User-Based Channel Assignment Algorithm in a Load-Balanced IEEE 802.11 WLAN

66 Int. J. of Interdisciplinary Telecommunications and Networking, 1(2), 66-81, April-June 2009

User-based channel Assignment
Algorithm in a Load-balanced
IEEE 802.11 WLAN
Mohamad Haidar, Ecole de Technolgie Superieure, Canada
Hussain Al-Rizzo, University of Arkansas at Little Rock, USA
Yupo Chan, University of Arkansas at Little Rock, USA
Robert Akl, University of North Texas,USA

AbstrAct
A new load balancing algorithm is presented based on power management of Access Points (APs) to re-
duce congestion at hot spots in Wireless Local Area Networks (WLANs) and to assign channels to APs The
algorithm first finds the Most Congested Access Point (MCAP), then decreases its transmitted power in
discrete steps, and the process continues until the users’ assignment which leads to a high balance index is
reached. A new mathematical programming formulation is then applied to assign channels to the APs such
that the Signal-to-Interference Ratio (SIR) at the users’ level is maximized. Results show that the algorithm
is capable of reducing the overall congestion at hot spots in a WLAN and increases the SIR significantly for
cases involving relatively large WLANs. In the process, network throughput is increased. [Article copies
are available for purchase from InfoSci-on-Demand.com]

Keywords: Access Point; Channel Assignment; Congestion; Load Balancing; Signal-to-Interference
Ratio; Wireless Local Area Network

INtrODUctION with others, and access the World Wide Web
while roaming.
More industries, organizations, and offices Numerous research has been conducted
are installing WLANs to support the growing on load balancing (Akl & Park, 2005; Fang
demand of wireless users. The motivation is & Low, 2004; Haidar, et al., May 2007; Pa-
to reduce the cost of running cables and, more panikos & Logothetis, 2001; Velayos, et al.,
importantly, meet the demand of users who wish 2004) and channel assignment (Akl & Arepally,
to stay connected to the network, communicate 2007; Eisenblätter, et al., 2007; Haidar, et al.,
Sept 2007; Kulkrani & Shenoy, 2004; Lee,

Copyright © 2009, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global
is prohibited.

Int. J. of Interdisciplinary Telecommunications and Networking, 1(2), 66-81, April-June 2009 67

et al., 2002; Mishra, et al., 2005) separately. problem of AP placement (Eisenblätter, et al.,
The authors are aware of only one research 2007; Kulkrani & Shenoy, 2004; Lee, et al.,
that involved combining load balancing and 2002). In (Lee, et al., 2002), an algorithm was
channel assignment (Mishra, et al., 2006). proposed for channel assignment in hot-spot
The advantage of combining load balancing service areas using an ILP formulation. The
and channel assignment in one algorithm is objective was to avoid assigning non-overlap-
to increase network throughput by utilizing ping channels among neighboring APs, leaving
available resources efficiently. other available channels unutilized. The authors
In the load balancing related literature, the in (Eisenblätter, et al., 2007) noted that previous
authors of (Akl & Park, 2005; Fang & Low, AP placement and channel assignment were
2004) proposed minimizing AP congestion in always designed sequentially. An integrated
WLANs using an Integer Linear Programming model that addresses both issues concurrently
(ILP) formulation. In (Akl & Park, 2005; Fang was proposed. It was shown that through an
& Low, 2004), the load at the MCAP was mini- ILP formulation, AP placement and channel as-
mized once without applying power manage- signment could be combined, resulting in better
ment. A free space propagation model is used performance. In (Kulkrani & Shenoy, 2004),
in (Akl & Park, 2005; Fang & Low, 2004) to the authors presented a greedy heuristic algo-
associate demand points with candidate APs. rithm that provides maximum coverage while
The authors in (Papanikos & Logothetis, 2001) minimizing interference in the overlapping APs.
proposed a load balancing technique that allows A channel-assignment based on a Non-Linear
a wireless station to join an AP depending on Integer Program (NLIP) that minimizes channel
the number of existing users and the mean interference among neighboring APs was pro-
Received Signal Strength Indicator (RSSI). In vided in (Akl & Arepally, 2007). The authors in
(Velayos, et al., 2004), the authors proposed a (Mishra, et al., 2005) proposed a weighted vari-
load-balancing scheme for overlapping wireless ant of the coloring-graph algorithm to improve
cells. Load Balancing Agents (LBA) running the usage of wireless spectrum in WLANs. In
at each AP broadcast periodically the local (Haidar, et al., Sept 2007), we proposed a chan-
loads via the Ethernet backbone and determine nel assignment algorithm, based on minimizing
whether the AP is overloaded, under-loaded, or interference between APs. The algorithm was
balanced. Users in overloaded APs are forced invoked after the load has been balanced based
to dissociate from their corresponding AP and on the power management scheme presented
associate with an under-loaded AP. In (Haidar, in (Haidar, et al., May 2007). Results indicate
et al., May 2007), we proposed an algorithm that a different channel assignment is required
that decrements the transmitted power at the to meet users’ demands. Note that the channel
MCAP in discrete steps until any of the three assignment at the planning stage may result
following conditions is met: (1) at least one user in severe interferences due to users’ distribu-
can no longer associate with any potential AP, tion—even after power has been adjusted on
(2) the desired user’s data rate can no longer be APs. Finally, the authors in (Mishra, et al., 2006)
accommodated, or (3) the balance index value also proposed a client-based channel assignment
exceeds a predefined threshold. At that stage, and load balancing approach that lead to better
the network’s load is distributed efficiently usage of the wireless spectrum. They showed
over the network compared to traditional as- that a joint solution has significant advantages,
sociation based on the highest RSSI. This is such as capturing the effect of interference at
accomplished with a lower transmitted power the clients and a better channel re-use. Since
levels at the MCAPs leading to less co-channel the authors did not employ any specific indoor
and adjacent channel interferences. propagation model, their model was based on
At the planning stage, channel assign- minimizing channel conflicts between APs
ment is often considered in the context of the within one-hop distance from each other, as-

is prohibited.


suming all APs transmit at the same power. The remainder of this article is organized
However, in our case, the coverage, which is as follows. The power management algorithm
directly proportional to the transmitted power at is presented in the next section. The channel as-
the AP, is not fixed after a load-balanced state is signment algorithms are presented in the section
achieved. We tend to calculate interferences at after, followed by the summary of the proposed
users that even extend beyond one-hop distance algorithm and the assumptions involved. Then,
APs. Losses due to signal propagation are also the numerical results are presented, followed
accounted for. by the conclusion.
In this article, we extend our research re-
ported in (Haidar, et al., May 2007; Haidar, et
al., Sept 2007) by formulating a NLIP model tHE POWEr MANAGEMENt
to maximize the total SIR at the user level, ALGOrItHM
taking into account adjacent and co-channel
interference. The authors in (Boulmalf, et al., The power management algorithm is based on
2005) showed that an increase in the SIR of the iteratively decrementing the transmitted power
network would be reflected in an increase in the at the MCAPs in discrete steps. The received
network throughput. After the network achieves power at each user’s location is evaluated using
a balanced state, as measured by the balance the No Line of Sight (NLOS) Path Loss model
index (Balachandran, et al., 2002; Chiu, et al., (Lei, et al., 2005):
1989), the SIR of all users is used to reassign
channels to the APs. The algorithm in (Haidar, et P L ( d ) = P L 0+ 2 9 . 4 L o g 1 0 ( d ) +
al., May 2007) balances the load on the WLAN 6.1xαLog10(d)+2.4y + 1.3xy (1)
in order to utilize the available bandwidth more
efficiently. This is accomplished by reassigning
users to different APs, while decrementing the x ≤ 1.3,
transmitted power of the MCAPs. After achiev-
ing a balanced-load state, channels are assigned xs ≤ 1.5,
to APs based on minimizing the interference y ≤ 1.5
among neighboring APs ((Haidar, et al., Sept
2007). Although assigning channels after load
Here, PL0 is the free-space path-loss in
balancing provided better results than assign-
dB, d is the distance between user i and APj in
ing channels without load balancing, it still
meters, and xα, xs, and y are mutually indepen-
does not take advantage of the users’ presence
dent Gaussian random variables of zero mean
on the network to improve the channel assign-
and unit variance.
ment. Therefore, we propose another channel
Once the power received at a user from an
assignment that leads to the maximum SIR for
AP exceeds the receiver’s predefined sensitivity
the users. In other words, channels assigned
threshold, that user becomes a candidate for
are determined by the user’s SIR rather than
association with that AP. Thus, a user can be a
interference between neighboring APs. One
candidate for association with several APs.
advantage of such a scheme is that if a certain
The WLAN under consideration consists
“assigned” channel suffers frequency-selective
of a grid of M APs distributed in a single-floor
fading, the effect is recorded at the user(s) af-
indoor environment. A set of N randomly
fected in terms of a sudden drop in their SIR
distributed users seek to associate with an AP
values below a certain threshold. This triggers
each. A user is defined by its randomly assigned
the central server to run the algorithm again
position and data rate. After the initial channel
in order to assign different channels based on
assignment, which is based on minimizing the
maximizing the SIR.
interference between neighboring APs, we seek

is prohibited.


to redistribute users’ associations in order to at each iteration based on the new bandwidth
minimize the overall congestion in the network. utilization (Cj’s of all APs), and decrements
We then assign channels to APs based on the its power assuming no changes are occurring
final association of users to APs. in the channel environment during the course
As mentioned, this is achieved by first of simulation. In other words, users’ data rates
identifying the MCAP and decrementing its suffer minimal fluctuations and the average data
transmitted power in discrete steps. This is done rate is considered constant over the simulation
such that each user is associated with one and time, which depends on the variables involved
only one AP. The congestion factor at APj, Cj and computer processing time. The final solu-
,is defined as: tion provides the power level of the individual
APs and the final users’ associations such that
Nj
each user is connected to one AP.
∑R i

Cj = i =1 , (2)
BWj
tHE cHANNEL AssIGNMENt
where Nj is the number of users associated ALGOrItHMs
with APj , Ri is the data of user i, and BWj is
the maximum bandwidth for each AP (54 Mbps In IEEE 802.11 b/g WLAN, there exists 14
for IEEE 802.11g). The commercial software channels. Channels 1, 6 and 11 are non-overlap-
package LINGO (www.lindo.com) is used to ping, as shown in Figure 1. Only 11 channels
solve the following NLIP (Haidar, et al., May are used in the US. Each channel spreads over
2007). 22 MHz due to the Direct Sequence Spread
Spectrum (DSSS) technique employed by
min max {C1(x), C2(x),...,CM(x)},, IEEE 802.11b/g. For instance, channel 1 ranges
from 2.401 GHz to 2.423 GHz and its center
(3) frequency is 2.412 GHz. The center frequency
subject to, of two adjacent channels is separated by 5
M MHz. The overlapping channel interference
∑x ij =1 (3.1) factor wjk is defined as follows (Haidar, et al.,
j =1 Sept 2007):
Nj

∑R i• xij
1 − Chj − Chk × c if wjk 0 ,
≥
Cj ( x ) = i =1 , (3.2) w jk = 
BWj  0 otherwise
for j ∈{1, …, M}. (4)

Objective (3) minimizes the congestion where Chj is the channel assigned to APj, Chk
at the MCAP in each iteration. Constraint is the channel assigned to APk and c is the non-
(3.1) states that each user must be assigned to overlapping portion of two adjacent channels,
one and only one AP at any time. The binary expressed as a fraction of the frequency spec-
variable, xij, is 1 when user i is assigned to APj trum of a channel. For instance, channel 1 and
and 0 otherwise. Constraint (3.2) defines the channel 2 do not overlap from 2.401 GHz to
congestion factor at the APs as a function of 2.406 GHz, as shown in Figure 1. Normalizing
the assignment. the overlap of 5 MHz over the spectrum of 22
It should be noted that as the users’ as- MHz, c is approximately equal to 1/5. When
sociations are changing due to the decrease of the channels are far apart, as is the case with
the transmitted power at the MCAP, the algo- channels 1 and 6, wjk = 0 (i.e., no interference).
rithm appropriately relocates the new MCAP When the two channels are the same, wjk = 1

is prohibited.


Figure 1. The three non-overlapping channels

(i.e., maximum interference). Channels should • Djk is the distance between APj and APk.
be assigned to APs such that overlapping chan- • Qj is the cardinality of set Aj.
nel interference is minimized. However, there • Pik is the power received by user i associ-
are limited channel resources (11 channels in ated with APk.
IEEE 802.11 b/g). If the same channel is to be • Pij is the power received by user i from the
assigned to two or more APs that are located interfering APj.
far enough from each other, the overlapping • Interfjk is the interference on APj from
channel interference signal detected by each AP APk.
should be less than a given threshold. • Iij is the total interference experienced by
We present two channel-assignment algo- user i due to all other APs j (where j ≠ k).
rithms, to be executed in sequence. The first is • Pk is the power transmitted from APk.
commonly used to assign channels to APs in
the planning stage, prior to users’ connection channel Assignment Algorithm 1
in the network. In this algorithm, the objective
is to minimize the total interference among The mathematical formulation of the first chan-
neighboring APs. The second assigns channels nel assignment algorithm, based on minimizing
to APs when users enter the network. Remem- interference between APs, is given by the fol-
ber the objective of the algorithm developed lowing NLIP formulation:
in this article is to maximize SIR at the user
level. The following variables are defined for min
both algorithms: for all j (5.1)
subject to
• Aj is the set of neighboring APs to APj.
• K is the number of available channels, 11 wjkPk
Interfjk =
in IEEE 802.11 b/g. PL( Djk ) (5.2)

is prohibited.


1 − Chj − Chk × c if wjk 0≥ of all SIR values at the individual users in the
w jk =  network. Constraint (6.2) is the overlapping
 0 otherwise
channel interference factor. Constraint (6.3)
(5.3) defines the interference at user i due to all
APs j except APk to which user i is connected.
for k ∈ Aj , Constraint (6.4) defines the SIR at user i. The
for j ∈{1, …, M} , NLIP determines the integer variables Chj and
for Chj, Chk ∈{1, …, K}. Chk corresponding to the assignment of chan-
nels to APs.
Objective (5.1) minimizes the total inter- It should be emphasized that the scenarios
ference at each APj. Constraint (5.2) defines presented in this article demonstrate the im-
the interference experienced by APj due to provements achieved by combining our load
neighboring APk. Constraint (5.3) is the overlap- balancing algorithm based on power manage-
ping channel interference factor wjk. The NLIP ment with the proposed channel assignment at
determines the integer variables Chj and Chk the user level as compared to the existing ap-
corresponding to the assignment of channels proaches based solely on interference between
to APs. This algorithm has been used in (Akl APs. If the algorithm is to be executed in real
and Arepally, 2007; Haidar, et al., Sept 2007) time, each user i must update the serving APk
to assign channels in the planning stage, before with its associated SIRi(k) = ∑j SIRij(k) upon
user connections exist. registering with it. Then each AP, synchronized
with the other APs, periodically requests SIR
channel Assignment Algorithm 2 from its users to be transferred to a central
server where the proposed algorithm would
The mathematical-programming formulation be running. Similarly, in case of a change in
of the second channel assignment algorithm, the current user distribution, resulting from
based on maximizing the SIR for the users, is users joining or exiting the network, the APs
given below as an NLIP: would also transfer the SIRij(k) information to
N M
the central server that runs the load balancing
max ∑∑ SIR (k ) for all k, where k ≠ j,
i =1 j =1
ij
and channel assignment algorithm to reassign
channels to the APs.
(6.1) One can think of our model as representing
the scenario in a time slice, for a particular user
subject towjk = max (0, 1– |Chj – Chk| × c) distribution. It is important to note that user-
(6.2) to-user interference was assumed negligible
due to its low transmitted power compared to
N M the AP’s transmitted power. The model can
Iij = ∑∑ ( P ij • wjk ), k ≠ j, (6.3) be executed for all time slices sequentially in
i =1 j =1 which the SIRij(k) information is updated for
each time slice.
Pik , ∀ i, j, j ≠ k (6.4)
SIRij (k ) =
Iij
sUMMArY OF tHE
for j, k ∈{1,...,M},
for i ∈ {1,...,N},
PrOPOsED ALGOrItHM
for Chj, Chk ∈ {1,...,K}.
The 11-step algorithm, which assigns channels
in a power managed WLAN, is summarized
Objective (6.1) maximizes the total SIR
as follows:
for all N users. The total SIR refers to the sum

is prohibited.


1. Assign channels to the M APs based on convergence of the load balancing algorithm.
the NLIP model in (5), which is based on This metric is referred to as the balance index,
minimizing the total interference between which appeared for the first time in (Balachan-
APs. dran, et al., 2002; Chiu, et al., 1989). The bal-
2. Generate N randomly distributed users i. ance index reflects the capacity used in each
3. Compute the RSSI at each user i from each AP. Let Tj be the total traffic of APj. Then the
APj using the path loss model in (1). balance index β is:
4. Generate a binary matrix that assigns 1 in
the corresponding (i, j) if a user’s RSSI
exceeds a given threshold value or 0 oth- β=
∑T )
( j
2
, j ∈ {1,...,M} (7)
(n × ∑ T j
2
)
erwise.
5. Solve the user-AP assignment NLIP, us-
ing the objective function and constraints where n is the number of cells with overlapped
defined in (3), (3.1), and (3.2). coverage area. Thus, when an average user can
6. Identify the new MCAP. tie into multiple APs, n becomes large. With
7. Decrement the power level of the new this definition, the denominator is the small-
MCAP by a discrete step size of 1 dBm. est when all the Rj are equal, and there is no
8. Repeat step 7 until at least one user can no overlap in coverage (n is small). It assumes its
longer associate with an AP. largest value when Rjs are disparate, and there
9. Compute the interference caused by neigh- is excess overlap in service areas (n is large).
boring APs to each user. The balance index has the property that it ap-
10. Compute the SIR for each user. proaches 1 when each AP has its equal share of
11. Reassign channels to the APs by invoking user traffic. Conversely, it gets closer to a small
the model in (6). If a different user distribu- fraction (such as 1/n) when n is large and there
tion is deployed, go back to step 3. is a huge disparity among Rjs, leading to the
situation that APs are heavily unbalanced.
Note that the following assumptions are
imposed in the course of the simulations de-
scribed in the next section: NUMErIcAL rEsULts

1. AP coordinates are given. In order to validate the algorithm proposed in this
2. The average data rate of each user is article, we initially considered two benchmark
considered constant during the simulation problems. In the first, all users are distributed in
period. the overlapping region between APs, as shown
3. User-to-user interference is ignored. in Figure 2. In this case, the best channel as-
4. Users are stationary. signment leading to the maximum SIR among
5. User-AP association is assumed to be fixed users is the same as that leading to minimum
during each iteration until a new association interference between APs. This is obvious since
is obtained in step 4. all users fall in the overlapping coverage region
6. Receiver sensitivity is -85 dBm. between the two APs.
7. Receiver detection threshold is -110 In the second, all the users are in the cor-
dBm. responding AP coverage zone with none in
8. Initial transmitted power of each AP is 20 the overlap region, as shown in Figure 3. In
dBm. this case, the best channel assignment, based
on maximizing the SIR, could be any set of
Another key metric, in addition to the combinations of channels since the users are
congestion factor (Cj), was used to quantify not in the common overlapping region and
how balanced is the network and determine the adjacent channel or co-channel interference

is prohibited.


Figure 2. Users in the overlapping region 2. A particular user’s data rate can no longer
be accommodated.
3. The balance index value exceeds 0.97.

The simulations were carried out with
service areas consisting of 4, 6, 9 and 12 APs
and 20, 30, 45 and 60 users, respectively. APs
are placed 60 meters from each other, 20 me-
ters from adjacent walls and the service area’s
lengths and widths vary with the number of
APs.

Scenario 1: 4 APs and 20 Users
Figure 3. Users not in the overlapping region
In scenario 1, we consider a grid of 4 APs over
a 100 m × 100 m area and 20 randomly distrib-
uted users as shown in Figure 4. We generate a
binary matrix based on a receiver’s sensitivity
threshold of -85 dBm, as shown in Table 1.
As can be seen from Table I, a user can be
a candidate for association with multiple APs
if its signal level exceeds the threshold value.
For instance, user 1, U1, can connect to AP1,
AP2, and AP3, U10 can connect to AP2 and AP4,
and U19 can connect only to AP3.
We ran the 11-step algorithm until step 8,
right after the power management step. This
yields the final power level at each AP, which
in turn leads to the final received power at the
has no impact at the users. Therefore, in this users, and the final association matrix of users.
case, even the same channel can be assigned The final power level map and the user-AP as-
to all APs. These observations were validated sociations’ matrix are depicted in Figure 5 and
by our algorithm when it was run under these
two conditions.
Next, the scalability of the algorithm is
Figure 4. Power level map for scenario 1 at
verified by applying it to scenarios with different
ground level (0 meters)
numbers of APs and users. In each scenario, we
compare the results of our algorithm to previ-
ously published algorithms. It is important to
mention that the convergence time of any of the
scenarios presented ranged from few seconds
to few minutes based on meeting any of the
aforementioned three conditions that terminate
the simulation:

1. At least one user can no longer associate
with a candidate AP.

is prohibited.


Table 1. Initial associations’ matrix (scenario Table 2. Final users’ assignment with power
1) management (Scenario 1)
Users AP1 AP2 AP3 AP4
Users AP1 AP2 AP3 AP4
U1 1 1 1 0
U1 0 1 0 0
U2 0 0 1 0
U2 0 0 1 0
U3 0 0 0 1
U3 0 0 0 1
U4 0 0 0 1
U4 0 0 0 1
U5 1 1 1 1
U5 0 1 0 0
U6 1 1 0 0
U6 1 0 0 0
U7 0 1 0 0
U7 0 1 0 0
U8 0 0 1 1
U8 0 0 0 1
U9 0 0 1 0
U9 0 0 1 0
U10 0 1 0 1
U10 0 1 0 0
U11 1 0 1 0
U11 0 0 1 0
U12 0 0 1 1
U12 0 0 0 1
U13 1 0 0 0
U13 1 0 0 0
U14 0 1 1 1
U14 0 0 1 0
U15 1 1 0 1
U15 0 0 0 1
U16 1 0 1 0
U16 1 0 0 0
U17 1 1 0 0
U17 1 0 0 0
U18 1 1 0 0
U18 0 1 0 0
U19 0 0 1 0
U19 0 0 1 0
U20 1 1 0 0
U20 0 1 0 0

Figure 5. Power level map for scenario 1 after Figure 6. User to AP association
power adjustment

Table 2, respectively. At the end of step 8, the and AP4 to 3 dBm. Table II confirms that each
transmitted power was dropped on AP1 from 20 user is associated with one AP, which is the
dBm to 11 dBm, AP2 to 9 dBm, AP3 to 4 dBm first constraint in the NLIP model (3). Figure
6 depicts such a user-AP association.

is prohibited.


Table 3 displays the Congestion Factor improvement. The average SIR is defined as the
(CF) for each AP and the balance index for total SIR of the users divided by the number of
three cases: No Power Management (NPM) users in the network.
where each user connects to the AP with the
highest RSSI, the CF for each AP based on Scenario 2: 6 APs and 30 Users
the ILP model (3), where only the load at the
initial MCAP is reduced but NPM, and the CF In scenario 2, we considered 6 APs over 160
achieved at each AP by our proposed Power m × 100 m and 30 randomly distributed users.
Management (PM) algorithm. We ran the 11-step model until step 8 to get the
As can be seen from Table 3, the NLIP final transmitted power level at each AP and the
model (3) by itself was able to reduce initial final user-AP association. The transmitted power
congestion at AP1 only. However, the proposed remained the same on AP1 and AP2, whereas
PM algorithm was able to share the load more the power of AP3 dropped to 18 dBm, AP4 to 9
evenly across the entire network. This also dBm, AP5 to -1 dBm and AP6 to 18 dBm. Table
can be seen through the values of the balance 5 summarizes the CF for each AP.
index, which clearly indicates that our pro- Again, the NLIP (3) by itself was not able
posed algorithm leads to the most balanced to equalize the load at the APs. It is interesting
load. Finally, we proceeded with our algorithm to note that the CF of AP2 is very low because
through steps 9 to 11 to assign channels to the users associated with it happened to be at a short
APs based on maximizing the SIR at the users. distance such that their RSSI from other APs
Table IV documents the initial channel assign- are considered to be noise and there is no way
ment, based on minimizing AP interference they can associate to another AP. Therefore, the
with NPM, and channel assignment based on following users will remain connected to their
maximizing the SIR at the user level after PM, respective AP no matter how long we run the
based on model (6). algorithm. The same applies to AP1. Under the
Table 4 shows that if we were to start with PM algorithm, the remaining APs achieved a
a channel assignment in the planning stage more balanced load distribution (around 89%
and keep that channel assignment unchanged balanced load distribution), compared with
after users enter the network, the average SIR the default association and the NLIP model
would be 4.48 per user. However, by applying (3) by itself. It is noticed” that the algorithm
our algorithm at the balanced state, the average converged before exceeding a balance index
SIR, expressed as a dimensionless ratio, was value of 95%. This is because it was forced to
increased to 5.83, which corresponds to 30% quit based on the other two constraints (at least a

Table 3: comparison between congestion fac- Table 4: comparison between model (5) and
tors (scenario 1) model (6) (scenario 1)
CF Model CF with Initial Channel Final Channel
CF : NPM
(3): NPM PM Assignment Assignment
AP1 0.7323 0.5416 0.4404 AP1 11 1
AP2 0.4735 0.5378 0.4155 AP2 1 6
AP3 0.2283 0.3547 0.4559 AP3 8 11
AP4 0.2393 0.2393 0.3615 AP4 3 2
Balance Avg. SIR 4.48 5.83
80.47% 91.43% 99.27%
Index(β)

is prohibited.


Table 5. Comparison between congestion factors (Scenario 2)

CF : CF Model (3):
CF with PM
NPM NPM
AP1 0.1279 0.1279 0.1880
AP2 0.0791 0.0791 0.0791
AP3 0.4296 0.4296 0.4194
AP4 0.5365 0.6034 0.4095
AP5 0.7556 0.6887 0.4419
AP6 0.2171 0.2171 0.5308
Balance Index (β) 68.95% 70.23% 89.00%

user can no longer associate to an AP or a user’s the service area. Results for the CF and chan-
data rate can no longer be accommodated) and nel assignment are depicted in Tables 7 and 8,
that is the best balance index achieved without respectively.
violating the other constraints. We proceed to Model (3) decreased the load at the
steps 9 through 11, assigning channels to the MCAP (AP3) but increased it greatly on AP8.
APs, the results of which are shown in Table Our PM algorithm distributed the load evenly
6. The average SIR was improved by almost (around 97% balanced). This load distribution
6%. In this case, both AP3 and AP4 used chan- was achieved at the following power levels
nel 6, which means there were no users in their at the APs: 7 dBm, 1 dBm, -4 dBm, 2 dBm,
overlapping region. -5 dBm, -4 dBm, 6 dBm, -3 dBm and 2 dBm,
respectively.
Scenario 3: 9 APs and 45 Users Table 8 shows that the average SIR of the
network was improved to 1.93 (almost 74%
In this scenario, we deploy 9 APs over 160 m improvement). This significant improvement
× 160 m area. The APs are distributed in a 3 can be attributed to the fact that after load
× 3 grid with 45 users randomly distributed in distribution some users that were close to their
original AP are now redirected to a farther AP
that reduces the AP congestion. Therefore, a
user may suffer interference from its nearest
APs, yet have enough RSSI to associate with
Table 6. comparison between model (5) and an AP other than the original AP.
model (6) (scenario 2)

Initial Channel Final Channel Scenario 4: 12 APs and 60 Users
Assignment Assignment
Finally, we apply our algorithm to a 12-AP
AP1 6 2
service area. The 12 APs are deployed over
AP2 1 11 a 3 × 4 grid. This time 60 users are generated
AP3 6 6 randomly in the service area. Similar steps are
AP4 11 6 followed for comparison between the channel
AP5 1 8
assignment algorithm based on the minimum
interference between APs and our proposed
AP6 11 1
PM algorithm. Comparison of results is listed
Avg. SIR 2.64 2.80 in Tables 9 and 10.

is prohibited.


Table 7. Comparison between congestion factors (scenario 3)
CF Model
CF :NPM CF with PM
(3): NPM
AP1 0.2650 0.2650 0.2858
AP2 0.5457 0.5457 0.4002
AP3 0.5536 0.4332 0.3700
AP4 0.1867 0.1867 0.4736
AP5 0.5396 0.5396 0.3160
AP6 0.3885 0.3885 0.3288
AP7 0.0509 0.0509 0.4179
AP8 0.4695 0.5899 0.3477
AP9 0.2215 0.2215 0.2812
Balance Index (β) 81.03% 80.48% 97.24%

Table 8. comparison between model (5) and model (6) (scenario 3)

Initial Channel Final Channel
Assignment Assignment
AP1 4 6
AP2 9 1
AP3 1 11
AP4 11 8
AP5 1 11
AP6 11 4
AP7 6 6
AP8 11 8
AP9 6 11
Avg. SIR 1.11 1.93

Model (3) decreased the load at the MCAP ping users, which caused the average SIR to
(AP8) but increased it greatly on AP3. Our PM improve greatly. The total SIR of the 60 users
algorithm distributed the load efficiently among has improved from 0.4090 to 1.8508 (around
APs (around 98% balanced). The load distribu- 350%). Figure 7 graphs the channels assigned
tion is achieved at the following transmitted to all 12 APs in scenario 4 and the AP positions
power levels at the APs: 17 dBm, 4 dBm, 3 followed in the previous four scenarios.
dBm, 15 dBm, 14 dBm, 4 dBm, 10 dBm, 1
dBm, 3 dBm, 4 dBm, 7 dBm, and 10 dBm,
respectively. cONcLUsION
It is noticed from the results in Table 10
that our algorithm assigned the same channels In this article, a new channel assignment algo-
to some APs where there were no overlap- rithm has been proposed based on maximizing

is prohibited.


Table 9. Comparison between congestion factors (scenario 4)
CF Model
CF :NPM CF with PM
(3): NPM
AP1 0.2622 0.2622 0.5199
AP2 0.8017 0.8017 0.6348
AP3 0.8243 0.9307 0.5417
AP4 0.3337 0.3337 0.6318
AP5 0.4391 0.4391 0.6161
AP6 0.6932 0.6932 0.6444
AP7 0.3969 0.3969 0.5023
AP8 0.9326 0.6930 0.7209
AP9 0.5591 0.6923 0.4058
AP10 0.7083 0.7083 0.5614
AP11 0.4816 0.4816 0.5003
AP12 0.5110 0.5110 0.5110
Balance Index (β) 89.08% 89.80% 97.92%

Table 10. Comparison between model (5) and model (6) ( scenario 4)
Initial Channel Final Channel As-
Assignment signment
AP1 1 1
AP2 11 1
AP3 1 6
AP4 6 1
AP5 11 6
AP6 6 1
AP7 1 11
AP8 6 1
AP9 11 5
AP10 1 1
AP11 9 8
AP12 4 1
Avg. SIR 0.4090 1.8508

the SIR at the users in combination with a is based on minimizing interference between
scheme that reduces congestion at the APs based APs only, with no consideration given to users.
on power management. The algorithm has been We have shown that our proposed algorithm
shown to provide superior results compared is scalable with different network sizes. The
to previous work where channel assignment algorithm showed great improvement in the

is prohibited.


load distribution and average SIR. Channel Eisenblätter, A., Geerdes, H. F., & Siomina, I. (2007).
assignment is better determined at the end of Integrated Access Point Placement and Channel
the balanced load state rather than keeping the Assignment for Wireless LANs in an Indoor Office
initial channel assignment from the planning Environment. 8th IEEE Intl. Symposium on a World
of Wireless, Mobile and Multimedia Networks.
stage. This will improve the throughput of the
network. Extensive computations suggest the Fang, C. & Low, C. P. (2004). Efficient Algorithms
11-step algorithm leads toward more stable for the Load-Balanced Demand Point Assignment
and better User-based channel assignment and Problem in Large-Scale WLAN. 12th IEEE Confer-
load-balance ence on Networks, Vol. 1, pp.196-200.
Since this is a prototype study, there is Haidar, M., Akl, R., Al-Rizzo, H. M., Chan, Y., &
room for further research. Work is in progress Adada, R. (2007). Optimal Load Distribution in
to identify a more efficient heuristic algorithm to Large Scale WLAN Networks Utilizing a Power
solve the optimization problems, and to include Management Algorithm. Proceedings of IEEE
dynamic user behavior and mobility. The latter Sarnoff Symposium.
will determine how fast both algorithms can Haidar, M., Al-Rizzo, H. M., Akl, R., & Chan, Y.
adapt to the dynamic environment. With these (2007).Channel Assignment and Load Distribution
features in place, a more rigorous understanding in a Power-managed WLAN. Proceedings of 18th
of the convergence behavior of the algorithm Annual IEEE International Symposium on Personal,
can be obtained. Indoor and Mobile Radio Communications, Athens,
Greece.

Kulkrani, P. & Shenoy, P. (2004). Exploiting Over-
rEFErENcEs lap for Provisioning of Access Points in Wireless
Networks (Tech. Report). Boston, Massachusetts:
Akl, R., & Arepally, A. (2007). Dynamic Channel University of Massachusetts.
Assignment in IEEE 802.11 Networks. Proceedings
of IEEE Portable 2007: International Conference Lee, Y., Kim, K., Choi, Y. (2002). Optimization of
on Portable Information Devices, Orlando, FL, AP placement and Channel Assignment in Wireless
USA, pp. 1-5. LANs. LCN 2002. 27th Annual IEEE Conference on
Local Computer Networks, pp. 831-836.
Akl, R. & Park, S. (2005). Optimal Access Point
selection and Traffic Allocation in IEEE 802.11 Lei, J., Yates, R., Greenstein, L., & Liu, H. (2005).
Networks. Proceedings of 9th World Multiconference Wireless Link SNR Mapping Onto an Indoor
on Systemics, Cybernetics and Informatics (WMSCI Testbed. Proceedings of IEEE Tridentcom 2005,
2005): Communication and Network Systems, Tech- Trento, Italy.
nologies and Applications, vol. 8, pp.75-79.
Mishra, A., Banerjee, S., & Arbaugh, W. (2005).
Balachandran, A., Bahl, P., Voelker, G. M. (2002). Weighted Coloring Based Channel Assignment for
Hot-Spot Congestion Relief in Public-area Wireless WLANs. ACM SIGMOBILE Mobile Computing and
Networks. Proceedings Fourth IEEE Workshop on Communications Review, vol.9, pp.19-31.
Mobile Computing Systems and Applications, pp.
70-80. Mishra, A., Brik, V., Banarjee, S., Srinivasan, A.,
& Arbaugh, W. (2006). A Client-Driven Approach
Boulmalf, M., El-Sayed, H., & Soufyane, A. (2005). for Channel Management in Wireless LANs. 25th
Measured Throughput and SNR of IEEE 802.11g in International Conference on Computer Communi-
a Small Enterprise Environment. 61st IEEE Vehicular cations, pp.1-12.
Technology Conference (VTC’05), UAE, Vol. 2, pp.
1333-1337. Papanikos, I. & Logothetis, M. (2001). A Study on
Dynamic Load Balance for IEEE 802.11b Wireless
Chiu, D. & Jain, R. (1989). Analysis of the Increase LAN. Proc. 8th International Conference on Ad-
and Decrease Algorithms for Congestion Avoidance vances in Communication & Control, COMCON 8,
in Computer Networks. Journal of Computer Net- Rethymna, Crete/Greece.
works and ISDN, vol. 17, pp. 1-14.

is prohibited.


Velayos, H., Aleo, V., & Karlsson (2004). Load
Balancing in Overlapping Wireless LAN Cells.
Proceedings of IEEE ICC 2004, Paris, France.

Mohamad Haidar received his BS degree in electrical engineering from the Lebanese American University,
Byblos, Lebanon in 2000. The MS degree was received from Roosevelt University in telecommunications,
Chicago, IL, in 2002 and PhD in applied science from the University of Arkansas at Little Rock (UALR),
Little Rock, AR, in 2007. His major field of research is in wireless networking and communications. Previ-
ously, he worked as a development engineer at IBM, Durham, NC, teacher assistant at UALR, Little Rock,
AR, and WiMesh/WiMax RF engineer analyst on a UALR-Alltel-Motorola project. His research interests
are load balancing, channel assignment, MAC and PHY layers design in WLAN, cognitive wireless mesh
networks, ad hoc networks, and WiMax environments.

Hussain Al-Rizzo received his BSc in electronics and communications (1979) (High Honors), postgraduate
diploma in electronics and communications (1981) (high honors) and MSc in microwave communication
systems (1983) from the University of Mosul, Mosul, Iraq. From May 1983 to October 1987 he was working
with the Electromagnetic Wave Propagation Department, Space and Astronomy Research Center, Scientific
Research Council, Baghdad, Iraq. On December, 1987, he joined the Radiating Systems Research Labora-
tory, Electrical and Computer Engineering Department, University of New Brunswick, Fredericton, NB,
Canada where he obtained his PhD (1992) in electromagnetic, wireless communications, and the global
positioning system. For his various academic achievements he won the nomination by the University of
New Brunswick as the best doctoral graduate in science and engineering. Since 2000, he joined the Systems
Engineering Department, University Arkansas at Little Rock where he is currently a tenured associate
professor. He has published over 25 peer-reviewed journal papers and 54 conference presentations. His
research areas include implantable antennas and wireless systems, smart antennas, WLAN deployment
and load balancing, electromagnetic wave scattering by complex objects, design, modeling and testing of
high-power microwave applicators, design and analysis of microstrip antennas for mobile radio systems,
precipitation effects on terrestrial and satellite frequency re-use communication systems, field operation of
NAVSTAR GPS receivers, data processing, and accuracy assessment, effects of the ionosphere, troposphere
and multipath on code and carrier-beat phase GPS observations and the development of novel hybrid
Cartesian/cylindrical FD-TD models for passive microwave components..

Yupo Chan received his PhD from Massachusetts Institute of Technology in 1972. After 28 years of post-doc-
toral experience in industry, universities and government, he became the founding chair of the Department
of Systems Engineering at the University of Arkansas at Little Rock (UALR) Donaghey College of Informa-
tion Science and Systems Engineering (CyberCollege) in 2000. Before coming to the UALR CyberCollege,
Chan worked at The Air Force Institute of Technology, Washington State University, the State University of
New York at Stony Brook, Pennsylvania State University, and Kates, Peat Marwick. Additionally, he was
a congressional fellow in the Office of Technology Assessment in Washington, DC. Chan’s training and
research focus on transportation systems, telecommunications, networks and combinatorial optimization,
multi-criteria decision-making and spatial-temporal information. Chan has published numerous books
and monographs, including Location Theory and Decision Analysis (Thomson/South-Western); Location,
Transportation, and Land-Use: Modeling Spatial-temporal Information (Springer); Urban Planning and
Development Applications of GIS (ASCE Press).

Robert Akl received the BS degree in computer science from Washington University in St. Louis, in 1994,
and the BS, MS and DSc degrees in electrical engineering in 1994, 1996, and 2000, respectively. He also

is prohibited.


received the Dual Degree Engineering Outstanding Senior Award from Washington University in 1993.
He is a senior member of IEEE. Akl is currently an associate professor at the University of North Texas,
Department of Computer Science and Engineering. In 2002, he was an assistant professor at the University
of New Orleans, Department of Electrical and Computer Engineering. From October 2000 to December
2001, he was a senior systems engineer at Comspace Corporation, Coppell, TX. His research interests
include wireless communication and network design and optimization.

is prohibited.

User-Based Channel Assignment Algorithm in a Load-Balanced IEEE 802.11 WLAN

More Related Content

What's hot (19)

Viewers also liked (6)

Similar to User-Based Channel Assignment Algorithm in a Load-Balanced IEEE 802.11 WLAN (20)

Recently uploaded (20)

User-Based Channel Assignment Algorithm in a Load-Balanced IEEE 802.11 WLAN