaluations.
1) Detailed investigations of characteristics
2) Development of a distributed algorithm
A
This work was partly supported by the “New generation
network R&D program for innovative network virtualization
platform and its application,” National Institute of Information
and Communications Technology, Japan; JSPS KAKENHI
Grant Numbers 26280032, 26420367; Project Research Grants,
Graduate School of Information Sciences, Hiroshima City Uni-
versity; and Grant for Special Academic Research, Hiroshima
City University.
R
[1] IEEE 802.11 Working Group, “Part 11: Wireless LAN medium access
control (MAC) and physical layer (PHY) specifications,” ANSI IEEE
Std. 802.11, Sep. 1999.
[2] M.S. Gast, “802.11 Wireless network: The Definitive Guide,”
O’REILLY, 2002.
[3] A. Balachandran, P. Bahl, and G.M. Voelker, “Hot-spot congestion
relief in public-area wireless networks,” Proc. WMCSA 2002, pp.70–
82, 2002.
[4] M. Heusse, F. Rousseau, G. Berger-Sabbatel, and A. Duda, “Per-
formance anomaly of 802.11b,” Proc. IEEE INFOCOM 2003, vol.2,
pp.836–843, 2003.
[5] T. Ikenaga, F. Miki, D. Nobayashi, and Y. Fukuda, “Performance eval-
uation of single-channel multi-rate communication in wireless LANs,”
International Journal of Research and Reviews in Computer Science,
vol.2, no.1, pp.168–172, 2011.
[6] Y. Fukuda, T. Abe, and Y. Oie, “Decentralized access point selection
architecture for wireless LAN,” Proc. Wireless Telecommunications
Symposium 2004, pp.137–145, 2004.
[7] H. Gong, K. Nahm, and J.W. Kim, “Distributed fair access point
selection for multi-rate IEEE 802.11 WLAN,” Proc. IEEE CCNC 2008,
pp.528–532, 2008.
[8] X. Wan, X. Wang, U. Heo, and J. Choi, “New AP-selection strategy
for high density IEEE802.11 WLANs,” Proc. International Conference
on CyberC, pp.52–58, 2010.
[9] S. Miyata, T. Murase, and K. Yamaoka, “Novel access-point selection
for user QoS and system optimization based on user cooperative
moving,” IEICE trans. on Commun., Vol.E95-B, No.6, pp.1953-1964,
2012.
[10] S. Kaneda, Y. Akinaga, N. Shinagawa, and A. Miura, “Tra c Control
by Influencing Users’Behavior in Mobile Networks,” Proc. 19th ITC,
pp.583–592, 2005.
[11] R. Schoenen, H. Yanikomeroglu, and B. Walke, “User in the loop:
mobility aware users substantially boost spectral e ciency of cellular
OFDMA systems,” IEEE Communications letters, vol.15, no.5, pp.488–
490, 2011.
[12] G. Motoyoshi, Y. Sudo,T. Murase, T. Masuzawa, “Advantages of
optimal longcut route for wireless mobile users,” Proc. IEEE ICC 2011,
pp.1–6 2011.
[13] T. Kakehi, R. Shinkuma, T. Murase, G. Motoyoshi, K. Yamori, and
T. Takahashi, “Route instruction mechanism for mobile users leveraging
distributed wireless resources,” IEICE Trans. on Commun., vol.E95-B,
no.6, pp.1965–1973, 2012.
[14] K. Kanai, Y. Akamatsu, J. Katto, and T. Murase, “QoS characteristics
on a longcut route with various radio resource models,” Proc. IEEE
PerCom 2012, pp.419–422, 2012.
[15] K. Medepalli and F.A. Tobagi, “Throughput analysis of IEEE 802.11
wireless LANs using a average cycle time approach” Proc. IEEE
GLOBECOM 2005, pp.3007–3011, 2005.
2015 IFIP/IEEE International Symposium on Integrated Network Management (IM2015): Short Paper
929