Load Distribution and Channel
Assignment in IEEE 802.11
Wireless Local Area Networks
Ph.D. Dissertation Defense
Presented by Mohamad Haidar
Department of Applied Science
George W. Donaghey College of Engineering and
Information Technology,
University of Arkansas at Little Rock
November 9, 2007
Presentation Outline
Introduction
Wireless Local Area Networks (WLANs)
Access Points (APs) Congestion
Channel Assignment
Related Work
Contributions
Problems Statements
1. Congestion Problem
Proposed Solution
Problem Formulation
Algorithm
Numerical Analysis and Results
Simulations (OPNET)
11/09/2007 Ph.D. Defense 2
Presentation Outline (Cont’d)
2. Channel Assignment Problem
Proposed Solution
Problem Formulation
Algorithm
Numerical Analysis and Results
Simulations (OPNET)
Dynamic Model
Scenario 1 (variable data rate)
Scenario 2 (dynamic user distribution)
Conclusion
Future Work
11/09/2007 Ph.D. Defense 3
Introduction
Wireless Local Area
Networks (WLANs)
Airports
Hotels
Campuses
WLANs are divided into 3
categories:
IEEE 802.11a in the 5 GHz
band (54 Mbps)
IEEE 802.11b in the 2 GHz
band (11 Mbps)
IEEE 802.11g in the 2 GHz
band (54 Mbps)
Example of WLAN
11/09/2007 Ph.D. Defense 4
Introduction (Cont’d)
What is Access Point (AP)
congestion?
Some times referred to as “Hot
Spot”
CAP= (R1+ R2+..+ RN)/BW
CAP: Congestion at AP
R : Data rate of a user connected to the AP
BW: Bandwidth (11 Mbps for IEEE
802.11b)
Channel Assignment
Minimize interference
To improve QoS (less delay and
higher throughput)
3 non-overlapping channels in IEEE
802.11b/g (1, 6, and 11) Frequency Spectrum for IEEE 802.11b/g
11/09/2007 Ph.D. Defense 5
Limitation of Previous Research
AP Placement
The main objective was to use a minimum number of APs for
adequate coverage of the desired area.
Did not account for channel assignment and/or load distribution.
Channel Assignment
Based on minimizing co-channel interference.
Limited to either minimizing total interference between APs or
maximizing the sum of interference at a given AP.
When integrated and applied simultaneously with AP placement,
better results were achieved than dealing with them sequentially.
User distribution was not accounted for in the channel assignment.
11/09/2007 Ph.D. Defense 6
(Cont’d)
Load Balancing/Distribution
Balancing the load based on the number of active users
performs poorly because the data rate of users was not
taken into consideration.
Minimizing the congestion at the most congested AP by
redistributing users.
Improves the load ONLY at the MCAP.
Load balanced agents installed at the APs that broadcast
periodically their load. APs are either under-loaded,
balanced, or overloaded.
Static user distribution and no power management.
All APs involved should be equipped with the LBA software.
Cell breathing technique used to reduce the cell size to
achieve a better load distribution.
Connects to the next higher RSSI: is not always the best
choice.
Static user distribution.
No channel assignment was considered. Interference was
not accounted for.
11/09/2007 Ph.D. Defense 7
Contributions of the Current
Research
A new Load Balancing scheme based on Power
Management.
As long as the received power exceeds a certain
threshold, that AP is a potential for association.
Channel Assignment based on Maximizing the
SIR at the users.
Users involved in the assignment of channels.
Different user distributions will lead to different
channel assignment.
(Cont’d)
Combining both load balancing based on power
management and the channel assignment based
on SIR: A Novel Scheme.
Verified the performance predicted from
optimization versus realistic OPNET-based
network simulations: New contribution
Developed a realistic dynamic model approach
that accounts for variable users’ data rates and
users’ behavior: New contribution
11/09/2007 Ph.D. defense 9
A New Heuristic Algorithm
Initial Channel Assignment
Users enter to network
Load Balancing based on PM
Re-Assign channels based on SIR
Sort arriving users and departing users
in ascending order in a list
Check list
Arrive Depart
End of list
Results
Add user to list Remove user from list
11/09/2007 Ph.D. Defense 10
1st Problem
AP Congestion Problem
Degrades network throughput
Slowest station will make other stations wait longer.
Unfair load distribution over the network
causes bottlenecks at hot spots.
Inefficient bandwidth utilization of the network.
11/09/2007 Ph.D. Defense 11
Proposed Solution
Reduce congestion at the hot spots by decrementing the
power transmitted by the Most Congested AP (MCAP) in
discrete steps until one or more users can no longer
associate with any AP or their data rate can no longer be
accommodated.
The final transmitted power of each AP is set to the best
balance index, , achieved.
( Tj )2
Advantages: (n * Tj )
2
Load is fairly distributed.
Increase in data rate throughput per user.
Less adjacent and co-channel interference.
11/09/2007 Ph.D. Defense 12
Problem Formulation
MCAP NLIP formulation
min max{C1, C 2,..., CM }
xij 1 i M 1 j N
N
x
i 1
ij 1
M
U x
j 1
i ij
Cj
BWj
for j= 1,…, M
for i= 1,…,N
11/09/2007 Ph.D. Defense 13
Algorithm
Compute Received Signal Strength Indicator (RSSI)
at each user.
Generate a binary matrix that assigns “1” if a user’s
RSSI exceeds the threshold value or “0” otherwise.
Invoke LINGO to solve the NLIP.
Identify the MCAP and compute .
Decrement its transmitted power by 1 dBm.
Repeat previous steps until one or more user can no
longer associate with an AP or their data rate can no
longer be accommodated.
Observe the power levels at each AP and the best
user’s association at the best .
11/09/2007 Ph.D. Defense 14
Numerical Analysis and Results
User-AP candidate association
User Number AP1 AP2 AP3 AP4 Receiver Sensitivity at the user is
1 1 1 1 0 -90 dBm
2 0 0 1 0 Transmitted Power at each AP is
3 0 0 0 1
20 dbm
4
4 0 0 0 1
5 1 1 1 1 1
6 1 1 0 0
2
7 0 1 0 0
8 0 0 1 1
9 0 0 1 0
10 0 1 0 1
11/09/2007 Ph.D. Defense 15
Numerical Analysis and Results
(Cont’d)
Data rate of users
Traffic is randomly generated between 1 Mbps
User Number Traffic (Kbps) and 6 Mbps for each user
1 1752 Service Area Map
2 5698
3 4265
4 1994
5 3558
6 3176
7 5319
8 1559
9 2982
10 2263
11/09/2007 Ph.D. Defense 16
Numerical Analysis and Results
Optimal user-AP association
User Number AP1 AP2 AP3 AP4
Each user is associated to
1 0 1 0 0
one and ONLY one AP.
2 0 0 1 0
3 0 0 0 1
4 0 0 0 1
5 0 1 0 0 1
6 1 0 0 0
7 0 1 0 0 1
8 0 0 0 1
9 0 0 1 0
10 0 1 0 0 1
11/09/2007 Ph.D. Defense 17
Numerical Analysis and Results
(Cont’d)
Congestion Factor comparison
Initial Congestion factor: Congestion factor solution Congestion factor with
(No Power Mgmt) according to [2] Power Mgmt
AP1 0.6319 0.5234 0.3793
AP2 0.4100 0.4100 0.3617
AP3 0.2117 0.2117 0.3167
AP4 0.2026 0.3110 0.3985
81.15% 90.84% 99.31%
Load is distributed fairly among APs.
Final transmitted power levels at each AP is: 12 dBm, 18
dBm, 20 dBm and 17 dBm, respectively.
11/09/2007 Ph.D. Defense 18
Numerical Analysis and Results
(Cont’d)
Service area map after Power Mgmt
Different radii sizes
after power
adjustment
Users do NOT always
associate to the closest
AP.
11/09/2007 Ph.D. Defense 19
Numerical Analysis and Results
4 APs (Cont’d) 9 APs
16 APs
*published at IEEE Sarnoff Conference
11/09/2007 May'07 20
Simulation Scenarios (OPNET)
Unbalanced Load v.s.
Balanced Load
20 dBm Transmitted power
-90 dBm Receiver threshold
FTP clients and APs
are stationary
File of 50 Kbytes uploaded
continuously.
Simulation time is 40 mins
Steady state after 15 mins
WLAN scenario in OPNET, 4 APs and 20 Users
11/09/2007 *Not published yet 21
Simulation Results (OPNET)
Overall load on the network
was reduced by “load
balancing” Reduced overall
congestion
Overall load at the network
After applying load balancing,
client 9 associated with BSS2,
and improved its throughput.
Throughput of FTP client 9
11/09/2007 Ph.D. Defense 22
2nd Problem
Channel Assignment
Careful consideration must be given to
assigning channels to APs. Otherwise the
followings may result:
High interference between APs’ overlapping zones.
Users in the overlapping region of two or more
interfering APs will suffer:
Delay This is due to the huge increased requests by the user
Low data rates in retransmitting damaged/unsuccessful packets.
11/09/2007 Ph.D. Defense 23
Proposed Solution
Two folds:
Assign channels at the design stage (no
users) with the objective to minimize the total
sum of interference between neighboring APs.
Re-Assign channels when users exist on the
network.
11/09/2007 Ph.D. Defense 24
Problem Formulation (initial stage)
Objective
M M
min WSUM Iij WMAX max i , j {Iij} for each i j
i 1 j 1
wijPj
Subject to Iij
PL( dij )
1 |Chi Chj| 0.2, for wij 0
where wij =
0 otherwise
i = 1, …, M Chj , Chk {1,.., K }
j = 1,…, M
ij K {1,..,11}
11/09/2007 *Formulation not yet published 25
Problem Formulation (with users)
Objective
N M
Max
i 1 j 1
SIRij (k )
M
Subject to Iij ( Pij wjk ), j k
j 1
1 |Chi Chj| 0.2, for wij 0
where wij =
0 otherwise
Pik
SIRij (k ) i, j
Iij
i {1,..N} Chj , Chk {1,.., K }
j, k {1,.., M } K {1,..,11}
11/09/2007 Ph.D. Defense 26
Heuristic Algorithm
Apply initial channel assignment
Users enter the network
Apply load balancing algorithm based on
power management.
Save final transmitted powers at APs.
Re-compute received signal at users.
Compute SIR.
Apply Channel Assignment algorithm based on
SIR.
11/09/2007 Ph.D. Defense 27
Numerical Analysis and Results
Initial Approach (based
on min AP interference)
Scenario 1: 4 APs (12, 18, 20, 17 (dBm))
AP2 AP3
FCA: Power
AP Number FCA: Equal Power
Management
AP1 1 7
AP2 8 1
4% AP1 AP4
AP3 3 11
AP4 11 3
Interference (dB) -21.17 -22.02
4 APs
Scenario2: 6 APS (16, 16, 11, 6, 6, 1 (dBm))
FCA: Power AP2 AP3 AP6
AP Number FCA: Equal Power
Management
AP1 11 11
AP2 1 1
AP3 8 7 AP1 AP4 AP5
AP4 4 5
AP5 11 2
AP6 1 10
Interference (dB) -19.15 -25.49 6 APs
11/09/2007 33% Ph.D. Defense 28
Numerical Analysis and Results
Initial Approach (Cont’d)
Scenario 3: 9 APs (4, 12, 20, 16, 20, 16, 17, 8, 19 (dBm))
FCA: Power
AP Number FCA: Equal Power
Management
AP7 AP8 AP9
AP1 11 11
AP2 4 1
AP3 8 6
AP4 1 1 AP2 AP3 AP6
AP5 11 10
AP6 4 1
AP7 11 11
AP1 AP4 AP5
AP8 1 1
AP9 11 11
Interference (dB) -17 -19.86
9 APs
17%
* Published at IEEE ICSPC conference Nov’07
* Published at IEEE PIMRC conference
11/09/2007 Jun'07 29
Numerical Analysis and Results
Second Approach (based on max SIR
at users)
Two special cases:
Many users in the overlapping zone
Users are not in the overlapping zone
11/09/2007 Ph.D. Defense 30
Numerical Analysis and Results
Scenario 1: 4 APs (12, 18, 20, 17 (dBm))
Scenario2: 6 APS (16, 16, 11, 6, 6, 1 (dBm))
FCA: No Users FCA: With Users
AP Number (minimize interference (Maximize SIR at the AP Number FCA: No users FCA: with users
between APs) Users)
AP1 11 2
AP1 1 6
AP2 1 11
AP2 8 11
AP3 8 6
AP3 3 2
AP4 4 6
AP4 11 1
Avg. SIR (dB) 6.51 7.66 AP5 11 8
AP6 1 1
AP Number FCA: No users FCA: With Users
17% Avg. SIR (dB) 4.22 4.47
AP1 11 6
AP2 4 1
AP3 8 11
6%
AP4 1 8
AP5 11 11
Scenario 3: 9 APs (4, 12, 20, 16, 20, 16, 17, 8, 19 (dBm))
AP6 4 4
AP7 11 6
AP8 1 8
540%
AP9 11 11
Avg. SIR (dB) 0.44 2.86
*Submitted to IEEE WCNC conference
11/09/2007 Apr'08 31
Simulation Scenarios (OPNET)
4-AP WLAN
Summary of the 4 Scenarios
Scenario Scenario Scenario Scenario
1 2 3 4
AP1 1 1 1 6
AP2 2 6 8 11
AP3 3 1 3 2
AP4 4 11 11 1
4-AP WLAN
11/09/2007 Ph.D. Defense 32
Simulation Results (OPNET)
Same assumptions from the load balancing scenarios
apply EXCEPT for the channel assignment.
Overall Throughput Overall Upload Response Time Zoomed in View
11/09/2007 *Results not yet published 33
Dynamic Model
Background
No such application of a dynamic user behavior model
on a full scale dynamic network.
Published work related to user behavior reported the
user behavior through monitoring network traffic and
behavior for long periods of time (10 months or
more).
Such a model is significant for future researchers in
the WLAN field or industry where load distribution
and channel assignment algorithms can be
implemented and tested on a dynamic scale .
11/09/2007 Ph.D. Defense 34
Dynamic Scenario 1
Scenario 1: Varying data rate with time
4 APs and 20 users.
Data rate of users vary with time according to a normal
distribution (= 4 Mbps, = 2 Mbps).
Data rate is captured every 5 minutes.
All users are continuously active.
All APs and users are stationary.
Default AP transmitted power is 20 dBm.
Receiver’s threshold is -90 dBm.
Simulation period is 2 hours.
11/09/2007 Ph.D. Defense 35
Numerical Analysis and Results
Iteration 1
Final
transmitted
Initial CF Final CF Final FCA
power
(dBm)
AP1 0.2563 0.2464 16 1
AP2 0.0669 0.2721 20 6
AP3 0.3752 0.2502 12 11
AP4 0.3445 0.2741 11 11
82.49% 99.77%
Initial user-AP association
Last iteration
Final
transmitted
Initial CF Final CF Final FCA
power
(dBm)
AP1 0.2454 0.3023 20 1
AP2 0.1275 0.3979 16 6
AP3 0.7240 0.3968 5 6
AP4 0.3703 0.3703 18 11
72.94% 98.89%
Final user-AP association
11/09/2007 *Results not yet published 36
Dynamic Scenario 2
Scenario 2: Dynamic User Behavior
Same assumptions as before apply EXCEPT that the
data rate now is fixed over simulation time.
Users arrive to the WLAN according to a Poisson
distribution with an arrival rate of .
varies with time. However, in this scenario has a
constant value over the simulation period (2 hours).
e n
Pr(n)
n!
11/09/2007 Ph.D. Defense 37
Dynamic Scenario 2 (Cont’d)
Session lengths of each user is
characterized by a Bi-Pareto distribution.
When a user’s session is over, the user is
assumed as either no longer active or left
the network.
i.e. the user no longer has a data rate it does not
constitute any load at its AP.
P( x) k (1 c) x ( 1) ( x kc) 1 ( x kc), x k
11/09/2007 Ph.D. Defense 38
Numerical Analysis and Results
= 4 Departure 26 1.42
User Number Arrival times(hrs)
Time(hrs)
21 0.10 10 1.62
22 0.15 27 1.66
Arrival and Departure time Table
23 0.70 3 1.93
24 0.76 28 1.87
25 1.13 29 2.00
4 APs, 20 Users
11/09/2007 Ph.D. Defense 39
Numerical Analysis and Results
(Cont’d)
FCA and Load Balancing results
Final Tx Final Tx Final Tx Final Tx
FCA: FCA: FCA: FCA: Power Power Power Power
Arrive 4 Arrive 5 Arrive 6 Leave 1 (dBm): (dBm: (dBm): (dBm):
Arrive 4 Arrive 5 Arrive 6 Leave 1
AP1 1 1 1 1 17 20 15 18
AP2 6 6 6 6 11 14 19 18
AP3 11 11 11 11 12 5 19 12
AP4 6 6 1 1 15 13 13 13
98.75% 99.14% 96.89% 99.62%
Avg. SIR
6.46 6.27 6.08 6.05
(dB)
11/09/2007 Ph.D. Defense 40
Numerical Analysis and Results
(Cont’d)
FCA and Load Balancing results
Final Tx Final Tx Final Tx Final Tx
FCA: FCA: FCA: FCA: Power Power Power Power
Arrive 7 Leave 2 Arrive 8 Arrive 9 (dBm): (dBm): (dBm): (dBm):
Arrive 7 Leave 2 Arrive 8 Arrive 9
AP1 1 1 1 1 20 19 20 18
AP2 6 6 6 6 18 17 15 18
AP3 11 11 11 11 8 15 16 15
AP4 6 6 1 1 10 18 11 9
99.01% 97.69% 98.92% 99.42%
Total SIR
5.92 5.94 5.77 5.71
(dB)
11/09/2007 *Results not yet published 41
Numerical Analysis and Results
(Cont’d)
FCA and Load Balancing results
-- Added users
-- Removed users
-- Existing users
11/09/2007 Ph.D. Defense 42
Conclusion
A new load balancing algorithm based on power
management was developed.
A new channel assignment algorithm based on
maximizing SIR was developed.
Results were validated using OPNET simulation
to show the effectiveness of the developed
algorithms.
Dynamic data rate and user behavior were
introduced to verify the ability of the developed
models to adapt to these dynamic behaviors.
11/09/2007 Ph.D. Defense 43
Future Work
Extension of the dynamic model to
combine both variable data rate and users’
behavior.
Application of this work to WiMAX (IEEE
802.16).
Integration of smart antenna technology
at the AP.
Expand developed work to larger WLANs.
11/09/2007 Ph.D. Defense 44
Special Thanks
Ph.D. Advising Committee:
Dr.. Hussain Al-Rizzo Network Administrator
(Advisor) Greg Browning
Dr. Robert Akl
Dr. Yupo Chan
Dr. Hassan El-Salloukh
Dr. Seshadri Mohan OPNET Technical Support
Dr. Haydar Alshukri LINGO Technical Support
Ph.D. Candidates
Rami Adada
Rabindra Ghimire
Graduate Student
TJ Calvin
11/09/2007 Ph.D. Defense 45