You are on page 1of 7

Centralized channel allocation algorithm for IEEE

802.11 networks
Helga Balbi, Natalia Fernandes, Felipe Souza, Ricardo Carrano,
Celio Albuquerque, Debora Muchaluat-Saade, Luiz Magalhaes
Universidade Federal Fluminense (UFF)
Niter oi, RJ - Brazil
E-mails: {helgadb, natalia, ferolim, carrano, celio, debora, schara}@midiacom.uff.br
AbstractThe sharing of the wireless spectrum is a major
concern of network administrators. Access points in the same
network interfere with each other, degrading the aggregate
performance of stations. Moreover, wireless networks usually
coexist with others applications that share the same spectrum
and negatively impact the packet transmission. To overcome these
issues, we propose the channel allocation algorithm designed for
central controllers of infra-structured IEEE 802.11 networks.
Our algorithm reduces the interference in controlled access points
through the dynamic choice of their operating channels and,
unlike other proposals, was designed to operate in a network
composed of low cost devices from different brands, and open
source software. Furthermore, we also consider the interference
caused by unmanaged networks, adjusting the settings of the
managed access points according to the wireless environment.
The proposal was implemented and evaluated in an open testbed,
and the results show that our controller efciently manages
the spectrum with low cost equipment and a low complexity
algorithm.
I. INTRODUCTION
With the popularity of IEEE 802.11, scale issues began to
emerge. One of these issues is the difculty to individually
manage each access point (AP) when the network includes
several APs. Another problem is the low management support
for coordinating APs. Lower cost devices have only a web
management interface, which, usually, do not consider other
APs, even those that are under the same management domain.
In order to provide a good throughput in a wireless network,
the administrator must reduce the interference among APs.
Nevertheless, even if the administrator manually allocates a
proper channel to each AP in order to minimize interference,
a static conguration will be obsolete in a fast changing
environment where new networks can be installed or excluded.
Likewise, even if there are no other on-site wireless networks,
the conguration and maintenance of each device must be
frequently performed, which is a hard task for larger networks.
To address these issues, the Project SCIFI - Intelligent Con-
trol System for Wireless Networks was created. The system
comprises a central controller which performs automatic and
dynamic congurations on low cost APs that run OpenWrt
embedded system [1].
This paper presents the main algorithm implemented in
SCIFI, responsible for dening the channel allocation of APs.
The algorithm was implemented and tested on an open testbed
in the School of Engineering of the Universidade Federal
Fluminense (UFF). Unlike many other proposals, SCIFI is ca-
pable of running with standard clients, increasing the systems
compatibility with current devices. Furthermore, the system
considers the interference generated by APs belonging to other
networks, not only by managed APs, with the only require-
ment of using APs compatible with OpenWrt [1]. Hence, it
reduces costs when compared to other proprietary solutions
that demands special hardware. The tests results show that
our controller efciently allocates channels to APs, increasing
the overall performance for both controlled and uncontrolled
networks that share the same radio medium.
This paper is organized as follows: Section II presents
prior related work on channel allocation techniques for 802.11
networks; Section III presents the channel allocation algorithm
implemented in SCIFI; Section IV presents the evaluation of
the algorithm, and nally Section V presents the conclusions
and future perspectives for SCIFI.
II. RELATED WORKS
At present, two unlicensed spectrum bands are available
to 802.11 standards: 2.4 GHz and 5 GHz. The 2.4 GHz
band, which is used by 802.11 b and g standards, has up to
14 channels available according to the region of the world.
However, in most countries, only three of these 14 channels
do not overlap [2]. In Brazil, for instance, the orthogonal
channels are 1, 6 and 11, since 12, 13 and 14 are not available.
When installing a wireless network, it is recommended that
neighboring APs are congured to operate in non-overlapping
channels in order to reduce interference. However, due to the
lack of such channels, the channel reuse becomes necessary in
dense networks. In this case, channel allocation techniques can
be used to determine a channel conguration that minimizes
the interference caused by the APs.
Proposals for channel allocation are divided in two main
groups: proposals that consider only the interference observed
by the APs, such as [3][8], and proposals that also consider
the interference observed by the client devices, such as [9]
[11]. The use of client data reveals interference scenarios that
cannot be detected with AP data. However, this client data
depends on special clients that perform spectral scanning to
nd other clients or APs that may interfere. Considering that
SCIFI requires compatibility with the full range of current
client devices, it does not depend on specic characteristics of
GIIS'12 1569662601
1
978-1-4673-5216-1/12/$31.00 2012 IEEE
client devices nor introduces changes to them. Therefore, our
channel allocation algorithm only considers the interference
caused between APs.
In some proposals, the channel allocation method allows
near APs to use non-orthogonal channels, i.e., channels whose
spectrum overlaps, as in [3][6], [9][11]. Mishra et al is said
to be the rst proposal to explore this area, indicating that it is
more interesting that two APs operate in overlapping channels
than on the same channel [2]. Since in the rst case the
interference radius of APs is reduced compared to the second,
it enhances the spatial reuse in a WLAN, increasing the
frequency with which the same channel can be used by other
APs. Riihijarvi et al, however, shows through experimental
results that, in cases where the channel interval between APs
is smaller than 4, using them on the same channel is more
interesting than in partially overlapping channels [6]. In fact,
the 802.11 MAC works better for APs operating on the same
channel than in nearby overlapping channels. Given that the
use of partially overlapping channels can cause signicant
reduction in network throughput, in our proposal, as well as [7]
and [8], we chose to not use overlapping channels, although
our algorithm can be extended to consider the use of such
channels when including an appropriate weighting factor.
As highlighted by Bernaschi et al, another issue that can
be addressed during the channel allocation in a WLAN
is the interference caused by APs that are not under the
same management domain and, therefore, will not have their
channels changed by the algorithm [4]. We consider this
interference important for our system because, unlike most of
the proposals found in the literature, our algorithm will operate
in a controlled network that will share the medium with APs
that are not under the controller management. Therefore, our
proposal considers the interference caused by the operation of
nearby APs belonging to different management domains.
The proposals found in literature also differ in how the
computation of channel allocation is performed, and may use
linear programming or a heuristic. Proposals based on linear
programming, as in [11], have the disadvantage of requiring
considerable computational processing due to their complexity.
Because of their large processing time, these proposals are
more applicable to static cases, i.e., cases where the channel
allocation is performed only once or at long intervals. On the
other hand, proposals based on heuristics, such as [3], [5]
[10], are interesting because of their reduced computational
complexity. In these cases, a suboptimal solution is acceptable.
Indeed, because of its fast computation, the channel allocation
can be performed at shorter intervals, in a dynamic and
adaptive way. As our solution aims to be adaptive, we chose
a heuristic to compute channel allocation.
Among the proposals based on heuristics, [3] and [7] use
propagation models for calculating the signal attenuation be-
tween interfering APs. The use of such models is not desirable
in our proposal because they do not accurately characterize
certain environments, such as indoor environments, in which
SCIFI system is intended to be used. The proposals that most
closely aligned with our requirements are [5], [6], [8]. In
[8] the authors showed that graph coloring techniques can be
used as theoretical basis to channel allocation algorithms for
802.11 networks. After verifying that the coloring problem
is NP-hard, the authors propose the use of DSATUR [12]
heuristic to its solution. In [5] the authors present algorithms
that use partially overlapping channels and adaptively perform
the coloring according to changes in network topology. In
[6] the authors complement DSATUR heuristics inserting a
new mechanism for using partially overlapping channels and
dene the implementation architecture of the algorithm, which
includes a central server, on which the algorithm is executed,
and a protocol for exchanging messages between the server
and the APs and among APs.
The graph coloring through DSATUR heuristic is easily
implemented and presents short execution time. Hence, we
choose DSATUR as the basis of the channel allocation al-
gorithm implemented in SCIFI. To meet all the SCIFI re-
quirements, we modied the original algorithm, as can be
seen in the next section. The intended goal was to create
a centralized algorithm with easy implementation and fast
execution, which was able to dene the channel allocation
in an WLAN reducing the interference caused between APs
that shares the same radio medium, including the managed and
unmanaged ones, without the need of deep changes in IEEE
802.11 standard nor in client devices, just by using a central
controller and APs compatible with OpenWrt rmware [1].
We opted for a centralized algorithm because, as the authors
conclude in [4], it prevents transient channel settings on the
network that would be caused if each access point could make
independent decisions in a decentralized manner. Moreover,
the convergence of the algorithm becomes faster when center-
ing channel allocation. On the other hand, as the authors in
[13] conclude, in terms of scalability decentralized algorithms
are considered more scalable, however, SCIFI aims to meet
small/medium sized institutional networks that do not have
large scale.
III. CHANNEL ALLOCATION ALGORITHM
This section describes the channel allocation algorithm
implemented on SCIFI, which is a system that aims to make
automatic and dynamic congurations of APs, improving the
manageability of IEEE 802.11 infra-structured networks. In
SCIFI, interference reduction is accomplished through the use
of centralized techniques for channel allocation, transmission
power control and client balancing among APs. The controller
orders each AP to execute certain tasks, such as performing
a spectral scanning or collecting statistics about associated
clients. Based on the collected data, the controller executes
the proposed algorithms. SCIFI is compatible with APs that
support OpenWRT operating system, and requires no deep
modication of the standard IEEE 802.11 protocols nor in
client devices, enabling the creation of large networks com-
posed by low cost devices from different brands.
Our channel allocation algorithm is modeled by an inter-
ference graph in which nodes (vertices) represent APs and
if theres interference between them, they must be connected
2
by unidirectional edges. Thus, the channel allocation problem
becomes the classic graph coloring problem, in which colors
represent the potential channels that APs can use. As the graph
coloring problem is classied as NP-hard, a heuristic is needed
for it to be solved in feasible time, such as DSATUR [8], [12].
The original DSATUR algorithm rst calculates a saturation
degree for each vertex, i.e., the number of different colors
that are already in use by neighboring vertices. If theres only
one vertex, this is chosen to be colored. If there are more
than one vertex, they are ordered decreasingly according to
their saturation degree. Thus, the vertex with higher degree is
colored with the rst available color.
In SCIFI, we improved this heuristic DSATUR according
to our requirements. First, to reduce the algorithm complexity
and take advantage of the global information of network, we
dened a centralized deployment, in which a central controller
collects information from APs and use them as basis for the
channel allocation. Moreover, we introduced the calculus of
the interference caused by networks that are not under the
same administrative domain. Also, the color choice is not made
in a greedy manner, like in original algorithm [8], [12]. Instead,
it is based on the quality of interfering signal received from
other APs. Furthermore, our algorithm considers the number
of associated clients with each controlled AP as a tiebreaker
for cases where there are more than one AP with the same
saturation degree.
The purpose of SCIFIs channel allocation algorithm is to
dene a set of channels for the controlled APs that comes
close to ideal in terms of minimizing the interference generated
by neighboring APs. In centralized deployment conducted for
validation and testing, the channel choice is restricted to one
of the three non-overlapping channels of 802.11g spectrum,
that is channels 1, 6 or 11, but the algorithm could easily be
adapted to 802.11a.
Figure 1 presents the basic pseudo code of the channel
allocation algorithm. Initially the central controller collects
data from APs, such as spectral scanning information (APs-
DataList[]). Based on this data, the algorithm determines the
interfering neighbors for each controlled AP, including APs
belonging to other management domains. Next, the controller
constructs the interference graph (Graph in line 1). This
graph consists of vertices representing the APs, and weighted
edges, which represent the interference between them. The
directional edges, which weighting value represents the quality
of received signal, are created if an access point is able to
receive signal from other. APs that do not belong to SCIFIs
administrative domain are represented by already colored
vertices. Subsequently, the controller creates a list containing
the vertices that are not colored (DiscoloredVerticesList[] in
line 2). At rst, discolored vertices are the controlled APs.
Each of discolored vertices will be colored in the block
while that begins in line 3 of Figure 1. Initially, the vertices
are ordered by priority in channel selection (line 4). Priority
is given to the vertex that has the highest saturation degree.
This degree indicates the number of different colors used by
neighboring vertices and does not consider vertices that were
not colored. If more than one vertex has the same saturation
degree, among them, the one with more associated clients will
have preference in channel selection. If there is a tie again,
priority is given based on the IP address of APs, so that a
lower IP has priority.
In our proposal, the interference generated by neighborhood
in channels different from 1, 6 or 11 is considered as it was oc-
curring in one of these channels (the nearest), since the spectra
of these channels partially overlaps. Given that this practice
can insert a small error on the interference calculating, in the
future we intend to implement a more rened mechanism for
interference estimates caused by overlapping channels. At rst,
the interference generated by controlled APs is not considered
in the calculation of the saturation degree, as its new channel
will be determined only at the end of the algorithm execution.
But when an AP gets a new channel, it begins to affect the
saturation degree of its neighbors.
After determining which vertex will be colored, its color
must be dened. This task is accomplished by the function
ColoringVertice (Figure 1 - line 6), which simplied pseudo
code can be seen in Figure 2. Initially, a list containing all
interference edges of the vertex is created (line 1). Next, the
controller traverses this list to verify which of the possible
channels are occupied, i.e., which of them have an operating
AP. Such channels are removed from the list, leaving only
unoccupied channels (line 2). If there are unoccupied channels,
rst channel of the list will be selected (line 4). If there are no
unoccupied channels, the controller will search for the channel
with less interference. For this, it traverses the edges list again
and performs the sum of edge qualities for each channel. The
channel that achieves the smallest sum of qualities will be
chosen (line 8).
We proposed this heuristic to select the channels because
we consider that the lower the observed signal quality of a
neighbor AP, the smaller the interference area of that neighbor
AP. Indeed, if the signal to noise ratio of the channel is low,
it indicates a greater distance between the APs, as shown
in Figure 3. This gure shows two scenarios that exemplify
our considerations on the relationship between the quality of
the received signal and the area of interference between two
access points. In scenario A, AP A receives signal from AP
B with high quality, and vice versa. In scenario B, AP A
receives signal from AP B with low quality, and vice versa.
In the latter scenario, as the lower quality indicates greater
distance between access points, the interference area between
them becomes smaller compared to the interference area in
scenario A. Thus, the controller must select the channel with
lower quality to reduce interference area.
If there is more than one neighboring AP on the same
channel, our algorithm considers the sum of all the neighbor
qualities in the channel. This sum reects both the sum of the
interference areas and the number of APs that share medium
access. Furthermore, this metric also facilitates power control
and load balancing mechanisms, which can be executed in
SCIFI to increase the network performance, but these mecha-
nisms will not be discussed here for lack of space.
3
AllocateChannels(APsDataList[], PossibleChannelsList[])
1: Graph BuildInterferenceGraph(APsDataList[])
2: DiscoloredV erticesList[] ListDiscoloredV ertices(Graph)
3: while sizeofDiscoloredV erticesList[] > 0 do
4: OrderedV erticesList
OrderV erticesByPriority(DiscoloredV erticesList[])
5: SelectedV ertice OrderedV erticesList[0]
6: ColoringV ertice(SelectedV ertice, PossibleChannelsList[])
7: DiscoloredV erticesList[]
RemoveV ertice(SelectedV ertice, DiscoloredV erticesList[])
8: end while
9: SetChannelsOnAPs();
Fig. 1. Basic pseudocode of SCIFIs channel allocation algorithm.
ColoringVertice(SelectedV ertice, PossibleChannelsList[])
1: EdgesList[] ListEdgesOfV ertice(SelectedV ertice)
2: ListOfFreeChannels[]
RemoveUsedChannels(EdgesList[], PossibleChannelsList[])
3: if sizeofListOfFreeChannels[] > 0 then
4: ChosenColor ListOfFreeChannels[0]
5: DefineV erticeColor(SelectedV ertice, ChosenColor)
6: return
7: else
8: ChosenColor
ChooseChannel(EdgesList[], PossibleChannelsList[])
9: DefineV erticeColor(SelectedV ertice, ChosenColor)
10: return
11: end if
Fig. 2. Basic pseudocode of SCIFIs channel allocation algorithm.
Returning to Figure 1, after being colored (line 6), the
vertex is removed from the list of discolored vertices and is
marked as colored, so that its interference is considered by
the other discolored vertices. The process is repeated until
all vertices are colored. Finally, the channels of APs are
physically congured in accordance with the selected channels
(line 9). In order to adjust the channel allocation in front of
changes in the environment, such as the rise of new interfering
APs, the algorithm runs with time interval dened by network
administrator. Executions are scheduled by a timer in the
controller.
IV. ALGORITHM EVALUATION
We evaluated our algorithm through tests conducted on an
open infra-structured 802.11g testbed developed in Universi-
dade Federal Fluminense (UFF). This testbed was composed
of a personal computer that ran SCIFI controller, seven low
cost OpenWRT-based APs, and a switch that connected the
controller and the APs. In addition, seven laptops were used
as network clients, each one associated with a respective AP.
In the implementation of SCIFI used in tests, communication
between APs and controller was performed via SSH (Secure
Shell) and registration of APs to be controlled was done
manually, but future work include automatically discover of
APs.
Fig. 3. The quality of the received signal indicates the interference area
between APs
Our test shows an example of a simple scenario among
several that can benet with the use of SCIFIs channel alloca-
tion algorithm. We demonstrate SCIFI benets by comparing
the throughput obtained in the testbed operating with and
without our controller. In tests without SCIFI, we used a
channel conguration with minimal interference between the
controlled APs. This conguration, however, did not consider
neighboring APs of other administrative domains.
Figure 4 shows part of the plant of the area where the
testbed was set. Controlled APs are represented by a dark gray
circle and those that are not controlled are represented by a
circle containing an x. Each client laptop was positioned
in a radius distance not greater than 2 meters from the AP
to which it was associated. Two models of APs were used
in the network, both manufactured by Ubiquiti [14]. One of
them was NanoStation Loco M2, which can be distinguished
in the gure by the presence of arrows representing the
direction of its 60

and 8 dBi sectorial antenna. Another


model was PicoStation 2, that has a 6 dBi omnidirectional
antenna. In all tests, the channels of the non-controlled APs
(represented by 04, 05, 06 and 07 in Figure 4) were 6, 6,
11 and 11, respectively. In the test without SCIFI, we set
the channels of controlled APs (C01, C02 and C03) as 6,
11 and 1, respectively, as shown in Figure 5. Notice that
this conguration avoids interference between controlled APs,
because it allocates a non-overlapping channel for each AP. By
using the described network structure, throughput tests were
performed to validate the implemented algorithm.
Iperf [15] was used to generate UDP trafc from each AP
to its respective associated client, using datagrams with 1470
bytes. UDP trafc was generated simultaneously from the
seven network APs and throughput samples were collected
once every 5 seconds. The rst and last samples were dis-
carded, considering the small error that could be caused by
the non-simultaneously activation of Iperf in each AP. Each
sample reported by Iperf is the average throughput on that
interval.
In tests performed without SCIFI, 3626 samples were col-
lected and, in tests with SCIFI, 3430 samples were collected.
Each test was performed on two different days. The average
4
values of samples obtained by each access point are presented
in Figure 6 with error margins given by a 95 % condence
interval. The error bars were small because of the large number
of samples. Comparing the results shown in gure 6 (a) and
(b), we note that SCIFI improves the throughput per AP in
most of the cases for both controlled and unmanaged APs.
This shows that the implemented algorithm and its interference
metric are effective for dening a better channel conguration,
in terms of throughput.
The main reasons for the SCIFI results are explained in fol-
lowing. In the tests with SCIFIs channel allocation algorithm,
the controller collected the interference information contained
in table I. This table shows which access points were found by
each controlled AP after performing a spectral scanning and
the quality of the received signal from each found AP. For
example, analyzing the intersection between column C01 and
line 05, we conclude that AP C01 receives signal from AP
05 with the highest quality (70/70). Cells with a - represent
that there is no communication between the APs, as occurs
between C01 and C02. The range of qualities, which varies
from 0 to 70, is determined by the wireless interface driver.
The access points used in the tests work with Madwi or
Ath9k drivers and the quality value reported by these drivers
represents the average Signal to Noise Ratio calculated for the
last received frames [16].
With the information from Table I the controller builds the
interference graph presented in gure 7, which is the basis for
the algorithm execution. Finally, gure 8 shows the channel
allocation calculated by SCIFI. As can be seen, the channels
1, 6 and 1 were assigned to the controlled APs C01, C02 and
C03, respectively. Although it seems worthwhile to use the
three managed APs in channels 6, 11 e 1 respectively (gure 6
- a), results were better when two of them operated on channel
1 (gure 6-b), since in this conguration they suffered less
interference.
Figure 9 shows the average aggregate throughput for both
the scenarios with and without SCIFI, considering error mar-
gins given by a 95% condence interval. Figure 9 (a) shows
the aggregate throughput of the network, including the seven
APs. In this graph, we can note the signicant increase of
network throughput (more than 29% in this scenario) when
SCIFI was used. Part of this increase was caused by the
throughput improvement of managed APs, as shown in gure
9 (b). Another part was due to the throughput improvement of
unmanaged APs, as shown in gure 9(c). These results show
that the network as a whole, including neighboring networks
that are not under the same administrative domain, may benet
from the use of the algorithm described in this paper.
V. CONCLUSION
This paper presented a new centralized channel allocation
algorithm for infra-structured 802.11 networks designed to op-
erate in a central controller for low cost APs called SCIFI. The
main objective of the algorithm is to reduce the interference
among APs and adjust their settings automatically according
to the ambient conditions, considering the interference caused
Fig. 4. Position of APs in testbed
Fig. 5. Channel conguration used in test performed without SCIFI
(b) Average LhroughpuL for each A
neLwork conflgured by SClll
13,27
13,60
20,98
14,09
9,73
8,40
9,41
0
3
10
13
20
23
M
e
g
a
b
l
L
s

p
e
r

s
e
c
o
n
d
(a) Average LhroughpuL for each A
neLwork conflgured wlLhouL uslng SClll
10,30
3,10
21,40
8,42
9,87
8,07 7,64
0
3
10
13
20
23
M
e
g
a
b
l
L
s

p
e
r

s
e
c
o
n
d
Fig. 6. Graphs of average throughput per AP
by APs belonging to other administrative domains whose
5
TABLE I
QUALITY OF RECEIVED SIGNAL
Channel
`
`
`
`
`
`
`
`
`
`
`
``
Found APs:
AP which
executes scan:
C01 C02 C03
- C01 x - 38/70
- C02 - x 50/70
- C03 35/70 46/70 x
6 04 39/70 34/70 70/70
6 05 70/70 - 29/70
11 06 - 70/70 49/70
11 07 31/70 49/70 70/70
Fig. 7. Interference Graph built by SCIFI
Fig. 8. Channel conguration used in test performed with SCIFI
channels cant be changed by the controller.
Among the other proposals in the literature, ours is distin-
guished mainly by having focus on its applicability in current
networks, without requiring major changes in 802.11 standard
or client devices. To quantify the interference, the algorithm
uses a metric based on the quality of communication channel
among neighboring APs, whose value can be obtained through
beacon frames.
Finally, the algorithm was validated on a network composed
by seven APs, in which throughput tests were performed
comparing the implemented solution with another that does
not consider the interference caused by neighboring APs that
belongs to other administrative domains. The results showed
that the network as a whole, including neighboring unmanaged
networks, can benet by the use of the algorithm described in
70,80
91,30
0
20
40
60
80
100
neLwork conflgured
wlLhouL uslng SClll
neLwork conflgured
by SClll
M
e
g
a
b
l
L
s

p
e
r

s
e
c
o
n
d
(a) Average AggregaLe 1hroughpuL of neLwork lncludlng lLs seven As.
34,00
41,63
0
10
20
30
40
30
neLwork conflgured
wlLhouL uslng SClll
neLwork conflgured
by SClll
M
e
g
a
b
l
L
s

p
e
r

s
e
c
o
n
d
(c) Average AggregaLe 1hroughpuL of nonconLrolled As (04, 03, 06, 07)
36,80
49,83
0
10
20
30
40
30
60
neLwork conflgured
wlLhouL uslng SClll
neLwork conflgured
by SClll
M
e
g
a
b
l
L
s

p
e
r

s
e
c
o
n
d
(b) Average AggregaLe 1hroughpuL of conLrolled As (C01, C02, C03)
Fig. 9. Graphs of average agregated throughput
this paper. Currently, a network of 60 nodes controlled by
SCIFI is being installed in UFF for large-scale testing. In the
future, the system will be available under the GPL license.
ACKNOWLEDGMENT
The work described in this paper was supported by RNP
(Rede Nacional de Ensino e esquisa).
REFERENCES
[1] (2012) OpenWRT website. [Online]. Available: https://openwrt.org/
[2] A. Mishra, E. Rozner, S. Banerjee, and W. Arbaugh, Exploiting
partially overlapping channels in wireless networks: turning a peril into
an advantage, in Proceedings of the 5th ACM SIGCOMM conference
on Internet Measurement - IMC 05. USENIX Association, Oct. 2005,
pp. 2929.
[3] R. Akl and A. Arepally, Dynamic channel assignment in IEEE 802.11
networks, in IEEE International Conference on Portable Information
Devices - PORTABLE07, May 2007, pp. 15.
[4] M. Bernaschi, F. Cacace, A. Davoli, D. Guerri, M. Latini, and L. Vollero,
A CAPWAP-based solution for frequency planning in large scale
networks of WiFi hot-spots, Computer Communications, vol. 34, no. 11,
pp. 12831293, Jul. 2011.
[5] J. Riihijarvi, M. Petrova, and P. Mahonen, Frequency allocation for
WLANs using graph colouring techniques, in Second Annual Confer-
ence on Wireless On-demand Network Systems and Services - WONS
2005, Jan. 2005, pp. 216 222.
[6] J. Riihijarvi, M. Petrova, P. Mahonen, and J. Barbosa, Performance
evaluation of automatic channel assignment mechanism for IEEE 802.11
based on graph colouring, in IEEE 17th International Symposium on
Personal, Indoor and Mobile Radio Communications, Sep. 2006, pp.
15.
6
[7] K. Leung and B.-J. Kim, Frequency assignment for IEEE 802.11
wireless networks, in IEEE 58th Vehicular Technology Conference -
VTC, vol. 3, Oct. 2003, pp. 14221426.
[8] P. Mahonen, J. Riihijarvi, and M. Petrova, Automatic channel allocation
for small wireless local area networks using graph colouring algorithm
approach, in 15th IEEE International Symposium on Personal, Indoor
and Mobile Radio Communications - PIMRC, vol. 1, Sep. 2004, pp.
536539.
[9] A. Mishra, V. Brik, S. Banerjee, A. Srinivasan, and W. Arbaugh, A
client-driven approach for channel management in wireless LANs, in
Proceedings of the 25th IEEE International Conference on Computer
Communications - INFOCOM 2006, Apr. 2006, pp. 1 12.
[10] A. Mishra, S. Banerjee, and W. Arbaugh, Weighted coloring based
channel assignment for WLANs, ACM SIGMOBILE Mobile Computing
and Communications Review, vol. 9, no. 3, pp. 1931, Jul. 2005.
[11] M. Haidar, R. Ghimire, H. Al-Rizzo, R. Akl, and Y. Chan, Channel
assignment in an IEEE 802.11 WLAN based on signal-to-interference
ratio, in Canadian Conference on Electrical and Computer Engineering
CCECE, May 2008, pp. 11691174.
[12] D. Br elaz, New methods to color the vertices of a graph, Communi-
cations of the ACM, vol. 22, no. 4, pp. 251256, Apr. 1979.
[13] S. Chieochan, E. Hossain, and J. Diamond, Channel assignment
schemes for infrastructure-based 802.11 wlans: A survey, IEEE Com-
munications Surveys Tutorials, vol. 12, no. 1, pp. 124 136, rst quarter
2010.
[14] (2012) Ubiquiti networks. [Online]. Available: http://www.ubnt.com
[15] (2012) Iperf. [Online]. Available: http://iperf.sourceforge.net
[16] (2012) RSSI in MadWi. [Online]. Available: http://madwi-
project.org/wiki/UserDocs/RSSI
7

You might also like