Association Control based

Load Balancing in Wireless

Cellular Networks using
Preamble Sequences

by

Ankit Chopra

A thesis
submitted to the Victoria University of Wellington
in fulfilment of the
requirements for the degree of
Master of Engineering
in Network Engineering.

Victoria University of Wellington

2012
 Abstract
The efficient allocation and use of radio resources is crucial for achieving
the maximum possible throughput and capacity in wireless networks. The
conventional strongest signal-based user association in cellular networks
generally considers only the strength of the signal while selecting a BS,
and ignores the level of congestion or load at it. As a consequence, some
BSs tend to suffer from heavy load, while their adjacent BSs may carry
only light load. This load imbalance severely hampers the network from
fully utilizing the network capacity and providing fair services to users.
In this thesis, we investigate the applicability of the preamble code se-
quence, which is mainly used for cell identification, as an implicit infor-
mation indicator for load balancing in cellular networks. By exploiting the
high auto-correlation and low cross-correlation property among preamble
sequences, we propose distributed load balancing schemes that implicitly
obtain information about the load status of BSs, for intelligent association
control. This enables the new users to be attached to BSs with relatively
low load in the long term, alleviating the problem of non-uniform user
distribution and load imbalance across the network.
Extensive simulations are performed with various user densities con-
sidering throughput fair and resource fair, as the resource allocation poli-
cies in each cell. It is observed that significant improvement in minimum
throughput and fair user distribution is achieved by employing our pro-
posed schemes, and preamble sequences can be effectively used as a lever-
age for better cell-site selection from the viewpoint of fairness provision-
ing. The load of the entire system is also observed to be balanced, which
consequently enhances the capacity of the network, as evidenced by the
simulation results.
ii
Acknowledgments

I would like to express my deepest gratitude to my supervisor Dr. Winston

K.G. Seah for his unrelenting motivation and criticism, without which, this
work would not have been possible. I also thank Peter Sam Raj for his help
with simulations and very thoughtful suggestions.

iii
iv
Contents

1 Introduction 1
1.1 Cellular Wireless Networks . . . . . . . . . . . . . . . . . . . 1
1.1.1 Challenges in Cellular Networks . . . . . . . . . . . . 5
1.1.2 Issue of Non-Uniform User Distribution . . . . . . . . 6
1.2 Thesis Objective . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3 Thesis Contributions . . . . . . . . . . . . . . . . . . . . . . . 8
1.4 Thesis Organization . . . . . . . . . . . . . . . . . . . . . . . . 9

2 Related Work 11
2.1 Load Balancing in Cellular Networks . . . . . . . . . . . . . . 11
2.1.1 Channel Assignement Schemes . . . . . . . . . . . . . 11
2.1.2 Dynamic Load Balancing Schemes . . . . . . . . . . . 12
2.1.3 Schemes based on Association Control . . . . . . . . 14
2.1.4 Relay Assisted Traffic Transfer Schemes . . . . . . . . 15
2.2 Load Balancing in Wireless Local Area Networks . . . . . . . 17
2.3 Recent Research Work . . . . . . . . . . . . . . . . . . . . . . 18
2.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

3 System Model 21
3.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.1.1 Preamble Sequences . . . . . . . . . . . . . . . . . . . 21
3.1.2 Fairness in Wireless Networks . . . . . . . . . . . . . 23
3.2 Utilizing Preamble Sequences for Load Balancing . . . . . . 25

v
vi CONTENTS

3.2.1 Load Balancing parameter . . . . . . . . . . . . . . . . 27

3.3 System Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

4 Auto-Correlation based Implicit Load Indication Scheme 31

4.1 The Auto-Correlation Property . . . . . . . . . . . . . . . . . 31
4.2 Preamble Modification Algorithm . . . . . . . . . . . . . . . 34
4.3 MS Cell Decision using AutoCorrelation . . . . . . . . . . . . 35
4.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

5 Cross-Correlation based Implicit Load Indication Scheme 39

5.1 The Cross-Correlation Property . . . . . . . . . . . . . . . . . 39
5.2 Preamble Allocation using Cross-correlation . . . . . . . . . 41
5.3 BS Selection Algorithm using Implicit Indication . . . . . . . 42
5.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

6 Simulation Results and Discussion 45

6.1 Simulation Setup . . . . . . . . . . . . . . . . . . . . . . . . . 45
6.1.1 Channel Characteristics . . . . . . . . . . . . . . . . . 46
6.1.2 Simulation Environment . . . . . . . . . . . . . . . . . 46
6.2 Simulation Results and Discussion . . . . . . . . . . . . . . . 47
6.2.1 Load Variance Test . . . . . . . . . . . . . . . . . . . . 49
6.2.2 Resource Fairness Test . . . . . . . . . . . . . . . . . . 51
6.2.3 Spectral Efficiency Test . . . . . . . . . . . . . . . . . . 52
6.2.4 Evaluation using Jain’s Fairness Index . . . . . . . . . 55
6.3 Performance under Different Environments . . . . . . . . . . 57
6.3.1 Hata Model for a Small City . . . . . . . . . . . . . . . 58
6.3.2 Hata Model for a Large City . . . . . . . . . . . . . . . 61
6.4 Comparsion against RSSI-based Schemes . . . . . . . . . . . 65
6.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

7 Conclusion 69
CONTENTS vii

8 Future Work 71
viii CONTENTS
List of Figures

1.1 System architecture of a conventional cellular system . . . . 2

1.2 Frequency reuse . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2.1 Channel borrowing without locking (CBWL). . . . . . . . . . 13

2.2 Cell breathing (decrease in coverage area with increase in
subscribers) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

3.1 A sample pseudo-noise sequence . . . . . . . . . . . . . . . . 21

3.2 M-Sequence generator . . . . . . . . . . . . . . . . . . . . . . 23
3.3 At central controller the metric is the number of peaks in the
auto-correlation plot. . . . . . . . . . . . . . . . . . . . . . . . 26
3.4 At central controller the metric is the value of cross-correlation
with particular sequence. . . . . . . . . . . . . . . . . . . . . . 26
3.5 A two cell model showing the concept of ”r”. . . . . . . . . . 27
3.6 Architecture of the Korean WiMax Broadband [30] . . . . . . 28

4.1 Auto-correlation function of the sinewave . . . . . . . . . . . 32

4.2 Auto-correlation plot of a preamble of length 284 bits made
with 4 repeated sequences of length 71. . . . . . . . . . . . . 33

5.1 Two time series x,y. . . . . . . . . . . . . . . . . . . . . . . . . 40

5.2 The cross-correlation series with a maximum delay of 4000
showing a strong correlation at a delay of about 40. . . . . . 41

6.1 A conventional 12-cell model with focus on cells 0,1 and 2 . 47

ix
x LIST OF FIGURES

6.2 Distribution of users in cells at r = 10 . . . . . . . . . . . . . . 48

6.3 Load Variance test with 100 users in cell 0, 80 users in cell 1,
and 20 users in cell 3. . . . . . . . . . . . . . . . . . . . . . . . 50
6.4 Load Variance test with 60 users in cell 0, 40 users in cell 1,
and 10 users in cell 2. . . . . . . . . . . . . . . . . . . . . . . 50
6.5 Resource Fairness with initially 100 users in cell 0, 80 in cell
1 and 20 users in cell 2. . . . . . . . . . . . . . . . . . . . . . . 51
6.6 Resource Fairness with initially 50 users in cell 0, 40 in cell
1 and 10 users in cell 2. . . . . . . . . . . . . . . . . . . . . . . 52
6.7 Throughput Fairness with initially 100 users in cell 0, 80 in
cell 1 and 20 users in cell 2. . . . . . . . . . . . . . . . . . . . . 53
6.8 Throughput Fairness with initially 50 users in cell 0, 40 in
cell 1 and 10 users in cell 2. . . . . . . . . . . . . . . . . . . . . 54
6.9 Load Balancing Index vs r . . . . . . . . . . . . . . . . . . . . 56
6.10 Jain’s Fairness Index vs r . . . . . . . . . . . . . . . . . . . . . 56
6.11 Throughout Fair vs r for a small city . . . . . . . . . . . . . . 59
6.12 Resource Fair vs r for a small city . . . . . . . . . . . . . . . . 59
6.13 Load Variance vs r for a small city . . . . . . . . . . . . . . . 60
6.14 Load Balance Index vs r for a small city . . . . . . . . . . . . 61
6.15 Throughout Fair vs r for a large city . . . . . . . . . . . . . . 62
6.16 Resource Fair vs r for a large city . . . . . . . . . . . . . . . . 63
6.17 Load Variance vs r for a large city . . . . . . . . . . . . . . . . 64
6.18 Load Balance Index vs r for a large city . . . . . . . . . . . . 65
6.19 Minimum Throughput gain vs r . . . . . . . . . . . . . . . . 66
6.20 Jain’s Fairness Index vs r . . . . . . . . . . . . . . . . . . . . . 66
List of Tables

4.1 Preamble Modification Algorithm (PMA) . . . . . . . . . . . 35

4.2 MSCD Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . 36

6.1 Channel parameters for simulation . . . . . . . . . . . . . . . 48

6.2 Standard coverage parameters for the Hata model . . . . . . 57

xi
xii LIST OF TABLES
List of Abbreviations

AP Access Point
AMC Adaptive Modulation and Coding
ASN Access Network
Autocorr Autocorrelation
BS Base Station
CAZAC Constant Amplitude Zero Autocorrelation Waveform
CDMA Code Division Multiple Access
CBWL Channel Borrowing Without Locking
CBWLCR CBWL with Channel Rearragement
CBWLnR CBWL without Channel Arrangement
CSN Connectivity Service Network
DCA Dynamic Channel Assignment
FCA Fixed Channel Assignment
FDMA Frequency Division Multiple Access
FFT Fast Fourier Transform
HCA Hybrid Channel Assignment
LBSB Load Balancing with Selective Borrowing
LFSR Linear Feedback Shift Register
MACA Mobile Assisted Call Admission
MCN Multihop Cellular Networks
MLS Maximum Length Sequences
MS Mobile Station
MSCD MS Cell Decision

xiii
xiv LIST OF TABLES

MTSO Mobile Telephone Switching Office

PAC Preamble Allocation using Cross-correlation
PMA Preamble Modification Algorithm
PN Pseudo Noise
PSTN Public Switched Telephone Network
QOS Quality of Service
RAS Radio Access Stations
RF Resource Fair
RSSI Received Signal Strength Indicator
SNR Signal to Noise Ratio
TF Throughput Fair
UCAN Unified Cellular and Ad hoc Network
WiBro Wireless Broadband
WLANSs Wireless Local Area Networks
WiMAX Worldwide Interoperability for Microwave Access
WSM WiMAX System Manager
Chapter 1

Introduction

The popularity of wireless communication has made great gains over the
past two decades as witnessed by the widespread adoption of Wireless
Local Area Networks (WLANSs), cellular networks, and worldwide in-
teroperability for microwave access (WiMAX) [3]. Users of these wireless
access networks expect the highest quality, reliability, and ease of access
to high-speed services. In terms of mobile communications, cellular net-
works have been a dominant player in the recent years and stands as the
focus of our research in this thesis.

1.1 Cellular Wireless Networks

A cellular network provides mobile stations (MSs) or cell phones, with
wireless access to the public switched telephone network (PSTN). The ser-
vice area of a cellular network is divided into many smaller areas, called
cells, each of which is served by a base station (BS) [1]. The BS is fixed
and is connected to the mobile telephone switching office (MTSO), also
known as the mobile switching center. An MTSO is in charge of a cluster
of BSs and it is, in turn, connected to the PSTN. With the wireless link be-
tween the BS and MS, MSs such as cell phones are able to communicate
with wireline phones in the PSTN. Both BSs and MSs are equipped with a

1
2 CHAPTER 1. INTRODUCTION

transceiver. Figure 1 illustrates a typical cellular network, in which a cell

is represented by a hexagon and a BS is represented by a triangle.

Figure 1.1: System architecture of a conventional cellular system

The success of todays cellular network is primarily due to the frequency

reuse concept. The frequency spectrum allocated for cellular communica-
tions is very limited, however. Each BS (or cell) is assigned a group of fre-
quency bands or channels. For the purpose of avoiding radio co-channel
interference, the group of channels assigned to one cell must be different
from the group of channels assigned to its neighboring cells. However,
the same group of channels can be assigned to the two cells that are far
enough apart such that the radio cochannel interference between them is
within a tolerable limit.
Typically, seven neighboring cells are grouped together to form a clus-
ter, as shown in Figure 1.2. The total available channels are divided into
seven groups, each of which is assigned to a cell. In Figure 1.2, the cells
marked with the same alphabet have the same group of channels assigned
to them. Furthermore, the cells marked with different numbers must be
assigned different groups of channels.
1.1. CELLULAR WIRELESS NETWORKS 3

Figure 1.2: Frequency reuse

The channels assigned to a cell are generally used either for voice or for
control. A voice channel is used for an actual conversation, whereas a
control channel is used for setting up conversations. Both voice and con-
trol channels are further divided into downlink and uplink (forward or
reverse). A forward channel is used to carry traffic from the BS to the MS,
and a reverse channel is used to carry traffic from the MS to the BS. The
channels assigned to a cell are shared by MSs located in the cell.
Every MS in the network has a home, which is the MTSO where the
mobile user originally subsribed for wireless services and this location is
stored in it. If an MS moves out of the home MTSO area, it is roaming. A
roaming MS is required to register for services in the MTSO visited. An MS
needs to be authenticated first before it can get any service by the network.
The process of making a call is as follows. The MS first requests for service
using a reverse control channel in the current cell. The request then needs
to be granted, by the MTSO, a pair of voice channels assigned for the call,
4 CHAPTER 1. INTRODUCTION

one for receiving and one for transmitting. Making a call to an MS is a little
more complex, however. The call is first routed to the home MTSO or the
visited MTSO if it is roaming. The MTSO needs to know the cell in which
the MS is currently located. Here, the concept of location management
comes in to locate the residing cell of an MS. Once the MTSO knows the
residing cell of the MS, a pair of voice channels assigned in the cell for the
call.
If a call is in progress when the MS moves into a neighboring cell, the MS
needs to get a new pair of voice channels from the BS to the neighboring
cell so the call can be continued. This process is known as handoff.

Cellular networks offer a number of advantages over other solutions :

• Increased Capacity : The increased capacity in a cellular network,

compared with a network with a single transmitter, comes from the
fact that the same radio frequency can be re-used in a different area
for a completely different transmission, which is one of the key char-
acteristics of it.

• Reduced Interf erence f rom Other Signals : In order to distinguish

signals from several different transmitters, frequency division multi-
ple access (FDMA) and code division multiple access (CDMA) were
developed. In FDMA, the transmitting and receiving frequencies
used in each cell are different from the frequencies used in each
neighbouring cell. In CDMA, spread spectrum techniques are used
by which a signal generated in a particular bandwidth is deliberately
spread in the frequency domain, resulting in a signal with a wider
bandwidth.

• Larger Coverage Area : The concept of dividing the land area into
cells and strategically placing base stations in them, results in a larger
coverage area as compared to other alternate solutions and is a sig-
nificant feature of cellular networks.
1.1. CELLULAR WIRELESS NETWORKS 5

• Reduced Cost : Although the cost of deploying the BSs and setting up
the network is substantial, taking into account the advantages that
cellular networks offer over alternate solutions, the cost of operation
is much less considering its use in the long term.

1.1.1 Challenges in Cellular Networks

Although cellular networks present many benefits, there are some inher-
ent challenges and limitations on coverage and cell capacity. In addition
to this, they also suffer from the problem of dead spots and then there is
the problem of limited capacity which gives rise to the hot spot problem
as described below. Lastly, there is the issue of radio resource utilization,
which is the focus of our research in this thesis.
We discuss these limitations and issues as follows before narrowing down
on the problem of non-uniform user distribution that causes a severe im-
balance of load in the system.

• The dead spot problem : Although, mobile users are within the com-
munication range of the BS, there are still some areas where coverage
is not available at all. These areas are often referred to as dead spots
such as indoor environments and underground areas like basements.

• Limited capacity : In cellular networks, the capacity of a cell is lim-

ited by the number of channels allocated to the cell. The larger the
number of channels, the greater the number of users that can be
served. The number of channels is limited by the available frequency
spectrums and by the frequency reuse factor [3]. A smaller cell size
allows higher frequency reuse and, thus, a higher capacity can be
achieved. The cell capacity is not only limited by the available fre-
quency spectrums, but also by the interference among mobile nodes
and BSs. The higher the interference, the lower the cell capacity is.

• The problem of hot spots : Due to the limited capacity, mobile users
6 CHAPTER 1. INTRODUCTION

tend to experience higher call blocking, i.e., call requests are denied,
in dense areas known as hot spots, such as downtown areas and
amusement parks . This is because of the fact that, in hot spots, there
are more mobile users than the number of available channels.

• The issue of radio resource utilization : The hot spot problem in

turn raises the issue of radio resource utilization. While there are not
enough channels or capacity in a hot spot for serving mobile users,
the cells neighbouring the hot spot may still have available channels,
leaving system resources under-utilized.

• Limitations on coverage : the coverage of cells is limited by the com-

munication range or transmission power of the BS. Mobile users,
which are outside the coverage of the BSs, are not able to access the
networks due to this.

1.1.2 Issue of Non-Uniform User Distribution

One of the major performance degradation factors in cellular networks,
which is the focus of our research in this thesis, is the issue of non-uniform
user distribution due to received signal strength indicator (RSSI) based
user association [3]. In cellular networks, a mobile station (MS) generally
scans and associates itself with the base station (BS), which yields the max-
imal RSSI obtained by reading the power of the preamble signal, while be-
ing oblivious to the load of the BS. As the distribution of MSs are, typically,
not even, some BSs tend to suffer from heavy load while the BSs adjacent
to them may carry only light load, leading to under-utilized resources in
the system.
In this kind of scenario, the boundary or cell-edge users in the net-
work suffer the most from low throughput due to this uneven distribution
which severely hampers the performance of the network. In order to al-
leviate this problem, new schemes need to be investigated that take into
1.2. THESIS OBJECTIVE 7

account the load of BSs, in addition to the conventional association met-

rics.

1.2 Thesis Objective

The main objective of this thesis is to explore the applicability of the pream-
ble code sequence, which is used mainly for cell identification, as an im-
plicit load indicator, towards the goal of designing distributed load balanc-
ing schemes to tackle the problem of imbalanced load due to non-uniform
user distribution in cellular networks.
The motivation to explore and use preamble sequences comes from the
fact that, with the growth of wireless subscribers, the service providers
now face the challenge of how to maximize their network capacity with
their existing infrastructure, and preamble sequences are already employed
by almost every network for the pupose of cell identification, synchroniza-
tion or channel measurement, thereby justifying the basis for our work.
Following are the goals that we aim to acheive that can be classified as
sub-objectives :

• Minimize the traffic load variances to achieve load balancing : The main
goal, as stated forehand, is to develop schemes that consider not only
the conventional metric RSSI for user association, but also the traffic
load or congestion at the BS, allowing users to make a more intel-
ligent decision on which BS to associate with. This would result in
minimizing the variation in traffic load to achieve a well-balanced
network in terms of load.

• Maximize overall network throughput and capacity : Once we are able to

achieve the goal of load balancing, we should be able to witness an
increase in the overall network and capacity as a result. In addition
to this, we aim to develop ways to focus on the boundary or cell-edge
users, which are the major targets of every load balancing scheme,
8 CHAPTER 1. INTRODUCTION

to effectively reduce the deterioration in cell edge throughput, by

getting more boundary users to participate in load balancing.

• Maximize Jain’s fairness to achieve fairness among the mobile users : Fair-
ness is a major criteria in evaluating how well the resources are be-
ing shared among the users equally. Therefore, our schemes should
yield good results under popular fairness measures such as max-min
fairness and proportional fairness and evaluation indexes like Jain’s
fairness index.

• Maximize robustness to perform well under different environments : Our

schemes should be able to perform well under different channel con-
ditions and be robust.

1.3 Thesis Contributions

In this thesis, we propose the following schemes to address the problem of
non-uniform user distribution due to conventional RSSI-based association
and confirm the applicability of the preamble code sequence as an implicit
load indication tool for load balancing in cellular networks.

• Auto-correlation based implicit load indication scheme [4] modifies

the preamble code sequence and proposes the auto-correlation oper-
ation as a way to decipher the load at a BS. This scheme enables the
MSs to make an intelligent decision while choosing a BS for service,
by not only considering the RSSI, but the level of load at the BS.

• Cross-correlation based implicit load indication scheme [4] expolits

the cross-correlation property among preamble sequences to implic-
itly obtain the load state of BSs for load balancing. This enables the
users while joining the network to perform smart cell-site selection
based on the level of congestion at the BS along with the conven-
tional association metrics.
1.4. THESIS ORGANIZATION 9

A part of this work titled ” Utilizing the Inherent Properties of Preamble

Sequences for Load Balancing in Cellular Networks” has been accepted by the
7th International Conference on Intelligent Sensors, Sensor Networks and
Information Processing (ISSNIP 2011), Adelaide, Australia, Dec 6-9, 2011.

1.4 Thesis Organization

• Chapter 2 - Related Work gives a brief overview of the various load
balancing philosphies that exist in the literature along with the recent
research work on load balancing using preamble sequences.

• Chapter 3 - System Model describes the employment of the IEEE 802.16e

WiBRo system for the purpose of illustration and gives a brief de-
scription of our concept of utilizing preamble sequences for load bal-
ancing including the defintions.

• Chapter 4 - Auto-Correlation based Implicit Load Indication Scheme de-

scribes the design and implementation of the auto-correlation based
implicit load indication scheme.

• Chapter 5 - Cross-Correlation based Implicit Load Indication Scheme de-

scribes the design and implementation of the cross-correlation based
implicit load indication scheme.

• Chapter 6 - Simulation Results and Discussion presents the results ob-

tained through simulations with a breif discussion on the achieve-
ments.

• Chapter 7 - Conclusion and Future Work talks about a breif summary

of what we have acheived in our research and their potential along
with the road ahead.
10 CHAPTER 1. INTRODUCTION
Chapter 2

Related Work

2.1 Load Balancing in Cellular Networks

Traditionally, load balancing has been widely studied for the efficient uti-
lization of the network resources. By doing so, users can enjoy better
throughput while the operator may support more subscribers by utiliz-
ing the deployed network resources more efficiently. In this section, we
provide an overview of the load balancing and capacity maximization
schemes that exist in the literature.

2.1.1 Channel Assignement Schemes

There are three main channel assignment strategies proposed in the litera-
ture [2]. The first one is fixed channel assignment (FCA) strategy, wherein
each cell is allocated a predetermined set of voice channels. If all the chan-
nels in a cell are occupied, any new calls or handover requests within the
cell will be rejected. In variants of the FCA strategy, a cell is allowed to
borrow channels from a neighboring cell if all of its channels are already
occupied. The mobile switching center (MSC) supervises such borrowing
procedures and ensures that borrowing of a channel will not affect any
of the calls already in progress in the cell which is lending. Simplicity is

11
12 CHAPTER 2. RELATED WORK

one of the main advantages of such a scheme along with frequency reuse
maximization, as opposed to the other schemes which have complex pro-
cedures [2].
The second one is the dynamic channel assignment (DCA) strategy, in
which voice channels are not permanently allocated to the cells [2][5][6][7].
Everytime a call request is made, the serving base station requests a chan-
nel from the MSC, instead. The switch then allocates a channel to the re-
questing cell following an algorithm that takes into account the likelyhood
of future call blocking within the cell, the frequency of use of the candi-
date, the reuse distance of the channel, and other cost functions. Although
DCA schemes improve the call blocking probability of the system sub-
stantially, for high traffic intensity, the computational load on the system
increases greatly.
The third type is hybrid channel assignment (HCA), which is a blend
of the concepts of fixed and dynamic channel assignement schemes. This
basically means, there are channels in a global pool for use in case there is
a shortage, in addition to the fixed set of channels assigned to each cell [8].
To conclude, although the objective of all the schemes is better utiliza-
tion of the available channels, causing reduced call blocking probability,
most of the schemes do not consider the non-uniformity of channel de-
mand, which is the variation of channel demand from time to time and/or
area to area.

2.1.2 Dynamic Load Balancing Schemes

The dynamic load balancing schemes proposed in the literature can be

broadly classified into two groups:
i) Strategies based on channel borrowing from cooler cells (lightly loaded)
such as simple borrowing [8]-[9], channel borrowing without locking (CBWL)
[10], load balancing with selective borrowing (LBSB) [11], etc.
ii) Strategies based on traffic transfer to cooler cells such as directed
2.1. LOAD BALANCING IN CELLULAR NETWORKS 13

retry [12]-[13], hierarchical macrocell overlay systems [14], etc.

Figure 2.1: Channel borrowing without locking (CBWL).

In simple borrowing scheme, every cell is assigned a fixed set of chan-

nels and a seperate set of channels that are allowed to be borrowed by
the neighboring congested cells [8]-[9]. When a channel is borrowed, the
co-channels within the reuse distance are locked to avoid co-channel in-
terference.
In CBWL, the use of borrowed channels with reduced transmission
power is proposed, when the set of available channels in a cell gets ex-
austed [10]. This is done to avoid interference with other co-channels. Tak-
ing advantage of transmission power control, however, to induce hand-
off’s into less loaded cells and reducing it, may decrease the link capacity
for others, who still remain in the cell.
There are two types of CBWL: i) CBWL without channel arrangement
(CBWLnR); and ii) CBWL with channel rearragement (CBWLCR). In CB-
WLnR, only the new call requests within the fraction of cellular area in
14 CHAPTER 2. RELATED WORK

which borrowed channels can be used (this area is limited due to reduced
transmission power over the borrowed channels) are granted service us-
ing a borrowed channel. However, in CBWLCR, even if the new call re-
quest is not in the channel-borrowing area, it can still get service if there is
at least one active user within the channel-borrowing area. In this case, the
active user will borrow a channel from a neighboring cell and start trans-
mission at a reduced power and the channel released by that user will be
assigned to the new user requesting service.
In selective channel borrowing [11], load balancing is achieved by a
structured borrowing mechanism where a hot cell can borrow a fixed num-
ber of channels only from adjacent cells in the next outer ring. In this way,
unused channels of the cold cells are moved to the hot spot area. This
mechanism reduces the amount of interference between the borrower cell
and the co-channel cells of the lender.
In directed retry scheme, the traffic load is shared in the overlap of
the neighbouring cells. In this, there is no concept of borrowing channels
from the neighboring cells, however, the subscribers who can communi-
cate with more than one base station are moved from the congested cell to
a neighboring cell [12].
In the hierarchical macrocell overlay system with microcells and over-
laid macrocells [14], high traffic areas are covered by microcells while over-
laying macrocells cover low-traffic areas and provide overflow groups of
channels for clusters of microcells. In other words, the excess traffic of
the microcells is served by the overlaying macrocells. As a consequence,
macrocell overlays inherently achieve dynamic load balancing by trans-
ferring the otherwise blocked microcell users to the macrocells.

2.1.3 Schemes based on Association Control

The most widely implemented association procedure uses the RSSI as a

metric for association, which causes non-uniform user distribution across
2.1. LOAD BALANCING IN CELLULAR NETWORKS 15

cells, as the load of the BS is not considered. This often causes the entire
network to be under-utilized.
To combat this, several studies [15]-[21] have proposed a variety of as-
sociation metrics instead of using the RSSI as the sole association criteria.
These metrics typically consider factors such as, the number of users cur-
rently associated with a BS, and the bandwidth that a new user will be
able to get if it is associated with a BS [15]-[16]. Balachandran et al. [17]
proposed to associate a user with the AP that can provide a minimal band-
width required by the user. If there exist many such APs, the one with the
strongest signal strength is selected.
Velayos et al. [18] introduced a distributed load balancing architecture
where the load of an AP is defined as the aggregated downlink and uplink
traffic through the AP. Kumar et al. [19] proposed an association selection
algorithm which is based on the concept of proportional fairness to bal-
ance between throughput and fairness. Most of these work heuristically
determine only the association of newly arrived users with some excep-
tions [20] [21] . Tsai and Lien [20] proposed to reassociate users when the
total load exceeds a certain threshold or the bandwidth allocated to users
drops below a certain threshold. An on-line scheme that periodically op-
timizes the user-AP association has also been proposed [21].

2.1.4 Relay Assisted Traffic Transfer Schemes

Recently, there has been a lot of interest in the usage of relays and relay-
ing techniques to expand coverage and capacity in cellular networks. Re-
searchers have looked at the problem of load balancing from the perspec-
tive of relaying and have proposed schemes to induce user handoff’s from
heavily loaded cells into adjacent cells that may be lighly loaded.
iCAR [35] and PARCelS [36] were the first two load balancing schemes
introduced for balancing the load among cells through relaying. In iCAR,
low cost limited mobility, ad-hoc relay stations (ARSs) are placed in hot
16 CHAPTER 2. RELATED WORK

spots for relaying traffic out of the hot spots. This strategy is still costly
and not flexible enough to handle the highly dynamic load situation in
3G networks. PARCelS uses mobile nodes for relaying. When a BS is
congested, mobile nodes search best routes to other non-congested cells.
Route information is forwarded to BSs for selection. This strategy requires
considerable routing overhead in mobile nodes and does not take advan-
tage of the presence of powerful BSs. In addition, both schemes do not
take into account the load balancing among relaying nodes which could
greatly affect the load balancing performance.
Some other schemes have also been introduced to increase the capacity
in cellular networks with the schemes based on the same idea and differ-
ing only in implementation.
Mobile Assisted Call Admission (MACA) [23], is a dynamic load balanc-
ing scheme proposed to improve call blocking probability performance
in cellular networks with the idea of forwarding the excess traffic of a
”hot” cell to its ”cooler” neighboring cells via mobile agents in the net-
work. For the purpose of forwarding the call, either relay channel can be
used, in-band (i.e., cellular band) channel or out-of-band (i.e., ISM-band
or any other band other than cellular band) channel. In the case of in-band
MACA, a portion of the fixed channels assigned to each cell is saved for
forwarding calls. Whereas, in out-of-band MACA, the mobile agents use
channels from a frequency band other than the cellular band such as the
ISM-band to forward calls.
In Multihop Cellular Networks (MCN) [22], the range of the BSs and mo-
bile devics are reduced and mobile users within the same cell are able to
communicate with each other over multi-hops, and on the other hand,
mobile users in different cells need to forward their traffic via BSs. The
objective in this scheme is to increase the capacity of the cellular network
by replacing single-hop communication with multi-hop between the BSs
and mobile users.
2.2. LOAD BALANCING IN WIRELESS LOCAL AREA NETWORKS 17

In Unified Cellular and Ad hoc Network (UCAN) [24], instead of improv-

ing the call blocking probability, the use of ad hoc wireless connection
to enhance a mobile user’s access to the 3G cellular infrastructure is pro-
posed. If the throughput is low (i.e., the direct wireless link from the BS to
MS is bad), a multi-hop link from the mobile user to the BS is formed via
other mobile users in the system that have better channel quality.

2.2 Load Balancing in Wireless Local Area Net-

works
Many of the schemes based on association contol described previously
also apply to WLANs, as there is not much fundamental difference when it
comes to applying load balancing between WLANs and celluar networks.
The only difference is in the channel allocation schemes which are not ap-
plicable to WLANs as each AP normally uses one channel and channel
allocation is fixed.
In WLANs, generally the concept of Cell Breathing [37]-[38], some-
times also referred as dynamic cell re-sizing, has been used for load bal-
ancing, in which the geographical area covered by the cell tower is con-
stantly changed by controlling the transmission power of an AP’s bea-
con packets. The heavily loaded cells shrink their coverage area, and the
lightly loaded cells expand their coverage area to attract clients previously
associated with the heavily loaded cells.
Both [26] and [27] take advantage of transmit power control to induce
the users handoff into less loaded cells. However, reducing the transmis-
sion power of BS may decrease the link capacity for other users, who still
remain in the cell with reduced transmit power. In order to deal with this
problem, S. Das et al. [29] proposed to control the power of common pilot
channel without changing the power allocated to data channel. However,
this is not practical considering the recent trend that the transmission rate
18 CHAPTER 2. RELATED WORK

Figure 2.2: Cell breathing (decrease in coverage area with increase in sub-
scribers)

is adapted via adaptive modulation and coding (AMC) according to the

channel condition, which is typically known from measuring the power
of common pilot channel.

2.3 Recent Research Work

In a recent study by Na, et al. [30], it has been shown that preamble se-
quence can be used as an implicit load indication tool. The work exploited
comparative high and low cross correlation among certain sets of pream-
ble sequences, to indicate similarity in terms of lightly loaded and heavily
loaded state of the BSs for cell selection. However, the proposed scheme
fails to address all the permutations of BS loads, and in situations where
certain PN sequences outside the set would yield high correlation with
sequences within it.
The scheme also does not consider the case of two heavily loaded and
one lightly loaded cell situation, and although this may not be a degra-
dation and the MS would select a BS based on the conventional received
signal strength (RSSI) association, it exposes the limitations of the scheme.
We, therefore, propose an alternative approach in our cross-correlation
based load balancing scheme to use preamble sequences in such a way
2.4. SUMMARY 19

that makes the implicit indication absolute as opposed to relative, thereby,

removing the inherent difficulties.

2.4 Summary
In this chapter, a brief overview of the all the load load balancing schemes
proposed for cellular networks, was presented. We discussed the various
channel assignment schemes which allocate the channels to cells using dif-
ferent strategies with the aim of maximizing the utilization of the available
channels for an increase in capacity and performance.
We talked about the dynamic load balancing schemes, wherein the ca-
pacity is increased via dynamic load/traffic transfer among cells. A brief
note on load balancing schemes based on association control is also pre-
sented, in which, several schemes have been proposed that use different
metrics to associate users with BSs in addition to the conventional RSSI-
based association.
Schemes based on relay assisted traffic transfer are also discussed which
use relays and relaying techniques to expand coverage and capacity. We
have also presented load balancing schemes designed for WLANs, where
the cell size is dynamically changed for load balancing, a phenomina called
as cell breathing, along with schemes relying on transmit power to induce
user handoffs into cells which are comparatively less loaded.
Finally, recent research work into load balancing using preamble sequences
was presented with its features and limitations. In the next chapter, we
present our system model, where we present our concept of load balanc-
ing using preamble sequences.
20 CHAPTER 2. RELATED WORK
Chapter 3

System Model

3.1 Definitions

3.1.1 Preamble Sequences

Preamble sequences, also known as pseudo-noise sequences (PN sequences),
are sequences that are deterministically generated and yet possess some
properties that one would expect to find in randomly generated sequences.
They are called PN sequences because of their low autocorrelation for pos-
itive delays. This makes them nearly random, even though they are deter-
ministic in nature.

Figure 3.1: A sample pseudo-noise sequence

They are useful for cell identification in cellular networks because of

their uniqueness, i.e. different PN sequences are highly uncorrelated with
each other, but the cross correlation of a PN sequence with an identi-

21
22 CHAPTER 3. SYSTEM MODEL

cal copy of itself produces a peak at the origin. The definitions of auto-
correlation and cross-correlation are defined subsequently. Applications
of PN sequences include signal synchronization, navigation, radar rang-
ing, random number generation, spread-spectrum communications, mul-
tipath resolution, cryptography, and signal identification in multiple-access
communication systems [39].
In the design of the preamble, the following must be taken into con-
sideration. The preamble sequence for cellular systems should have a low
peak-to-average power ratio (PAPR), which is is the ratio of peak signal
power to the average signal power used for estimating the backoff re-
quired for an radio-frequency/microwave power amplifier to exhibit ac-
ceptable intermodulation distortion. Along with this, it should also pos-
sess good (low) cross-correlation and (impulse-like) auto-correlation char-
acteristics all with the aim of serving well for the purpose of cell iden-
tification, time and frequency synchronization and channel measurement.
Well-known preamble sequences for cellular systems include PN sequence,
Gold sequence, and CAZAC sequence [46].
Various pseudo random codes are generated using LFSR (Linear Feed-
back Shift Register). The generator polynomial governs all the characteris-
tics of the generator. For a given generator polynomial, there are two ways
of implementing LFSR. Galois feedback generator uses only the output bit
to add (in Galois field) several stages of the shift register and is desirable
for high speed hardware implementation as well as software implementa-
tion.
The other one is the Fibonacci feedback generator which can generate
several delays of sequences without any additional logic. Shift-register
sequences having the maximum possible period for an r-stage shift reg-
ister are called maximal length sequences or m-sequences. A primitive gen-
erator polynomial always yields an m-sequence. The maximum period of
an p-stage shift register can be proven to be 2p − 1. The m-sequences has
three important properties, i.e., balance property, run-length property and
3.1. DEFINITIONS 23

shift-and-add property. An MLS-generating system with a shift register of

length 4 is shown in Figure 3.2. It can be expressed using the following
recursive relation:
(
a0(n) + a1(n), k=3
ak (n + 1) =
ak+1 (n), otherwise

where n is the time index, k is the bit register position, and + represents
modulo-2 addition. As MLS are periodic and shift registers cycle through
every possible binary value (with the exception of the zero vector), regis-
ters can be initialized to any state, with the exception of the zero vector.

Figure 3.2: M-Sequence generator

Polynomial Interpretation

A polynomial over GF(2) can be associated with the linear feedback shift
register. It has degree of the length of the shift register, and has coefficients
that are either 0 or 1, corresponding to the taps of the register that feed
the xor gate. For example, the polynomial corresponding to Figure 3.1 , is
x4 +x1 + 1. A necessary and sufficient condition for the sequence generated
by a LFSR to be maximal length is that its corresponding polynomial be
primitive.

3.1.2 Fairness in Wireless Networks

Fairness is an important property of a resource allocation algorithm. Fair-
ness measures or metrics are used in network engineering to determine
24 CHAPTER 3. SYSTEM MODEL

whether users or applications are receiving a fair share of system resources.

When network resources are insufficient to satisfy demand, they should be
divided f airly among network users [42]. There are two main approaches
to model fairness in resource allocation : max-min fairness [43] and pro-
portional fairness [44]. Max-min fairness is achieved by allocating avail-
able resources to the maximum extent possible to the disadvantaged users,
while not necessarily wasting resources. On the other hand, proportional
fairness approaches the problem from another perspective; proportional
fairness is realized when an objective function is maximized that repre-
sents the overall utilization of all users while respecting the total available
resource constraint.

Both proportional fairness and max-min fairness possess optimality

properties and are P areto optimal (i.e., if one cannot increase the assign-
ment to one source i without strictly decreasing an assignment to another
source j) and can be implemented under the convex optimization [45].

While considering the fact that in a wireless environment, because of

the random channel variations , we should be able to distinguish between
ef f ort (radio resource given to user) and outcome (actual useful through-
put achieved by the user). Normally, effort will be equal to the outcome,
but in a wireless environment, they can be substantially different. Both
max-min and proportional fairness can be defined based on either effort
or outcome.

Due to the time varying nature of wireless channels, fairness is usually

defined and implemented over a period of time. Based on the length of
the period, two kinds of fairness are considered : short term fairness and
long term fairness. Short term fairness means the ability to provide equal
allocation of resources to all active connections over short time scales; and
long term fairness is the opposite, and is concerned with the amount of
resources assigned over a longer time-scale.
3.2. UTILIZING PREAMBLE SEQUENCES FOR LOAD BALANCING 25

3.2 Utilizing Preamble Sequences for Load Bal-

ancing
In a cellular network, every BS is assigned a unique preamble sequence.
All the MSs in the network are embedded with the all the preamble se-
quences that exist in a particular technology and are cognizant about the
assignment of the sequences to BSs. Therefore, when they enter the net-
work and recieve the preamble signals from BSs, the simply correlate these
sequences with the all the sequences they know about to find a match and
identify the cell. This process if known as Cell Identification.

In this thesis, we configure the preamble sequences to implicitly indi-

cate the load experienced by the BS they are associated with, to MSs seek-
ing association and service. This allows the MSs to make decisions that
are more efficient in terms of fairness, depending on the quality of service
(QOS) measure we take as a criteria. We use the properties of high auto-
correlation and low cross-correlation to define certain metrics first, which
will be used to sort all the preamble sequences of the system.

Once the sequences have been sorted, we link the load information to
the value of the metric associated with the preamble sequence. This is
done by mapping the sequences to BSs according to the metric values and
the load experienced by the BS. This implies, for a highly loaded BS b, the
corresponding metric defined on b’s preamble sequence → −p b , would have
a higher value and for a relatively lightly loaded BS, the value would be
small.
We consider the following preamble configurations towards the design
of our schemes:

→
−0
• A sequence p formed by repetition of a part of a preamble sequence
on autocorrelation forms a number of peaks in the autocorrelation
26 CHAPTER 3. SYSTEM MODEL

plot which will be proportional to the load experienced by the BS

→
−0
corresponding to this sequence p .

• A preamble sequence of a lightly loaded BS will be highly correlated

with a particular code sequence →
−p 0 (known to both the MS and the
controller) and vice-versa.

Using the above described metrics, the sequences corresponding to the

BSs are sorted and assigned in a predefined order as shown in Figure 3.3
for the auto-correlation based load indication scheme and Figure 3.4 for
absolute cross-correlation scheme.

Figure 3.3: At central controller the metric is the number of peaks in the
auto-correlation plot.

Figure 3.4: At central controller the metric is the value of cross-correlation

with particular sequence.
3.2. UTILIZING PREAMBLE SEQUENCES FOR LOAD BALANCING 27

3.2.1 Load Balancing parameter

In each of the above mentioned configurations, we propose two algo-
rithms which work in parallel at the MS and at the central network con-
troller. For the algorithms to implement the load balancing feature, we
propose a design parameter r which is basically a measure of how much
greater the strongest RSSI is over the next strongest RSSI in the system,
when our algorithm stops coming into effect. This physically implies the
measure of area near the border between two cells where load balancing
comes into effect. As r increases, more area near the boundary is consid-
ered and vice-versa. Here, r = 1 will be the limiting case, where the ratio
occurs at the cell boundary.

RSSIstrongest
α= =1 (3.1)
RSSInextstrongest

Figure 3.5: A two cell model showing the concept of ”r”.

From this limiting case onwards, when r > 1, larger areas near the
boundary are considered for load balancing. Additionally, the conven-
tional RSSI based cell selection can be invoked by setting r = 0, when
28 CHAPTER 3. SYSTEM MODEL

the algorithm skips the load balancing steps and adopts the conventional
method using RSSI since α is always positive.
Thus, when r = 0, the condition for α < 0 will never be satisfied as can
be seen later in the MSCD AutoCorr and MSCD CrossCorr algorithms,
and can be better understood from Figure 3.5 above.

3.3 System Model

Figure 3.6: Architecture of the Korean WiMax Broadband [30]

For the illustration purpose, we in this thesis, consider the IEEE 802.16e/Mobile
3.3. SYSTEM MODEL 29

WiMAX system. Figure 3.6 shows the network architecture of WiBro (Wire-
less Broadband) system, which is a Korean version of Mobile WiMAX.
The WiBro network is composed of MSs, access network (ASN) and
connectivity service network (CSN). Among the network components, a
WiMAX system manager (WSM) is devised to manage multiple radio ac-
cess stations (RASs or BSs) and access control routers (ACRs or ASN-
G/W) for the purpose of network optimization. ACR is the central sys-
tem of the Wibro network, which connects the CSN and RAS. It enables
multiple RASs to internetwork with CSN and IP networks and sends and
receives traffic between the external network and MS. Therefore, it can
be used as the centralized controller over WiBro system. We also empha-
size that our proposed schemes can be used in any other wireless system,
which employs preamble code sequences.
The IEEE 802.16e specification [31] defines 114 pseudo noise (PN) se-
quences to be used as the preamble sequences in the 1k Fast Fourier Trans-
form (FFT) mode. These sequences are 284 bits long, and have the char-
acteristic properties that make them appealing for applications like cell
identification and synchronization.
In our schemes, we work with decisions involving cross-correlation be-
tween the PN code sequences, as the average cross-correlation between the
different PN code sequences have been found to be ranging from a mini-
mum of 0.04 to a maximum of 0.16 [30]. Also, a recent study [33] shows,
it is possible to use just a part of the whole PN code sequence to practi-
cally achieve the goals of synchronization and cell identification, mainly
because of this cross-correlation property among different sequences.
In a practical cellular network, a station generally selects the BS, which
yields the maximal received signal strength indicator (RSSI) obtained by
reading the preamble code sequence. In a model where B is the set of
all the base stations, gbs is the channel gain from BS b to an MS s, Pb is the
transmission power of BS b, → −
p b is the preamble code sequence correspond-
→
−
ing to base station b, and x s (t) is the signal received by an MS s at time t
30 CHAPTER 3. SYSTEM MODEL

from all BSs; the decision made by the MS s is given in Eq. 5.1, where b̂ is
the chosen BS.

b̂ = arg max RSSIs (t, b) (3.2)

b∈B

RSSI refers to the signal component of base station b, and →

−
p b ∈ P refers
to the preamble sequences, such that P is the set of all BS preamble se-
quences.

RSSIs (t, b) = →
−
p b .→
−
x s (t) (3.3)

In addition, the k th sequence in the set can be referred to as →

−
p (k) and
will be used in the later sections.

→
− gbs Pb (t)→
−
X
x s (t) = pb (3.4)
b∈B

3.4 Summary
In this chapter, we have presented a brief definition of preamble sequences,
how they are generated, and their applications, along with a note on fair-
ness in cellular networks. We introduced our proposed concept of how
preamble sequences can be utilized for load balancing with a model/design
parameter to target the boundary users. Finally, we presented our sys-
tem model, and used the IEEE 802.16e/WIMAX system for illustration,
with a note on its architectural components and how we use the pream-
ble sequences in it for load balancing. In the next chapter, we present our
proposed auto-correlation based implicit load indication scheme with the
algorithms discussing how it is realized.
Chapter 4

Auto-Correlation based Implicit

Load Indication Scheme

4.1 The Auto-Correlation Property

Auto-correlation is defined as the cross-correlation of a signal with itself. It
is the similarity between observations as a function of the time separation
between them. It is a mathematical tool for finding repeating patterns,
such as the presence of a periodic signal which has been buried under
noise, or identifying the missing fundamental frequency in a signal im-
plied by its harmonic frequencies [40]. It is often used in signal processing
for analyzing functions or series of values, such as time domain signals.
In the case of m-sequences, there are two types of auto-correlation : pe-
riodic auto-correlation and aperiodic auto-correlation. If the synchronization
window is only one period long or less, then the correlation is aperiodic,
otherwise it will be always be periodic if it is any longer. As an example,
we consider the auto-correlation of a sine wave with a frequency of 1000
Hz (Hertz) and a sampling frequency of 8000 Hz. In the auto-correlation
plot, the peaks are clearly visible which are used for the applications men-
tioned above and in this thesis, we use them to indicate the level of load
to the MS while cell selection.

31
32CHAPTER 4. AUTO-CORRELATION BASED IMPLICIT LOAD INDICATION SCHEME

Figure 4.1: Auto-correlation function of the sinewave

In this scheme, we propose two algorithms working side by side to

provide BS load indication using the preamble sequence. In this method,
we modify the preamble code sequence and propose the autocorrelation
operation as a way to decipher the load in a BS. This approach works at
two levels: at the MS and at the central network controller.
Here, B is the set of all BSs and a particular BS is b with load lb . First,
the central controller creates a list of BSs sorted in order of their loads.
Then, we define p intervals of the load value which can be denoted by
L1 , L2 , ...Lp . Thus, the load values lb pertaining to the base station b fall in
one of these ranges.
In a general case where the preamble sequence length is M bits, the
autocorrelation plot of the sequence will give a peak at the origin and neg-
ligible values for all other delays. This is a characteristic feature of the
PN sequences due to their pseudo random nature. We modify the PN se-
4.1. THE AUTO-CORRELATION PROPERTY 33

Figure 4.2: Auto-correlation plot of a preamble of length 284 bits made

with 4 repeated sequences of length 71.

quence by selecting the first Mj bits of the PN sequence and then repeating
it j times where j = arg{Lj } and this will grouped into one of Lj . This will
give us a sequence of M bits which would be periodic in the time domain
and would have j periods in the M bit length.
Figure 4.2 shows the auto-correlation plot of a preamble of length 284
bits which is made up of four repeated preamble sequences of length 71.
After this operation, the autocorrelation plot for the modified pream-
ble code shows j peaks corresponding to the number of repetitions in the
M bit sequence. We propose that this value j be the metric which we use
for load indication of a particular BS. Hence, the more the repetitions in
34CHAPTER 4. AUTO-CORRELATION BASED IMPLICIT LOAD INDICATION SCHEME

the preamble sequence, the more will be the number of peaks in the auto-
correlation plot and that would indicate a more loaded BS.
This modification to the preamble code sequence by introducing pe-
riodicity in the time domain does cause concerns over its effect on cell
identification and synchronization which the preamble sequence actually
has to fulfill. These are valid concerns as due to the periodic nature of the
modified preamble, it exhibits lower separation of magnitude between the
autocorrelation peak and other values on the autocorrelation plot. These
concerns have been addressed in a recent study [33], which shows that a
considerably small part of the preamble sequence is practically sufficient
in terms of achieving the required goals of identifiability and also synchro-
nization. Larger number of repetitions would make the errors higher, but
below a maximum number of allowed repetitions p, it would be practi-
cally sufficient. We now present two algorithms working side by side. The
preamble modification algorithms works at the central controller and MS
cell decision algorithm works at the user end.

4.2 Preamble Modification Algorithm

The modification of the preamble code would happen at the central net-
work controller where it will assign the j value (number of repetitions) to
the BSs according to their loads.

This will be done by first identifying the maximally loaded BS in the

set of BSs B (line 3). Then, the group number of the load value is extracted
(line 4, 5) and accordingly, the preamble sequence is modified (line 7 - 9)
by taking the first Mj bits of →
−p and repeating it j times (line 8). Thus, the
BSs will then transmit using preamble sequences that have j repetitions
of the first Mj bits of the original sequence. We refer to this as Preamble
Modifying Algorithm (PMA).
4.3. MS CELL DECISION USING AUTOCORRELATION 35

Pseudo-code for PMA

1: procedure PMA(p, B)
2: while B 6= ∅ do
3: b́ ←− arg max lb
b∈B
4: if lb́ ∈ Lj then
5: j ←− arg{Lj }
6: endif
→
−0 →
−0
7: p ←− → −
p b́ where p ∈ RM
→
−0 →
−0 →
−0 →
−0
8: p ←− [ p ( Mj ) p ( Mj )... p ( Mj )] (repeated j times)
→
−0
9: →
−p ←− p
b́
10: B ←− B/{b́}
11: end while
12: end procedure
Table 4.1: Preamble Modification Algorithm (PMA)

4.3 MS Cell Decision using AutoCorrelation

Then, at the MS, the MS Cell Decision (MSCD) algorithm will be used to
detect load values of BSs which are of interest to the MS and consequently
make cell choice decision. We denote this MSCD algorithm using auto-
correlation as MSCD AutoCorr. The MS first correlates the preamble code
with the copy of the original preamble sequence code list it possesses to
identify the BSs (this step is involved in detecting the three strongest BS
signals in lines (2 - 6) of MSCD AutoCorr) and then uses the autocorre-
lation property (line 8 - 10) to find multiple peaks (if applicable) instead
of the single peak which would have been the normal case. Detecting the
number of peaks (line 11-13), it arrives at the value j which conveys infor-
36CHAPTER 4. AUTO-CORRELATION BASED IMPLICIT LOAD INDICATION SCHEME

mation about the load, which is then used to decide on the choice between
the three BSs of interest.Prior to the Preamble Modifying Algorithm, the
grouping intervals of the load values are assumed to have been decided
by the central network controller using uniform grouping.

Pseudo code for the MSCD AutoCorr Algorithm

1: procedure MSCD AutoCorr
2: →
−
p (k1 ) ←− arg −
→
max RSSIs (→
−
p (k))
p (k)∈P
3: P ←− P/{→ −
p (k1 }
→
−
4: p (k2 ) ←− arg −max RSSIs (→
−
p (k))
→ p (k)∈P
5: P ←− P/{→ −
p (k2 }
→
−
6: p (k3 ) ←− arg −max RSSIs (→
−
p (k))
→ p (k)∈P
RSSIs (−
→
7: if −
→ < r && RSSIs (→
p (k1 ))
RSSIs ( p (k2 ))
−
p (k2 )) > δ, then
→
−
8: a(k1 , τ ) ←− autocorr( p (k1 ) where τ is delay
9: a(k2 , τ ) ←− autocorr(→
−p (k2 )
10: a(k , τ ) ←− autocorr(→
3
−
p (k ) 3

11: j(k1 ) ←− peakf ind(a(k1 , τ ))

12: j(k2 ) ←− peakf ind(a(k2 , τ ))
13: j(k3 ) ←− peakf ind(a(k3 , τ ))
14: ḱ ←− arg min {j(k)}
k∈{k1 ,k2 ,k3 }
15: {b∗i } ←− {b|→
−
pb=→
−
p (ḱ)}
16: else
17: {b∗i } ←− {b|→
−
pb=→
−
p (k1 )}
18: end if
19: end procedure

Table 4.2: MSCD Algorithm

The MSCD algorithm using autocorrelation (denoted as MSCD Autocorr)

is given as above. The peakf ind() function finds the number of peaks in
the argument plot for different values of τ and returns a value which is
4.4. SUMMARY 37

equal to the number of peaks in it. Here we define two design parameters
r and δ, which specifies the area to be influenced by load balancing algo-
rithm and the minimum RSSI requirement, respectively.

4.4 Summary
In this chapter, we first presented a brief defintion of the auto-correlation
property of preamble sequences, which is the cross-correlation of a signal
sequence with itself along with its use and applications. Based on this
property, we have developed and defined two algorithms working side
by side, one on the central network controller and one on the user side to
associate MSs with BSs considering the level of load at them.
In the first algorithm, at the central network controller, the load values
of all BSs is collected first and then the preamble sequences of all of them
are modified. Then the controller assigns a specific value obtained by the
modification to BSs according to their loads. The BSs will then transmit
using these modified preamble sequences in the Preamble Modifying Al-
gorithm. The MS then receives and correlates the preamble code with the
copy of the original preamble it had and finds the modification which im-
plicitly conveys the load level to it and it chooses the BS which is lightest
loaded in the MSCD algorithm, which runs on the user’s side.
In the next chapter, we present our proposed cross-correlation based
implicit load indication scheme and how it is applied for load balancing.
38CHAPTER 4. AUTO-CORRELATION BASED IMPLICIT LOAD INDICATION SCHEME
Chapter 5

Cross-Correlation based Implicit

Load Indication Scheme

5.1 The Cross-Correlation Property

Cross-correlation is defined as a measure of similarity of two waveforms
as a function of a time-lag applied to one of them. This is also known as
a sliding dot product or sliding inner-product. It is commonly used for
searching a long-duration signal for a shorter, known feature. It also has
applications in pattern recognition, single particle analysis, electron tomo-
graphic averaging, cryptanalysis, and neurophysiology.

Cross-correlation is similar in nature to the convolution of two func-

tions. Whereas convolution involves reversing a signal, then shifting it
and multiplying by another signal, correlation only involves shifting it
and multiplying (no reversing) [41].

R∞
(f ? g)(t) = f ∗ (τ )g(t + τ )dτ (5.1)
−∞

where f ∗ denotes the complex conjugate of f .

39
40CHAPTER 5. CROSS-CORRELATION BASED IMPLICIT LOAD INDICATION SCHEME

In the case of dicrete functions, the cross-correlation is defined as :

m=∞
X
(f ? g)[n] = f ∗ [m]g(n + m) (5.2)
m=−∞

The cross-correlation is similar in nature to the convolution of two

functions.

Figure 5.1: Two time series x,y.

As an example, Figure 5.1 shows two time series xy, and Figure 5.2, a
cross-correlation between the two time series performed with a maximum
delay of 4000. The strong correlation can be observed from the peak at the
delay of about 40.
5.2. PREAMBLE ALLOCATION USING CROSS-CORRELATION 41

Figure 5.2: The cross-correlation series with a maximum delay of 4000

showing a strong correlation at a delay of about 40.

5.2 Preamble Allocation using Cross-correlation

In this approach, what we call the Preamble Allocation using Cross-correlation
(PAC) algorithm is performed at the centralized controller. Here, we use
the maximum correlation value of the preamble with a particular prede-
fined sequence → − p 0 . This sequence can be an arbitrary preamble sequence
taken from one of the 114 PN codes which are used in the 1k FFT mode
in IEEE 802.16e. When we correlate the BS’s preamble code with this par-
ticular code → −p 0 , the result can then be used as a metric to which we can
implicitly link the load of the BS. This is largely dependent on getting dif-
ferent such cross-correlation values for different preamble sequences. In
this method, the set Φs contains the sorted list of various preamble se-
quences sorted with respect to their cross-correlation value with → −
p 0 . In
Φs , the first preamble sequence has the lowest cross-correlation with → −p0
and the last one has the highest cross-correlation with p 0 .→
−

Before the commencement of this algorithm, the central network con-

troller will obtain the load values of all the base stations. This load could
42CHAPTER 5. CROSS-CORRELATION BASED IMPLICIT LOAD INDICATION SCHEME

be the number of users connected or the number of calls connected. Af-

ter this, the controller would create a list Φs which would contain the
preamble sequences sorted according to the value of their maximum cross-
correlation with the reference preamble sequence → −p 0 . The algorithm then
searches the most loaded BS in the set B (line 4). It then assigns the first
sequence in Φs , which is the one with the least correlation with → −p 0 to it
(line 5). Then, the pointer to the top of the list Φs is incremented to the
sequence with the next lowest correlation with → −p 0 and the whole process
is repeated till all the BSs are exhausted.

Pseudo-code for the PAC Algorithm

1: procedure PAC(B)
2: i=1
3: while B 6= ∅ do
4: b́ ←− arg max lb
b∈B
5: pb́ ←− p (i) s.t. →
→
− →
− −
p (i) ∈ Φs
6: i ←− i + 1
7: B ←− B/{b́}
8: end while
9: end procedure

5.3 BS Selection Algorithm using Implicit Indi-

cation
Together with the PAC algorithm, the MSCD using Cross-Correlation (de-
noted as MSCD CrossCorr) algorithm is employed by the MS if it is near
5.3. BS SELECTION ALGORITHM USING IMPLICIT INDICATION 43

the boundary between three BSs. Firstly the algorithm identifies the three
nearest BSs by using their RSSI (line 2 - 6). Then, a condition is tested using
a design parameter r which determines whether the MS is at the boundary
area between the two identified BS’s cells by calculating the ratio of their
RSSI.

Pseudo code for the MSCD CrossCorr Algorithm

1: procedure MSCD CrossCorr
2: →
−
p (k1 ) ←− arg −
→
max RSSIs (→
−
p (k))
p (k)∈P
3: P ←− P/{→ −
p (k1 }
→
−
4: p (k2 ) ←− arg −max RSSI (→
−p (k))
→ s
p (k)∈P
5: P ←− P/{→ −
p (k2 }
6: →
−
p (k3 ) ←− arg −
→
max RSSIs (→
−
p (k))
p (k)∈P
RSSIs (−
→
7: if RSSI (→
s
− < r && RSSIs (→
p (k1 ))
p (k ))
2
−p (k2 )) > δ, then
8: cc(k1 ) ←− crosscorr(→−p (k1 ), →
−
p 0)
9: cc(k2 ) ←− crosscorr(→−p (k2 ), →
−
p 0)
10: cc(k3 ) ←− crosscorr( p (k3 ), →
→
− −
p 0)
11: Q ←− {→ −
p (k ), →
−
p (k ), →
1
−
p (k )}2 3

12: ḱ ←− arg max cc(k)

k=arg −
→
p (k)s.t.−
→
p (k)∈Q
13: {b∗i } ←− {b|→
−
pb=→
−
p (ḱ)}
14: else
15: {b∗i } ←− {b|→
−
pb=→
−
p (k1 )}
16: end if
17: end procedure

Also, a minimum RSSI criteria is tested using a parameter δ (line 7). When
44CHAPTER 5. CROSS-CORRELATION BASED IMPLICIT LOAD INDICATION SCHEME

either of these two conditions fail, the usual RSSI based selection is fol-
lowed (line 15). Otherwise, the load balancing algorithm comes into effect.
Cross correlations of the preamble sequences of the three BSs are calcu-
lated (line 8 - 10) and the BS having highest maximum cross-correlation is
chosen.

5.4 Summary
In this chapter, we first presented a brief defintion of the cross-correlation
property of preamble sequences, along with its use and applications. We
have developed and defined two algorithms working side by side on the
basis of this property, one on the central network controller and one on the
user side to associate MSs with BSs considering the level of load at them.
In the first algorithm that we call Preamble Allocation using cross-
correlation (PAC), at the central network side, the load value of all BSs
is collected first and then the preamble sequences are correlated with an
arbitary sequence to get different values which are then assigned to im-
plicitly indicate the load to the users seeking association and service. On
the user’s side runs the MSCD algorithm, where the users correlate the
sequences present in them with the sequence of the incoming stations to
implictly know the load of the BS for smart cell-selection.
In the next chapter, we perform extensive simulations of our proposed
schemes and discuss the results.
Chapter 6

Simulation Results and

Discussion

In this chapter, simulations results are presented on the two load balancing
schemes described in Chapter 4 and 5. The main objective is to investigate
the performance of the schemes to substantiate their effectiveness using
popular performance metrics such as throughput, load variance and fair
resource sharing. First, Section 6.1 discusses the channel characteristics
in wireless systems that affect the performance and following that a breif
note on the simulation environment. Subsequently, Section 6.2 presents
the performance results using the performance metrics mentioned earlier.
After this, in Section 6.3, the performance of our proposed schemes is then
evaluated under different environments, namely, the Hata for urban areas.
Finally, in Section 6.4, we present a comparison of our proposed schemes
against RSSI-based schemes.

6.1 Simulation Setup

In this section, we first define the channel characteristics and emphasize
how they affect the performance of a system and following it, the simula-
tion environment.

45
46 CHAPTER 6. SIMULATION RESULTS AND DISCUSSION

6.1.1 Channel Characteristics

In a typical outdoor wireless propogation environment, when a mobile
wireless user communicates with a BS, the signal transmitted from the
mobile user reaches the BS either directly (line-of-sight) or through multi-
ple reflections on local scatterers such as buildings, mountains, etc. This is
because the received signal strength is affected by multiple random atten-
uations and delays. Moreover, the mobility of either the mobile users or
the scatterers may cause these random fluctuations to vary over time. Fi-
nally, in a shared wireless environment, the transmitted signal may come
across interference due to concurrent transmissions.
The attenuation incurred in wireless propogation can be classified into
three main elements : a signal attenuation due to the distance between
communicating nodes (path-loss), attenuation effects due to absorption
in local structures such as buildings (shadowing), and rapid signal fluc-
tuations due to constructive and destructive interfeerence of multiple re-
flected radio wave paths (fading).
Variations due to path-loss and shadowing occur over relatively large
distances that happens as mobile users move through a distance of the
order of the cell. Path-loss and shadowing are typically frequency depen-
dent. Variation due to multi-path fading occurs over very short distances,
on the order of the signal wavelength and is frequency dependent. The
channel characteristics depend on the combination of all three propoga-
tions effects mentioned, but for simplicity we only consider a noisy chan-
nel with path-loss attenuation for our simulations.

6.1.2 Simulation Environment

We consider a 12-cell model with a frequency reuse factor of 1, as illus-
trated in Figure 6.1, and focus on the three cells in the center with numbers
0,1 and 2 highlighted.
For the initial distribution of users, we assume that 100 stations are lo-
6.2. SIMULATION RESULTS AND DISCUSSION 47

Figure 6.1: A conventional 12-cell model with focus on cells 0,1 and 2

cated in cell 0, 80 in cell 1, 20 in cell 2, and all the other cells have 45 stations
randomly distributed over their cell region. Our algorithm is then applied
to the boundary stations. In our simulations, we consider the channel gain
due to path loss effect with a path loss exponent of 4, which is ideal for a
typical cellular network.

6.2 Simulation Results and Discussion

In this section, simulation results are presented along with the underlying
assumptions. The simulations implement the proposed algorithms at the
decision making level where depending on the load of BSs, the stations
in a certain area (defined by r) near the boundary choose the BS to be
associated with. Hence, as both algorithms look at the same quantity i.e.
load to help MS’s make the cell choice, the simulation results are identical.
48 CHAPTER 6. SIMULATION RESULTS AND DISCUSSION

Channel Prameters Value

Bandwidth (MHz) 10
Transmission Power (db) 20
Path Loss Exponent 4

Table 6.1: Channel parameters for simulation

The differences in the two algorithms lie in the method of detection of

load information in the preamble, and thus when this is accurately done as
assumed in the simulations, both algorithms give identical results. Figure
6.2 shows that the number of users becomes more evenly distributed as r
increases. This is where the red MSs are associated to heavily loaded cell
0, blue MSs are associated to cell 2 with heavy load again and the green
MSs are associated with the lightly loaded cell 1.

Figure 6.2: Distribution of users in cells at r = 10

At r = 10 case, one can see how our algorithm leads many of the MSs near
the boundary of cell 0 to be handed over to neighbouring BSs and thus
6.2. SIMULATION RESULTS AND DISCUSSION 49

becoming green/blue/yellow, as cell 0 is heavily loaded. Similarly, blue

MSs are also handed over to cells depending on their loads. Green MSs,
however, are mostly those who have been handed over to BS of cell 1 by
other heavily loaded BSs.

6.2.1 Load Variance Test

As a quantitative measure of how the load is distributed among the vari-

ous BS’s cells, we propose a metric called the Load Variance lvar which is
given by:

(lb − lb )2
P
b∈B
lvar = (6.1)
n(B)

where n(B) is the total number of BSs, lb is the load experienced by each
BS and lb is the mean of all the load values.
For this test, we take the number of users as 100 in cell 0, 80 in cell 1, and
20 in cell 3. Therefore, as the load gets distributed more evenly among the
different BSs, we see a gradual decrease in lvar , as seen in Figure 6.3, which
shows its variation with different values of our design parameter r.

We now vary the number of user in cells and simulate by having 60 users
in cell 0, 40 users in cell 1, and 10 users in cell 2. We observe, at r =
7, in Figure 6.4, the load gets minimal and then slightly increases before
stabilizing after r = 9. This is because the total population decreases a bit
as r increases in cell 2 and go into cell 10 and 11 as their serving cells. Thus,
this metric is useful in measuring the even-ness of distribution of MSs in a
collection of cells and shows the effectiveness of our proposed schemes in
terms of fair user distribution.
50 CHAPTER 6. SIMULATION RESULTS AND DISCUSSION

Figure 6.3: Load Variance test with 100 users in cell 0, 80 users in cell 1,
and 20 users in cell 3.

Figure 6.4: Load Variance test with 60 users in cell 0, 40 users in cell 1, and
10 users in cell 2.
6.2. SIMULATION RESULTS AND DISCUSSION 51

6.2.2 Resource Fairness Test

We evaluate the effect of load balancing over multi-cells in terms of the

corresponding fairness criteria when the resource allocation in every cell
follows resource fair (RF) scheduling policy. Since the resource allocated
to each user in each single cell is inversely proportional to the number
of users in that cell, it is good load balancing from the viewpoint of RF
scheduling, resulting in users in every cell being distributed as evenly as
possible.
As we see in Figure 6.5, at r = 10, all three BSs of cells 0 to 2 have
number of associated MSs between 50 to 60 and initially the number of
users were 100 in cell 0, 80 in cell 1, and 20 in cell 2.

Figure 6.5: Resource Fairness with initially 100 users in cell 0, 80 in cell 1
and 20 users in cell 2.
52 CHAPTER 6. SIMULATION RESULTS AND DISCUSSION

We now vary the number of user in cells and simulate by having 50

users in cell 0, 40 users in cell 1, and 10 users in cell 2 in Figure 6.6. We
again observe that the number of users get even in cells 0 to 2. This can
be interpreted that our load balancing scheme leads to an improved con-
sequence in terms of fair resource sharing.

Figure 6.6: Resource Fairness with initially 50 users in cell 0, 40 in cell 1

and 10 users in cell 2.

6.2.3 Spectral Efficiency Test

For minimum throughput simulations, we consider the lower bound case
for throughput in a cell that occurs when all the users are at the maximum
distance from the BS, and hence, have the lowest possible SNR (Signal to
Noise Ratio) at the MS. In such cases, the bandwidth is equally distributed
among the MSs, but the capacity reaches its minima because of lowest
6.2. SIMULATION RESULTS AND DISCUSSION 53

SNR. This is clear from the expression for channel capacity [34], where R
is the channel capacity, W is the bandwidth assigned to each MS, P is the
power received by the MS and N0 /2 is the noise.

R P
= log(1 + ) (6.2)
W N0 W

Figure 6.7: Throughput Fairness with initially 100 users in cell 0, 80 in cell
1 and 20 users in cell 2.

Thus, this lower bound of throughput in a cell is calculated for differ-

ent variations of r and it is seen that the minimum throughput improves
considerably for the heavily loaded BS on application of the proposed al-
gorithms for load balancing. This is because the initially heavily-loaded
BS hands over some of its boundary users to the adjoining cells and hence,
achieves better throughput.
54 CHAPTER 6. SIMULATION RESULTS AND DISCUSSION

On the other hand, the other two lightly loaded BS’s throughput de-
teriorates slightly because of the increase in number of users managed by
the BS, as seen in Figure 6.7.
We now vary the number of user in cells and simulate by having 50
users in cell 0, 40 users in cell 1, and 10 users in cell 2 in Figure 6.8. We
again observe that the minimum throughput of users in cell 0 increases
along with users in cell 2, while the minimum throughput of users in cell
1 experience a deterioration in their throughput due to the new handoffs.

Figure 6.8: Throughput Fairness with initially 50 users in cell 0, 40 in cell

1 and 10 users in cell 2.

This can be interpreted that our load balancing scheme leads to an im-
proved consequence in terms of max-min throughput fairness.
6.2. SIMULATION RESULTS AND DISCUSSION 55

6.2.4 Evaluation using Jain’s Fairness Index

The most simple way to evaluate how load balanced the system is, is to
calculate the average load of the whole system:

Pn
i=1 Ui
L= (6.3)
n
where n is the number of BSs, and Ui is the resource utilization of BS i
and compare this average to the individual resource utilizations Ui of each
BS. To describe the loading state of the whole system with one value the
following load balance index has been defined [42]:

( ni=1 Ui )2
P
B= 2 (6.4)
n ∗ ni=1 Ui
P

Figure 6.9 shows the load balance index, where values near 1 depict a
load balanced system and 0, no load balancing.
To quantify and measure the fairness acheived in the system, we use
Jains0 s F airness Index again described as follows :

Pn
( i=1 xi )2
J(x1 , x2 , ..., xn ) = P n 2 [42] (6.5)
n∗ i=1 xi

Equation 6.4 rates the fairness of a set of values where there are n users
and xi is the throughput for the ith connection. The result ranges from 1/n
(worst case) to 1 (best case), and it is maximum whenall users receive the
same allocation. This index is k/n when k users equally share the resource,
and the other n-k users receive zero allocation. This metric identifies un-
derutilized channels and is not unduly sensitive to atypical network flow
patterns.
Figure 6.10 shows, as r increases, the system gets fairer and at r = 6, we
are able to obtain fairness greater than 0.9 on the index. This also proves
that there is a very strong correlation between load balancing and fairness.
56 CHAPTER 6. SIMULATION RESULTS AND DISCUSSION

Figure 6.9: Load Balancing Index vs r

Figure 6.10: Jain’s Fairness Index vs r

6.3. PERFORMANCE UNDER DIFFERENT ENVIRONMENTS 57

6.3 Performance under Different Environments

In wireless communication, the Hata Model for Urban Areas, also known
as the Okumura-Hata model for being a developed version of the Oku-
mura Model, is the most widely used radio frequency propagation model
for predicting the behaviour of cellular transmissions in built up areas.
This model incorporates the graphical information from Okumura model
and develops it further to realize the effects of diffraction, reflection and
scattering caused by city structures. We consider this model to evaluate
the robustness of our schemes in different environments. Below men-
tioned are the coverage parameters considered which are a standard for
the Hata model.

Coverage Value
Frequency (MHz) 150 - 1500
Mobile Station Antenna Height (Meters) 1 - 10
Base station Antenna Height (Meters) 30 - 200
Link distance (Kms) 1 - 20

Table 6.2: Standard coverage parameters for the Hata model

The Hata model for Urban Areas [49] is formulated as :

LU = 69.55 + 26.16 log f − 13.82 log hB − CH + [44.9 − 6.55 log hB ] log d(6.6)

where,
LU = Path Loss in Urban Areas. Unit: decibel (dB)
hB = Height of Base Station Antenna. Unit: meter (m)
hM = Height of Mobile Station Antenna. Unit: meter (m)
f = Frequency of Transmission. Unit: megahertz (MHz)
CH = Antenna Height Correction Factor
58 CHAPTER 6. SIMULATION RESULTS AND DISCUSSION

d = Distance between the Base and Mobile Stations. Unit: kilometers

(km)
In the next section, we will evaluate our schemes under the small and
large city models to test how they perform under varying levels of path
loss.

6.3.1 Hata Model for a Small City

We now simulate our schemes under the Hata - small city model with
the equation for path loss being the same as above, but the value of the
antenna height correction factor being different which is calculated as fol-
lows [49]:

CH = 0.8 + (1.1 log f − 0.7)hM − 1.56 log f (6.7)

We first simulate for throughput fairness versus our design parameter r.

Figure 6.11 shows the fairness of throughput under the channel conditions
mentioned above. It can be observed that there is an improvement in the
throughput of the cell 0 as the cell edge users leave the cell and are handed
off to cell 1 and 2. Similarly, the throughput of cell 1 deteriorates as it
accomodates new users from cell 0 and gets stabilized as r increases. Due
to the very less number of users in cell 2, the throughput improvement is
visible, but is minimal and stabilizes along with increasing r.
This shows that our proposed schemes fair well in terms of throughput
fair scheduling, even though the improvement in the minimum through-
put in not very substantial.
As seen in Figure 6.12, our schemes perform well to induce user hand-
offs into the lightest loaded cell, which is cell 2 and gains users from both
cells 0 and 1 showing a well balanced user distribution under these chan-
nel conditions. This is mostly due to the effectiveness of our design param-
eter r that targets the boundary users that the schemes are able to balance
user distribution well.
6.3. PERFORMANCE UNDER DIFFERENT ENVIRONMENTS 59

Figure 6.11: Throughout Fair vs r for a small city

Figure 6.12: Resource Fair vs r for a small city

60 CHAPTER 6. SIMULATION RESULTS AND DISCUSSION

Figure 6.13: Load Variance vs r for a small city

Figure 6.13 shows the varinace of load in the system. Due to the sever-
ity of noisy conditions, when our load balancing algorithm comes into
effect at r = 1, we see a sharp decline in the load variance initially until
r = 2 because of the user handoffs. Then the variance gradually stabilizes
at r = 8.

We now evaluate our schemes in terms of the load balance index of the
whole system [42] now. It can be observed from Figure 6.14 that due to
the noisy channel conditions, at r = 2, when our load balancing scheme
comes into effect, we see an increment initially but it is met by resistance
from r = 3 to 8, until it sharply increases again at r = 9. This shows the
robustness of our schemes.
6.3. PERFORMANCE UNDER DIFFERENT ENVIRONMENTS 61

Figure 6.14: Load Balance Index vs r for a small city

6.3.2 Hata Model for a Large City

We now simulate our schemes under the Hata - large city model with the
equation for path loss being the same as above, but the value of the an-
tenna height correction factor being different which is calculated as fol-
lows:

(
8.29(log(1.54hM ))2 − 1.1, 150 ≤ f ≤ 200
CH = (6.8)
3.2(log(11.75hM ))2 − 4.97, 200 < f ≤ 1500

We first simulate again for throughput fairness versus our design param-
eter r. Figure 6.15 shows the fairness of throughput under the channel
conditions mentioned above. It can be observed that there is a great im-
provement in the throughput of the cell 0 as the cell edge users leave the
62 CHAPTER 6. SIMULATION RESULTS AND DISCUSSION

cell and are handed off to cell 1 and 2 until r = 9 where noisy conditions
overpower it, but it recovers again at r = 10.
The throughput of cell 1 deteriorates as it accomodates new users from
cell 0 and gets stabilized as r increases, but due to the extremely noisy
channel conditions, it does not gain at all to be noticable. Also, due to the
very low number of users in cell 2, the throughput deterioration is barely
visible, but is minimal and stabilizes along with increasing r.
This shows that our proposed schemes are able to perform in the noisiest
of environments in terms of throughput fair scheduling, even though the
improvement in the minimum throughput in not very substantial.

Figure 6.15: Throughout Fair vs r for a large city

As seen in Figure 6.16, our schemes perform well again to induce user
handoffs into the lightest loaded cell, which is cell 2 and gains users from
6.3. PERFORMANCE UNDER DIFFERENT ENVIRONMENTS 63

Figure 6.16: Resource Fair vs r for a large city

both cells 0 and 1 showing a well balanced user distribution under these
channel conditions. This is again due to the effectiveness of our design
parameter r that targets the boundary users that the schemes are able to
balance user distribution well.
Figure 6.17 shows the varinace of load in the system. Due to the sever-
ity of noisy conditions in the large city model, when our load balancing al-
gorithm comes into effect at r = 1, we again see a sharp decline in the load
variance initially from r = 2 till r = 4 because of the user handoffs. Then,
the variance stabilizes despite the noisy channel conditions and mildly
climbs from r = 9 to r = 10.

We now evaluate our schemes in terms of the load balance index of the
whole system [42] now for the large city model. It can be observed from
64 CHAPTER 6. SIMULATION RESULTS AND DISCUSSION

Figure 6.17: Load Variance vs r for a large city

Figure 6.17 that due to the extremely noisy channel conditions, at r = 2,

when our load balancing scheme comes into effect, we see an increment
initially, but it is met by strong resistance due to the noise from r = 3 to 8,
until it sharply increases again at r = 9. This shows the resilience of our
proposed schemes in very noisy environments.

In the next section, we move on to comparing our schemes with the

conventional RSSI-based association to evaluate performance comparison.
6.4. COMPARSION AGAINST RSSI-BASED SCHEMES 65

Figure 6.18: Load Balance Index vs r for a large city

6.4 Comparsion against RSSI-based Schemes

We peform a comparison of our proposed schemes with schemes and liter-

ature based on conventional RSSI-based cell selection [47]-[48] to evaluate
the performance. We use our design parameter r to perform the compari-
son again. Figure 6.19 shows the result and demonstrates that employing
our proposed association control based schemes result in a substantial im-
provement in the throughput of users belonging to cells 0 to 2 as compared
to the conventional RSSI-based user association in each cell.
Thus, our schemes perform well when throughput fair scheduler is
adopted in each cell. Finally, we perform a comparison using Jain’s fair-
ness index in Figure 6.20 and it is clearly evident that our schemes outper-
form the conventional RSSI based schemes in terms of an overall fairness
66 CHAPTER 6. SIMULATION RESULTS AND DISCUSSION

Figure 6.19: Minimum Throughput gain vs r

Figure 6.20: Jain’s Fairness Index vs r

6.5. SUMMARY 67

to users index.

6.5 Summary
In this chapter, we performed extensive simulations on our two proposed
load balancing schemes. We first present a brief discussion on the channel
characteristics of wireless networks and the various degradation factors.
Then we discuss the simulation environment with the underlying assump-
tions. We first simulate using the peformance metric, load variance, which
shows that our proposed schemes peform well balancing the load of the
system, minimizing the traffic load variance. Then, the schemes are tested
for fairness to users in terms of resources under various user densities to
demonstrate their effectiveness in balanced user distribution across the
network. We then quantify the fairness acheived by the schemes using
Jain’s fairness index, which displays very high levels of fairness in terms
in terms of load and user distribution. We also compare the performance
of our schemes under different environments to test their robustness using
the Hata model. Finally, we compare our proposed schemes with schemes
based on RSSI-based assocation to compare their performance and make
our case.
68 CHAPTER 6. SIMULATION RESULTS AND DISCUSSION
Chapter 7

Conclusion

Wireless communications are gaining momentum in mobile applications

and the increasing demand for all types of wireless services (voice, data,
multimedia) is fueling the need for higher capacity. The conventional cel-
lular networks are experiencing several difficulties for reaching this apex
and are bounded by factors such as bandwidth, coverage area, or infras-
tructure costs. This thesis is devoted to the development of algorithms
that explore and make use of preamble sequences to alleviate the problem
of non-uniform user distribution in cellular networks due to the conven-
tional RSSI-based association. This problem is one of major degradation
factors in wireless cellular networks impacting capacity and performance
due to the imbalanced load across the network.
We have developed two schemes to tackle this problem in this the-
sis. We first developed a design parameter to work with our schemes that
targets the cell-edge users for load balancing as they are the most impor-
tant entities of any load balancing scheme. In our first scheme, which is
the auto-correlation based implicit load indication scheme, we modify the
preamble code sequence and propose the auto-correlation operation as a
way to implicitly decipher the load at the BSs. This scheme allows the mo-
bile users to make an intelligent decision while choosing a BS for service,
by not only considering the strength of the received signal from the BS,

69
70 CHAPTER 7. CONCLUSION

but also the level of congestion at it. The second scheme works using the
cross-correlation property among preamble sequences to implicitly obtain
the load state of BSs for load balancing. This enables the users while join-
ing the network to perform smart cell-site selection based on the level of
congestion at the BS along with the conventional association metrics.
We have performed extensive tests to evaluate the performance of the
schemes considering the scheduling disciplines RF and TF, as the resource
allocation policies. The proposed schemes are able to obtain a high degree
of improvement in the minimum throughput of users in the cells we focus
on, and apply our load balancing schemes to, and perform well in terms of
throughput fair scheduling. The schemes are then tested for fair user dis-
tribution and achieve great results in alleviating the problem of load vari-
ance by swiftly inducing user handoffs into lightly loaded cells from the
highly loaded ones to balance the load. We also test the schemes in terms
of Jain’s fairness index for resource fairness evaluation and for load vari-
ance and obtain great results. Finally, we tested the schemes for robustness
and compared them with schemes relying on conventional RSSI-based cell
selection to the conclude that, our proposed schemes are reasonably ro-
bust, performorming well in noisy environments and outpeforming the
schemes relying on the conventional association metric.
To conclude, we have confirmed the applicability of the preamble se-
quence as an implicit information indicating tool and stations can take
advantage of this in selecting a more appropriate BS, thereby maintaining
resource and throughput fairness across the network.
Chapter 8

Future Work

Probing deeper, the results in this thesis provide a strong foundation for
future work in the area of using of preamble sequences for load balanc-
ing in cellular networks. Some of the areas that can be considered for for
future work are as follows:

• Developing new preamble sequence structures that will be able to

accomodate and minimize degradation of the preamble sequence
in terms of detection when we try to concatenate repeated PN se-
quences to form premables, based on the proposed auto-correlation
scheme.

• There is a significant amount of computation involved in the cross-

correlation operation requiring much computation power. There-
fore, developing methods which are less computationally complex
will be a great benefit as they will improve the speed of cell-selection
decision making and handovers.

• Lastly, it will be very interesting to see how the schemes perform in

different topologies with a change in the scalability of the system,
which will give a better insight into the behaviour in new dimen-
sions.

71
72 CHAPTER 8. FUTURE WORK
Bibliography

[1] V . H. Mac Donald, ”The Cellular Concept,” Bell Sys. Tech. J, vol. 58,
no.1, pp. 15.45, Jan. 1979.

[2] T. Rapport,”Wireless Communications: Principles and Practice,” Pren-

tice Hall, 1996.

[3] A. S. Tanenbaum, Computer Networks, 4th edition, Prentice Hall, 2003.

[4] Ankit Chopra, Peter Sam Raj and Winston KG Seah, ”Utilizing the
Inherent Properties of Preamble Sequences for Load Balancing in Cellu-
lar Networks”, accepted by the 7th International Conference on Intelli-
gent Sensors, Sensor Networks and Information Processing (ISSNIP 2011),
Adelaide, Australia, Dec 6-9, 2011.

[5] I Katzela and M. Naghshineh, ”Channel Assignment Schemes for Cel-

lular Mobile Telecommunication Systems: A Comprehensive Survey,” IEEE
Pers. Commun., vol.3, no.3, pp.10-31, Jun 1996.

[6] J. C.- I Chuang, ”Performance Issues and Algorithms for Dynamic

Channel Assignment ,” IEEE J. Sel. Areas Commun., vol.11, no.6, pp.955-
963, Aug 1993.

73
74 BIBLIOGRAPHY

[7] M. Zhang, T.- S. P. Yum , ”Comparisons of Channel-Assignment Strate-

gies in Cellular Mobile Telephone Systems,” IEEE Trans. Vehic. Tech., vol.38,
no.4, pp.211-215, Nov 1989.

[8] T. J. Kahwa and N. D. Georganas, ”A Hybrid Channel Assignment

Scheme in Large-Scale, Cellular-Structured Mobile Communication Sys-
tems,”IEEE Trans. Commun., vol.26, no.4, pp. 432- 438, Apr 1978.

[9] S. M. Elnoubi, R. Singh, and S. C. Gupta, ”A New Frequency Chan-

nel Assignment Algorithm in High Capacity Mobile Communication Sys-
tems,” IEEE Trans. Vehic. Tech., vol.31, no.3, pp. 125- 131, Aug 1982.

[10] H. Jiang and S.S. Rappaport, ”CBWL: A New Channel Assignment

and Sharing Method for Cellular Communication Systems,” IEEE Trans.
Vehic. Tech., vol.43, no.2, pp.313-322, May 1994.

[11] S. K. Das, S. K. Sen, and R. Jayaram, ”A Novel Load Balancing Scheme

for the Tele-Traffic Hot Spot Problem in Cellular Networks.”Wireless Net-
works, vol. 4, no. 4, pp. 325-340, Jul 1998.

[12] B. Eklundh,”Channel Utilization and Blocking Probability in a Cellu-

lar Mobile Telephone System with Directed Retry,” IEEE Trans. Commun.,
vol.34, no.4, pp. 329- 337, Apr 1986.

[13] J. Karlsson and B. Eklundh, ”A Cellular Mobile Telephone System

with Load Sharing Enhancement of Directed Retry,” IEEE Trans. Commun.,
vol.37, no.5, pp.530-535, May 1989.

[14] S. S. Rappaport and L.-R. Hu, ”Microcellular Communication Sys-

tems with Hierarchical Macrocell Overlays: Traffic Performance Models
and Analysis,” in Proc. IEEE, vol.82, no. 9, pp. 1383-1397, Sep. 1994.
BIBLIOGRAPHY 75

[15] I. Papanikos and M. Logothetis, ”A Study on Dynamic Load Balance

for IEEE 802.11b Wireless LAN” In Proc. COMCON, Crete, Greece, pages
1-7, 14-17 June, 2001.

[16] I. Tinnirello and G. Bianchi,”A Simulation Study of Load Balancing

Algorithms in Cellular Packet Network” In Proc. MSWIM, pages 73-78,
Rome, Italy, 16-21 July, 2001.

[17] A. Balachandran, P. Bahl, and G. M. Voelker, ”Hot-spot Congestion

Relief and Service Guarantees in Public-Area Wireless Networks”, ACM
SIGCOMM Computer Communication Review., vol. 32, no.1, page 59, Jan-
uary, 2002.

[18] H. Velayos, V. Aleo and G. Karlsson,” Load balancing in overlapping

wireless LAN cells”, In Proc. ICC, Vol. 7, page 3833- 3836, 20-24 June, 2004.

[19] A. Kumar and V. Kumar, ”Optimal Association of Stations and APs

in an IEEE 802.11 WLAN ” In Proc. MSWIM, vol.1, page 1-5, Kharagpur,
India, 28-30 January, 2005.

[20] T-C. Tsai and C-F. Lien., ”IEEE 802.11 Hot Spot Load Balance and QoS
Maintained Seamless Roaming”, In Proc. NCS, page 1-8, Taiwan, R.O.C,
24-26 November, 2003.

[21] Y. Bejerano S-J. Han and L. E. Li. ”Fairness and Load Balancing in
Wireless LANs Using Association Control”, In Proc. MOBICOM, pages
315-329, Philadelphia, PA, USA, 26 September- 01 October, 2004.

[22] Y. D. Lin and Y. C. Hsu, ”Multihop cellular: A new architecture for

wireless communication”, in Proc. IEEE INFOCOM, page 1273- 1282, Tel-
76 BIBLIOGRAPHY

Aviv, Israel, 26-20 March, 2000.

[23] X. Wu, B. Mukherjee, and S. -H. G. Chan, ”MACA - An Efficient Chan-

nel Allocation Scheme in Cellular Networks,” in Proc. IEEE GLOBECOM,
vol. 3, pp. 1385-1389, San Francisco, California, USA, 27 November- 1 De-
cemeber, 2000.

[24] H. Luo, R. Ramjee, P. Sinha, L. Li, and S. Lu, ”UCAN: A Unified Cellu-
lar and Ad-hoc Network Architecture,” in Proc. MOBICOM, pp. 353-367,
San Diego, California, USA , 14-19 September, 2003.

[25] G. N. Aggelou and R. Tafazolli, ”On the Relaying Capability of Next-

Generation GSM Cellular Networks,”IEEE Pers. Comm., pp. 40-47, Feb.
2001.

[26] R. Yates and C.-Y. Huang, Integrated Power Control and Base Station
Assignment, IEEE Trans. on Veh. Technol., vol. 44, no. 3, pp. 638-644, Aug.
1995.

[27] S. V. Hanly, ”An Algorithm for Combined Cell-Site Selection and Power
Control to Maximize Cellular Spread Spectrum Capacity,” IEEE J. Select.
Areas in Comms., vol. 13, no. 7, pp.1332-1340, Sep 1995.

[28] Aimin Sang, et al., ”Coordinated Load Balancing, Handoff/Cell-site

Selection and Scheduling in Multi-Cell Packet Data Systems,” in Proc. ACM
MobiCom, Philadelphia, PA, USA, Sep 26-Oct 1, 2004.

[29] S. Das, et al. ”Dynamic Load Balancing through Coordinated Schedul-

ing in Packet Data Systems,” in Proc. IEEE INFOCOM, San Francisco, CA,
USA, Apr 1-3, 2003.
BIBLIOGRAPHY 77

[30] Minsoo Na, Youngkyu Choi, Dongmyoung Kim, Kwanghun Han, and
Sunghyun Choi, ”On Utilizing Preamble Code Sequence as Implicit Load
Indicator,” in Proc. IEEE APWCS, Sendai, Japan, Aug 21-22, 2008.

[31] IEEE 802.16e-2006, ”Part 16: Air Interface for Fixed and Mobile Broad-
band Wireless Access Systems-Amendment for Physical and Medium Ac-
cess Control Layers for Combined Fixed and Mobile Operation in Licensed
Bands,” IEEE std., Feb. 2006.

[32] Byeong Gi Lee and Sunghyun Choi, Broadband Wireless Access & Local
Networks: Mobile WiMAX and WiFi, Aretech House, 2008.

[33] Peng Cheng, et al., ”A study on cell search algorithms for IEEE 802.16e
OFDMA systems,” in Proc. IEEE WCNC, Hong Kong, Mar 11-15, 2007.

[34] Shannon, et al.,”Communication in the Presence of Noise,” in Proc.

IEEE, Sept ,1984.

[35] De S, Tonguz O, Wu H, Qiao C.”Integrated Cellular and Ad Hoc Relay

(iCAR) Systems pushing the performance limits of conventional wireless
networks,” In Proc. IEEE HICSS, 3899–3906, January 2002.

[36] Zhou J and Yang Y. PARCelS,”Pervasive Ad-Hoc Relaying for Cellular

Systems”, in Proc. Med-Hoc-Net, page 1-7, Sardegna, Italy, 4-6 September
2002.

[37] T. Togo, I. Yoshii and R. Kohno, ”Dynamic Cell-Size Control Accord-

ing to Geographical Mobile Distribution in a DS/CDMA Cellular Sys-
tem”, In Proc. IEEE PIMRC, Vol. 2, pages 677-681, Boston, MA, USA, 8-11
September 1998.
78 BIBLIOGRAPHY

[38] A. Jalali,”On Cell Breathing in CDMA Networks”,In Proc. IEEE ICC,

Vol. 2, pages 985 - 988, Atlanta, Georgia, USA, 1 June 1998.

[39] Jerry D. Gibson, ”The Mobile Communications Handbook”, chap. 8,

1999.

[40] Bracewell, R. ”The Autocorrelation Function.” The Fourier Transform

and Its Applications. New York: McGraw-Hill, pp. 40-45, 1965.

[41] Asrar U. H. Sheikh, ”Wireless communications: Theory and Tech-

niques”, chap. 10, 2004.

[42] R. Jain, D. Chiu, and W. Hawe, ”A Quantitative Measure Of Fairness

And Discrimination For Resource Allocation In Shared Computer Sys-
tems”, DEC Research Report TR-301, September 1984.

[43] B. Radunovic and J.-Y. L. Boudec, ”A Unified Framework for Max-Min

and Min-Max Fairness with Applications,” in Proc. Allerton, Monticello,
Illinois, USA, 3 October, 2002.

[44] H. J. Kushnerand P. A. Whiting, ”Convergence of Proportional-Fair

Sharing Algorithms under General Conditions”, IEEE Trans. Wireless Com-
mun., vol.3, no. 4, pp. 12501259, Jul 2004.

[45] D. Julian, M. Chiang, D. O’Neill, and S. Boyd, ”QoS and Fairness Con-
starined Convex Optimization of Resource Allocation for Wireless Cellular
and Ad-hoc Networks,” in Proc. of IEEE INFOCOM, vol. 2, pp. 477-486,
San Francisco, California USA, 30 March, 2003.

[46] H. J. Zepernick and a. Finger, ”Pseudo Random Signal Processing” ,

John Wiley and Sons Ltd., Chichester, UK, 2005.
BIBLIOGRAPHY 79

[47] A. Balachandran, G.M. Voelker, P. Bahl, and P.V. Rangan, Character-

izing User Behavior and Network Performance in a Public Wireless LAN,
ACM SIGMETRICS 2002, June 2002.

[48] D. Kotz and K. Essien, Characterizing Usage of a Campus-wide Wire-

less Network, Technical Report TR2002-423, Dartmouth College, March
2002.

[49] Seybold, John S., ”Introduction to RF propagation” ,John Wiley and

Sons, 2005.