You are on page 1of 108

A Spatial Decision Support System for Planning

Broadband, Fixed Wireless Telecommunication Networks



by

Kevin Paul Scheibe




Dissertation submitted to the Faculty of the
Virginia Polytechnic Institute and State University
in partial fulfillment of the requirements for the degree of



Doctor of Philosophy
in
Business




Dr. Loren Paul Rees, Chairman

Dr. Laurence J. Moore

Dr. Cliff T. Ragsdale

Dr. Terry R. Rakes

Dr. Christopher W. Zobel




April 4, 2003
Blacksburg, Virginia

Keywords: Spatial Decision Support Systems, Broadband Wireless Network Planning

Copyright by Kevin Paul Scheibe
ALL RIGHTS RESERVED
A Spatial Decision Support System for Planning Broadband,
Fixed Wireless Telecommunication Networks

Kevin Paul Scheibe
(ABSTRACT)

Over the last two decades, wireless technology has become ubiquitous in the United
States and other developed countries. Consumer devices such as AM/FM radios, cordless and
cellular telephones, pagers, satellite televisions, garage door openers, and television channel
changers are just some of the applications of wireless technology. More recently, wireless
computer networking has seen increasing employment. A few reasons for this move toward
wireless networking are improved electronics transmitters and receivers, reduced costs,
simplified installation, and enhanced network expandability.
The objective of the study is to generate understanding of the planning inherent in a
broadband, fixed wireless telecommunication network and to implement that knowledge into an
SDSS. Intermediate steps toward this goal include solutions to both fixed wireless point-to-
multipoint (PMP) and fixed wireless mesh networks, which are developed and incorporated into
the SDSS.
This study explores the use of a Spatial Decision Support System (SDSS) for broadband
fixed wireless connectivity to solve the wireless network planning problem. The spatial
component of the DSS is a Geographic Information System (GIS), which displays visibility for
specific tower locations. The SDSS proposed here incorporates cost, revenue, and performance
capabilities of a wireless technology applied to a given area. It encompasses cost and range
capabilities of wireless equipment, the customers’ propensity to pay, the market penetration of a
given service offering, the topology of the area in which the wireless service is proffered, and
signal obstructions due to local geography.
This research is both quantitative and qualitative in nature. Quantitatively, the wireless
network planning problem may be formulated as integer programming problems (IP). The line-
of-sight restriction imposed by several extant wireless technologies necessitates the incorporation
of a GIS and the development of an SDSS to facilitate the symbiosis of the mathematics and
geography.
The qualitative aspect of this research involves the consideration of planning guidelines
for the general wireless planning problem. Methodologically, this requires a synthesis of the
literature and insights gathered from using the SDSS above in a “what-if” mode.


Dedication

I dedicate this dissertation to my beautiful wife and family, without whom I could not
have accomplished any of this.

iii
Acknowledgements

I would first like to thank my Lord and Savior, Jesus Christ, through whom all things are
possible – especially this research. Thank you to my dissertation chairman, Professor Loren Paul
Rees. I am extremely grateful to have been able to work with you, and I truly appreciate your
mentoring and tutelage. You have been instrumental in my academic development.
The guidance and support of my committee members are also very much appreciated.
Dr. Laurence J. Moore for your direction, wisdom and many great books, Dr. Cliff T. Ragsdale
for your friendship, analytical mind, and immortalizing me in one of the examples of your book,
Dr. Terry R. Rakes for your insight and ideas, and Dr. Christopher W. Zobel for your friendship,
encouragement, and great racquetball games.
Thank you to Dr. William L. Carstensen and George E. Morgan for your helpful
guidance in the areas of wireless communications and geographic information systems.
Finally, I want to thank my wife and best friend, Mary. I cannot even begin to express
how thankful I am for you. You have supported me in countless ways, and I thank God for you
every day. I love you dearly.

iv
Table of Contents

Chapter 1:

Introduction..................................................................................................................................... 1
Broadband, Fixed Wireless Telecommunication Networks ....................................................... 1
Planning ...................................................................................................................................... 3
Spatial Decision Support Systems .............................................................................................. 3
Statement of the Problem............................................................................................................ 4
Objective of the Study ................................................................................................................ 5
Research Methodology ............................................................................................................... 6
Scope and Limitations................................................................................................................. 6
Contributions of the Research..................................................................................................... 6
Unification of Chapters............................................................................................................... 6
Plan of Presentation .................................................................................................................... 7

Chapter 2:

LITERATURE REVIEW............................................................................................................ 8
BROADBAND TELECOMMUNICATION NETWORKS .......................................................... 9
Fixed Wireless Networks............................................................................................................ 9
PLANNING.................................................................................................................................. 10
SOLUTION METHODOLOGIES ............................................................................................... 11
Mathematical Programming...................................................................................................... 11
Fixed-Charge Network Flow Models ....................................................................................... 11
Decision Support Systems ........................................................................................................ 12
Geographic Information Systems ............................................................................................. 13
Spatial Decision Support Systems ............................................................................................ 13

v
Chapter 3:

A MATHEMATICAL-PROGRAMMING AND GEOGRAPHIC-INFORMATION-
SYSTEM FRAMEWORK FOR WIRELESS BROADBAND DEPLOYMENT IN
RURAL AREAS.......................................................................................................................... 16
ABSTRACT.................................................................................................................................. 17
MOTIVATION............................................................................................................................. 18
BACKGROUND.......................................................................................................................... 19
Wireless Telecommunication.................................................................................................... 19
Different Wireless Systems................................................................................................... 19
Costs...................................................................................................................................... 20
Differences Among Wireless Systems.................................................................................. 20
Geographic Information Systems and Decision Support Systems ........................................... 20
(NEAR) LINE-OF-SIGHT OPTIMIZATION MODELS ............................................................ 21
Notation..................................................................................................................................... 22
View Sheds ............................................................................................................................... 23
Pedagogical Mini-Example....................................................................................................... 23
Profit Model .......................................................................................................................... 24
Pedagogical Example............................................................................................................ 25
Maximizing Exposure Model ............................................................................................... 26
Additional Constraints .......................................................................................................... 26
A Spatial Decision Support System.......................................................................................... 26
REACHING THE LAST MILE ................................................................................................... 27
GETWEBS Data ....................................................................................................................... 27
Wireless Parameters and Costs ................................................................................................. 27
Range .................................................................................................................................... 28
Tower Costs .......................................................................................................................... 28
Customer Premise Equipment Cost ...................................................................................... 28
Revenue Model ..................................................................................................................... 28
Results....................................................................................................................................... 28
LMDS at $50 per month ....................................................................................................... 29
802.16 (WirelessMAN) at $50 and $100 per month............................................................. 29
Wi-Fi at $50 per month......................................................................................................... 29
DISCUSSION............................................................................................................................... 29
CONCLUSIONS........................................................................................................................... 30
REFERENCES ............................................................................................................................. 31

vi
Chapter 4:

A CAPACITATED, FIXED-CHARGE, NETWORK-FLOW MODEL FOR
SOLVING HOP CONSTRAINED, WIRELESS MESH NETWORKS................................ 43
ABSTRACT.................................................................................................................................. 44
INTRODUCTION........................................................................................................................ 45
BACKGROUND.......................................................................................................................... 46
Mesh Networks ......................................................................................................................... 46
Fixed-Charged Network Flow Models ..................................................................................... 47
METHODOLOGY ....................................................................................................................... 48
Preprocessing............................................................................................................................ 48
Step 1 –Potential Equipment Placement and Line-of-Sight Determination ......................... 48
Step 2 – Artificial Directional Node Determination............................................................. 49
Network Flow Model................................................................................................................ 49
EXAMPLE.................................................................................................................................... 50
IMPLICATIONS .......................................................................................................................... 51
CONCLUSIONS AND FUTURE WORK................................................................................... 51
REFERENCES ............................................................................................................................. 53

v ii
Chapter 5:

ADDRESSING UNIVERSAL-BROADBAND-SERVICE IMPLICATIONS WITH
WIRELESS MESH NETWORKS ............................................................................................ 62
ABSTRACT.................................................................................................................................. 63
INTRODUCTION........................................................................................................................ 64
BACKGROUND.......................................................................................................................... 65
Technology Options.................................................................................................................. 65
LMDS ................................................................................................................................... 66
MMDS .................................................................................................................................. 66
Wi-Fi ..................................................................................................................................... 66
Free Space Optics (FSO) ...................................................................................................... 66
Network Configuration Options ............................................................................................... 67
Point to Point (PtP) ............................................................................................................... 67
Point to Multipoint (PMP) .................................................................................................... 67
Mesh...................................................................................................................................... 68
Mesh Network Terminology..................................................................................................... 68
WIRELESS MESH PLANNING METHODOLOGY................................................................. 69
Step 1 – Potential Equipment Placement and Line-of-Sight Determination ............................ 69
Step 2 – Feasible Path Enumeration ......................................................................................... 70
Step 3 – Network-Flow Model Formulation............................................................................. 70
Step 4 – Solution Generation.................................................................................................... 70
EXPLORATORY STUDY........................................................................................................... 70
Random Problem Generation.................................................................................................... 70
MANAGERIAL IMPLICATIONS .............................................................................................. 71
Effect of Customer Density ...................................................................................................... 72
Effect of Obstructions............................................................................................................... 72
Effect of Quality of Service: Hops............................................................................................ 73
Effect of Quality of Service: Bandwidth .................................................................................. 73
Effect of Cost: Bandwidth ........................................................................................................ 74
Cost Reductions Necessary to Achieve Profitability................................................................ 74
CONCLUSIONS........................................................................................................................... 74
REFERENCES ............................................................................................................................. 75
APPENDIX I: THE WIRELESS MESH PROBLEM MODEL................................................... 77
APPENDIX II: PROBLEMATIC CORNERS ............................................................................. 78

v iii
Chapter 6:

A GENERAL FRAMEWORK FOR PLANNING BROADBAND, FIXED
WIRELESS TELECOMMUNICATION NETWORKS ........................................................ 84
ABSTRACT.................................................................................................................................. 85
INTRODUCTION........................................................................................................................ 86
BACKGROUND.......................................................................................................................... 87
SDSS......................................................................................................................................... 87
Network Topologies.................................................................................................................. 87
Technology Issues..................................................................................................................... 87
METHODOLOGY ....................................................................................................................... 87
Database/GIS ............................................................................................................................ 87
For the geographic area under consideration:....................................................................... 87
For each viable technology: .................................................................................................. 87
Model Base ............................................................................................................................... 87
Marketing models ................................................................................................................. 88
Decision models.................................................................................................................... 88
Helper models ....................................................................................................................... 88
GUI Base................................................................................................................................... 88
Possible Synthesis/Flow of Above Factors in the Planning Methodology............................... 88
Possible Products/Outcomes/Outputs of the Methodology ...................................................... 89
Charts/Graphs ....................................................................................................................... 89
Rules ..................................................................................................................................... 89
Decision Trees ...................................................................................................................... 89
EXAMPLES ................................................................................................................................. 90
MANAGERIAL IMPLICATIONS .............................................................................................. 90
CONCLUSIONS AND FUTURE WORK................................................................................... 90

REFERENCES............................................................................................................................ 91

VITA............................................................................................................................................. 96


ix
List of Figures

Figure 2.1. Virginia Tech Center for Wireless Telecommunications BTA Regions shown
shaded on a map of Virginia (United States) ........................................................ 15
Figure 3.1. A GETWEBS screen shot indicating the view shed (in magenta) of a
tower placed at the green dot near the center of the top of the screen
and with range indicated by the green circle. ....................................................... 33
Figure 3.2a. A fictitious town, divided into an east and west region by a railroad
track that runs north/south through the center of town. ........................................ 34
Figure 3.2b. An elevation map of the fictitious town................................................................ 34
Figure 3.2c. A transmission tower’s signal can reach diagonally to all households
two cells away....................................................................................................... 34
Figure 3.3. The exposure matrix E and the view shed matrix V for the fictitious
town depicted in Figure 3.2. ................................................................................. 35
Figure 3.4. A fictitious mini-town, divided into six cells........................................................ 36
Figure 3.5. The wireless spatial DSS architecture................................................................... 37
Figure 3.6. Cells may be individually excluded from (red) or included (blue) for
tower placement as desired. .................................................................................. 38
Figure 3.7. Tower placement and profit sheds (magenta) for Wi-Fi (802.11b)
service offered at $50/month. ............................................................................... 39
Figure 3.8. Profit, households, and towers in Montgomery County as a function
of percent coverage with Wi-Fi broadband. ......................................................... 40
Figure 4.1. Providing broadband service to a neighborhood with a wireless mesh
network. ................................................................................................................ 55
Figure 4.2. Nodes represent potential wireless mesh equipment locations............................. 56
Figure 4.3. Nodes shown linked together with line-of-sight capabilities................................ 56
Figure 4.4. Viable paths for customer C1 to communicate with the backhaul,
given the line-of-sight configuration of Figure 4.3............................................... 57
Figure 4.5. Customer C1 viable paths expressed in terms of artificial directional nodes. ...... 57
Figure 4.6. Re-linking nodes to their artificial directional nodes............................................ 58
Figure 5.5. Effect of Customer Density .................................................................................. 72
Figure 5.6. Visibility ............................................................................................................... 72
Figure 5.7. Quality of Service (latency) .................................................................................. 73
Figure 5.8. Bandwidth Offerings............................................................................................. 73
Figure 5.9. Bandwidth Cost..................................................................................................... 74
Figure 5.1. Providing broadband service to a neighborhood with a wireless
mesh network. ....................................................................................................... 80
Figure 5.2. Nodes represent potential wireless mesh equipment locations............................. 81
Figure 5.3. Nodes shown linked together with line-of-sight capabilities................................ 81
Figure 5.4. Viable paths for customer C1 to communicate with the backhaul,
given the line-of-sight configuration of Figure 5.3............................................... 82


x
List of Tables

Table 2.1. Some wireless communications examples............................................................ 14
Table 3.1. Some wireless applications and their allocated frequencies. ................................ 41
Table 3.2. The percentage participation (in Montgomery County) as a
function of average household annual income, given the price
per month for wireless service. ............................................................................. 42
Table 4.1. Source and neighbor nodes ................................................................................... 59
Table 4.2. Viable paths for customer C1 to communicate with the backhaul ....................... 59
Table 4.3. Viable paths for node C1 with artificial directional nodes ................................... 59
Table 4.4. Source and immediate destination nodes with artificial directional
nodes included ...................................................................................................... 60
Table 4.5. Re-linking nodes to their artificial directional nodes............................................ 60
Table 4.6. List of sources and sinks for direct inclusion into the network flow model ......... 61
Table 4.7. Some Results from the Example’s Preprocessing Step ........................................ 61
Table 4.8. Results from the Example’s Capacitated, Fixed-Charge,
Network-Flow Problem........................................................................................ 61
Table 5.1. “Typical” Wireless Assumptions.......................................................................... 83
Table 5.2. Factor Levels......................................................................................................... 83


xi
CHAPTER 1:
INTRODUCTION
Over the last two decades, wireless technology has become ubiquitous in the United
States and other developed countries. Consumer devices such as AM/FM radios, CB radios,
cordless and cellular telephones, pagers, satellite televisions, car alarm signalers, garage door
openers, and television channel changers are just some of the applications of wireless
technology. More recently, wireless computer networking has seen increasing employment,
including applications meeting the Institute of Electrical and Electronic Engineers (IEEE) 802.11
standard for local area networks (LAN) and local multipoint distribution service (LMDS),
another wireless technology using higher frequencies and providing greater data transfer rates. A
few reasons for this move toward wireless networking are improved electronics transmitters and
receivers, reduced costs, simplified installation, and enhanced network expandability.
In an op-ed piece in the New York Times, Kornbluh (2001, p. 21) describes the hundreds
of thousands of miles of fiber-optic cable buried in the United States as the “digital equivalent of
fallow farmland.” She points out that although investors plowed $90 billion into a cross-
continental fiber-optic broadband network, today merely 3% of that backbone is in use. The
problem, she states, is that entrepreneurs failed to foresee the enormous cost of upgrading the
“last mile – copper telephone wires that connect individual homes and small businesses to the
broadband backbone” (p. 21). She then encourages both Congress and the president to subsidize
sparsely populated regions of the country, low-income users or both, suggesting that an
ambitious broadband strategy can help revitalize the economy.
One popular alternative strategy to “going the last mile” with copper is to provide
wireless services, whereby the signal is transmitted from a tower through various alternative
mechanisms to equipment at business and residential customer sites. Willebrand and Ghuman
(2001) cite an example where a fiber cost of $400,000 was reduced to $60,000 with a wireless
system. Advances in the technologies of wireless systems provide new opportunities for service
providers. Nokia has recently announced a national initiative to bring broadband wireless
connectivity to business and residential customers via their Nokia RoofTop solution
(BusinessWire, 2002).
Broadband, Fixed Wireless Telecommunication Networks
Comer (3
rd
edition, p. 603) defines “broadband technology” as the term to describe a
networking technology that uses a large part of the electromagnetic spectrum to achieve higher
throughput rates. Cable TV, for example, uses broadband transmission. This is as contrasted
with “baseband technology,” the term used to describe a networking technology that uses a small
part of the electromagnetic spectrum and sends only one signal at a time over the underlying
medium (Comer, p.602). What constitutes a “large part” or a “small part” of the electromagnetic
spectrum means different things to different people. For example, according to Ben Macklin, a
broadband analyst at eMarketer, broadband means: a downstream connection of 256 kilobits per
second (kbps) and higher to Jupiter Communications; access speeds greater than 144 kbps, and
200 kbps in at least one direction to the US Federal Communications Commission.
Fundamentally, however, broadband has come to mean fast connectivity whether by a wired or
wireless medium (Macklin, 2001).
Chapter 1: Introduction 1
Under the umbrella of broadband wireless communications, there are several different
technologies. The distinguishing factor separating the technologies is frequency. Varying
spectral frequencies will determine the distance of transmission, whether the transmission needs
to be line of sight, and whether the FCC requires licensing. Wireless communications use
frequencies ranging from radio level to optical.
Two general categories of wireless connectivity exist: fixed wireless and portable
wireless. Portable wireless refers to devices such as pagers, cellular phones, and personal digital
assistants (PDA), where there are fixed towers but the receivers are mobile. Fixed wireless
means that both the sending and receiving components of the network are situated in fixed
locations and do not move without reconfiguring the network. The focus of this research is on
fixed wireless connections.
There are three different types of connection methodologies in fixed wireless
communication. The first is point-to-point (PTP). This is the oldest of the three methods and is
most commonly used in campus environments, where it is desirable to connect one point to
another without the expense of laying cable. Of the three, PTP typically offers the greatest
transmission range. The second method is point-to-multipoint (PMP). This method has a single
transmitting tower and multiple receivers. Instead of sending a narrow beam to a single receiver,
the transmitter has a broader beam reaching multiple receivers and, consequently a shorter
distance capability from the transmitter to the receivers than PTP. With higher frequency
wireless methods, every receiver must be visible to the transmitter. The third fixed wireless
method is Mesh, where every receiver is also a transmitter. A mesh network will have a so-
called “insertion point,” which acts as a transmitter, and is connected to the network backbone.
The other nodes in the mesh network act as receivers and as re-transmitters, capable of passing
signal along to other nearby nodes. With a mesh, it is thus unnecessary for every node to be
visible to the insertion point, so long as a path exists from a node to the backbone via other
nodes.
Some current fixed wireless technologies that are one or more of the three connection
methodologies are free space optics (FSO), Local Multipoint Distribution Services (LMDS),
Multipoint Microwave Distribution Services (MMDS), and the IEEE 802.11 standards for
wireless connectivity (See Table 1.). Free space optics (FSO), also known as "open-air
photonics," "optical wireless" or "infrared broadband," transmits data using low-powered
infrared lasers. FSO do not currently require FCC licensing; however, certain power restrictions
must be observed. FSO is a line-of-site (LOS) technology, and to remain in the unlicensed
territory, the laser must not be too highly powered. Consequently, the maximum range of FSO is
a few kilometers. Furthermore, fog can corrupt the transmission as water particles act as prisms
to the laser and dissipate the light beam. The data transfer rate of FSO is between 155 and 622
Megabits per second (Mbps).
Local Multipoint Distribution Services is a PMP service. It operates in the 28 GHz band
and is LOS. The max range of LMDS is approximately five kilometers and can transfer data up
to 2 Gigabits per second (Gbps); however, LMDS behaves more reliably when transferring data
in the Mbps range.
Multipoint Microwave Distribution Services are also called Multi-channel Multipoint
Distribution Systems and wireless cable. MMDS uses both unlicensed and licensed channels.
This technology uses multiple channels simultaneously and, therefore, by aggregation creates
large pathways between the sender and receiver. MMDS can transfer data on the unlicensed
channels up to 27 Mbps and up to 1 Gbps on licensed channels. MMDS is also LOS.
Chapter 1: Introduction 2
In the 1990s, IEEE adopted the 802.11 standard for wireless Ethernet. In 1999, the
802.11b standard was approved. 802.11b, also known as Wi-Fi (for wireless fidelity), uses the
unlicensed portion of the spectrum at 2.4GHz. Wi-Fi was originally intended for local area
networks (LAN), but is now also used in metropolitan area networks (MAN). The reported
range of 802.11b is approximately 100 meters, seemingly insufficient for larger scale
connectivity, but it is possible to connect points over greater distances with the same technology.
In fact, Wi-Fi networks are being implemented with distances of up to 15 kilometers (Carstensen
and Morgan, 2002), and data transfer rates up to 11 Mbps. This Wi-Fi frequency band is part of
the Industrial, Scientific, and Medical (ISM) bands. These bands are unlicensed and are usually
used for household appliances such as microwaves to Wi-Fi routers. Another family of bands is
the Unlicensed National Information Infrastructure (UNII) band. UNII provides 300MHz of
bandwidth in the 5GHz range. 802.11a uses the UNII bands and provides 54 Mbps data transfer
rate.
Planning
One common meaning of planning used in the literature is to understand the necessity of
building a system (Dennis, Wixom, and Tegarden, 2002). This is not the meaning used in this
research, however. Rather, this work uses a definition of planning closer to that used in artificial
intelligence (AI). AI planning systems assume an initial (current) state, a goal state, and a set of
allowable actions. The planning system’s purpose is to specify the actions that will take the
system from its initial state to its goal state, if possible (Rolston, 1988; Fikes, Hart, and Nilsson,
1972). Although planning as used here will not be utilized in a formal way to derive a sequence
of actions, still the general sense of the term here will be to specify those activities that will meet
a service provider’s goals of maximizing profit or minimizing cost within the context of
geographic, financial, and other constraints.
In general, this dissertation will not generate long-range plans (say 10 to 20 years), but
rather more intermediate plans in the context of three to ten years. Plans will include the choice
of wireless technology plus the time sequencing of sending and receiving antenna placement.
Spatial Decision Support Systems
Decision Support Systems (DSS), a branch of information systems, has been a topic of
ongoing research for the past 30 years. They used data, models, and user interface components
to help decision makers solve semi-structured or unstructured problems (Bennett, 1983).
Unstructured decisions are defined as those for which no algorithm can be written, whereas
algorithms can be specified for structured decisions. Semi-structured decisions fall between the
other two (Keen and Scott-Morton, 1978). Keen and Scott-Morton (1978), Bennett (1983), or
Turban (2001) are excellent resources for more information on DSS.
Geographic Information Systems (GIS) are a type of relational database management
system used to collect, store, retrieve, and analyze spatial data. GIS link tabular data to graphical
data by relating graphical layers to database tables. GIS are widely used to aid decision makers
in solving spatial problems. GIS’ capabilities of spatial visualization often simplify difficult
problems. For example, fast food chains considering new locations can use GIS to determine
where their nearest competitors are located. GIS may also be used to show demographics such
as residents’ income. Furthermore, the GIS can show the road network so the decision maker
can determine possible traffic near their store. Any of these pieces of information may be critical
in determining the financial success of the new store. The capabilities of the GIS, taken in
Chapter 1: Introduction 3
concert with the knowledge of the decision maker, furnish a tremendous capability in solving
this semi-structured spatial problem.
Marrying DSS and GIS creates Spatial Decision Support Systems (SDSS). The
advantage of an SDSS is it is able to integrate the model portion of the DSS with the graphical
representation of the GIS, thereby aiding decision makers with semi-structured or unstructured
spatial problems. In their research, Crossland, Wynne, and Perkins (1995) determined that SDSS
enabled decision makers to complete their tasks quicker, more efficiently, and with greater
understanding of the problem through visualization.
SDSS is a relatively new research area, primarily because GIS software has historically
needed a great amount of computing power, memory, and hard disk space. Since such
equipment was very expensive, GIS software required large budgets. Because computers have
become significantly cheaper and more powerful, GIS packages can now run on desktop
computers, which creates more SDSS research possibilities. SDSS have been applied to siting
problems (Maniezzo, Mendes, and Paruccini, 1988; Vlachopoulou, Silleos, and Manthou, 2001),
land planning (Nehme and Simoen, 1999), and vehicle routing (Keenan, 1998; Tarantilis and
Kiranoudis, 2002).
Statement of the Problem
Since the “last mile” of wiring has become the prohibitive factor for connectivity,
wireless solutions are being aggressively pursued. However, each of the wireless technologies
introduced earlier has inherent strengths and weaknesses, and it is necessary to overcome or
compensate for the weakness of a specific technology in order to make it a viable method of
reaching customers. Technologies that use relatively lower wireless frequencies are not as
restricted by LOS problems as higher frequencies, but they are limited to short transmission
distances. Furthermore, as more people use these wireless technologies, bandwidth will fill
causing other problems such as interference. The situation will possibly become similar to the
days when garage door openers first became popular, and it was possible to drive around town
and use one remote to open other garages. Conversely, fixed wireless technologies that use
relatively higher frequency bands are able to propagate data over greater distances, but are
limited to LOS. That means that they are unable to transfer through walls, mountains, trees, etc.
The LOS constraint requires careful planning of the wireless network. For customers to be
included in a wireless network, they must maintain a clear path between themselves and a signal
propagator.
One fundamental issue with high frequency, fixed wireless, PMP technologies such as
LMDS is that while it may be possible to calculate the physical coverage of a tower,
transmission may not be economical. Providers must still determine whether that coverage is
profitable. For example, in an area of potential wireless customers under a realistic scenario that
is capitally constrained, providers must determine where towers should be placed to maximize
profit or alternatively to minimize cost. The solution may or may not contain locations that
reach the most customers. This may be true if a minority of customers has the greatest
propensity to pay for services provided, and the majority of customers are only willing to pay a
minimal amount. Carstensen, Bostian, and Morgan (2001) defined the physical area that is
visible from a tower as a view shed; part of this research determines that physical region
providing profitable coverage from a tower, and labels it a profit shed. Since different tower
placements will lead to varying levels of profitability, providers will want to determine the profit
sheds yielding the greatest financial return. Factors affecting the optimal profit shed include the
Chapter 1: Introduction 4
cost of wireless towers, the cost of receivers, the propensity to pay of the customers reached, the
terrain of the region, and the range of a tower signal.
Mesh networks are able to work around the problem of a customer being too far away
from a single tower, or in such a location that they cannot maintain a clear line-of-sight to the
tower. By allowing each node in the network to function as a transmitter as well as a receiver,
signal is passed from one node to another – theoretically allowing more customers than a PMP
network. Factors that must be considered in planning a mesh network are the number of
allowable hops from a node to the backbone, the capacitation of arcs between nodes, the
placement of additional nodes to fill in gaps on the mesh, LOS between nodes, and the density of
the neighborhood. Mesh networks do not have profit sheds, as do the PMP networks. Instead,
there are cost incurring nodes in the network. The objective of this research is to minimize the
cost of the network by using as few cost incurring nodes while maintaining a desired level of
service.
This dissertation addresses the problem of generating overarching guidelines for a wide
range of possible broadband, fixed wireless use. In particular, factors are delineated that
differentiate the specification of present wireless technologies, and solution models are
developed that determine optimal profit sheds for PMP and optimal cost for Mesh
implementations.
Objective of the Study
The objective of this study is to explore the use of an SDSS for broadband fixed wireless
connectivity to solve the wireless network-planning problem. The spatial component of the DSS
is the locally developed (Carstensen et al, 2001) GIS tool Geographic-Engineering Tool for
Wireless: Evaluation of Broadband Systems (GETWEBS). One of the functions of the
GETWEBS program is the display of view sheds for specific tower locations. GETWEBS,
however, has no ability to calculate the profitability of a specific region, but specifies only
whether a tower signal can be received. The first SDSS proposed here incorporates cost,
revenue, and performance capabilities of a wireless technology applied to a given area. It
encompasses cost and range capabilities of a particular tower, the cost of receivers, the
propensity to pay of customers, the market penetration of a given service offering, the topology
of the area in which the wireless service is proffered, and signal obstructions due to local
geography.
A major difficulty of the wireless tower location problem is the number of possible
locations a tower can be placed; an exhaustive search over possible sites is prohibitive. One
method explored here for solving this problem is to cast it as a set-covering problem, a classical
mathematical programming problem formulation. The formulation requires integer solutions and
is NP-hard. Moreover, the set covering formulation must be augmented to incorporate
geographic constraints discovered by the GIS in order to determine profitable configurations and
regions.
The second portion of this research is to develop a SDSS that incorporates a capacitated,
fixed-charge network flow model to achieve a minimum cost for a Mesh network architecture.
Four factors – visibility, density, hops, and bandwidth – are addressed in the planning of such
networks.
The overall objective of the study is to generate understanding of the planning inherent in
broadband, fixed wireless telecommunication networks and to implement that knowledge into a
spatial decision support system. As steps along the way in obtaining this understanding,
Chapter 1: Introduction 5
solutions to both fixed wireless PMP and fixed wireless mesh connectivity are developed and
incorporated into the SDSS. A case study is undertaken to illustrate the general procedure and to
aid in refinement of the process and use of the SDSS.
Research Methodology
This research is both quantitative and qualitative in nature. Quantitatively, the wireless
network-planning problem may be formulated as integer programming problems (IP), in
particular, the so-called set covering problem and capacitated, fixed-charge network flow
problem. An IP formulation is developed for both the PMP and the mesh wireless network
problems. The line-of-sight restriction imposed by several extant wireless technologies
necessitates the incorporation of a GIS and that a spatial DSS be built to facilitate the symbiosis
of the mathematics and geography.
The problem we are dealing with we will refer to as the general wireless planning
problem, and is defined as providing wireless broadband services to residential and small home
office customers. The qualitative aspect of this research involves the generation of planning
guidelines for the general wireless planning problem. Methodologically, this requires a synthesis
of the literature and insights gathered from using the SDSS above in a “what-if” mode. A
limited case study is undertaken to support development and understanding of the guidelines.
Scope and Limitations
The general wireless planning problem is largely unexplored in the literature. To keep
the exploration manageable, this research will focus mainly on planning to ensure profitability
and economic feasibility of wireless connectivity. It will not discuss network security issues
intrinsic to wireless data transfer or wireless network protocols. In general, engineering issues in
the design of antennas or the propagation of radio signals at uncharted frequencies are also
beyond the realm of this study. Moreover, solution strategies such as parallel computation to
solve larger problems are not addressed here.
Contributions of the Research
This research makes three primary contributions.
• This dissertation provides an investigation into “going the last mile” in reaching
customers with broadband fixed wireless networks.
• The research develops mathematical models embedded in an SDSS to develop profit
sheds which are solutions to the equipment location problem for broadband, fixed PMP
wireless networks maximizing profitability.
• This study also derives a model embedded in an SDSS to minimize the cost associated
with equipment location for broadband, fixed mesh wireless networks.
Unification of Chapters
This dissertation, under the guidance and tutelage of Dr. Loren Paul Rees, has been
written as a series of four separate journal articles all under the thematic umbrella of spatial
planning of broadband wireless networks. Consequently, Chapters 3 through 5 are formatted as
journal articles and are meant to stand on their own. Furthermore, each has its own title page,
abstract, and references. Chapter 6 is another journal article, in preliminary form. References
from each chapter have been alphabetically compiled at the end of the dissertation.
Chapter 1: Introduction 6
The overall theme of the dissertation is planning broadband wireless telecommunication
networks, and while each chapter either describes a different methodology in application or
general description, they are all thematically related. Wireless technologies may be delivered by
three different network topologies, and the latter two are discussed in this dissertation. The first
approach is omitted as it only allows for connecting two customers. These topologies are
formulated (and solved) mathematically in Chapters 3 and 4. Chapter 5 develops managerial
conclusions from the latter topology. Moreover, there are some necessary factors for planning
the wireless networks that are common to all four papers. A crucial one is the utilization of a
GIS that accounts for wireless technological characteristics. The method in Chapter 3 uses such
a GIS, whereas Chapter 5 describes the entire process, but instead of using the GIS, it simulates
the information the GIS would provide. The model presented in Chapters 4 and 5 would use the
GETWEBS program for effective “real world” problem solving. Another common factor is the
semi-structured nature of wireless network planning. This factor gives cause for using a decision
support system. Depending upon the wireless methodology, some models are more appropriate
than others in determining maximum profitability or minimum cost.
Chapter 6 would use the methods proposed in the earlier chapters, which includes the
application of the GETWEBS program, to help decision makers address the larger picture of
network planning, such as which methodology(s) would be most appropriate given the
characteristics of the area of a desired wireless network. Therefore, while each chapter was
written as a separate article and contains its own introduction, literature review, methodology,
conclusions and references, they are closely related in their address of reaching the “last mile”
using a spatial decision support system and wireless telecommunication networks.
Plan of Presentation
This chapter has served as an introduction to fixed wireless communications, planning,
and spatial decision support systems (including geographic information systems). Additionally,
it has identified a need for the application of SDSS in wireless technology – specifically in the
areas of profitability, area coverage, and resolution of line-of-sight issues.
Chapter 2 is a literature review in the areas of fixed, wireless telecommunication
networks, planning, and spatial decision support systems. This review establishes the need for
computer support in the placement of antennas and receivers to maximize profit, minimize cost,
or guarantee coverage.
Chapter 3 presents an SDSS with an embedded mathematical programming model based
on the set-covering formulation in the management science literature. Two different objective
functions are considered and a detailed example using a GIS in concert with a mathematical
programming model is presented.
Where Chapter 3 proposes the set-covering formulation to solve PMP networks, Chapter
4 introduces a capacitated, fixed-charge network flow problem as a means of minimizing the cost
of wireless mesh networks. Chapter 5 provides a major study developing managerial
implications of mesh networks in rural / suburban settings a mechanism for addressing the “last-
mile” problem.
Future work is outlined in chapter 6, which summarizes the research and draws
conclusions by outlining an overarching planning methodology for fixed, wireless broadband
telecommunications. The methodology is “delivered” to the planner in a spatial decision support
system.

Chapter 1: Introduction 7
CHAPTER 2
LITERATURE REVIEW
The growth rate of networking, and specifically of the Internet, over the last twenty years
is staggering. According to Comer (2001), only two decades ago the Internet was just a research
project involving a dozen locations, but now it has grown to reach millions of people in every
populated country of the world. The Internet has created a paradigm shift in the way companies
do business, schools educate, consumers shop, people interact, and governmental agencies
operate. The impact of the Internet has been far greater than just allowing people to send email.
It has created a global community of organizations and individuals. In the United States the
Internet connects companies, schools, government, and people. It has become a rare thing for a
company not to have at least one network that is connected to the Internet. Individuals are also
connecting either by low speed methods such as modems or high speed methods such as DSL,
cable modems, satellite, or wireless. Moreover, as computers have become faster, more capable,
and less expensive, owning a PC and being “on-line” is now the norm. Additionally, consumers
are less satisfied with small bandwidth. They want faster connections, and to meet the demand
cable and telephone companies are racing to be broadband service providers (Wagner, 2002).
However, a fundamental problem exists in that cable companies or telephone companies cannot
reach many areas because their infrastructure cannot handle the level of throughput required for
broadband speeds. Companies such as PSINet and Quest have spent millions in laying a
transcontinental fiber optic backbone (PSINet, 1988; Quest, 1998), however, connecting from
homes and companies to the backbone is a much larger problem than anybody initially seemed to
realize.
The need for “last mile” connectivity has never been greater. Kornbluh (2001) goes so
far as to argue for governmental subsidization of connecting lower income or sparsely populated
areas to revitalize the economy. She states that the underestimation of the cost of connecting end
users to the backbone of the fiber-optic network has left the majority of the capacity of the
network unused.
Wireless connectivity is now a rapidly growing field with many companies racing to
establish a market niche for themselves. Since the wireless spectrum ranges from radio to laser,
there is a myriad of possible wireless solutions to the “last mile” problem. While each alternate
solution may be drastically different from another, a commonality to many, if not all, is the need
for transmitters and receivers. It is in dealing with these factors that planning is essential.
Decisions are costly. More specifically, hardware such as transmitting towers is costly and great
prudence is required in determining their placement. Appropriate or inappropriate tower
placement can mean the difference between financial success and disappointment.
This dissertation deals with the intermediate planning of broadband, fixed wireless
telecommunication networks. The term broadband refers to the utilization of a large part of the
spectrum to achieve high throughput ranges (Comer, 2001); this is in distinction to baseband
networks, where a small part of the spectrum is used and signals are sent one at a time. The term
fixed wireless means that neither transmission towers nor receiving towers are mobile. This case
is as opposed to cellular telephones, for example, whereby the receivers move around with the
customer. Planning, as it is used here is concerned not with implementation details, nor with the
electrical engineering development of improved wireless technologies; rather, it refers to
Chapter 2: Literature Review 8
determining a proper wireless infrastructure and its deployment. Furthermore, the word
intermediate in the phrase intermediate planning stipulates that the horizon is not necessarily
long-range in nature, that is, of 20 to 30 years duration. Instead the focus here is upon the three
to ten year horizon whereby the infrastructure to reach communities or businesses, the capability
to go “the last mile,” is determined. Thus, this research deals with the intermediate-range
infrastructure planning of broadband, fixed connectivity, wireless telecommunication networks.
BROADBAND TELECOMMUNICATION NETWORKS
As introduced in Chapter 1, wireless communications encompasses many different
technologies differentiated by frequency. These frequencies determine the distance of
transmission, whether the transmission needs to be line of sight, and whether the FCC requires
licensing. They range from radio level to optical. Table 2.1 shows some of the radio frequencies
and their applications.
Fixed Wireless Networks
Also mentioned in Chapter 1 are the three types of connection methodologies in fixed
wireless communication: point to point (PTP), point to multipoint (PMP), and Mesh (Willis,
Hasletad, Friisø, and Holm, 2001). Theoretically, one could use any of the connection
methodologies at any of the frequencies to bridge the “last mile.” Practical issues, however,
limit the matrix of possibilities. Factors such as the frequencies available, licensing, and
transmission properties reduce the realistic combinations to a few. This research examines those
frequency-connectivity choices that are being considered now or are on the horizon (Nokia,
2002; Radiant, 2001; VTCWT, 2002).
The primary focus of this research is on the Wi-Fi (802.11b) and LMDS systems.
802.11b systems operate in the 2.4 GHz range, are near line-of-sight (LOS) operation and are
license free, whereas LMDS frequencies are around 28-31 GHz, require FCC licensing, and are
strongly LOS. Near LOS means that signal may pass through some obstructions such as foliage,
but cannot pass through mountains and such. The main advantages of LMDS over Wi-Fi are
bandwidth and coverage distance. LMDS has a bandwidth that is more than twice that of
AM/FM Radio, VHF/UHF Television and cellular telephones combined (VTCWT, 2002).
LMDS is also able to transmit a signal over several miles versus Wi-Fi’s range of about 300 feet
(Gibbs, 2001). The advantages of Wi-Fi over LMDS are that no FCC licensing is required and
Wi-Fi is near-LOS, and, therefore, is not as concerned with objects blocking the tower signal.
Simply stated, higher frequency means greater range, LOS, and FCC licensing. Lower
frequency means shorter range, less LOS, and that FCC licensing may not be required. Although
LMDS’s transmission range is so limited as to be economically prohibitive, this research
examines Wi-Fi systems with a greater range, as a means of exploring possible future near-LOS
systems.
The Center for Wireless Telecommunications (CWT) at Virginia Tech is an
interdisciplinary research group designed to aid client companies develop new products and
services using wireless technologies. Participating colleges in the CWT are Arts and Sciences,
Business, and Engineering. Furthermore, CWT “has succeeded in its objective of bidding and
acquiring licenses in the Federal Communications Commission's (FCC's) first LMDS auction by
winning licenses to provide 1150 megahertz of wireless bandwidth for the Greater Roanoke,
Danville, Martinsville and Kingsport-Johnson City market areas” (VTCWT, 2000).
Chapter 2: Literature Review 9
Approximately 609,000 people live within the basic trading area (BTA). See Figure 2.1 for a
map of CWT’s BTA region.
Each CWT college brings specific capabilities to the table. The College of Engineering
has technical expertise with wireless transmitters and receivers. The College of Arts and
Sciences is experienced with GIS. The College of Business has financial know-how, and has
created Monte Carlo simulations for various economic scenarios that determine the cost and
profit of a given wireless solution. However, there does not exist a single unifying system that
incorporates the capabilities and limitations of the towers, the topology of the landscape, and the
economic analysis. Each component currently exists in a somewhat isolated form, and there is
no integrated computer system available to do planning.

PLANNING
The American Heritage
®
Dictionary of the English Language (2000) defines planning as
formulating “a scheme or program for the accomplishment, enactment, or attainment of,” or
alternatively, as “to have as a specific aim or purpose; intend.” Yet, in the literature, planning is
a term with many different uses. For example, a search on “planning” using ScienceDirect for
the years 1999 through 2002 on titles alone (not including abstracts and keywords) results in
1863 articles. Planning in an information systems context refers to the first of four phases in the
information systems development life cycle; it is crucial in understanding why an information
system should be built. During this phase, the feasibility of the project is assessed, tasks are
identified, and time is estimated for the software system’s completion (Dennis, Wixom and
Tegarden, 2002). Recent articles may also be found on network planning (Wen, Wu, and Shyur,
2002), radiation therapy planning (Hamacher and Kufer, 2002), workload planning (Lewis and
Slotnick, 2002) and vehicle planning (Horng and Li, 2002). Not only are the application areas in
which planning is applied varied, but the fundamental activities included may be quite different.
For example, planning may be proactive such as when producing blueprints for the construction
of a house, or reactive, as when preparing a retreat from a lost battle.
This work uses a definition of planning closer to that used in artificial intelligence (AI).
AI planning systems assume an initial (current) state, a goal state, and a set of allowable actions.
The planning system’s purpose is to specify the actions that will take the system from its initial
state to its goal state, if possible. A famous example of planning is STRIPS, whereby a group of
wooden blocks on a table in a particular initial orientation is transformed into an alternate
configuration by a computerized robot with certain well-defined actions (Rolston, 1988).
Although planning as used here will not be utilized in a way to derive a sequence of automated
actions, still the planning will be executed to specify those activities that meet a service
provider’s goals of maximizing profit or minimizing cost within the context of geographic,
financial, and other constraints. For example, in Chapter 6 a general planning methodology will
be outlined that maps existing infrastructure and goals into a specification of technology and
equipment providing wireless capability.

Chapter 2: Literature Review 10
SOLUTION METHODOLOGIES
As this research encompasses topics that range over different optimal specification of
wireless transmitters and receivers the solution methodologies will be disparate. However, in
general, each methodology will incorporate the optimization abilities of mathematical
programming, the visualization facilities of GIS, and the integrating capabilities of spatial
decision support systems.
Mathematical Programming
The term mathematical programming is a misnomer in the sense that the user of the
approach does not program the computer by writing code in Basic, C, or another language.
Rather, program is used in the sense of a set of activities or a schedule or a managerial activity
for which a mathematical or quantitative solution approach will be advanced. Dantzig first used
the approach to solve logistic activities during World War II (Dantzig, 1963).
Mathematical programming is used to solve constrained optimization problems, that is
problems in which an objective (such to maximize profit) is pursued, but is limited by constraints
(such as cash on hand). Mathematical programming problems (MP) are classified by the nature
of the objective and constraints. For example, those problems in which the objectives and
constraints are linear are termed linear programming problems (LP) (Moore, Lee, and Taylor,
1993).
Some MP can be shown to have optimal solutions that can be found algorithmically
regardless of the size of the problem. For example, LP in which the constraints form a convex
region can be shown to have optimal solutions that can be found by examining the so-called
corner points of the region (Moore, Lee, Taylor, 1993). Conversely, many MP cannot be
guaranteed to have optimal solutions that can be found quickly. In fact, many MP can be shown
to require exponential solution time in n, the number of variables in the problem.
This research will show that the wireless PMP equipment location problem may be cast
as a well-known MP formulation, the set-covering problem. For a description of the set-covering
problem, see Taha (1975). In the simplest wireless case to be examined here, one may solve for
the location of transmitters and receivers directly as integer variables in the set-covering
problem. In general, however, solutions cannot be guaranteed, and heuristic solutions must be
substituted. In fact, for the case of line-of-sight wireless systems, this research will embed both
mathematical programming solution techniques and Geographic Information Systems into a
spatial decision support system, whereby the GIS determines LOS satisfiability, and the MP
solution approach generates answers to the spatially constrained optimization problem. A
general wireless equipment location spatial DSS will require heuristic solution approaches as
well.
Fixed-Charge Network Flow Models
Mesh networks are made up of nodes and arcs where multiple nodes at different locations
all transmit signals to perhaps a single node (Willis, Hasletad, Friis∅, and Holm, 2001). A node
is a point to which a signal is transmitted or received. Mesh networks may use a variety of
frequencies such as 2.4GHz (Nokia, 2002), 26 GHz to 40 GHz (Fowler, 2001), and laser
(Acampora and Krishnamurthy, 1999). This research uses a network flow formulation to solve
the mesh network problem. The fixed-charge network flow problem (FCNFP) is one of a large
class of network design problems, which have been used in many applications including
Chapter 2: Literature Review 11
telecommunications (Balakrishnan et at, 1991, Gavish, 1991), logistics and production planning
(Minoux, 1989), and transportation (Magnanti and Wong, 1984). FCNFP problems are known to
be NP-hard, and much research has been devoted to creating better and more efficient solutions
(Kim and Pardalos, 1999, Cruz et al, 1998). In these problems, the objective is to seek the most
efficient way to move flow (in our case bandwidth) on a network in order to satisfy demand
between origin and destination nodes and to minimize the overall cost. A fixed cost is incurred
for using arcs between nodes; therefore, it behooves the objective to use a few arcs in which
costs are associated. When the network is capacitated then upper bounds exist for the amount of
flow over an arc. Another consideration for some network design problems is the number of
hops from a source (transmitting node) to a sink (receiving node) (Pirkul and Soni, 2002, Soni,
2001, Girish, Zhou and Hu, 2000, Gouveia and Requejo, 2000). The principal reason for
constraining the number of hops in a network is to maintain a level of quality of service.
Decision Support Systems
Decision Support Systems were first developed four decades ago, primarily at MIT. The
rationale for this new approach was that computers could be utilized not just to automate tasks,
but also to support managers at their jobs. This new DSS era followed on the heels of Electronic
Data Processing Systems in the 1950s and Management Information Systems in the 1960s.
Keen and Scott Morton (1978) noted that most managerial tasks of significance (Alter,
1980) were semi-structured or unstructured, meaning that no algorithm could be written to
specify the path to task solution. This was problematic, because computer programs of that time
required algorithms; hence, the computer was being asked to solve problems that it could not
solve. The answer advanced was to encourage the manager and computer into a symbiotic
relationship, whereby the human provided algorithmic, real-time direction at the prompting of
the machine. The machine would, in turn, perform the requested operations in a representational
form consistent with the preferences of the manager.
From a builder’s perspective, DSS were organized into three primary, independent
modules: a database, a model base, and a dialog generation component (Sprague and Carlson,
1982). The dialog module generally consisted of over half of the code and served the purpose of
communication with the manager. The model base acted as a repository of models that managers
might want or need in working their way through a particular problem. Supposedly, the manager
could ask for any model and/or data needed for any aspect of the entire decision-making process
(Simon, 1963), and the system would respond.
Early DSS failed in the sense that they were not used, in general, once developed. A
variety of reasons was responsible and enhancements were made (Sprague and Carlson, 1982;
Bennett, 1983). However, DSS did not reach widespread acceptance until software development
introduced two fundamental capabilities: (1) the graphical user interface, whereby managers
could now do dialog generation in a natural manner, and (2) the development of easily-utilized
model-base competence, through such packages as Excel with its built-in functions, modules,
and macros.
Today, DSS are developed routinely, with hundreds of articles in the literature, and whole
conferences organized around such systems. The reader interested in further detail may consult,
for example, Turban and Aronson (2001).
Chapter 2: Literature Review 12
Geographic Information Systems
Geographic Information Systems are a type of relational database management system
(RDBMS) used to collect, store, retrieve, and analyze spatial data. GIS link tabular data to
graphical data by relating graphical layers to database tables. For example, a demographic layer
of a town map may have sections linked to economic information such as household income or
school zoning information (ESRI, 1999).
GIS are excellent tools for presenting spatial data in a format that is easy to manipulate
and understand. The layering of different data makes GIS extremely powerful. However, it
takes more than just layering data to extract meaningful information. For example, West and
Hess (2001) point out that it often takes someone with expert knowledge in GIS to make
adequate use of the system. In fact, they suggest using software agents to format the information
in a manner most helpful to the user. GIS often require the user not only to be facile in
manipulating the GIS, but also to possess expert domain knowledge.
Spatial Decision Support Systems
Spatial Decision Support Systems (SDSS) combine DSS and GIS. The advantage of an
SDSS is the seamless integration of the model portion of the DSS with the graphical
representation of the GIS, thereby aiding decision makers with semi-structured or unstructured
spatial problems. Crossland, Wynne and Perkins (1995) determined that SDSS, through
visualization, enabled decision makers to complete their tasks quicker, more efficiently, and with
greater understanding of the problem.
SDSS is a relatively new research area, primarily because GIS software has historically
needed a great amount of computing power, memory, and hard disk space. Since such hardware
was expensive, GIS software required large budgets. As computers have become cheaper and
more powerful, GIS packages can now run on desktop computers, and the SDSS research stream
is beginning to flow. SDSS have been applied to siting problems (Maniezzo, Mendes, and
Paruccini, 1988; Vlachopoulou, Silleos, and Manthou, 2001), land planning (Nehme and
Simoen, 1999), and vehicle routing (Keenan, 1998; Tarantilis and Kiranoudis, 2002). GIS have
been combined with LP models to solve cropland allocation and land-use modeling (Campbell,
Radke, Gless, & Wirtshaffer, 1992; Chuvieco, 1993), and to perform multicriteria analysis
(Laaribi, Chevallier, & Martel, 1997; van der Merwe, Lohrentz, 2001)
This research will develop an SDSS to solve the line-of-sight tower location problem for
a PMP network. Objective functions considered include maximization of profit, minimization of
cost, and the maximization of area coverage. The dissertation will also utilize the SDSS to solve
mesh-connectivity wireless problems utilizing its mathematical programming and GIS
components.
In conclusion, a review of the literature indicates a plethora of knowledge on
mathematical programming/set covering and GIS, but a dearth of work utilizing the techniques in
concert. By building an SDSS to combine these symbiotically, this research develops a tool to
address important variations of the wireless equipment location problem (in Chapters 3 and 4).
However, once the new tool is assembled, an additional host of problems becomes amenable to
solution, including issues such as infrastructure development and recovery, for example with
developing countries, Native American lands, and both urban and rural disaster recovery. This
future research subject is introduced in Chapter 6.

Chapter 2: Literature Review 13
Table 2.1. Some wireless communications examples.
Frequency Band Applications
170 - 190 kHz
VLF band radios
Beacons
550 - 1600 kHz
AM broadcast communications
Low power voice and data
27 MHz
CB radios
Low power voice and data
49 MHz
Remote control
Cordless telephones
Low power voice and data
88 - 108 MHz
FM radios
Low power voice and data
150 - 170 MHz
Commercial two-way voice
Pager services
260 - 470 MHz
303 MHz garage door openers
Keyless entry systems
Security alarms
824 - 849 MHz and 869 - 894 MHz Cellular phones
902 - 928 MHz
ISM band Wireless LAN's
Part 15 devices (spread spectrum
cordless phones, etc.)
Military radiolocation systems
Federal mobile communications
930 MHz
Pager services with high transmitter
power
2.4 - 2.4385 GHz
Amateur satellite
Part 15 devices
Microwave ovens and systems
Army packet radio development
802.11b
5.15-5.25GHz
5.25-5.35GHz
U-NII
802.11a
5.727 - 5.875 GHz
Amateur satellite
Part 15 devices
Naval radar systems
Test range instrumentation radars
802.11a
802.16 – (Tsunami equipment)
24 - 24.25 GHz Radio navigation
28 – 31 GHz
Local multipoint distribution service
(LMDS)

Source: The Virginia Tech Center for Wireless Communications.
(http://www.cwt.vt.edu/wireless_faq/default.htm).

Chapter 2: Literature Review 14

Figure 2.1. Virginia Tech Center for Wireless Telecommunications BTA Regions shown
shaded on a map of Virginia (United States)


Chapter 2: Literature Review 15

CHAPTER 3

A MATHEMATICAL-PROGRAMMING AND
GEOGRAPHIC-INFORMATION-SYSTEM FRAMEWORK
FOR WIRELESS BROADBAND DEPLOYMENT IN RURAL AREAS






























submitted to Decision Sciences
July 31, 2002

Chapter 3: A Mathematical Programming… 16
A MATHEMATICAL-PROGRAMMING AND
GEOGRAPHIC-INFORMATION-SYSTEM FRAMEWORK
FOR WIRELESS BROADBAND DEPLOYMENT IN RURAL AREAS




ABSTRACT
Organizations and individuals are calling for the universal availability of broadband as a
national imperative by 2010. Although the United States has the requisite cross-continental
fiber-optic backbone infrastructure in place, only 2% to 5% of that nationwide network is used
today. This is because of the prohibitive cost necessary to extend the backbone to many small
businesses and homes, even for a relatively short distance. For example, nine out of ten
American small businesses are less than one mile from the backbone, but still do not have
service – the “last mile” problem.
This research integrates a mathematical programming model and a specially developed
geographic information system to examine the potential of “going the last mile” using current
wireless telecommunication costs in a rural county in the Mid-Atlantic United States. The
appropriateness of current wireless technologies in both for-profit and government subsidized
scenarios is discovered. Implications for the national scene are discussed.

Keywords: Wireless telecommunications, Broadband, Geographic information systems,
Decision support systems, Set covering problem, Mathematical programming optimization
Chapter 3: A Mathematical Programming… 17
A MATHEMATICAL-PROGRAMMING AND
GEOGRAPHIC-INFORMATION-SYSTEM FRAMEWORK
FOR WIRELESS BROADBAND DEPLOYMENT IN RURAL AREAS

MOTIVATION
TechNet, a bipartisan group comprising more than 300 chief executive officers and senior
partners of the major companies in the fields of information technology, biotechnology, venture
capital, investment banking and law, claims that (1) “widespread adoption of true broadband will
increase the efficiency and productivity of Americans at work and at home – with a potential
$500 billion impact on the United States economy. The benefits to quality of life are
immeasurable,” and that (2) true broadband is the key to the next generation of communications
and Internet services (TechNet, 2002, p. 1). However, Kornbluh (2001, p.21) points out that this
potential will remain unrealized without government action. She states in an op-ed piece in the
New York Times, that although investors have plowed $90 billion into a cross-continental fiber-
optic broadband network, today merely 3% of that backbone is in use – it is the “digital
equivalent of fallow farmland.” The problem, she states, is that entrepreneurs failed to foresee
the enormous cost of upgrading the “last mile – copper telephone wires that connect individual
homes and small businesses to the broadband backbone” (Kornbluh, 2001, p. 21). In an article in
Scientific American, Acampora (2002) elaborates by noting that for nine out of 10 American
businesses with more than 100 workers, the backbone is less than a mile away.
TechNet has called on the President and policymakers to make broadband a national
priority and to set a goal of making an affordable 100-megabits per second (Mbps) broadband
connection available to 100 million American homes and small businesses by 2010 (TechNet,
2002). They note that to achieve the 100 million homes goal will require network providers to
invest hundreds of billions of dollars to upgrade infrastructures and increase bandwidth capacity
in the last mile, primarily by providing new fiber connections to homes and offices. Today,
virtually no homes have connections with such bandwidth (TechNet, 2002). An intermediate
goal is the “availability of affordable broadband at speeds of at least 6 Mbps from 2 or more
providers to at least 50% of U.S. households and small businesses by 2004” (TechNet, 2002, p.
6). TechNet developed 6 principles to address roadblocks and provide a guide to a national
broadband policy. Principle 5 reads as follows (pages 2-3):

Investment incentives, potentially including targeted tax incentives, should
encourage broadband deployment to underserved communities and businesses. …
In a market-oriented environment that encourages the deployment of broadband
networks, there may still be a segment of the U.S. population that does not have
broadband availability. Public policies should seek to narrow the current and
future disparity in the level of high-speed access to the Internet, to ensure that all
Americans can enjoy the benefits of broadband.

Similarly, Kornbluh encourages both Congress and the president to subsidize sparsely
populated regions of the country, or low-income users or both, suggesting that an ambitious
broadband strategy can help revitalize the economy. (Kornbluh, 2001, p.21)
Chapter 3: A Mathematical Programming… 18
The focus of this research is on the sparsely populated segment of this problem. For
reasons that will be explained in the next section, this paper examines the wireless provision of
broadband service to rural homes and small businesses using current costs and technologies.
Because the wireless technologies require “line of sight” or “near line of sight” operation, it will
be necessary to develop a special-purpose geographic information system (GIS) and then
integrate it with the mathematical programming formulation we derive. Solutions will be
generated for both “for profit” and government subsidized scenarios. The procedure will be
demonstrated for a rural county in the mid-Atlantic region of the United States, and several
implications will be drawn for the rural “last mile” problem as a whole.
BACKGROUND
TechNet defines broadband as the capacity to deliver Internet access with a continuous
‘always on’ connection and the ability to both receive and transmit digital content or services at
high speeds (TechNet, 2002). Although other definitions of the term abound, most agree with
this one – except as to what precisely is meant by “fast.” Today, approximately only 4.4% of
American households have speeds approaching 400 Kbps (TechNet, 2002). In order to facilitate
telecommuting, it is estimated that speeds of 10 Mbps will be required. Many experts have
defined 100 Mbps as the speed at which the web’s true potential can be achieved. (TechNet,
2002)
Copper wires and coaxial cables connecting buildings do not possess the gigabit per
second capacity necessary to carry advanced bandwidth-intensive services and applications,
whereas optical fiber bridges needed to connect millions of users to the optical-fiber backbone
would cost too much to install (between $100,000 and $500,000 a mile). As these costs are
prohibitive, service providers are looking to other transmission media, and in particular to
wireless telecommunications. Willebrand and Ghuman (2001) cite an example whereby a fiber
potential cost of $400,000 was reduced to $60,000 with a wireless system. Nokia has recently
announced a national initiative to bring broadband wireless connectivity to business and
residential customers via their Nokia RoofTop solution. (Nokia, 2002)
Wireless Telecommunication
For years, many consumer devices such as AM/FM radio, cordless and cellular phones,
satellite television, CB radios, pagers, car alarm signalers, garage door openers, and television
channel changers have applied wireless technology. More recently, wireless computer
networking has seen increasing employment, including applications meeting the IEEE 802.11
standard for local area networks (LAN), the IEEE 802.16 standard, and local multipoint
distribution service (LMDS). These standards are described below. Table 3.1 indicates radio
frequencies allocated for several different applications.
Different Wireless Systems
Operationally, wireless radio systems may be classified along several dimensions. The
systems we explore in this chapter are fixed, point-to-multipoint systems. “Fixed” wireless
networks are those where the transmitter and receiver locations are fixed, as opposed to mobile
networks, such as utilized with cellular telephones. Point-to-multipoint (PMP) networks are
those where a single source (in our case an antenna or antennas on top of a tower) connected to a
backbone network propagates signal to multiple customers (in our case, receiving antennas
Chapter 3: A Mathematical Programming… 19
located at customer premises). Multipoint-to-multipoint networks (MPM), such as mesh
networks (Fowler, 2001; Willis, Hasletad, Friis∅, and Holm, 2001), whereby multiple sources at
different locations all transmit signals to perhaps the same customers, will be discussed in
Chapters 4 and 5.
The three wireless technologies mentioned above (IEEE 802.11, 802.16, and LMDS)
share similarities in characteristics and properties and are the leading contenders for fixed PMP
service. They are all either line-of-sight, or near line-of-sight, broadband wireless
communications, but vary in the degree of throughput. LMDS offers very high-speed
connectivity, but is the most costly of the three. The 802.16 technology (also known as
WirelessMAN) is less expensive than LMDS, but the customer premise equipment (CPE) costs
are still high. Finally, 802.11b (Wi-Fi) service is gaining acceptance and market share, thus
allowing for a greater number of choices in equipment and, consequently, lower prices. Tower
costs may be similar to 802.16, but in some cases customer premise equipment is one-fourth that
of the WirelessMAN.
Costs
As this paper considers only fixed PMP systems, there is a tower cost at the hub (the
“point”) as well as customer premise equipment costs (the “multipoints”). Tower costs can vary
widely, depending on location and terrain, and include components such as cost of land/right of
way, structure, connecting the tower to the backbone, power, bringing power to the tower,
antennas, and annual maintenance. At each CPE site there are receiving antenna costs and
installation costs (so called “truck rolls”, Schrick and Riezenman, 2002), which include antenna
alignment and positioning expenses.
Differences Among Wireless Systems
It might be supposed that one could apply directly a solution methodology for cell phone
networks to the wireless network determination problem. Such is not the case, however, because
of three primary differences between the two wireless technologies. First, the receivers in the
cell phone scenario are mobile, allowing users to switch from one tower to the next as they
travel; utilizing the same frequencies at adjacent towers is not a problem. Such is not the case
with broadband. Second, the expectations in terms of service level are much higher for
broadband (the so-called “five nines” – 99.999% reliability), as opposed to cell phone users
accustomed to breakup, callbacks, and frequent fading and/or loss of signal. Third, the
transmission range from towers for cell phones is much greater than for broadband because cell
phone frequencies are much lower. Consequently, to determine a broadband distribution
network requires the ability to position towers in locations that guarantee reliable line-of-sight
transmission. To make this determination, we turn to geographic information systems and
decision support systems.
Geographic Information Systems and Decision Support Systems
Decision Support Systems (DSS), a branch of information systems, has been a topic of
ongoing research for the past 30 years. Using data, model and user interface components, DSS
aid decision makers to solve semi-structured or unstructured problems (Bennett, 1983) and cover
all phases of Simon’s (1960) decision-making process. According to Keen and Scott-Morton
(1978), unstructured decisions are defined as those for which no algorithm can be written,
whereas algorithms can be specified for structured decisions. Semi-structured decisions fall
Chapter 3: A Mathematical Programming… 20
between the other two. The reader interested in further detail may consult, for example, Keen
and Scott-Morton (1978), Bennett (1983), or Turban and Aronson (2001).
Geographic Information Systems (GIS) are used to collect, store, retrieve, and analyze
spatial data and can provide decision support. GIS link tabular data to graphical data by relating
graphical layers to database tables. For example, a demographic layer of a town map may have
sections linked to economic information such as household income or school zoning information
(ESRI, 1999).
Marrying DSS and GIS creates Spatial Decision Support Systems (SDSS). The
advantage of an SDSS is its ability to integrate the model portion of the DSS with the graphical
representation of the GIS, thereby aiding decision makers with semi-structured or unstructured
spatial problems. In their research, Crossland, Wynne and Perkins (1995) determined that SDSS
enabled decision makers to complete their tasks quicker, more efficiently, and with greater
understanding of the problem through visualization.
SDSS is a relatively new research area, primarily because GIS software has historically
needed a great amount of computing power, memory, and hard disk space. Since such
equipment was very expensive, GIS software required large budgets. As computers have
become cheaper and more powerful, GIS packages can run on desktop computers, which creates
more SDSS research possibilities. SDSS have been applied to (among others) siting problems
(Maniezzo, Mendes, and Paruccini, 1988; Vlachopoulou, Silleos, and Manthou, 2001), land
planning (Nehme and Simoen, 1999), and vehicle routing (Keenan, 1998; Tarantilis and
Kiranoudis, 2002).
As part of the National Mapping Program, the United States Geological Survey (USGS)
produces Digital Elevation Models (DEMs). A DEM is a digital file containing terrain elevation
information that has been sampled at regularly spaced intervals, and stored in raster (matrix of
rows and columns) format. There are five DEM products for sale by the USGS, 7.5-minute, 7.5-
minute Alaska, 15-minute, 2-arc-second, and 1-degree. These five products fall into one of three
different scale categories, large, medium, and small. Larger scale means greater map detail. The
large-scale category includes 7.5-minute and 15-minute maps with ratios of 1:24,000 and
1:63,360 respectively; the medium-scale category includes 2-arc-second maps with a ratio of
1:100,000; and the small-scale category includes the 1-degree case with a ratio of 1:250,000.
The 7.5-minute DEM provides the greatest level of detail from the USGS with 30x30
meter data spacing. These maps are produced either by digitizing cartographic map contour
overlays or by scanning photographs from the National Aerial Photography Program (USGS,
2002).
ESRI, now based in Redlands, CA, was founded as Environmental Systems Research
Institute in 1969 as a privately held consulting firm that specialized in land use analysis projects.
ArcView GIS is a desktop geographic information system from ESRI, now the leading global
provider of GIS software (ESRI, 1999). In this research we use ArcView version 3.2 to develop
our own special-purpose GIS software.
(NEAR) LINE-OF-SIGHT OPTIMIZATION MODELS
As a simplified introduction to the problem, consider wireless service that does not
require line of sight. Assume for pedagogical purposes that service is to be provided to a
rectangular-shaped geographic area, and divide the rectangle into a grid of n equi-sized smaller
rectangles, called cells.
Chapter 3: A Mathematical Programming… 21
Notation
Define the following four n x 1 vectors:
• Let x represent a decision variable indicating tower placement in cells; in
particular,

let
¹
´
¦
=
. , 0
; ) ( , 1
grid the of i cell in anywhere placed is tower no if
grid the of i cell of center the at placed is tower a if
x
i
• Let c represent a vector of costs incurred by placing towers in the grid, namely c
i

equals the cost of placing a tower in cell i. Note that this allows costs to vary from
cell to cell.

• Let r be a vector of revenues, where r
i
indicates revenue generated in cell i by a
tower placed somewhere in the grid (not necessarily in cell i) that provides signal to
cell i.

• Let e be an n by 1 vector of ones such that e
t
= (1 1 … 1), where e
t
represents the
transpose of e.

In order to express mathematically a tower’s ability to transmit signals successfully to
cells in the grid (i.e., “cover” the cell), we define an n x n binary matrix E, which we call the
exposure matrix. In particular,

¹
´
¦
=
i. cell to signal a furnish not does j cell of center at the placed tower a if 0,
i cell to signal a provides j cell of center at the placed tower a if , 1
ij
E
In general, note that E
ij
might equal zero for either of two basic reasons: A tower placed
at j cannot transmit a strong enough signal to a distant cell i, and/or a signal between i and j
might not have line-of-sight (due to terrain). But since in the simplified non-line-of-sight case
being considered at the moment it does not matter whether the receiving antenna can see the
tower, the second reason is of no concern, an assumption we will relax shortly.
With these definitions the following should be observed:

• The number of towers placed equals the vector dot product e
t
· x = . ∑
i
x
• The cost of placing towers is c
t
· x = . ∑
i i
x c
• If we denote the ith row of the E matrix as the 1 x n vector E
i
·
, then the dot
product E
i
·
· x equals the number of signals cell i will receive from all towers
actually placed.

This latter observation leads to the formulation of several important potential constraints:
• The specification E
i
·
· x ≥ 1 specifies that cell i must receive at least one signal.
Chapter 3: A Mathematical Programming… 22
• The matrix stipulation E · x ≥ e mandates that each cell in the grid must receive at
least one signal.
Finally, note that E
i
·
· x takes on the value zero if cell i receives no signal and equals a
positive integer value if a signal or signals are received. Consequently, we may write total
revenue generated in the entire grid as

revenue in cell i = (1)
¹
´
¦ >

.
,
otherwise 0,
0 x · E if r
i i
View Sheds
Introducing the line-of-sight requirement necessitates a fundamentally different analysis
because terrain, buildings, and (sometimes) even trees must be considered. “Near” line-of-sight
is less restrictive. Signals still cannot go through large obstructions such as mountains; trees and
smaller hindrances are not a problem. This means that DEM maps contain the topographic detail
necessary to ascertain near line-of-sight capability.
To exploit these maps, Bill Carstensen wrote a special-purpose geographic information
system (Carstensen, Bostian, and Morgan, 2001) and introduced the view shed concept. The
modified GIS, the Geographic-Engineering Tool for Wireless: Evaluation of Broadband Systems
(GETWEBS), can determine the visibility over an area of any tower signal, given the height and
tower location on a DEM, and the transmission signal strength. Figure 3.1 shows (superimposed
on a DEM) the view shed, i.e., area that can see a tower’s transmission given the tower’s location
and height (40 feet) in a rural county in the mid-Atlantic United States. One thing we are
interested in is determining the profit shed, i.e., the region that can both see a tower’s signal and
generate a profit for a service provider. To determine this, we integrate the GETWEBS view
shed with a mathematical programming formulation, as will now be explained.
Enhance the notion of E, the exposure matrix, by defining an n x n binary matrix V,
which we term the view shed matrix. Whereas the exposure matrix assumes that an antenna
within the tower’s range can receive any signal sent from a tower, we define V
ij
such that

¹
´
¦
=
i. cell to signal a furnish not does j cell of center at the tower a terrain, given the if, 0,
i cell to signal a provides j cell of center at the placed tower a terrain, given the if, , 1
ij
V

That is, values of one for elements V
ij
in the matrix V indicate which cells j actually
receive signal from a tower placed at the center of cell i. GETWEBS must make this
determination. This is done in practice by making a point-to-multipoint determination within the
GIS of what can be seen. A simple example illustrates the concept.
Pedagogical Mini-Example
Consider a fictitious town (shown in Figure 3.2a) whose elevation profile is indicated in
Figure 3.2b. This town is chosen so we may calculate both exposure and view shed matrices
without relying on GETWEBS. If the town is divided into a grid of rectangular cells 1 kilometer
on a side, and planners consider tower placements using line-of-sight LMDS technology with an
omni directional antenna array reaching 2.5 2 kilometers (see Figure 3.2c for the pedagogic
rationale for this “unusual” choice), then the exposure matrix E ignoring line-of-sight issues and
Chapter 3: A Mathematical Programming… 23
the view shed matrix including those considerations may be written as indicated in Figure 3.3.
(We have assumed a tower height of zero for pedagogic reasons.) Note that since the east part of
town is flat (no hills or valleys), the portion of the view shed matrix corresponding to that part of
town is identical to the exposure matrix E. The same is not true on the other side of the tracks, as
can be seen in Figure 3.3, because the hill prevents signals from a tower on one side of it from
reaching the other side.
Profit Model
A wireless Internet, etc., provider will most likely want to maximize profit subject to
financial capabilities. This suggests an integer mathematical programming formulation similar
to the set-covering problem (Taha, 1975). We develop the wireless formulation by way of a
simple example.
Consider another fictitious small, flat town in which we want to consider providing
wireless service. Since the town is completely flat, we note that the V matrix is equivalent to the
E matrix. For pedagogical purposes we divide the town into only 6 cells (see Figure 3.4) and
assume that a tower placed at the center of any cell will cover that cell and any cell it touches
horizontally and vertically, but not in a diagonal direction. For example, a tower in cell 2 will
cover cells 1, 2, 3, and 5. This may be seen in the second row of the V matrix:

V = .
|
|
|
|
|
|
|
.
|

\
|
1 1 0
1 1 1
0 1 1
1 0 0
0 1 0
0 0 1
1 0 0
0 1 0
0 0 1
1 1 0
1 1 1
0 1 1
Recall that V
1
·
indicates the first row of the V matrix; it specifies which towers, if placed,
will reach cell 1. That is, cell 1 will get coverage if towers are placed in cells 1, 2, or 4.
Although the expression V · x ≥ e guarantees coverage of all cells, maximized revenue
cannot be determined without defining a new binary variable that indicates whether a cell is
covered:
Let z
i
=
¹
´
¦
. , 0
; ) ( , 1
all at signal a receive not does i cell if
signal one least at receives i cell if
Furthermore, stipulate that z
1
≤ x
1
+ x
2
+ x
4
(or, equivalently, z
1
≤ V
1
·
· x).
Now if x
1
= x
2
= x
4
= 0, then z
1
= 0. This says (as desired) that if there is no tower in cells 1, 2, or
4, then cell 1 receives no signal. But if one or more of x
1
, x
2
, and x
4
≠ 0, then z
1
may equal either
0 or 1, whereas we want z
1
=1. To fix this, we say

max z
1

subject to z
1
≤ x
1
+ x
2
+ x
4
;

Chapter 3: A Mathematical Programming… 24
then if x
1
= x
2
= x
4
= 0, then z
1
still equals 0, and if one or more of x
1
, x
2
, and x
4
≠ 0, then z
1
= 1,
as desired. The objective max r
t
· z subject to z ≤ V x allows us to write the vector version
including all cells.
If we let ρ
i
= r
i
− c
cpe
(where c
cpe
is the customer-premise-equipment cost), then in
general, with a budget of M, we may write:

max ρ
t
· z – c
t
· x (2a)
st/ V x ≥ z (2b)
c
t
· x + c
t
cpe
· z ≤ M (2c)
x
i
, z
i
∈ {0,1} ∀i. (2d)

If everyone must receive a signal, then constraint (2e) below is added to the system:

V x ≥ e (2e)

If there is no budgetary restriction, then M is allowed to approach infinity, and constraint (2c) is
removed from the model.
Pedagogical Example
To illustrate the model for our six-celled, fictitious town, we further specify the
following: Assume M = $500K, c
i
= $200K (∀i), c
cpe
= 0, there is a 5-year economic horizon,
and all cells have the same number of households, penetration, and propensity to pay, namely:

N
i
= 70 households;
propensity to pay = $50/month or $3000/horizon, and
penetration = 1/3.

Then, revenue = r
i
= 70 * (1/3) * $3000 = $70,000 for each cell if a signal is received in that cell.
Equations (2) are written as (dollars are in thousands):
Max 70(z
1
+ z
2
+ z
3
+ z
4
+ z
5
+ z
6
) – 200(x
1
+ x
2
+ x
3
+ x
4
+ x
5
+ x
6
)
st/ 200(x
1
+ x
2
+ x
3
+ x
4
+ x
5
+ x
6
) ≤ 500
x
1
+ x
2
+ x
4
≥ z
1

x
1
+ x
2
+ x
3
+ x
5
≥ z
2

x
2
+ x
3
+ x
6
≥ z
3

x
1
+ x
4
+ x
5
≥ z
4

x
2
+ x
4
+ x
5
+ x
6
≥ z
5

x
3
+ x
5
+ x
6
≥ z
6

x
i
, z
i
ε [0,1] for all i.
Chapter 3: A Mathematical Programming… 25
The solution is x
2
= x
5
= 1; x
1
= x
3
= x
4
= x
6
= 0, which says to place towers in cells 2 and
5 only. Moreover, z
1
= z
2
= z
3
= z
4
= z
5
= z
6
= 1, which states that all cells are covered. The total
revenue generated is $420,000; the cost of placing towers is $400,000; and the profit is $20,000.
Of the $500,000 budget, $100,000 remains unspent.
Maximizing Exposure Model
In addition to the profit maximization model formulated above, an important model
maximizes exposure within a prescribed budget. This model is interesting in examining regions
where profit cannot be made, but government subsidy is considered at a specified level. If h
i
is
the number of households in cell i, then the exposure model is

max h
t
· z (3a)
st/ V x ≥ z (3b)
c
t
· x + c
t
cpe
· z ≤ M (3c)
x
i
, z
i
∈ {0,1} ∀i. (3d)
Additional Constraints
There are other constraints not indicated above that may be written when appropriate.
These include: budget constraints for any given time period, the exclusion of a particular cell
from tower placement, the insistence that a tower be placed in a given cell, the specification of
an existing tower in a cell, the requirement that a specified fraction of total households be
serviced, the demand that k out of r cells (where k and r are specified constants) be given service,
etc. The reader interested in incorporating such constraints should consult a basic mathematical
programming text, such as Taha (1975).
A Spatial Decision Support System
We have built an SDSS that implements the mathematical modeling features described
above with the GETWEBS geographic information system. The point noted here with this SDSS
is that general modules have been/can be written that allow various equipment, forecasting, and
financial modeling to be intertwined in the appropriate mix. A graphic description of the SDSS
architecture we employed is shown in Figure 3.5.
Figure 3.5 shows three paths to the (output) profit-shed module: the visibility module, the
revenue module, and the evaluation module. The visibility module consists of the equipment
definition sub module, and the view shed sub module. In addition to descriptive information on
each type of equipment, the equipment sub module contains technical capabilities, including the
range, angle of transmission, and shape of the region covered by each antenna (e.g., circular,
elliptical, etc.). An example of a tower specification screen is shown in Figure 3.6. The view
shed sub module calculates exposure (i.e., the view shed matrix V) given the antenna specified
and the GETWEBS program. The revenue module includes the census data access and various
marketing revenue sub modules, and the evaluation module contains both the max profit and
max exposure mathematical programming models as well as requisite financial sub modules.

Chapter 3: A Mathematical Programming… 26
REACHING THE LAST MILE
Having developed a mathematical programming model with an embedded geographic
information system, we now return to the issue posed at the beginning of this paper, namely the
lack of broadband infrastructure going “the last mile.” Recall that Kornbluh (2001) noted that
97% of the broadband backbone is not in use, with only 10% of households and small businesses
connected to it.
To demonstrate proof of concept, we develop profit and exposure models for various near
line-of-sight technologies for a region approximately 40 km (25 miles) on a side. We use USGS
census maps in the GETWEBS program. We base our wireless cost data on actual systems with
which we are familiar, although some cost specifications are difficult and approximate, as will be
noted.
GETWEBS Data
Montgomery County is a medium-sized county in the mid-Atlantic United States. It has
a population of 77,500 (about 29,000 households), consists of 393 square kilometers in a rural
setting, and has an active interest in low-cost, high–speed Internet service, partly due to the
presence of a large land-grant university. Figure 3.1, seen earlier, shows a screenshot of this
county displayed in the GETWEBS program.
The county is approximately a 40 km by 40 km rectangle, which we subdivide into 394
cells (arranged roughly as 20 rows with 20 columns), each cell being a square of side 2
kilometers. As mentioned, census data down to squares 30 meters on a side is available, and the
process described below would be unchanged if analysis in greater detail were desired, although
computational complexity would be significantly multiplied. For simplicity, we assume
transmission towers may be placed only in the middle of any 2-kilometer cell.
GETWEBS considers a transmission tower centrally placed in a cell, and then determines
whether line-of-sight exists to each of the other (“receiving”) cells. Three notes should be made
with respect to this calculation: (1) this is a point-to-multipoint GIS calculation, from the tower
to each locale in the receiving cell; (2) the software reports the percentage of the receiving cell
able to get a signal; and (3) the search over the receiving cells can be significantly limited by the
range of the signal. For example, in one scenario considered here, the signal is such that any one
transmitting cell can reach at most only 28 other cells, and transmitting cells on edges of the
region have an even further reduced set of possible receiving cells.
The output from our view shed software is a 394 by 394 matrix filled with percentages.
We arbitrarily specify that a cell with greater than 50% coverage from a tower is covered, and
enter a “1” in the corresponding cell for the view shed matrix. Obviously the 50% threshold may
be adjusted to meet particular circumstances.
Wireless Parameters and Costs
We examine the three fixed, point to multipoint technological options currently under
most consideration: local multipoint distribution systems (LMDS), the 802.16 standard, and the
802.11b option. The parameters we used are based on published data.
Chapter 3: A Mathematical Programming… 27
Range
We use a range (i.e., transmission signal from the tower) of 5 2 ≈ 7.1 kilometers. This
value is within the scope of all three systems and is chosen because a tower placed at the center
of a cell can reach diagonally to all households two cells away.
Tower Costs
We assume four 90º sectored antennas at each tower, providing omni-directional
coverage. Costs for each tower, beyond the antennas and the structure, include the cost of land
purchase and/or right-of-way, getting electricity to the site, and connecting (or transmitting) to
the backbone. As the non-antenna costs dominate, it is a fair approximation to say the tower
costs of all three technologies are approximately the same. The tower costs, even for
Montgomery County, will vary considerably depending on the location and accessibility to other
infrastructure. Talks with “experts” suggested that we model tower costs probabilistically; we
chose a triangular distribution with a minimum value of $100,000; a maximum of $500,000; and
a most likely value of $150,000.
Customer Premise Equipment Cost
The receiver antenna costs do vary widely by technology, and depend primarily on the
electronics costs of each system. The costs we used were $7,900 for LMDS; $1295 for the
WirelessMAN (IEEE 802.16); and, $230 for the Wi-Fi system.
Revenue Model
Actual year 2000 number of households and per household annual income are available
or can be calculated from United States census data for each cell in Montgomery County. We
assume the following functional form to calculate revenue: %P = f(I | p
m
), where %P is the
percent participation, I is the average annual household income for a cell, and p
m
is the price per
month for the wireless broadband offering. That is, the percentage of households in a cell
purchasing (“participating”) in the broadband offering is a function of the average household
income in that cell given the monthly price of that service. These functional relationships were
estimated from information on “related” services in the local area, namely, the price charged for
DSL and for Adelphia Internet service. At a similar price, one would expect the wireless
broadband service to do at least as well as the two alternatives mentioned, as the wireless
bandwidth would be an order of magnitude or so better, depending on the wireless technology
employed. The particular functional relationships employed in the study are shown in Table 3.2.
Note that in that table, I is the average per-household annual income (in 000s of dollars), p
m
is
expressed in dollars, and %P is a number between 0 and 100.
Results
The profit model developed above is run for each of the three technologies. First, the
view shed matrix V is determined by GETWEBS and substituted into equation (2b). Then the
system given by equations (2a), (2b), and (2d) is solved.
Chapter 3: A Mathematical Programming… 28
LMDS at $50 per month
This technology, though touted for several years until the fairly recent past, is
substantially too expensive for deployment in a county similar to Montgomery County. The
system solution declares that no profit can be made and that, hence, no towers should be placed.
802.16 (WirelessMAN) at $50 and $100 per month
The WirelessMAN reduces the customer premise equipment to $1295. Nonetheless, at a
price of $50/month, once again wireless service is not profitable. At a price of $100/month, the
optimal solution suggests that one tower be placed reaching approximately 700 households (less
than three percent of the county total). Moreover, only (approximately) $25,000 in profit is
generated. The one tower placed is positioned in the northern portion of the county (see Figure
3.1). Regions in the profit shed (i.e., in a view shed that generates profit) are shaded.
Wi-Fi at $50 per month
With CPEs down to $230, the 802.11b Wi-Fi alternative generates a significant profit.
Almost 80% of the county is projected to participate, requiring 6 towers (see Figure 3.7),
generating a profit of $5M on a cost of $6.2M.
DISCUSSION
The previous section examined three wireless technologies in a “for profit” context.
Even with the Wi-Fi scenario, universal service was not nearly achieved. We now examine
Kornbluh’s (2001) and Technet’s (2002) calls for government subsidization to see at what cost
Montgomery County can reach more households with broadband availability. We consider only
the Wi-Fi (802.11b) alternative.
We first consider the case where a for-profit service provider has found a limited level of
service profitable, but the government wishes to increase the number of households reached. For
example, in the Wi-Fi example given above, the maximum-profit solution provides service to
80% of the households in Montgomery County. If the government decides it wishes service to
be provided to (say) 90% of the households in the county, then the for-profit provider will not
want to service the extra households because less profit will be obtained. If the government
agrees to a subsidy to the provider for reaching the extra homes, the amount of the subsidy could
be studied using the spatial decision support system – one merely adds a constraint to the profit
maximization model requiring service to the desired percentage of homes. (See Figure 3.8 for
some results at different levels of coverage.) For example, if the government agrees to make up
for the provider’s loss in profit, the subsidy would be $500K. Alternatively, if the government
agreed to make up for the difference in total costs (not shown), the subsidy would be $1.7M.
Moreover, further analysis with the tool can be performed to determine the incremental benefit
of reaching additional homes. For example, data indicate that the profit begins to drop quite
sharply beyond 90% coverage, where tower costs increase dramatically. What is happening is
that, with the easiest-to-reach people having been served, towers must be placed that cover
relatively few individuals. So, for Montgomery County, a government subsidy to provide
service beyond 90% would give relatively little marginal benefit for the marginal expenditure.
A second subsidization alternative beyond the cooperative venture between the
government and a for-profit firm is one in which no for-profit wishes to provide broadband
service to a region. In such a case the government may believe it is necessary to subsidize the
Chapter 3: A Mathematical Programming… 29
project. In this situation the maximum exposure model of equations (3) could be utilized,
specifying the subsidization amount in equation (3c). For example, for a small subsidy of $500K
in Montgomery County, 1670 households can be reached, with one tower being placed.
CONCLUSIONS
This paper has shown how mathematical programming can be integrated with an
enhanced geographic information system to build a spatial decision support system for analyzing
options in the deployment of wireless broadband alternatives in rural areas. Both profit-
maximization and subsidization scenarios were presented in addressing the “last-mile” problem
in a particular county in the mid-Atlantic United States region.
It is not possible to generalize from one county to draw conclusions about broadband in
the United States as a whole. Nonetheless, it is safe to draw several general conclusions. It is
seen that at current technology levels and costs, providing widespread broadband service to
many rural regions will require significant government subsidization. The cost of customer
premise equipment is the largest problem, and is still prohibitive for sparsely populated, hilly
terrain. Tower costs also need addressing, although little can be done to lower several
components of these costs, including the cost of land access, provision of power, and connecting
from the tower to the backbone. However, the spatial decision support system developed in this
paper can be used to determine the magnitude of the last-mile problem and analysis of different
policy options, although significant effort may be required to generate reliable estimates. In
particular, census data maps are available with household data, and DEMs can be analyzed to
develop view sheds. But what is hard to specify are tower costs and revenue projections. To the
extent that appropriate probabilistic models can be formulated to specify local tower costs and
revenue receipts, one could generate answers for the entire United States, given sufficient effort.

Chapter 3: A Mathematical Programming… 30
REFERENCES
Acampora, A. (2002). Last Mile by Laser. Scientific American, 287(1), 49-53.

Bennet, J.L. (Ed.) (1983). Overview. Building Decision Support Systems. Reading, MA:
Addison-Wesley, 1-14.

Carstensen, L. W., Bostian, C.W., & Morgan, G.E. (2001). Combining Electromagnetic
Propagation: Geographic Information Systems, and Financial Modeling in a
Software Package for Broadband Wireless Wide Area Network Design. Proceedings of
the International Conference on Electromagnetics in Advanced
Applications, 799-810.

Crossland, M.D., Wynne, B.E., & Perkins, W.C. (1995). Spatial Decision Support Systems: An
Overview of Technology and a Test of Efficacy. Decision Support Systems, 3, 219-235.

ESRI (1999). Getting to Know Arcview GIS: The Geographic Information System (GIS) for
Everyone (3
rd
ed.). New York: Environmental Systems Research Institute, Inc.

Fowler, T. (2001). Mesh Networks for Broadband Access. IEE Review, 47 (1), 17-22.

Keen, P.G.W., & Scott Morton, M.S. (1978). Decision Support Systems: An Organizational
Perspective. Reading, MA: Addison-Wesley.

Keenan, P.B. (1998). Spatial Decision Support Systems for Vehicle Routing. Decision Support
Systems, 1, 65-71.

Kornbluh, K. (2001). “The Broadband Economy,” The New York Times, December 10, 2001,
Section A, page 21, column 2, editorial desk, dateline: Washington, D.C.

Maniezzo, V., Mendes, I., & Paruccini, M. (1998). Decision Support for Siting Problems.
Decision Support Systems, 3, 273-284.

Nehme, C.C., & Simões (1999) M. Spatial Decision Support System for Land Assessment. in
ACM GIS. Kansas City, MO: ACM Press.

Nokia, (2002). Nokia Demonstrates Wireless Mesh Broadband Market Leadership with More
Than 50 Network Customers. Press release. Retrieved July 30, 2002, from the World
Wide Web: http://press.nokia.com/PR/200202/848470_5.html.

Shrick, B., & Riezenman, M.J. (2002). Wireless broadband in a box. IEEE Spectrum, 39(6), 38 –
43.

Simon, H. (1960). The new science of management decision. New York: Harper & Row.

Chapter 3: A Mathematical Programming… 31
Sprague, J.R.H., & Carlson, E.D. (1982). Building Effective Decision Support Systems.
Englewood Cliffs, NJ: Prentice-Hall, Inc.

Taha, H.A. (1975). Integer Programming - Theory, Applications, and Computations. New York:
Academic Press.

Tarantilis, C.D., & Kiranoudis, C.T. (2002). Using a Spatial Decision Support System for
Solving the Vehicle Routing Problem. Information & Management, (5), 359-375.

TechNet (2002). A National Imperative: Universal Availability of Broadband by 2010.
Technical report. Retrieved July 30, 2002, from the World Wide Web:
http://www.technet.org.

Turban, E., & Aronson, J.E. (2001). Decision Support Systems and Intelligent Systems. New
Jersey: Prentice-Hall. 867.

USGS (2002). USGS Digital Elevation Model Data. United States Geological Survey.
http://edcwww.cr.usgs.gov/glis/hyper/guide/usgs_dem

Vlachopoulou, M., Silleos, G., & Manthou, V. (2001). Geographic Information Systems in
Warehouse Site Selection Decisions. International Journal of Production Economics, (1-
3), 205-212.

West, J., Lawrence A., & Hess, T.J. (2002). Metadata as a Knowledge Management Tool:
Supporting Intelligent Agent and End User Access to Spatial Data. Decision Support
Systems, (3), 247-264.

Willebrand, H.A., & Ghuman, B.S. (2001). Fiber Optics without Fiber. IEEE Spectrum, (8), 41-
45.

Willis, B., T. Hasletad, T. Friis∅, O.B. Holm (2001). Exploiting Peer-to-Peer Communications –
Mesh Fixed and ODMA Mobile Radio. Journal of the IBTE. 2(2), 48-53.

Chapter 3: A Mathematical Programming… 32











Figure 3.1. A GETWEBS screen shot indicating the view shed of a tower placed at the dot
near the center of the top of the screen and with range indicated by the circle.
Chapter 3: A Mathematical Programming… 33






5 km
5 km
10 km
East West

N N
10 km

Figure 3.2a. A fictitious town, divided into an east and west region by a railroad track that runs
north/south through the center of town.


















Figure 3.2b. An elevation map of the fictitious town.
elevation
above sea level
(meters)
West East
400
300










Figure 3.2c. A transmission tower’s signal can reach diagonally to all households two cells
away.


Chapter 3: A Mathematical Programming… 34
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50


Notes:
1. With regard to Figure 3.2, cell 1 is in the upper left corner of the town, and cell 50
the bottom right corner.
2. Black cells indicate a 1 in cell (i, j), wheras white cells indicte a 0.
3. Checkered cells indicate a 1 in the exposure matrix E and a 0 in the view shed
matrix V.

Figure 3.3. The exposure matrix E and the view shed matrix V for the fictitious town depicted
in Figure 3.2.
Chapter 3: A Mathematical Programming… 35








6 5 4
3 2 1

Figure 3.4. A fictitious mini-town, divided into six cells.
Chapter 3: A Mathematical Programming… 36













MATH PGM
VISIBILITY
EQUIPMENT
DEFINITION
VIEW SHED
GETWEBS
REVENUE
PROPENSITY
TO PAY
HOUSEHOLDS
PROFIT
SHED
INCOME, ETC. PENETRATION
Census data
EVALUATION CRITERIA




MATH PGM MODEL
• max profit
o budget
year by year
• max exposure
o budget
in/exclude
FINANCIAL MODEL
• economic life
• cost of money
• depreciation, etc.



Figure 3.5. The wireless spatial DSS architecture
Chapter 3: A Mathematical Programming… 37



Figure 3.6. Cells may be individually excluded from (e.g. 26, 38, and 74) or included (e.g. 2,
32, and 88 ) for tower placement as desired.
Chapter 3: A Mathematical Programming… 38


Figure 3.7. Tower placement and profit sheds (magenta) for Wi-Fi (802.11b) service offered
at $50/month.
Chapter 3: A Mathematical Programming… 39

igure 3.8. Profit, households, and towers in Montgomery County as a function of percent

0
10
20
30
40
50
60
50% 55% 60% 65% 70% 75% 80% 85% 90% 95%
Percentage of Households
Profit ($00000's)
Households (000's)
Towers
F
coverage with Wi-Fi broadband.
Chapter 3: A Mathematical Programming… 40
Table 3.1. Some wireless applications and their allocated frequencies.

Frequency Band Applications
170 - 190 kHz
VLF band radios
Beacons
550 - 1600 kHz
AM broadcast communications
Low power voice and data
27 MHz
CB radios
Low power voice and data
49 MHz
Remote control
Cordless telephones
Low power voice and data
88 - 108 MHz
FM radios
Low power voice and data
150 - 170 MHz
Commercial two-way voice
Pager services
260 - 470 MHz
303 MHz garage door openers
Keyless entry systems
Security alarms
824 - 849 MHz and 869 - 894 MHz Cellular phones
902 - 928 MHz
ISM band Wireless LAN's
Part 15 devices (spread spectrum
cordless phones, etc.)
Military radiolocation systems
Federal mobile communications
930 MHz
Pager services with high transmitter
power
2.4 - 2.4385 GHz
Amateur satellite
Part 15 devices
Microwave ovens and systems
Army packet radio development
802.11b
5.15-5.25GHz
5.25-5.35GHz
U-NII
802.11a
5.727 - 5.875 GHz
Amateur satellite
Part 15 devices
Naval radar systems
Test range instrumentation radars
802.11a
802.16 – (Tsunami equipment)
24 - 24.25 GHz Radio navigation
28 – 31 GHz
Local multipoint distribution service
(LMDS)

Source: The Virginia Tech Center for Wireless Communications.
(http://www.cwt.vt.edu/wireless_faq/default.htm).
Chapter 3: A Mathematical Programming… 41

Table 3.2. The percentage participation (in Montgomery County) as a function of average
household annual income, given the price per month for wireless service.

Percent Participation (%P) Average Annual
Household Income, in 000s
(I)

$50/mo price

$75/mo price

$100/mo price
$0 to $75 %P = 0.2 I %P = 0.1 I %P = 0.05 I
$75 to $100 %P = 0.4 I − 15 %P = 0.1 I %P = 0.05 I
>$100 %P = 25 %P = 10 %P = 5

Chapter 3: A Mathematical Programming… 42
CHAPTER 4


A CAPACITATED, FIXED-CHARGE, NETWORK-FLOW MODEL FOR
SOLVING HOP CONSTRAINED, WIRELESS MESH NETWORKS




































Chapter 4: A Capacitated, Fixed-Charge… 43

A CAPACITATED, FIXED-CHARGE, NETWORK-FLOW MODEL FOR
SOLVING HOP CONSTRAINED, WIRELESS MESH NETWORKS





ABSTRACT
Wireless mesh networks are proposed as a possible solution to the “last-mile” problem in
broadband communications. A solution procedure is developed here that generates all feasible
customer paths to the backhaul in a preprocessing step, then solves the resulting problem as a
fixed-charge, capacitated, network flow problem, whose solution is well known. In particular,
mesh networks with hop constraints that ensure transmission quality are examined. The model is
capable of handling differing bandwidth requirements for each customer.

Keywords: Network Flow Models, Math Programming, Broadband Wireless
Telecommunications

Chapter 4: A Capacitated, Fixed-Charge… 44

A CAPACITATED, FIXED-CHARGE, NETWORK-FLOW MODEL FOR
SOLVING HOP CONSTRAINED, WIRELESS MESH NETWORKS

INTRODUCTION
The popularity and rate of growth of the Internet over the last decade has been
prodigious; what was once a medium for government and academics to exchange ideas has now
become an almost ubiquitous entity that assumes many shapes – from personal communications,
to on-line shopping, to global information collection. As more people and companies today do
business on line, the need for greater bandwidth increases; it is not longer sufficient to connect to
the Internet with low speed modems. Broadband telecommunications have become a necessity,
whereby broadband is meant the capacity to deliver high speed, always “on” service. (TechNet,
2002)
In the 1990s, companies poured billions of dollars into building a transcontinental fiber
optic backbone (Kornbluh, 2001, p. 21). However, as Kornbluh states in her NY Times op-ed
article, only three percent of that backbone is currently in use. This is due, primarily, to
underestimation of the cost of delivering broadband service the “last-mile,” where the “last-mile”
is that part of the telecommunication network that connects the homes and businesses to the high
speed backbone. While most American homes and businesses exist relatively near the backbone,
they are not connected to it. In fact, Acampora (2002) states that for nine out of ten American
businesses with more than 100 workers, the backbone is less than a mile away.
This last-mile link to the customer has traditionally consisted of copper wiring, and has
been sufficient for voice communications for half a century. But it will not be adequate,
particularly financially, to meet for the broadband requirements of the new millennium. One
approach to solving this problem is wireless telecommunications (Willebrand and Ghuman,
2001). Several varieties of this approach exist (see, e.g., Scheibe, Carstensen, Rakes, and Rees,
2003, for a discussion of the point-to-multipoint (PMP) line of attack), including mesh, which we
will define shortly.
The purpose of this paper is to ascertain the minimum-cost, mesh network configuration
of a neighborhood. We will utilize Geographic Information Systems (GIS) in a preprocessing
analysis step to determine which neighbors can see each other electronically, as obstructions
such as buildings and terrain prohibit transmission. This discovery of viable communication
links is combined with an enforcement of number of hops from each source (customer) to sink
(backhaul), thereby ensuring a quality of service (QoS) level. We then enumerate every viable,
feasible path from customers to the backhaul, and formulate a mixed integer programming
problem, namely the capacitated, fixed-charge, network flow problem (FCNFP), to determine the
optimal network with minimal cost. After background material on meshes as well as on the
general capacitated FCNFP, the solution methodology (including the mixed integer formulation)
is presented. This is followed by an example, and then by implications and limitations. Finally,
the paper concludes with a summary of contributions and requisite future work.

Chapter 4: A Capacitated, Fixed-Charge… 45

BACKGROUND
Mesh Networks
As mentioned, the fundamental issue with the last mile problem is that capacity exists in
broadband network, but it is not economic to connect residential neighborhoods and/or
businesses by wire to that backbone. To facilitate discussion of our approach to providing such
capability in a wireless manner, we define some terms as they are used in practice or in the
literature.
In Figure 4.1 an existing backbone and neighborhood without broadband service are
shown. The backhaul point (BH in Figure 4.1) is that location along the backbone network
where the mesh network will connect. The backhaul is then generally "wired" (say with fiber
optics) to trunk connection points, shown as T
1
and T
2
in Figure 4.1; the transmission medium
changes to wireless at these trunk connection points. The wireless signal is not, in general,
transmitted from the trunk connections directly to customers, however. The signal is sent to
mesh insertion points (MIPs, labeled M
1
, M
2
, and M
3
in the figure) that are closer to customers.
These mesh insertion points do transmit and receive directly to/from customers. Generally, once
signal is received at a MIP back from a customer, it is only forwarded to a trunk connection point
and then to the backhaul. That is, signals are not transmitted from customers to MIPs, and then
back out to customers.
One of the key features of mesh networks is that customers are used to forward signals on
to other customers, up to the capacity limits of the receiver and transmitters. This can reduce
network costs as well as provide network survivability, that is, redundant paths. However,
sometimes it is necessary to insert repeater nodes into the network when there are not enough
customers nearby to get messages to every customer. Repeater nodes are generally identical to
customer equipment – there is just not a customer at the repeater location. Thus a repeater may
be thought of as a non-revenue generating customer. Mesh networks may use a variety of
frequencies, such as 2.4 GHz (Nokia, 2002), 26 GHz to 40 GHz (Fowler, 2001), and laser
(Acampora and Krishnamurthy, 1999).
Because mesh frequencies are in the gigahertz range, transmission between each
transmitter and receiver must be line of sight. That is, not only must equipment be proximate
and within range, there must be no obstruction between transmitter and receiver.
In a mesh network signals from a customer seeking the backhaul may be forwarded from
customer to customer (possibly including repeaters), until finally a mesh insertion point is
reached. The MIP then sends the signal to a trunk connection point, and finally to the backhaul.
Each link in the chain back to the backhaul is called a hop. Too many hops in a chain of
connectivity can cause inordinate delay, or latency. Latency is the time it takes for a packet to
travel from source to destination (Webopedia, 2003). To ensure quality of service, the
probability of a network meeting a given traffic contract (Wikipedia, 2003), it is sometimes
necessary to restrict the number of hops in each transmission path. A traffic contract may be
viewed as an agreed upon throughput rate during peak usage times. Although hop constraints are
not new to network design (Balakrishnan and Altinkemer, 1992), there has been little discussion
in the literature as to what should be the optimal number of links for a wireless mesh network.
Whitehead suggests that six to ten hops offer an acceptable level of service (Whitehead, 2000).
Wireless networks have traditionally been point-to-multipoint (PMP) networks
(Whitehead, 2000). Mesh networks overcome several limitations of point-to-multipoint (PMP)
networks. The former work with strategically placed towers, connected to the backhaul, and
Chapter 4: A Capacitated, Fixed-Charge… 46

within range of the maximal number of potential customers. That approach necessitates
purchase of/permission to use land, permitting, bringing electric power to the tower, etc. Mesh
networks obviate most of these concerns. Note also that with a mesh, a connection point does
not need to be optimally placed to reach every customer directly. This allows for greater
flexibility in choosing connection node locations, and should lower the initial investment cost
(Whitehead 2000). With mesh networks, it is not crucial to locate the single tower location that
will best serve the entire area; it is only necessary to determine a connection point location that
will bring signal into one section of the mesh. From that point, the other nodes will pass the
signal along. Moreover, the incremental addition of customers to the network is inexpensive;
often, the only cost is the customer premise equipment. In fact, by adding more customers, the
cost of the mesh network can decrease. This occurs when, as more customers join the network,
the need for repeater nodes decreases because paths to the backbone may be found through
revenue-paying customers (Fowler, 2001).
Fixed-Charged Network Flow Models
The fixed-charge network flow problem (FCNFP) is one of a large class of network
design problems, which has been used in many applications including telecommunications
(Balakrishnan, Magnanti, Shulman, and Wong, 1991; Gavish, 1991), logistics and production
planning (Minoux, 1989), and transportation (Magnanti and Wong, 1984). FCNFP problems are
known to be NP-hard, and much research has been devoted to creating better and more efficient
solution procedures (Kim and Pardalos, 1999; Cruz, Smith, and Mateus, 1998). In these
problems, the objective is to seek the most efficient way to move flow (in our case bandwidth)
on a network in order to satisfy demand between origin and destination nodes and to minimize
the overall cost. A fixed cost is incurred for using arcs between nodes; therefore, it behooves the
objective to use as few arcs as possible for which costs are associated. When the network is
capacitated, upper bounds exist for the amount of flow over an arc. Another consideration for
some network design problems is the number of hops from a source to a sink (Pirkul and Soni,
2002; Soni, 2001; Girish, Zhou, and Hu, 2000; Gouveia and Requejo, 2001). The principal
reason for constraining the number of hops in a network is to maintain a level of quality of
service. Simply stated, the greater the number of hops between nodes, the greater the amount of
time it takes for data to be transferred between them; the longer it takes, the lower quality of
service.
The mesh network can be cast as a FCNFP with the customers acting as sources, the
backhaul as the sink, and TCN, MIP, and repeaters as transshipment nodes. There is a fixed cost
associated with arcs that are connected to and from the transshipment nodes, and arcs are
constrained with an upper bound or bandwidth capacity.
The capacitated, fixed-charge, network-flow problem can be formulated mathematically
as follows: Given a directed graph G = (N, A) where N is a set of n nodes and A is a set of m
arcs, let (i,j) denote a directed arc from node i to node j. Every node is classified into one of
three categories: Source nodes, which produce flow; sink nodes; which consume flow; and
transshipment nodes, which pass flow through. We assume a balanced network where total
supply equals total demand (or that we have balanced the network by introducing either slack or
surplus variables to consume excess capacity or demand). For each node k, the design cost is
denoted f
k
, which is assumed to be nonnegative. The objective is to minimize the design costs
that occur when arcs are used. Let d
j
denote the supply or demand at node j, and u
i
> 0 be the
capacity of each node i.
Chapter 4: A Capacitated, Fixed-Charge… 47

In this formulation, we also define x
ij
to be the flow on arc (i,j), and y
i
to be a binary
variable equal to one when node i is in use and zero otherwise. The resulting formulation is



∈N k
k k
y f , min (1)

∑ ∑
∈ ∋ ∈ ∋
∈ ∀ = −
A k j k
j jk
A j k k
kj
N j d x x
) , ( ) , (
, (2)
(3) A j i N i y u x
i i
j i
ij
∈ ∈ ∀ ≤

) , ( ; ,
,
, ) , ( , 0 A j i x
ij
∈ ∀ ≥ (4)
} 1 , 0 { ∈
ij
y (5)

Constraint (2) is the flow conservation constraint and ensures that demand/supply is
satisfied at each node j, and that inflows equal outflows for transshipment nodes. If d
j
> 0, then j
is a sink node. If d
j
< 0, then j is a source node, and if d
j
= 0, then j is a transshipment node.
Constraint (3) is the flow capacity constraint, and it ensures that flow on all arcs leaving a node
will not exceed the capacity of the node at the start of that arc, and will be zero if the node is not
made operative in the design.
METHODOLOGY
The approach taken in this chapter is to introduce a preprocessing step whose main
purpose is to determine, given a hop constraint, all viable paths from each customer back to the
backhaul. It is possible to enumerate these paths, as the number of customers for any mesh
insertion point is bounded by the technology and since the number of hops on each possible path
is also constrained by quality of service considerations. Once all possible paths are enumerated,
a minimum cost solution may be found using integer programming solution packages.
Preprocessing
Step 1 –Potential Equipment Placement and Line-of-Sight Determination
Preprocessing begins with the specification of the backhaul point, trunk connection
points, and all possible mesh insertion points. These equipment locations are primarily a matter
of physical characteristics of a neighborhood, that is, where right-of-way/permissions are
available, terrain is not obstructive, etc. Again, with the methodology developed here, the
location of the single backhaul point and two trunk connection points are assumed. It is
unknown how many MIPs will be economic, but all potential locations are to be specified; which
locations are chosen is an economic decision to be determined by the model.
Next, all known customer locations are specified. The situation to this point is indicated
in Figure 4.2, where each circle (node) indicates the (geographic) location of the
equipment/customer indicated. Line-of-sight must be determined next; either a geographic
information system (GIS) or field visit is required. Figure 4.3 can then be generated, where a
dotted line between any two circles indicates no obstruction between the two nodes. (Lack of a
dotted line indicates no feasible transmission; e.g., customer C3 cannot transmit or receive
to/from customer C4 in Figure 4.3.) At this point it may occur that a customer desirous of
Chapter 4: A Capacitated, Fixed-Charge… 48

service cannot obtain a signal, either because transmission range limits are exceeded or because
obstructions block all possible paths to the customer. In such an eventuality, potential repeater
nodes (Figure 4.3) are defined at non-customer locations that will ensure an unobstructed path to
the backhaul for each customer. Finally in Figure 4.3, arrows are shown on the links from each
MIP to its closest trunk connection point, and from each trunk connection point to the backhaul.
Recall that signals received at MIPs from customers do not go back out to other customers; they
move on to trunk connection points. Information equivalent to Figure 4.3 may be stored in
tabular form (see Table 4.1). Once this information is in hand, all possible paths from each
customer to the backhaul may be enumerated, given a limit on number of hops. The possible
paths for customer C1 are shown in Figure 4.4.
Step 2 – Artificial Directional Node Determination
We use Excel to process feasible, hop-constrained paths. Excel has proven to be an
excellent tool for many problems of this sort (Ragsdale, 2001). With this approach the simplest
way to generate these paths is to define what we term artificial directional nodes (ADNs). A
simple example is presented in Table 4.2 that is based on Figure 4.4 and Table 4.1 and a
(temporary) hop limit of six that illustrates the process for customer C1.
The method for creating and naming the ADNs is as follows. Begin with the source node
(in this case, C1), count the number of hops from the source to the BH, and append that number
after a period. For example, the first row of Table 4.2 has node C1 reaching the backhaul in
three hops, so C1 becomes C1.3 in Table 4.3. Next, move from the source to the subsequent hop
in line to the BH; if the next node is not a MIP, a TCN, or the backhaul, then append the number
of hops to the BH from that point. On line two of our example, the next node from C1.4 is C2,
which becomes C2.3. Every number with a period represents an ADN. Following this
procedure creates a new (artificial) node for every possible number of links from the node back
to the backhaul. ADNs are generated in this fashion for every customer and repeater node; they
are not needed for MIPS, TCNs, or the backhaul, as the number of nodes from each is constant.
Shown in Table 4.3 and Figure 4.5 is the example with artificial directional nodes. Table 4.4
lists each node as a source and indicates the one-step-away nodes, one per row. With this list the
construction of any feasible path may be automated, as the connectivity and hop constraint
information is all included.
Figure 4.6 shows the re-integration of the nodes into the mesh by connecting them to
their artificial directional nodes. Since ADNs are not actual physical nodes in the network, their
addition does not violate any hop constraints. Even though it appears that node C1 must now
also pass through C1.3 – C1.6, they are all actually node C1 themselves. Therefore, there is no
hop constraint violation.
Table 4.5 presents the new mesh network with all the artificial directional nodes along
with the re-integrated nodes themselves. This is the last step before transforming the data into a
form that may be solved directly with the network flow model.
Network Flow Model
Once the artificial directional nodes and the actual nodes have been integrated into the
new mesh network, the network flow model may be formulated. The nodes may be viewed as a
series of sources and sinks, as shown in Table 4.6. Each row in the table represents an allowable
hop within the mesh network.
Chapter 4: A Capacitated, Fixed-Charge… 49

We modify the general FCNFP model of equations (1) – (5) as follows. Let N be the set
of all nodes and N
a
be the set of artificial nodes. Further designate T, M, R, and C as the sets of
trunk connection nodes, mesh insertion points, repeater nodes, and customer nodes respectively.
Finally, define the set of all arcs in the network by the letter A.
For each node k, let the cost of activating that node be f
k
, which is assumed to be
nonnegative; any artificial nodes have a cost of 0. The objective is to minimize total cost. Let d
j

denote the supply or demand at node j, all artificial nodes should be considered transshipment
nodes (with a supply of 0). Let u
i
> 0 represent the capacity on the arcs emanating from node i.
x
ij
is the flow on arc (i,j), and y
i
is a binary variable equal to one when node i is in use and zero
otherwise.
The model may be written as


∈N k
k k
y f , min (6)

∑ ∑
∈ ∋ ∈ ∋
∈ ∀ = −
A k j k
j jk
A j k k
kj
N j d x x
) , ( ) , (
, (7)
(8) ) ( ,
) , (
M T i y u x
i i
A j i j
ij
∪ ∈ ∀ ≤

∈ ∋
(9)
i i
j i j
ij
y u x ≤

∋ ) *, (
]} * [ ] *) , [(
)] ( ) ( {[
a
a
N i A i i
N i R C i i
∈ ∩ ∈ ∩
∉ ∩ ∪ ∈ ∋ ∀
, ) , ( , 0 A j i x
ij
∈ ∀ ≥ (10)
N i y
i
∈ ∀ ∈ }, 1 , 0 { (11)
Constraint (7) ensures flow conservation, whereas constraints (8) and (9) are flow
capacity constraints. Constraint (8) restricts flow from trunk nodes and mesh insertion points,
whereas (9) restricts flow emanating from customer or repeater nodes, as follows. Flow leaving
either a customer or a repeater node (i.e., the sum of flow over all arcs leaving such a node)
cannot exceed the capacity of the transmitter at that node; moreover, if the transmitter is
inoperative, no flow may traverse any arc departing that node.
EXAMPLE
To demonstrate the model a small example was constructed. The example consists of
three neighborhoods with a total of 25 houses requiring wireless broadband service. There are
two potential TCNs connected to the backhaul, eight potential MIPs, and fourteen potential
repeaters. The actual location of the aforementioned nodes is set; their potentiality exists in
whether or not they will be used. The example assumes a visibility of 60%, meaning that nodes
that are located near enough to each other for a signal to pass between may still have a 40%
chance that something obstructs line of sight. The lots are assumed to be one acre each, and the
houses are placed on the center of each lot. We consider two independent variables, the number
of hops and bandwidth. These factors have three levels each – low, medium, and high. The
levels of the hops are five, seven, and nine, and the levels for the bandwidth will be 2 Mbits/s,
Chapter 4: A Capacitated, Fixed-Charge… 50

5Mbits/s, and 10 Mbits/s. The arcs between nodes are capacitated at 30 Mbits/s between
customer nodes and repeaters, 80 Mbits/s between MIPS and any other nodes, and 200 Mbits/s
between TCNs and MIPS. The cost for using repeaters is $2,500 each; the cost for MIPs is
$5,000 a piece; and the cost for TCNs is $25,000 each. Shown in Tables 4.7 and 4.8 are some
results of the example.
Table 4.7 shows pertinent data from the example’s preprocessing step, in particular, the
number of source and sink nodes and viable paths created for each of the hop levels. Note from
this example that while the number of viable paths increased dramatically with each level of
hops, the number of source and sink nodes did not. This is due to high level of redundancy of
paths from a node to the backhaul. While the number of viable paths may be high, many of the
paths will pass through the same pairs of nodes. Table 4.8 shows some of the results of running
the mixed integer problem code, namely the total cost for each network. Since the network is
capacitated, as the number of Mbits increases, the bandwidth is split up over multiple paths, thus
increasing the cost as more repeaters and MIP nodes are required. Another noteworthy point is
that the cost does not necessarily decrease as the number of hops increases. While it may seem
intuitive that the cost would decrease by increasing the number of hops, and consequently, the
number of viable paths from a customer node to the backhaul, if the nodes are not located such
that they may avoid repeater nodes, the overall cost may actually increase.
Although this is a minimum cost solution procedure, results may be manipulated to
examine sundry revenue schemes. For instance, in the example given above and its assumed
cost structure, revenue of $127 per month for the basic 2 megabit service with a hop constraint of
seven proves profitable within a five-year window – a not very lucrative opportunity.
IMPLICATIONS
The solution procedure successfully generates answers to the wireless mesh broadband
service problem. Small problems may be run using Excel combined with Frontline’s Premium
Solver (Frontline, 2002). Larger problems may need a more sophisticated integer-programming
engine. The procedure works reasonably well at nine or fewer hops, but, as mentioned,
generates a very large number of viable paths in the preprocessing step. With each additional
hop added to the allowable number, the problem size grows dramatically, and consequently,
proves expensive to solve. This issue must be addressed in future work. Although mesh
networks should not indiscriminately route customer signals over too many hops, nine hops is
probably not excessive. However, it is approaching an upper bound on the number of hops.
Several simple heuristics can be attempted (such as limiting the search to a fixed number of
randomly chosen paths as the algorithm steps out from each node in search of additional hops to
the backhaul) to limit the number of paths and thus the execution time of the integer
programming part of the problem.
CONCLUSIONS AND FUTURE WORK
A methodology has been presented to determine minimum cost wireless mesh network
configurations. The methodology preprocesses data by first using a GIS and then manipulating
customer requests and equipment locales to form a list of feasible paths from which the optimal
set of paths may be found by solving a well-known network-flow problem. The network
generated is constrained by a given number of hops, or links, to ensure quality of service, and
Chapter 4: A Capacitated, Fixed-Charge… 51

allows differing bandwidths for each customer, as desired. An example was shown
demonstrating how an Internet service provider (ISP) would approach service provision
constrained at seven hops in a new neighborhood. Bandwidth requested was ten megabit per
second for all customers. Results were also shown when quality of service was loosened to nine
hops and tightened to five hops with three different bandwidth levels.
Future work examines managerial implications of wireless mesh technology.
Implications of customer density, cost of technology, and the effects of quality of service and
bandwidth are possible. This research is important in estimating the extent to which wireless
mesh broadband solutions may be expected to mitigate the “last-mile” problem. The tool
developed here may not only be used by ISPs, but by policy bodies to examine the wisdom of
subsidies, and by equipment manufacturers to examine the likely adoption of new transmitters
with improved ranges and lower costs.
Chapter 4: A Capacitated, Fixed-Charge… 52

REFERENCES
Acampora, A. (2002). Last Mile by Laser. Scientific American, 287(1), 49-53.

Acampora, A. and Krishnamurthy, S. (1999). A broadband wireless access network based on
mesh-connected free-spaced optical links. IEEE Personal Communications, 6(5), 62-65.

Balakrishnan, A., Altinkemer, K. (1992). Using a hop constrained model to generate alternative
communications netowrk design. ORSA Journal on Computing, 4,192-205.
Balakrishnan A., Magnanti T.L., Shulman A. and Wong R.T. (1991). Models for Planning
Capacity Expansion in Local Access Telecommunication Networks. Annals of
Operations Research. 33, 239-284.
Carstensen, L. W., Bostian, C.W., and Morgan, G.E. (2001). Combining Electromagnetic
Propagation: Geographic Information Systems, and Financial Modeling in a Software
Package for Broadband Wireless Wide Area Network Design. Proceedings of the
International Conference on Electromagnetics in Advanced Applications, 799-810.

Crossland, M.D., Wynne, B.E., & Perkins, W.C. (1995). Spatial Decision Support Systems: An
Overview of Technology and a Test of Efficacy. Decision Support Systems, 3, 219-235.

Cruz F.R.B., Smith J.M., and Mateus G.R. (1998). Solving to optimality the uncapacitated
fixed-charge network flow problem. Computers and Operations Research. 25(1), 67-81.
ESRI (1999). Getting to Know Arcview GIS: The Geographic Information System (GIS) for
Everyone (3
rd
ed.). New York: Environmental Systems Research Institute, Inc.

Fowler, T. (2001). Mesh Networks for Broadband Access. IEEE Review, 47 (1), 17-22.

Frontline. (2002). Frontline’s Premium Solver™, Frontline Systems, Inc., Incline Village, NV.
From the World Wide Web: http://www.solver.com.
Gavish B. (1991). Topological Design of Telecommunications Networks – Local Access Design
Methods. Annals of Operations Research. 33, 17-71.
Girish, M.K., Zhou B., and Hu, J. (2000). Formulation of the Traffic Engineering Problems in
MPLS Based IP Networks. IEEE. 214-219.
Gouveia, L. and Requejo C. (2001). A new Langrangean relaxation approach for the hop-
constrained minimum spanning tree problem. European Journal of Operational
Research. 132, 539-552.
Kim, D. and Pardalos P. (1999). A solution approach to the fixed charge network flow problem
using a dynamic slope scaling procedure. Operations Research Letters. 24, 195-203.
Chapter 4: A Capacitated, Fixed-Charge… 53

Kornbluh, K. (2001). “The Broadband Economy,” The New York Times, December 10, 2001,
Section A, page 21, column 2, editorial desk, dateline: Washington, D.C.

Magnanti, T.L. and Wong, R.T. (1984). Network Design and Transportation Planning: Models
and Algorithms. Transportation Science. 18(1), 1-55.
Minoux, M. (1989). Network Synthesis and Optimum Network Design Problems: Models,
Solution Methods and Applications. Networks. 19, 313-360.
Nokia, (2002). Nokia Demonstrates Wireless Mesh Broadband Market Leadership with More
Than 50 Network Customers. Press release. Retrieved July 30, 2002, from the World
Wide Web: http://press.nokia.com/PR/200202/848470_5.html.
Pirkul, H. and Soni, S. (2002). New formulations and solution procedures for the hop
constrained network design problem. European Journal of Operational Research, In
Press.
Ragsdale, C.T., Spreadsheet Modeling and Decision Analysis: A Practical Introduction to
Management Science, Third Edition, Southwestern College Publishing, Cincinnati, 2001.
Scheibe, K.P., Carstensen, L.W., Rakes, T.R., and Rees, L.P. (2003). A Mathematical-
Programming And Geographic-Information-System Framework For Wireless Broadband
Deployment In Rural Areas. Under review.
Soni, S. (2001). Hop Constrained Network Design Problem with Partial Survivability. Annals of
Operations Research. 106, 181-198.
Webopedia (2003). Retrieved March 10, 2003, from the World Wide Web:
http://www.webopedia.com/TERM/l/latency.html.
Whitehead, Philip (2000). Mesh Networks: A new Architecture for Broadband Wireless Access
Systems. RAWCON.
Wikipedia (2003). Retrieved February 10, 2003, from the World Wide Web:
http://www.wikipedia.org/wiki/Quality_of_Service.
Willis, B., T. Hasletad, T. Friis∅, O.B. Holm (2001). Exploiting Peer-to-Peer Communications –
Mesh Fixed and ODMA Mobile Radio. Journal of the IBTE. 2(2), 48-53.
Chapter 4: A Capacitated, Fixed-Charge… 54


Figure 4.1. Providing broadband service to a neighborhood with a wireless mesh network.
Chapter 4: A Capacitated, Fixed-Charge… 55

T1 T2
M1 M2
C2
C1
C3
C4
BH





Figure 4.2. Nodes represent potential wireless mesh equipment locations.
Key:
BH – backhaul
Ti – trunk i
Mj – mesh insertion point j
Rk – repeater k (not shown)
Cl – customer l






T1 T2
M1 M2
R1
C2
C1
C3
C4
BH

Figure 4.3. Nodes shown linked together with line-of-sight capabilities.


Chapter 4: A Capacitated, Fixed-Charge… 56

T1 T2
M1 M2
R1
C2
C1
C3
C4
BH

Figure 4.4. Viable paths for customer C1 to communicate with the backhaul, given the line-
of-sight configuration of Figure 4.3.






T1
T2
M1
M2
BH
C1.3
C2.5
C2.4
C2.3
C3.3
C4.4
R1.3
C1.4
C1.5
C1.6

Figure 4.5. Customer C1 viable paths expressed in terms of artificial directional nodes.


Chapter 4: A Capacitated, Fixed-Charge… 57

C2
C1
M1
C1.3
C2.5
C2.4
C2.3
C1.4
C1.5
C1.6

Figure 4.6. Re-linking nodes to their artificial directional nodes.

Chapter 4: A Capacitated, Fixed-Charge… 58

Table 4.1. Source and neighbor nodes
Source “Reachable” Neighbors
C1 M1 C2


C2 M1 C1 C3 C4
C3 M1 C2


C4 R1 C2


R1 M2 C4


M1 T1
M2 T2
T1 BH
T2 BH



Table 4.2. Viable paths for customer C1 to communicate with the backhaul
Source
Viable paths
C1
M1 T1 BH
C1 C2 M1 T1 BH
C1 C2 C3 M1 T1 BH
C1 C2 C4 R1 M2 T2 BH



Table 4.3. Viable paths for node C1 with artificial directional nodes
Source Viable paths
C1.3
M1 T1 BH
C1.4 C2.3 M1 T1 BH
C1.5 C2.4 C3.3 M1 T1 BH
C1.6 C2.5 C4.4 R1.3 M2 T2 BH


Chapter 4: A Capacitated, Fixed-Charge… 59


Table 4.4. Source and immediate destination nodes with artificial directional nodes included
Source
Immediate
Destinations
C1.3 M1
C1.4 C2.3
C1.5 C2.4
C1.6 C2.5
C2.3 M1
C2.4 C3.3
C2.5 C4.4
C3.3 M1
C4.4 R1.3
R1.3 M2
M1 T1
M2 T2
T1 BH
T2 BH



Table 4.5. Re-linking nodes to their artificial directional nodes
Source
Possible Immediate
Destinations
C1.3 M1
C1.4 C2.3
C1.5 C2.4
C1.6 C2.5
C2.3 M1
C2.4 C3.3
C2.5 C4.4
C3.3 M1
C4.4 R1.3
R1.3 M2
M1 T1
M2 T2
T1 BH
T2 BH
C1 C1.3 C1.4 C1.5 C1.6
C2 C2.3 C2.4 C2.5
C3 C3.3
C4 C4.4

Chapter 4: A Capacitated, Fixed-Charge… 60

Table 4.6. List of sources and sinks for direct inclusion into the network flow model
Source Sink
T1 BH
T2 BH
M1 T1
M2 T2
C1 C1.3
C1 C1.4
C1 C1.5
C1 C1.6
C1.3 M1
C1.4 C2.3
C1.5 C2.4
C1.6 C2.5
C2 C2.3
C2 C2.4
C2 C2.5
C2.3 M1
C2.4 C3.3
C2.5 C4.4
C3 C3.3
C3.3 M1
C4 C4.4
C4.4 R1.3
R1.3 M2

Table 4.7. Some Results from the Example’s Preprocessing Step
Source and Sinks Viable Paths
Five Hops 337 2,131
Seven Hops 832
18,928
Nine Hops 760 1,202,048


Table 4.8. Results from the Example’s Capacitated, Fixed-Charge, Network-Flow Problem
2 Mbits/s 5 Mbits/s 10 Mbits/s
Five hops $255,000 $255,000 $300,000
Seven hops $190,000 $205,000 $285,000
Nine hops $190,000 $210,000 $295,000


Chapter 4: A Capacitated, Fixed-Charge… 61

CHAPTER 5


ADDRESSING UNIVERSAL-BROADBAND-SERVICE IMPLICATIONS
WITH WIRELESS MESH NETWORKS

































Chapter 5: Addressing Universal Broadband Service… 62

ADDRESSING UNIVERSAL-BROADBAND-SERVICE IMPLICATIONS
WITH WIRELESS MESH NETWORKS

ABSTRACT
Demand is increasing for broadband telecommunications, but most areas lack adequate
means of delivering the desired services to residential customers and businesses – at least for the
last mile of connectivity. A new approach to providing broadband technology is wireless
communications, and this technology has been touted as a possible savior for broadband.
This paper proposes a methodology for supplying broadband services using a wireless
mesh network structure. In order to examine the potential of such networks to mitigate the “last-
mile” problem in rural/suburban areas, 270 “neighborhoods” are randomly generated, and then
each neighborhood is evaluated for fiscal promise under nine different cases of bandwidth
offerings and quality of service. Results are discouraging for almost all cases. In fact, costs
would have to decrease (through either technological advances or government subsidy/tax
breaks) by 45% just to break even. Stated differently, customer monthly charges would have to
be reduced by $62/month per customer over five years for wireless broadband to achieve
profitable status. A study exploring government options is suggested as future work.


Keywords: Mesh networks, Wireless telecommunications, Geographic information systems,
Spatial decision support systems, Planning, Broadband service


ISRL Categories:
Chapter 5: Addressing Universal Broadband Service… 63
ADDRESSING UNIVERSAL-BROADBAND-SERVICE IMPLICATIONS
WITH WIRELESS MESH NETWORKS
INTRODUCTION
Broadband is the capacity to deliver Internet access with a continuous connection and the
ability to both send and receive digital content or services at high speeds (TechNet, 2002).
Organizations such as TechNet, a bipartisan group comprising more than 300 chief executive
officers and senior partners of the major companies in the fields of information technology,
biotechnology, venture capital, investment banking and law, claim that the productivity of
Americans in their homes and at work will increase with widespread adoption of broadband.
This, in turn, will have a potential affect on the United States (US) economy by $500 billion, and
positively benefit quality of living. Secondly, TechNet says true broadband is the key to the next
generations of communications and Internet services. (TechNet, 2002, p. 1)
However, Kornbluh (2001, p.21) points out that this potential will remain unrealized
without government action. She states in an op-ed piece in the New York Times that although
investors have plowed $90 billion into a cross-continental fiber-optic broadband network, today
merely 3% of that backbone is in use. She says it is the “digital equivalent of fallow farmland,”
and the problem is entrepreneurs failed to foresee the enormous cost of upgrading the “last mile
– copper telephone wires that connect individual homes and small businesses to the broadband
backbone” (Kornbluh, 2001, p. 21). In an article in Scientific American, Acampora (2002)
elaborates by noting that for nine out of ten American businesses with more than 100 workers,
the backbone is less than a mile away.
The issue of broadband connectivity has attracted the attention of the highest levels of the
US government. The current (2003) presidential administration has made statements concerning
the importance of broadband connectivity, although they have not been as strong as some in
congress would like. Senator Tom Daschle (D - SD) has made a call to bring broadband to every
American by the end of the decade, and a joint letter to President Bush from Senator Daschle and
Senator Gephardt (D - MO) states that deploying the technology should be a “national
imperative” (Dreazen, 2002). Moreover, TechNet has called on the president and policymakers
to make broadband a national priority and to set a goal of making an affordable 100-megabits per
second (Mbps) broadband connection available to 100 million American homes and small
businesses by 2010 (TechNet, 2002). To reach this goal, network providers will need to invest
hundreds of billions of dollars to upgrade infrastructures and increase bandwidth capacity in the
last mile, primarily by providing new wired connections to homes and offices. Today, virtually
no homes have connections with such bandwidth (TechNet, 2002). An intermediate goal is the
“availability of affordable broadband at speeds of at least 6 Mbps from 2 or more providers to at
least 50% of U.S. households and small businesses by 2004” (TechNet, 2002, p. 6). TechNet
developed 6 principles to address roadblocks and provide a guide to a national broadband policy.
Principle 5 reads as follows (pages 2-3):

Investment incentives, potentially including targeted tax
incentives, should encourage broadband deployment to
underserved communities and businesses. … In a market-oriented
environment that encourages the deployment of broadband
networks, there may still be a segment of the U.S. population that
Chapter 5: Addressing Universal Broadband Service… 64

does not have broadband availability. Public policies should seek
to narrow the current and future disparity in the level of high-speed
access to the Internet, to ensure that all Americans can enjoy the
benefits of broadband.

Similarly, Kornbluh encourages both Congress and the president to subsidize sparsely
populated regions of the country, low-income users, or both, suggesting that an ambitious
broadband strategy can help revitalize the economy. (Kornbluh, 2001, p.21)
While broadband connectivity is becoming a technological necessity, and while the
United States has a fiber-optic backbone that is severely underutilized with members of both the
government and industry calling for greater broadband coverage, America continues to fall
behind other developed countries in deployment of broadband services. Five years after the
phone and cable companies began offering broadband services, a mere fifteen percent of US
households use them (Woolley, 2002). This is primarily due to prohibitive cost of services.
Countries in which products and services have generally been more expensive, such as Japan, are
offering broadband services at almost half of what American citizens pay. As a result, while
Japan previously lagged behind the US by a factor of twenty in broadband usage, they have not
only caught up but are surpassing them (Woolley, 2002).
One popular alternative, which has emerged for providing this “last mile” connectivity, is
wireless networks (Willebrand and Ghuman 2001). Prompted by the high cost of fixed-media
connections and the problems with obtaining permission to run cable trenches within urban
areas, wireless technologies have received renewed attention. As an example, one case
encountered by LightPointe Communications involving the tradeoff between fiber cable and
laser-based wireless for connecting three urban buildings to a backhaul hub clearly illustrates the
cost differential (Willebrand and Ghuman 2001). Fiber cost was calculated on the basis of 1220
meters required to interconnect the buildings to the hub. At $325 per meter for trenching, cable,
and installation, the cost came to $396,500. In comparison, the cost to interconnect wirelessly
was a mere $59,000 for the transmitting and receiving equipment across the four locations.
While relative costs obviously depend on the number of locations, distance involved, local labor
rates, and numerous other factors, it is clear that wireless connectivity provides an alternative
worth exploring in many network settings.
This paper proposes a mesh network planning methodology and examines mesh networks
as a possible solution to the last-mile problem. In particular, after discussion of some
background material, details of the planning methodology are explained, and then results of
applying the approach to approximately 2500 randomly generated neighborhoods are presented.
The effects of density, quality of service, equipment costs, and transmission ranges are
examined. The paper concludes with overarching thoughts on the practicality and likely success
of the wireless mesh approach to solving the last-mile problem.
BACKGROUND
Technology Options
Some current fixed wireless technologies are Local Multipoint Distribution Services
(LMDS), Multipoint Microwave Distribution Services (MMDS), the IEEE 802.11 standards for
wireless connectivity, and free space optics (FSO). Although other options exist, these four
provide a significant starting point for this research.
Chapter 5: Addressing Universal Broadband Service… 65

LMDS
Local Multipoint Distribution Services operates in the 28 GHz band and cannot be
transmitted through obstructions (such as buildings and mountains). The maximum range of
LMDS is approximately five kilometers, and though it can transfer data up to 2 gigabits per
second (Gbps), LMDS behaves more reliably when transferring data in the Mbps range.
MMDS
Multipoint Microwave Distribution Services are also called Multi-channel Multipoint
Distribution Systems and wireless cable. MMDS uses both unlicensed and licensed channels.
This technology’s key feature is that it uses multiple channels simultaneously and, therefore, by
aggregation creates large pathways between the sender and receiver. MMDS can transfer data
on the unlicensed channels up to 27 Mbps and up to 1 Gbps on licensed channels. It also cannot
transmit though obstructions.
Wi-Fi
The variations of the Institute of Electrical and Electronics Engineers (IEEE) wireless
standards and their usage are rapidly increasing, as, for example, the 802.11b standard shows.
This is more commonly known as Wi-Fi (short for wireless fidelity). The Wi-Fi was originally
intended for local area networks (LAN), but is now also used in metropolitan area networks
(MAN). The reported range of 802.11b is approximately 100 meters, seemingly insufficient for
larger scale connectivity, but it is possible to connect points over greater distances with the same
technology. In fact, Wi-Fi networks are being implemented with distances of up to 15
kilometers (Carstensen and Morgan, 2002). Other IEEE 802 wireless standards are also
available, such as 802.16 (Wireless MAN), 802.11a, 802.11g, and others.
Wi-Fi technology is becoming popular in an application related to, but distinct from, that
considered in this research. According to USA Today (Kessler, 2003, p.1), “big companies are
starting to pour big bucks into Wi-Fi technology – marking a turning point for the young
technology.” Companies such as Starbucks and McDonald's are putting so-called “public access
points” in their stores – Starbucks in 1200 stores across the country, and McDonald's in 10 stores
on a trial basis in Manhattan. There are currently 5,900 public access points in the United States;
this figure is projected to grow to 103,800 in 2006. Cell phone maker Ericsson is building 5,000
wireless hot spots in the United Kingdom (UK), Toshiba plans to install 10,000 more across the
United States by the end of 2003, and Cometa Networks (financed by IBM and AT&T, among
others) plans to have wireless networks installed in 50 US cities by year's end. Intel is also
squarely behind the push; it plans to build the Wi-Fi capability right into its chips. Intel will
spend over $300 million touting this advance, its biggest marketing blitz ever. (Kessler, 2003)
The use of Wi-Fi in coffee shops and McDonald's does not solve the last-mile problem.
But the availability of Wi-Fi on chips and across the US and UK will only help promote usage of
the technology and drive down prices and improve the electronics as time goes on.
Free Space Optics (FSO)
Free space optics (FSO), also known as "open-air photonics," "optical wireless" or
"infrared broadband," transmits data using low-powered infrared lasers. FSO do not currently
require FCC licensing; however, certain power restrictions must be observed. FSO is a line-of-
site (LOS) technology, and to remain in the unlicensed territory, the laser must not be too highly
powered. Consequently, the maximum range of FSO is a few kilometers. Furthermore, fog can
Chapter 5: Addressing Universal Broadband Service… 66

corrupt the transmission as water particles act as prisms to the laser and dissipate the light beam.
The data transfer rate of FSO is between 155 and 622 Megabits per second (Mbps).
Network Configuration Options
Wireless technologies have been deployed in three basic network configurations, point to
point, point to multipoint, and mesh (also known as multipoint to multipoint) – the subject matter
of this paper.
Point to Point (PtP)
Point-to-point is the oldest of the three configurations and consists of transmission from
one location to a second point. This is in distinction from technologies that broadcast from a
single point to multiple homes or businesses, or that broadcast from multiple points to multiple
homes and businesses.
This methodology is typically implemented when there are two buildings that need to be
connected as part of a LAN. Instead of laying wire or fiber optics between the edifices, it is
often more straightforward to place two small directed antennas on the tops of the buildings.
The advantage of this methodology is its simplicity for connecting two locations, and the cost of
the wireless equipment is often substantially less than that of a traditional wired network. The
disadvantages are potential security issues and possible weather disruptions. A very common
application area is campus (school) environments. Some applications of PtP connections use the
Free Space Optics technology. Such connections are often found in metropolitan areas when two
offices for the same company are in separate buildings, and it is desirable to have them part of
the same network. Since the technology of FSO is a fairly narrow laser beam, this introduces
another complication, namely the sway of the buildings involved. Several solutions have been
introduced to address this particular problem. It is also worth noting that since FSO's laser beam
is very high frequency and is very directed, security becomes a relative non-issue. It is not
possible to “eavesdrop” on a FSO signal without blocking the beam and thereby notifying the
sender and receiver.
Point to Multipoint (PMP)
PMP networks have a single source connected to a backbone network that propagates
signal to multiple receivers. This type of network may be envisioned with a single tower
blanketing an area with a signal, and everything falling under that blanket part of the network.
PMP often must deal with either line-of-sight (LOS) issues, where obstructions cause loss of
signal, or near-line of-sight complications, where some obstructions (buildings, e.g.) cause
failure, but lesser obstructions (e.g., trees) do not.
PMP networks operate somewhat similarly to those of cellular telephones in the sense
that when one is within a “cell,” coverage is provided by one tower. However, there are two
major differences between cell telephone provision and wireless PMP broadband coverage. The
first deals with line of sight. Broadband wireless typically operates in higher frequencies and
thus requires its receivers to be line-of-sight with the transmitter, whereas mobile wireless
(cellular) does not need this imposition. Second, to maintain expected quality of service (QoS)
for broadband, there must not be any interference with other signals. Cellular telephone does not
have this stipulation either. This is primarily because audio communication occurs at such low
data rates; moreover, people are very forgiving of interference (including minor static, lost
syllables or words) that mobile networks can operate satisfactorily where broadband networks
could not. Therefore, planning for PMP networks is significantly different from that of cell
Chapter 5: Addressing Universal Broadband Service… 67

phone towers and networks. The reader interested in additional information about the planning
of PMP networks is directed to Scheibe, Carstensen, Rakes, and Rees (2003).
Mesh
Since mesh (or multipoint to multipoint) networks do not have a tower or towers that
must reach all customers, several drawbacks of PtP and PMP are immediately overcome. The
tower-based networks require permission, purchase/lease of land, bringing electric power out to a
(perhaps) remote site, etc. With these non-mesh systems, it is also important to have already
generated a critical mass of customers that will pay for the service, because of the high cost of
the tower.
Mesh networks, the most recent of the network configurations and also the focus of this
research, obviate many of these issues by using smaller, cheaper, multiple transmitters. For
example, with mesh networks, as long as a customer is within range and view of another
customer that is within range and view of yet another customer or piece of equipment that is
eventually connected to the backbone, then the signal may be received. The incremental addition
of customers to the network is often inexpensive, with the only additional cost being that of the
customer premise equipment. In fact, increasing the number of customers can decrease the cost
of the mesh network, as will be seen later.
Radiant Networks, a Cambridge, England, based company announced plans to implement
a mesh wireless network in the South Wales region, planning to offer services equivalent to
asymmetric digital subscriber line (ADSL) networks (Willis et al., 2001). In February 2002,
Nokia announced that more than 50 customer were using their Nokia RoofTop™ Wireless
Routing mesh network solution. Nokia claimed this was the first commercially successful
deployment of a wireless mesh system (Nokia, 2002). Vista Broadband Networks has adopted
Nokia’s technology and has deployed equipment providing residential customers up to 512 Kbps
speeds. They also provide businesses with up to 3 Mbps access speeds (Wery, Kar, and
Woodrum, 2002). A small municipality in western England has taken upon itself to provide
broadband wireless access to its citizens by purchasing MeshBoxes and strategically placing
them throughout the area. Citizens need only to have Wi-Fi cards for their PCs to be part of the
mesh (Batista, 2003). As to future plans, companies such as Mesh Networks “envision
blanketing cities and highways with a peer network to provide continuous 802.11 presence”
(Gillmor, 2002). VoiceStream, a company offering broadband services over public access Wi-Fi
networks, cites figures from Gartner Inc. in Stamford, Connecticut, and predicts that by 2006,
more than 19 million mobile and remote workers will regularly use public access Wi-Fi networks
(Brewin, 2002). Finally, the need for and interest in wireless connectivity is increasing in the
global market. Former Soviet President Mikhail Gorbachev appealed to American companies
and investors to bring information technology outsourcing to Russia to strengthen their wireless
infrastructure. (Hamblen, 2002)
Mesh Network Terminology
In order to define terminology common to mesh implementations, Figure 5.1 shows a
neighborhood for which a mesh network has been provided. The backhaul point (BH in Figure
5.1) is that location along the backbone network where the mesh network will connect. The
backhaul is generally “wired” (say with fiber optics) to trunk connection points, shown as T
1
and
T
2
in Figure 5.1; the transmission medium changes to wireless at these trunk connection points.
The wireless signal is not, in general, transmitted from the trunk connections directly to
customers, however. The signal is sent to mesh insertion points (MIPs, labeled M
1
, M
2
, … M
7
in
Chapter 5: Addressing Universal Broadband Service… 68

the figure) that are closer to customers. These mesh insertion points do transmit and receive
directly to/from customers. Generally, once signal is received at a MIP back from a customer, it
is only forwarded to a trunk connection point and then to the backhaul. That is, signals are not
transmitted from customers to MIPS, and then back out to customers.
As mentioned, one of the key features of mesh networks is that customers are used to
forward signals on to other customers, up to the capacity limits of the receiver and transmitters.
However, sometimes it is necessary to insert repeater nodes into the network when there are not
enough customers nearby to get messages to every customer. Repeater nodes are generally
identical to customer equipment – there is just not a customer at the repeater location. Thus a
repeater may be thought of as a non-revenue generating customer.
In a mesh network signals from a customer seeking the backhaul may be forwarded from
customer to customer (possibly including repeaters), until finally a mesh insertion point is
reached. The MIP then sends the signal to a trunk connection point, and finally to the backhaul.
Each link in the chain back to the backhaul is called a hop. Too many hops in a chain of
connectivity can cause inordinate delay, or latency. Latency is the time it takes for a packet to
travel from source to destination (Webopedia, 2003). To ensure quality of service, that is, the
probability of a network meeting a given traffic contract (Wikipedia, 2003), it is sometimes
necessary to restrict the number of hops in each transmission path. Although hop constraints are
not new to network design (Balakrishnan and Altinkemer, 1992), there has been little discussion
in the literature as to what the optimal number of links should be for a wireless mesh network.
Whitehead suggests that somewhere from six to ten hops still offers an acceptable level of
service (Whitehead, 2000).
Since slightly different equipment and terminology are in use with different mesh
providers, and since cost data are confidential, Table 5.1 only gives a rough profile of current
mesh equipment costs. These numbers are an amalgamation of numbers we have heard
informally or seen in various internal publications/white papers, etc.
WIRELESS MESH PLANNING METHODOLOGY
The methodology proposed in this research to plan wireless mesh networks is a four-step
procedure. The first three steps are used to prepare (i.e., preprocess) the neighborhood or region
being analyzed for solution.
Step 1 – Potential Equipment Placement and Line-of-Sight Determination
Preprocessing begins with the specification of the backhaul point, trunk connection
points, and all possible mesh insertion points for a specified geographic region. (We will call
these neighborhoods, whether residential or commercial.) These equipment locations are
primarily a matter of physical characteristics of a neighborhood, that is, where right-of-
way/permissions are available, terrain is not obstructive, etc. With the methodology developed
here, the locations of the single backhaul point and two trunk connection points are assumed. It
is unknown how many MIPs will be economic, but all potential locations are to be specified;
which locations are chosen is an economic decision to be determined by the model.
Next, all known customer locations are specified. The situation to this point is indicated
in Figure 5.2, where each circle (node) indicates the (geographic) location of the
equipment/customer indicated. Line-of-sight must be determined next; either a geographic
information system (GIS) or field visit is required. Figure 5.3 can then be generated, where a
dotted line between any two circles indicates no obstruction between the two nodes. (Lack of a
Chapter 5: Addressing Universal Broadband Service… 69

dotted line indicates no feasible transmission; e.g., customer C
3
does not have line of sight with
customer C
4
.) At this point it may occur that a customer desirous of service cannot obtain a
signal, either because transmission range limits are exceeded or because obstructions block all
possible paths to the customer. In such an eventuality, potential repeater nodes (e.g., R
1
in
Figure 5.3) are defined at non-customer locations that will ensure an unobstructed path to the
backhaul for each customer. Finally in Figure 5.3, arrows are shown on the links from each MIP
to its closest trunk connection point, and from each trunk connection point to the backhaul.
Recall that signals received at MIPs from customers do not go back out to other customers; they
move on to trunk connection points.
Step 2 – Feasible Path Enumeration
Once the information from step 1 is in hand, all possible paths from each customer to the
backhaul may be enumerated, given a limit on number of hops. To generate these paths
efficiently, it is helpful to define artificial directional nodes; discussion of these is beyond the
scope of this paper. Interested readers may consult Scheibe, et al. (2003). Four possible paths to
the backhaul for customer C
1
may be traced in Figure 5.4.
Step 3 – Network-Flow Model Formulation
Once the feasible paths have been generated, they may be cast into the form of a
mathematical programming problem, namely the capacitated, fixed-charge, network-flow model,
(see Appendix I). The fixed-charge network flow problem has been used in many applications
including telecommunications (Balakrishnan, Magnanti, Shulman, and Wong, 1991; Gavish,
1991), logistics and production planning (Minoux, 1989), and transportation (Magnanti and
Wong, 1984). We have written an Excel (preprocessing) program that does this.
Step 4 – Solution Generation
The results from our Excel preprocessing program leads to data entered in Excel cells that
can be run directly in an integer programming package within Excel. We used Frontline’s
Premium Solver as this package (Frontline, 2002), and use the capacitated, fixed-charge network
flow formulation described in Chapter 4.
EXPLORATORY STUDY
To examine the potential of wireless mesh technology to reach previously unreached (and
perhaps unreachable) customers in relatively low-density areas – the case of most economic
uncertainty, we randomly generated 270 such neighborhoods, and invoked the procedure above
for nine different cases on each. The point was to study suburban and more rural areas as
candidates for profitable wireless mesh operation. The method of neighborhood generation is
outlined below.
Random Problem Generation
We randomly placed customer homes in a square grid to achieve differently configured
neighborhoods with three different rural/suburban densities; two hundred seventy such
neighborhoods were produced, ninety at each density. Low-density neighborhoods (L) were
constructed on 1.6-acre plots; medium-density areas were spawned with 1-acre lots; and high-
density regions (H) were created on ¼-acre lots. Three levels of visibility were specified in order
to introduce the effect of obstructions (i.e., terrain, buildings, etc.) and thereby portray
Chapter 5: Addressing Universal Broadband Service… 70

realistically line-of-sight issues among customers. A level of 40% visibility (low) infers that
only two-fifths of the possible connections among homes within the broadcast range of the
transmitter are viable, whereas a (medium) visibility level suggests that three fifths of homes
have such line of sight. The high level of visibility was set at 80%. Connections were randomly
assigned to the visible or obstructed categories according to their likelihoods. Table 5.2 shows
each factor and its corresponding level.
Each of the 270 neighborhoods was then evaluated on two different factors at three
levels, resulting in nine different cases. Each factor reflected different aspects of quality of
service (QoS); in particular, bandwidth and number of hops were each evaluated at low, medium,
and high levels. Bandwidth was studied at 2, 5, and 10 Mbps rates. The levels examined of
number of hops were five, seven, and nine. This means that, for example, at the medium hop
level, no path from a customer proceeding to the backhaul is allowed to exceed seven hops; this
keeps latency in check, thereby helping ensure a level of quality.
The particular cost and other assumptions made in the study are listed in Table 5.1.
MANAGERIAL IMPLICATIONS
The results from the 2,430 computer runs are economically disappointing. That is to say,
the use of Wi-Fi technology at current costs is economically infeasible.
The details from the study are shown below for each factor. We then show data
indicating the decline in costs necessary to “prove in” this technology for rural/suburban areas at
the densities examined in the study. Such drops in costs could be achieved either through
technological advances or government incentives/subsidies. Development of these scenarios is
beyond the scope of the present paper.
Each section below discusses in a similar manner one of the factors explored. Plots are
generated showing the monthly revenue required per customer to achieve breakeven under the
conditions given.
The data below are hindered in several cases by an inability to generate feasible
neighborhoods or solutions. When this occurred it sometimes became difficult to make
comparisons across factors. This suggests that for the journal article to be generated from these
results, factor levels be changed to include only feasible alternatives. Appendix II discusses
several of these anomalies.
Chapter 5: Addressing Universal Broadband Service… 71

Effect of Customer Density
Effect of Customer Density
$240,000.00
$245,000.00
$250,000.00
$255,000.00
$260,000.00
$265,000.00
Low Medium High
Density
T
o
t
a
l

C
o
s
t

Figure 5.5. Effect of Customer Density

It can be seen from Figure 5.5 that as the density of an area increases the cost of bringing
a broadband wireless mesh network decreases. Although the drop visually appears significant,
none of the three cases is financially feasible. If a five-year payback period is specified,
breakeven revenue is $174/month at a low density and $165/month at a high density. Both are
beyond the ordinary reach of residential customers.
Effect of Obstructions
Visibility
$240,000.00
$245,000.00
$250,000.00
$255,000.00
$260,000.00
$265,000.00
$270,000.00
$275,000.00
Low Medium High
Density
T
o
t
a
l

C
o
s
t
Low Visibiltiy
Med Visibility
High Visibility

Figure 5.6. Visibility

(The low visibility / low density point was excluded from this graph because no feasible
solution was found at those factor levels.)
Not surprisingly, the lower the visibility, the steeper the descent from high cost across the
density levels. This is because with many buildings and/or terrain to obstruct signal paths (i.e.,
low visibility), the harder it is at even medium customer densities to get from the customer’s
home to the backhaul; this necessitates more hops and often more repeaters. Managers will be
interested to note the magnitude of this change: for the low visibility case, the cost is increased
by 25% in going from a high customer density to a medium one.

Chapter 5: Addressing Universal Broadband Service… 72

Effect of Quality of Service: Hops
Quality of Service (latency)
$243,000.00
$248,000.00
$253,000.00
$258,000.00
$263,000.00
$268,000.00
$273,000.00
Low Medium High
Density
T
o
t
a
l

C
o
s
t
Low Hops
Med Hops
High Hops

Figure 5.7. Quality of Service (latency)

The message of Figure 5.7 is that, at low densities, there is a large price to pay in
reducing latency. For example, when a five-hop maximum is enforced (thereby maintaining
fairly low transmission delay), the cost is increased by $18,000 (14%) over a more lax nine-hop
maximum number of hops. This is a significant cost to pay for quality; it is not clear that
managers will enforce such restrictions once they become aware of these cost penalties.
Effect of Quality of Service: Bandwidth
Bandwidth Offerings
$-
$50,000.00
$100,000.00
$150,000.00
$200,000.00
$250,000.00
$300,000.00
$350,000.00
$400,000.00
$450,000.00
$500,000.00
Low Medium High
Density
T
o
t
a
l

C
o
s
t
Low BW
Med BW
High BW

Figure 5.8. Bandwidth Offerings

Figure 5.8 illustrates the changes in cost over the three density levels for each level of
bandwidth. With a five-year payback, monthly revenue charges for breakeven are $287/month
for the 10 Mbps case; $152/month for the 5 Mbps scenario; and, $73/month for the 2 Mbps
setting. This indicates a hefty charge for the fastest rate of service – one that is prohibitive for
almost all residential customers. The 5-Mbps rates are perhaps affordable for business
customers, but not for residential, and the 2-Mbps is only slightly more than what many
residential customers are currently paying for about 1/3 the throughput. Finally, note that there
is very little effect as customer density changes for any bandwidth.

Chapter 5: Addressing Universal Broadband Service… 73

Effect of Cost: Bandwidth
Bandwidth Cost
$-
$50,000.00
$100,000.00
$150,000.00
$200,000.00
$250,000.00
$300,000.00
$350,000.00
$400,000.00
$450,000.00
Low Medium High
Bandwidth
T
o
t
a
l

C
o
s
t

Figure 5.9. Bandwidth Cost

It can be seen from Figure 5.9 that averaged over all other factors, as the bandwidth
increases, the cost of bringing a broadband wireless mesh network also increases. The cost
increase is almost a factor of five – the difference between a profitable service offering and an
unaffordable one.
Cost Reductions Necessary to Achieve Profitability
To illustrate the need for cost reduction, consider a case of medium-level factors. That is
to say, each of the four factors in this study takes on a medium level: visibility is 60%, density is
one-acre lot sizes, QoS (latency) is restricted to seven hops, and bandwidth is 5 Mbps. The total
cost for providing mesh service to the average 25-customer neighborhood is $205,000. If the
desired time to recoup the investment were specified to be five years, then each customer would
be required to pay a monthly fee of approximately $137 for his or her service. While under
current conditions this rate may be very reasonable, we believe that very few end users would be
willing to pay that much. If we assume customers are willing to pay $75/month for such service,
then the total cost of equipment would need to be reduced to $112,500 to break even – a 45%
cost reduction. One way of viewing this result is to say that, on average and under
average/medium conditions, Wi-Fi costs need to drop by 45% to achieve breakeven.
CONCLUSIONS
A planning methodology has been developed that configures wireless mesh networks in
an optimal manner. This methodology was employed in a large computer study to assess
wireless technology as a mechanism for solving the last-mile problem of broadband provision in
rural/suburban areas.
The study finds that Wi-Fi service is not economically viable at current costs. The
analysis also concludes that, under average conditions, costs would have to drop by 45% (or
revenues be supplanted by $62/month per customer for five years) to become profitable. There
are those in government and industry who have called for the federal government to intervene
through subsidy and/or tax breaks to make broadband service “universally” available. Now that
a methodology is available for analysis of wireless mesh implementation, we view the next area
of future work to be an inquiry into various governmental options. Costs and penetration could
be assumed for the promising scenarios to see whether such expenditures are socially possible or
desirable.
Chapter 5: Addressing Universal Broadband Service… 74
REFERENCES
Acampora, A. (2002). Last Mile by Laser. Scientific American, 287(1), 49-53.
Balakrishnan, A., Altinkemer, K. (1992). Using a hop constrained model to generate alternative
communications netowrk design. ORSA Journal on Computing, 4,192-205.
Balakrishnan A., Magnanti T.L., Shulman A. and Wong R.T. (1991). Models for Planning
Capacity Expansion in Local Access Telecommunication Networks. Annals of Operations
Research, 33, 239-284.
Batista, E. (2003). Mesh Less Cost of Wireless. Wired News, Retrieved March 17, 2003, from
the World Wide Web: http://www.wired.com/news/print/0,1294,57617,00.html.
Brewin, B. (2002). New services spur growth of public access Wi-Fi. Computerworld, March
21, 2002.
Carstensen, L. W., and Morgan, G.E. (2002). Private communication.
Dreazen, Y.J. (2002). Tech Firms Bemoan Bush Talk --- ‘Broadband’ Policy is Viewed as
Lacking Significant Details. The Wall Street Journal, June 21, 2002, Section A4.
Frontline. (2002). Frontline’s Premium Solver™, Frontline Systems, Inc., Incline Village, NV.
From the World Wide Web: http://www.solver.com.
Gavish B. (1991). Topological Design of Telecommunications Networks – Local Access
Design Methods. Annals of Operations Research. 33, 17-71.
Gillmor, S. (2002). Riding the tiger. InfoWorld, May 27, 2002.
Hamblen, M. (2002). “Russia needs U.S. oursourcing dollars, Gorbachev says,”
Computerworld, March 21, 2002.
Kessler, M. (2003). “Wireless Net Technology Taking Off,” USA Today, March 12, 2003,
Section A, page 1, dateline: San Francisco, CA.
Kornbluh, K. (2001). “The Broadband Economy,” The New York Times, December 10, 2001,
Section A, page 21, column 2, editorial desk, dateline: Washington, D.C.
Magnanti, T.L. and Wong, R.T. (1984). Network Design and Transportation Planning: Models
and Algorithms. Transportation Science. 18(1), 1-55.
Minoux, M. (1989). Network Synthesis and Optimum Network Design Problems: Models,
Solution Methods and Applications. Networks. 19, 313-360.
Chapter 5: Addressing Universal Broadband Service… 75

Nokia, (2002). Nokia Demonstrates Wireless Mesh Broadband Market Leadership with More
Than 50 Network Customers. Press release. Retrieved July 30, 2002, from the World Wide
Web: http://press.nokia.com/PR/200202/848470_5.html.
Scheibe, K.P., Carstensen, L.W., Rakes, T.R., and Rees, L.P. (2003). A Mathematical-
Programming And Geographic-Information-System Framework For Wireless Broadband
Deployment In Rural Areas. Under review.
Scheibe, K.P., Ragsdale. C.T., Rakes, T.R., and Rees, L.P. (2003). A Capacitated, Fixed-
Charge, Network Flow Model for Solving Hop Constrained, Wireless Mesh Networks.
Chapter 4 of dissertation.
TechNet. (2002). A National Imperative: Universal Availability of Broadband by 2010.
Technical report. Retrieved July 30, 2002, from the World Wide Web: http://www.technet.org.
Webopedia. (2003). Retrieved March 10, 2003, from the World Wide Web:
http://www.webopedia.com/TERM/l/latency.html.
Wery, R., Kar S., and Woodrum J. (2002). “When Wi-Fi Meets Mesh – The Combination
Promises ‘Organic’-Like Network Growth and Low-Cost Equipment,” Broadband Wireless
Online, 3(6), Retrieved March 17, 2003, from the World Wide Web:
http://www.shorecliffcommunications.com/magazine/volume.asp?vol=29&story=290.
Whitehead, Philip (2000). Mesh Networks: A new Architecture for Broadband Wireless Access
Systems. RAWCON.
Wikipedia. (2003). Retrieved February 10, 2003, from the World Wide Web:
http://www.wikipedia.org/wiki/Quality_of_Service.
Willebrand, H.A., & Ghuman, B.S. (2001). Fiber Optics without Fiber. IEEE Spectrum, (8), 41-
45.
Willis, B., T. Hasletad, T. Friisø, O.B. Holm (2001). Exploiting Peer-to-Peer Communications
– Mesh Fixed and ODMA Mobile Radio. Journal of the IBTE. 2(2), 48-53.
Woolley, S. (2002). Bottleneck Breakers. Forbes, November 11, 2002.

Chapter 5: Addressing Universal Broadband Service… 76

APPENDIX I: THE WIRELESS MESH PROBLEM MODEL
The fixed-charge, network-flow problem (FCNFP) model as applied to the wireless mesh
problem is formulated as follows. Let N be the set of all nodes and N
a
be the set of artificial
nodes in the network. Further designate T, M, R, and C as the sets of trunk connection nodes,
mesh insertion points, repeater nodes, and customer nodes respectively. Finally, define the set of
all arcs in the network by the letter A.
For each node k, let the cost of activating that node be f
k
, which is assumed to be
nonnegative; any artificial nodes have a cost of 0. The objective is to minimize total cost. Let d
j

denote the supply or demand at node j; all artificial nodes should be considered transshipment
nodes (with a supply of 0). Let u
i
> 0 represent the capacity on the arcs emanating from node i.
x
ij
is the flow on arc (i,j), and y
i
is a binary variable equal to one when node i is in use and zero
otherwise.
The model may be written as



∈N k
k k
y f , min (A-1)

∑ ∑
∈ ∋ ∈ ∋
∈ ∀ = −
A k j k
j jk
A j k k
kj
N j d x x
) , ( ) , (
, (A-2)
(A-3) ) ( ,
) , (
M T i y u x
i i
A j i j
ij
∪ ∈ ∀ ≤

∈ ∋
(A-4)
i i
j i j
ij
y u x ≤

∋ ) *, (
]} * [ ] *) , [(
)] ( ) ( {[
a
a
N i A i i
N i R C i i
∈ ∩ ∈ ∩
∉ ∩ ∪ ∈ ∋ ∀
, ) , ( , 0 A j i x
ij
∈ ∀ ≥ (A-5)
N i y
i
∈ ∀ ∈ }, 1 , 0 { (A-6)
Constraint (A-2) ensures flow conservation, whereas constraints (A-3) and (A-4) are flow
capacity constraints. Constraint (A-3) restricts flow from trunk nodes and mesh insertion points,
whereas (A-4) restricts flow emanating from customer or repeater nodes, as follows. Flow
leaving either a customer or a repeater node (i.e., the sum of flow over all arcs leaving such a
node) cannot exceed the capacity of the transmitter at that node; moreover, if the transmitter is
inoperative, no flow may traverse any arc departing that node. Further details may be found in
Scheibe, Ragsdale, Rakes, and Rees (2003).
Chapter 5: Addressing Universal Broadband Service… 77

APPENDIX II: PROBLEMATIC CORNERS
In the study described in this chapter, one constant is the range of the wireless
transmission. The technology used is IEEE 802.11b or Wi-Fi. The range of Wi-Fi antennae can
very from 300 feet to 23,000 feet. For this design, the range is conservatively held constant at
900 feet. As initial tests were performed with these factors, certain “corners” of the design
proved to be problematic for analysis – specifically, low visibility / low density and high
visibility / high density / high hops.
When the density of the area in which a wireless mesh network may be desired is low, it
makes sense that some nodes in the network may be too far from others to be a part of the
network. When this is the case, then it may be necessary to add a repeater node to bridge the gap
and bring the customer node into the mesh. If an area’s density is very low, then it just may not
be feasible to create a mesh. Similarly, when the visibility is low, while nodes may be close
enough for the signal to reach, there is something blocking the visibility. If the visibility is too
low, then, again it may not be feasible for a mesh network. When the two factors are brought
together, low visibility and low density, then it becomes virtually impossible to create a mesh
network. During the pre-testing of these factors, the computer would randomly generate
neighborhoods based upon factor parameters. When the parameters were low visibility and low
density the computer could not generate a neighborhood that met the necessary criteria of each
customer node being able to see at least one other node. The visibility was set at 20% and the
density was set at 5-acre lots. The computer created over 100,000 neighborhoods, and not one
neighborhood met the necessary criteria. Therefore, it became necessary to increase the density
and visibility to a level where tests could be run. It is worth noting, however, that some
neighborhood configurations are very rare. For example, for the computer to generate twenty-
three neighborhoods with medium visibility and low density, it took approximately 214,000
generations. When the hop level was set at 5, it took another 225,000 generations just to create
the twenty-three. The implications are that neighborhoods that have medium visibility, low
density and require a high QoS are very rare and prove difficult to serve.
Neighborhoods with high density and high visibility created an entirely different, but
equally interesting condition. This is especially true when the number of hops is set at the
maximum level. It is very easy to generate a neighborhood where all the necessary criterion are
met, however, it was discovered, that the redundancy of paths from a node to the backhaul made
computation exceedingly difficult. To illustrate this issue consider a neighborhood with a
visibility of 50%, a density of one-acre lots, and a hop level of seven. The number of viable
paths for every node to the backhaul may range from 500 to more than 12,000. In its present
state, the algorithm used to generate every viable path may take a half of a minute to several
minutes. Now consider when the visibility is 80%, the density is ¼ acres, and the hops are nine.
The total number of viable paths may be in the 100 thousands to millions, and the time to
compute them exponentially increases. This is exclusively due to the high level of redundancy
within the mesh paths.
There are a few practical conclusions that may be drawn from the two corners just
discussed. The first is when an area has a density and/or visibility that are very low it may be
practically impossible to bring in a mesh network. There are many such neighborhoods in the
United States, and wireless service providers should be wary before attempting to penetrate such
areas. If the visibility is sufficient, but the area is not dense, then it may be possible to change
Chapter 5: Addressing Universal Broadband Service… 78

the wireless technology to one that is capable of transmitting over greater distances. Of course,
the costs change with different technologies.
Another practical conclusion is when the density is high, the visibility is high, and the
hops are high. While the mesh network will certainly work in neighborhoods with such
characteristics, there is a level of “overkill.” It may be worth considering using a wireless
technology that has a shorter signal distance, and, thereby, potentially reducing network costs.
Another means of addressing the overkill issue is to reduce the number of hops from a node to
the backhaul. This may or may not be desirable as the number of viable paths decreases so does
the network’s ability to simultaneously transmit multiple packets over different routes.

Chapter 5: Addressing Universal Broadband Service… 79



Figure 5.1. Providing broadband service to a neighborhood with a wireless mesh network.
Chapter 5: Addressing Universal Broadband Service… 80

T
1
T
2
M
1
M
2
R
1
C
2
C
1
C
3
C
4
BH





Key:
BH – backhaul
T
i
– trunk i
M
j
– mesh insertion point j
R
k
– repeater k (not shown)
C
l
– customer l
Figure 5.2. Nodes represent potential wireless mesh equipment locations.





T1 T2
M1 M2
R1
C2
C1
C3
C4
BH
T
1
T
2
M
1
M
2
R
1
C
2
C
1
C
3
C
4
BH

Figure 5.3. Nodes shown linked together with line-of-sight capabilities.

Chapter 5: Addressing Universal Broadband Service… 81

T1 T2
M1 M2
R1
C2
C1
C3
C4
BH
T
1
T
2
M
1
M
2
R
1
C
2
C
1
C
3
C
4
BH

Figure 5.4. Viable paths for customer C
1
to communicate with the backhaul, given the line-
of-sight configuration of Figure 5.3.
Chapter 5: Addressing Universal Broadband Service… 82


Table 5.1. “Typical” Wireless Assumptions

Equipment Cost Assumptions Capacity Assumptions
Trunk Connection
Nodes
$50,000 200 Megabits/second
Mesh Insertion Points $10,000 80Megabits/second
Repeater Nodes $5,000 30 Megabits/second
Customer Premise
Equipment
$5,000 30 Megabits/second
Antenna Range 900 Feet



Table 5.2. Factor Levels

Visibility Hops Density Bandwidth
Low 40% 5 1.6 Acre Lots 2 Mbps
Medium 60% 7 1 Acre Lots 5 Mbps
High 80% 9 .25 Acre Lots 10 Mbps

Note: We performed 30 replications in each cell.

Chapter 5: Addressing Universal Broadband Service… 83

CHAPTER 6


A GENERAL FRAMEWORK FOR PLANNING BROADBAND,
FIXED WIRELESS TELECOMMUNICATION NETWORKS





























Under the recommendation of my dissertation chair and committee, this dissertation
chapter merely outlines and sketches future work to be done in developing a planning
methodology for fixed, wireless broadband networks. As such, this future journal
submission is more developed in some sections than others; it is research in progress and a
document in progress.


Chapter 6: A General Framework… 84

A GENERAL FRAMEWORK FOR PLANNING BROADBAND,
FIXED WIRELESS TELECOMMUNICATION NETWORKS





ABSTRACT
Wireless broadband telecommunication networks have been touted as being the solution to the
“last-mile” problem. With this problem, 97% of the United States’ broadband network is laying
fallow because it is too expensive to go the remaining distance from the broadband backbone to
American residences and businesses. In two previous papers, the authors demonstrate how to
apply different wireless technologies in suburban and rural settings in a minimum-cost manner.
This paper presents a planning methodology that shows how and when to apply which
technologies, given parameters such as customer density, terrain, and infrastructure. The
methodology varies in scope from metropolitan areas to developing countries. A spatial decision
support system is proposed as a delivery mechanism for the planning methodology.

Keywords: Broadband Wireless Telecommunications, Geographic Information Systems,
Decision Support Systems, Network Planning


Chapter 6: A General Framework… 85

A GENERAL FRAMEWORK FOR PLANNING BROADBAND,
FIXED WIRELESS TELECOMMUNICATION NETWORKS

INTRODUCTION
Broadband communication refers to high speed, always “on” connections, (TechNet,
2002) and fixed, wireless telecommunications refers to those network delivery mechanisms that
depend on electromagnetic propagation to non-mobile customers, as opposed to copper, coaxial,
and fiber-optic connectivity. In previous work, the authors demonstrated two methodologies for
planning broadband, fixed wireless telecommunication networks (Scheibe, et al., 2003; Scheibe,
et al., 2003). These two methods of point to multipoint and mesh were presented as means of
bringing broadband communication to homes and small office / home offices (SOHO). Each
method shared a particular physical requirement due to the nature of the transmission and the
physical medium for broadband wireless telecommunications, namely a line-of-sight (LOS)
stipulation. LOS means that the path between transmitting and receiving antennas cannot be
obstructed; the antennas must be visible to each other for a signal to pass successfully. The
physical reason for the LOS requirement is the high frequency level of the signal; in fact, it is
generally correct to say that as frequency increases, so does the necessity of LOS. Lasers
provide a good example of this phenomenon; even fog and mist can disrupt a laser signal,
whereas they have very little effect on FM radio, which is lower in the frequency spectrum than
laser. Note that, by definition, a second shared characteristic of the two fixed wireless
methodologies is that each antenna is fixed in a specific location; that is to say, antennas do not
move as do mobile telephones.
Physically speaking, therefore, it is incumbent on a fixed wireless planning methodology
to consider directly the terrain over which the technology is to be applied. Geographic
information systems (GIS) lend themselves well to this class of problems. In each of the
methodologies mentioned above, we use a GIS to aid in the determination of optimal tower
placement in one case, and equipment visibility in the other. A particular GIS tool developed
locally (GETWEBS) takes topography into consideration as well as radio frequency
characteristics (Carstensen, Morgan, and Bostian, 2002) and is recommended in this planning
methodology. The power of this GIS is that it allows the decision maker to evaluate alternatives
in a straightforward, quick manner
A second necessary component for a fixed wireless broadband planning methodology is a
mechanism to deal with the fairly sophisticated mathematics (mixed integer programming
solution) necessary for minimum-cost solution of the problem. In this research we propose a
spatial decision support system (SDSS) to encapsulate the methodology. SDSS contain three
major components: a spatial database (which can be manipulated by GETWEBS in our case); a
model base, i.e., collection of requisite mathematical models (the mixed integer programming
models in our system); and a GUI dialog system. SDSS seem to fit naturally with the planning
methodology we are developing as the purpose of this research.
In this paper we advance an SDSS for planning fixed wireless broadband networks. The
next section provides background knowledge, particularly with respect to assumptions made in
our two previous papers dealing with broadband service via two different network topologies:
point-to-multipoint networks (PMP), and mesh networks. The subsequent section presents the
Chapter 6: A General Framework… 86

actual methodology. Two (large-scale) examples are presented next, and the implications then
discussed. Finally, conclusions are drawn and future work discussed.
BACKGROUND
SDSS
(This section will provide a very brief overview of spatial decision support systems, their
design, and limitations.)
Network Topologies
(Current major topologies – such as PMP and mesh – will be outlined. Emphasis will be
placed on limitations and conditions of their applicability, as is understood from our previous
work.)
Technology Issues
(This section will discuss issues such as the major technologies, the equipment they
demand, and limitations such as transmission ranges and capacities.)
METHODOLOGY
Database/GIS
Factors and items stored in the spatial (and non-spatial, as well) database will include:
For the geographic area under consideration:
• Developed/undeveloped
• Infrastructure
• Terrain visibility
• Customer density
• Customer clustering
• Disaster recovery an issue?
For each viable technology:
• Transmission range (for various technologies)
• Transmission capacities

The above issues will be incorporated into the methodology as root nodes in decision
trees, independent variables in charts, etc.
Model Base
This portion of the SDSS will contain three major model sections: a marketing
component, whereby customer demand is developed; a decision model section, in which
minimum-cost solutions are computed for a given technology/topology combination; and helper
models.
Chapter 6: A General Framework… 87

Marketing models
• Propensity to pay
• Demand models
Decision models
These models provide the mathematical wherewithal to solve a particular
topology/technology implementation at minimum cost, once those issues have been determined
by the system.
• Set covering problem. (The PMP solution requires this – see Scheibe, Carstensen,
Rakes, and Rees, 2003.)
• Capacitated, fixed-charge, network-flow problem. (The mesh solution requires this –
see Scheibe, Ragsdale, Rakes, and Rees, 2003.)
Helper models
This class of models, in general, provides support to the other SDSS components. For
example, some of these models would manipulate data and graphs from one format to another.
Also providing support are
• Forecasting models
• Tax data/models, etc.
GUI Base
The GUI base will contain presentation formats for the different aspects of the
methodology. Formats most appropriate to the task at hand are
• Decision trees
• Rules
• Charts
• Tables
Possible Synthesis/Flow of Above Factors in the Planning Methodology
To adequately address holistic planning of wireless broadband networks, it is crucial to
view the big picture. The foremost question should be, what is the desired service. Broadband
communications encompass a variety of services, including telephone, Internet, and cable.
Therefore, it is important to know not only the level of demand for a service, but also which
service is demanded. For example, individuals in a developing area/country that has no network
infrastructure whatsoever may demand basic services such as telephone before requesting high
speed Internet. However, when planning a network, future demand as well as current demand
should be considered. While voice may be the only current requirement, it would be
inappropriate to fail to plan for cable or Internet connectivity as well.
The next factor which should be considered in this planning phase is the characteristics of
the area for the desired service. There are several sub-factors, including topography, weather
characteristics, location and condition of nearest existing infrastructure, location of available
power, population and/or household density of the area, socio-political conditions of the area, is
it a new network or augmentation of an existing network, and is the area a new development or a
disaster recovery?
Chapter 6: A General Framework… 88

After the characteristics of the area under investigation have been determined, the next
set of facts to be examined is the wireless frequencies and their inherent capabilities and
weaknesses. This should be done in conjunction with the potential wireless topology such as
point-to-point, point-to-multipoint, or mesh. The characteristics of the area will play a large role
in determining which methodology should be used. Moreover, they will also determine which
frequency may best suit the area. For example, if the area were often very foggy, then higher
frequencies such as those used with free space optics would not be appropriate, or if the area had
much rain, then microwave frequencies would not be appropriate, etc.
Given the multitude of factors that affect the planning of a wireless network and how the
planning process, in general, lacks structure, a decision support system would certainly be
advantageous, and perhaps a necessary tool for a planner. Once the facts have been collected,
the next step is to recommend a technology/topology pair, and then to analyze them within the
SDSS. This research suggests using three decision aids to narrow the possible choices for a
broadband wireless network – charts/graphs, rules, and decision trees.
Possible Products/Outcomes/Outputs of the Methodology
Charts/Graphs
As an example of this potential SDSS output, consider which topology to utilize in
planning wireless networks. Based upon customer density and clustering, regions of topology
applicability can probably be derived. These regions may be appropriately displayed in charts or
graphs plotted against the key factors, say customer density on one axis and degree/type of
clustering on another. A graph could be generated for any cost profile assumed for the given
technology.
Rules
Another approach to reducing the number of planning choices to only the necessary few
is to employ rules in a top-down fashion. Three examples of rules are shown below.
Rule 1: if the area is undeveloped, and
there is no existing infrastructure close by,
then high bandwidth PtP is necessary from the backhaul

Rule 2: if the area is developed, and
a disaster has occurred
then high bandwidth PtP should be used to bridge network gaps

Rule 3: if the weather for area is foggy
then use lower frequency communications

This scenario suggests that it may be propitious to embed intelligence in the SDSS in the
form of an expert system (i.e., develop a knowledge base that integrates with the rest of the
system).
Decision Trees
A third possible beneficial SDSS output is a decision tree. Decision trees are visually
informative; as one prunes off branches one can see, for instance, how the search space has been
Chapter 6: A General Framework… 89

reduced. Decision trees also help enforce organization and complete consideration of all factors
to be considered in planning.
EXAMPLES
(Two detailed examples will be developed and analyzed. The current plan is for these
examples to possess widely different characteristics.)
MANAGERIAL IMPLICATIONS
(Implications will be drawn from the examples as to the utility and limitations of the
fixed wireless broadband planning methodology.)
CONCLUSIONS AND FUTURE WORK
While the research in this paper is yet untested, it is believed that a way of thinking
methodically about broadband such as has been suggested here would greatly benefit decision
makers at higher levels.

Chapter 6: A General Framework… 90

REFERENCES
Acampora, A. (2002). Last Mile by Laser. Scientific American, 287(1), 49-53.

Acampora, A. and Krishnamurthy, S. (1999). A broadband wireless access network based on
mesh-connected free-spaced optical links. IEEE Personal Communications, 6(5), 62-65

The American Heritage® Dictionary of the English Language, Fourth Edition. (2000). Houghton
Mifflin Company.

Alter, S.L. (1980). Decision Support Systems: Current Practice and Continuing Challenges.
Reading, MA: Addison-Wesley.

Bennet, J.L. (Ed.) (1983). Overview. Building Decision Support Systems. Reading, MA:
Addison-Wesley, 1-14.

Balakrishnan, A., Altinkemer, K., (1992). Using a hop constrained model to generate alternative
communications network design. ORSA Journal on Computing, 4, 192-205.
Balakrishnan A., Magnanti T.L., Shulman A. and Wong R.T. (1991). Models for Planning
Capacity Expansion in Local Access Telecommunication Networks. Annals of Operations
Research. 33, 239-284.
Batista, E. (2003). “Mesh Less Cost of Wireless,” Wired News. Retrieved March 17, 2003, from
the World Wide Web: http://www.wired.com/news/print/0,1294,57617,00.html.
Brewin, B. (2002). “New services spur growth of public access Wi-Fi,” Computerworld, March
21, 2002.
Carstensen, L. W., Bostian, C.W., and Morgan, G.E. (2001). Combining Electromagnetic
Propagation: Geographic Information Systems, and Financial Modeling in a Software
Package for Broadband Wireless Wide Area Network Design. Proceedings of the
International Conference on Electromagnetics in Advanced Applications, 799-810.

Comer, D.E. (2001). Computer Networks and Internets with Internet Applications. Third ed.
Upper Saddle River, NJ: Prentice Hall.

Crossland, M.D., Wynne, B.E., & Perkins, W.C. (1995). Spatial Decision Support Systems: An
Overview of Technology and a Test of Efficacy. Decision Support Systems, (3), 219-235.

Cruz F.R.B., Smith J.M., and Mateus G.R. (1998). Solving to optimality the uncapacitated
fixed-charge network flow problem. Computers and Operations Research. 25(1), 67-81.
Dennis A., Wixom, B. H., & Tegarden, D. (2002). Systems Analysis & Design: An Object-
Oriented Approach with UML. New York, NY: John Wiley & Sons, Inc.

References 91

Dreazen, Y.J. (2002). “Tech Firms Bemoan Bush Talk --- ‘Broadband’ Policy is Viewed as
Lacking Significant Details,” The Wall Street Journal, June 21, 2002, Section A4.
ESRI (1999). Getting to Know Arcview GIS: The Geographic Information System (GIS) for
Everyone. Third ed. New York: Environmental Systems Research Institute, Inc.
Fikes, R.E., Hart, P.E., and Nilsson, N.J. (1972). STRIPS: learning and executing generalized
robot plans. Artificial Intelligence, 3, 1-3, 251-288.
Fowler, T. (2001). Mesh Networks for Broadband Access. IEE Review, 47 (1), 17-22.
Frontline. (2002). Frontline’s Premium Solver™, Frontline Systems, Inc., Incline Village, NV.
From the World Wide Web: http://www.solver.com.
Gavish B. (1991). Topological Design of Telecommunications Networks – Local Access Design
Methods. Annals of Operations Research. 33, 17-71.
Gillmor, S., (2002). Riding the tiger. InfoWorld, May 27, 2002.
Girish, M.K., Zhou B., and Hu, J. (2000). Formulation of the Traffic Engineering Problems in
MPLS Based IP Networks. IEEE. 214-219.
Gouveia, L. and Requejo C. (2001). A new Langrangean relaxation approach for the hop-
constrained minimum spanning tree problem. European Journal of Operational Research.
132, 539-552.
Hamblen, M. (2002). “Russia needs U.S. oursourcing dollars, Gorbachev says,” Computerworld,
March 21, 2002.
Hamacher, H.W., & Kufer, K.-H. (2002). Inverse Radiation Therapy Planning -- a Multiple
Objective Optimization Approach. Discrete Applied Mathematics, (1-2), 145-161.

Horng, J.-H., & Tienyi Li, J. (2002). Vehicle Path Planning by Using Adaptive Constrained
Distance Transformation. Pattern Recognition, (6), 1327-1337.

Keen, P.G.W., & Scott Morton, M.S. (1978). Decision Support Systems: An Organizational
Perspective. Reading, MA: Addison-Wesley.

Keenan, P.B. (1998). Spatial Decision Support Systems for Vehicle Routing. Decision Support
Systems, (1), 65-71.

Kessler, M. (2003). “Wireless Net Technology Taking Off,” USA Today, March 12, 2003,
Section A, page 1, dateline: San Francisco, CA.
Kim, D. and Pardalos P. (1999). A solution approach to the fixed charge network flow problem
using a dynamic slope scaling procedure. Operations Research Letters. 24, 195-203.
References 92

Kornbluh, K. (2001). “The Broadband Economy,” The New York Times, December 10, 2001,
Section A, page 21, column 2, editorial desk, dateline: Washington, D.C.

Lewis, H.F., & Slotnick, S.A. (2002). Multi-Period Job Selection: Planning Work Loads to
Maximize Profit. Computers & Operations Research, (8), 1081-1098.

Macklin, Ben (2001). The Broadband Revolution: You Say You Want a Definition, eMarketer,
http://www.emarketer.com/analysis/broadband/20010306_bband.html.

Magnanti, T.L. and Wong, R.T. (1984). Network Design and Transportation Planning: Models
and Algorithms. Transportation Science. 18(1), 1-55.
Maniezzo, V., Mendes, I., & Paruccini, M. (1998). Decision Support for Siting Problems.
Decision Support Systems, (3), 273-284.

Minoux, M. (1989). Network Synthesis and Optimum Network Design Problems: Models,
Solution Methods and Applications. Networks. 19, 313-360.
Nehme, C.C., & Simões (1999) M. Spatial Decision Support System for Land Assessment. In
ACM GIS. Kansas City, MO: ACM Press.

Nokia, (2002). Nokia Demonstrates Wireless Mesh Broadband Market Leadership with More
Than 50 Network Customers. Press release. Retrieved July 30, 2002, from the World Wide
Web: http://press.nokia.com/PR/200202/848470_5.html.

Pirkul, H. and Soni, S. (2002). New formulations and solution procedures for the hop
constrained network design problem. European Journal of Operational Research, In Press.
PSINet. (1998). PSINet Begins Operating North American Transcontinental Fiber-Optic
Network. PSINet Press Release. http://www.psi.net/news/pr/98/aug1198.html

Quest. (1998). Qwest Awarded $20 Million Contract to Provide High Speed Internet Access to
Centurion Telecommunications. Quest Press Release.
http://www.qwest.com/about/media/pressroom/1,1720,121_archive,00.html

Radiant Networks Plc. (2001). http://www.radiantnetworks.com.

Ragsdale, C.T., Spreadsheet Modeling and Decision Analysis: A Practical Introduction to
Management Science, Third Edition, Southwestern College Publishing, Cincinnati, 2001.
Rolston, D. (1988). Principles of Artificial Intelligence and Expert Systems Development. New
York, NY: McGraw-Hill Book Company.

Scheibe, K.P., Carstensen, L.W., Rakes, T.R., and Rees, L.P. (2003). A Mathematical-
Programming And Geographic-Information-System Framework For Wireless Broadband
Deployment In Rural Areas. Under review.
References 93

Scheibe, K.P., Ragsdale. C.T., Rakes, T.R., and Rees, L.P. (2003). A Capacitated, Fixed-
Charge, Network Flow Model for Solving Hop Constrained, Wireless Mesh Networks.
Chapter 4 of dissertation.
Shrick, B., & Riezenman, M.J. (2002). Wireless broadband in a box. IEEE Spectrum, 39(6), 38 –
43.

Simon, H. (1960). The new science of management decision. New York: Harper & Row.

Soni, S. (2001). Hop Constrained Network Design Problem with Partial Survivability. Annals of
Operations Research. 106, 181-198.
Sprague, J.R.H., & Carlson, E.D. (1982). Building Effective Decision Support Systems.
Englewood Cliffs, NJ: Prentice-Hall, Inc.

Taha, H.A. (1975). Integer Programming - Theory, Applications, and Computations. New York:
Academic Press.

Tarantilis, C.D., & Kiranoudis, C.T. (2002). Using a Spatial Decision Support System for
Solving the Vehicle Routing Problem. Information & Management, (5), 359-375.

TechNet (2002). A National Imperative: Universal Availability of Broadband by 2010.
Technical report. Retrieved July 30, 2002, from the World Wide Web:
http://www.technet.org.
Turban, E., & Aronson, J.E. (2001). Decision Support Systems and Intelligent Systems. New
Jersey: Prentice-Hall. 867.

USGS (2002). USGS Digital Elevation Model Data. United States Geological Survey.
http://edcwww.cr.usgs.gov/glis/hyper/guide/usgs_dem

Vlachopoulou, M., Silleos, G., & Manthou, V. (2001). Geographic Information Systems in
Warehouse Site Selection Decisions. International Journal of Production Economics, (1-3),
205-212.

Virginia Tech Center for Wirless Telecommunications. (2002). http://www.cwt.vt.edu/.

Wagner, Jim. (2002). Teaming Up to Beat Cable. internetnews.com.
http://www.internetnews.com/isp-news/article/0,,8_1011151,00.html.

Webopedia (2003). Retrieved March 10, 2003, from the World Wide Web:
http://www.webopedia.com/TERM/l/latency.html.
Wen, U.-P., Wu, T.-L., & Shyur, C.-C. (2002). Bi-Directional Self-Healing Ring Network
Planning. Computers & Operations Research, (12), 1719-1737.

References 94

Wery, R., Kar S., and Woodrum J. (2002). “When Wi-Fi Meets Mesh – The Combination
Promises ‘Organic’-Like Network Growth and Low-Cost Equipment,” Broadband Wireless
Online, 3(6), Retrieved March 17, 2003, from the World Wide Web:
http://www.shorecliffcommunications.com/magazine/volume.asp?vol=29&story=290.
West, J., Lawrence A., & Hess, T.J. (2002). Metadata as a Knowledge Management Tool:
Supporting Intelligent Agent and End User Access to Spatial Data. Decision Support
Systems, (3), 247-264.

Whitehead, Philip (2000). Mesh Networks: A new Architecture for Broadband Wireless Access
Systems. RAWCON.
Wikipedia (2003). Retrieved February 10, 2003, from the World Wide Web:
http://www.wikipedia.org/wiki/Quality_of_Service.
Willebrand, H.A., & Ghuman, B.S. (2001). Fiber Optics without Fiber. IEEE Spectrum, (8), 41-
45.

Willis, B., T. Hasletad, T. Friis∅, O.B. Holm (2001). Exploiting Peer-to-Peer Communications –
Mesh Fixed and ODMA Mobile Radio. Journal of the IBTE. 2(2), 48-53.

Woolley, S. (2002). Bottleneck Breakers. Forbes, November 11, 2002.

References 95

VITA
KEVIN P. SCHEIBE


EDUCATION:

• Ph.D. Candidate in Business Information Technology, at Pamplin College of Business,
Virginia Polytechnic Institute and State University, Blacksburg, VA, May 2003

Dissertation Title: A Spatial Decision Support System for Planning of Broadband, Fixed
Wireless Telecommunication Networks

• M.B.A., California State University, San Marcos, CA, 1998
• B.S. in Computer Science, Biola University, La Mirada, CA, 1991

REFEREED CONFERENCE PROCEEDINGS:

1. Rakes, T., Rees, L.P., Scheibe, K.P. “A Coverage Model for Topology Design in
Wireless Mesh Networks,” Proceedings of the Decision Sciences Institute, San Diego,
CA, 2002.
2. Dong, X., Liu, Y., LoPinto, F.A., Scheibe, K.P., Sheetz, S.D. “Information Model for
Power Equipment Diagnosis and Maintenance,” Proceedings of Power Engineering
Society Winter Meeting, New York, 2002, pp.701-706.
3. Scheibe, K.P. “A Comparative Analysis of Heuristic Techniques for Estimating the
Parameters for a Nonlinear Time Series Model,” Proceedings of the Thirty-Seventh
Annual Meeting of Southeastern INFORMS, Myrtle Beach, SC, October 4-5, 2001, pp.
1056-1063.
4. Scheibe, K.P. “Non-Standard Methods for Simulation Optimization Including
Nonparametric Regression, Metamodeling and Knowledge-Based Approaches,”
Proceedings of the Thirty-First Annual Meeting of Southeast Decision Sciences Institute,
Charlotte, NC, February 21-23, 2001, pp. 334-336.

CONFERENCE PRESENTATIONS:

1. “From Decision Models to Decision Support: Excel with VBA,” Thirty-First Annual
Meeting of Southeast Decision Sciences Institute, Charlotte, NC, February 21-23, 2001.

PAPERS UNDER REVIEW:

1. Scheibe, K.P., Carstensen, L.W., Rakes, T.R., and Rees, L.P. (2003). A Mathematical-
Programming And Geographic-Information-System Framework For Wireless Broadband
Deployment In Rural Areas. Under review at Decision Sciences.

Vita 96

Vita 97
WORKING PAPERS:

1. Scheibe, K.P., Ragsdale, C.T., Rakes, T.R., Rees, L.P., “A Capacitated, Fixed-Charge,
Network Flow Model for Hop Constrained, Wireless Mesh Networks.”
2. Scheibe, K.P., Ragsdale, C.T., Rakes, T.R., Rees, L.P., “Addressing Implications for
Universal Broadband Service with Wireless Mesh Networks,”

WORK EXPERIENCE:

• 1999 – Present, Graduate Assistant, Department of Business Information Technology,
Virginia Tech, Blacksburg, VA
• 1996 – 1999, Director of Operations, Britton Capital Corporation, San Diego, CA
• 1994 – 1996, System Administrator / Software Engineer, Basic Research Corporation,
San Diego, CA
• 1992 – 1994, Software Engineer, Fisk Communications Inc., San Diego, CA
• 1991 – 1992, System Administrator, Britton Capital Corporation, San Diego, CA
• 1990 – 1991, Student Representative, Apple Computer, Inc., La Mirada, CA
• 1989 – 1991, Computer Programmer, The Lockman Foundation, La Habra, CA

TEACHING EXPERIENCE:

• BIT 4444: Computer-based Decision Support Systems (Virginia Tech)
• BIT 5495: Decision Support Systems Design and Implementation (Virginia Tech)

SPECIAL AWARDS and HONORS:

• R. B. Pamplin Doctoral Fellowship
• 2001 Southeastern INFORMS Ph.D. Student Paper Competition award winner

AFFILIATIONS:

• Association for Computing Machinery (ACM)
• Association for Information Systems (AIS)
• Decision Sciences Institute (DSI)
• Institute of Electrical and Electronics Engineers (IEEE)
• The IEEE Computer Society
• Institute for Operations Research and the Management Sciences (INFORMS)

COMPUTER SKILLS:

• Languages: C/C++, Java, PASCAL, HTML, ASP, VB.NET, BASIC, ADA, COBOL,
Assembly, SQL
• Systems: Windows XP/2000/NT/98/95/3.1, Macintosh, UNIX, LINUX, DOS,
VAX/VMS

A Spatial Decision Support System for Planning Broadband, Fixed Wireless Telecommunication Networks
Kevin Paul Scheibe (ABSTRACT) Over the last two decades, wireless technology has become ubiquitous in the United States and other developed countries. Consumer devices such as AM/FM radios, cordless and cellular telephones, pagers, satellite televisions, garage door openers, and television channel changers are just some of the applications of wireless technology. More recently, wireless computer networking has seen increasing employment. A few reasons for this move toward wireless networking are improved electronics transmitters and receivers, reduced costs, simplified installation, and enhanced network expandability. The objective of the study is to generate understanding of the planning inherent in a broadband, fixed wireless telecommunication network and to implement that knowledge into an SDSS. Intermediate steps toward this goal include solutions to both fixed wireless point-tomultipoint (PMP) and fixed wireless mesh networks, which are developed and incorporated into the SDSS. This study explores the use of a Spatial Decision Support System (SDSS) for broadband fixed wireless connectivity to solve the wireless network planning problem. The spatial component of the DSS is a Geographic Information System (GIS), which displays visibility for specific tower locations. The SDSS proposed here incorporates cost, revenue, and performance capabilities of a wireless technology applied to a given area. It encompasses cost and range capabilities of wireless equipment, the customers’ propensity to pay, the market penetration of a given service offering, the topology of the area in which the wireless service is proffered, and signal obstructions due to local geography. This research is both quantitative and qualitative in nature. Quantitatively, the wireless network planning problem may be formulated as integer programming problems (IP). The lineof-sight restriction imposed by several extant wireless technologies necessitates the incorporation of a GIS and the development of an SDSS to facilitate the symbiosis of the mathematics and geography. The qualitative aspect of this research involves the consideration of planning guidelines for the general wireless planning problem. Methodologically, this requires a synthesis of the literature and insights gathered from using the SDSS above in a “what-if” mode.

Dedication I dedicate this dissertation to my beautiful wife and family, without whom I could not have accomplished any of this.

iii

Terry R. You have been instrumental in my academic development. and Dr. I am extremely grateful to have been able to work with you. and I thank God for you every day. Ragsdale for your friendship. Moore for your direction. I love you dearly. You have supported me in countless ways. Zobel for your friendship. encouragement. wisdom and many great books. I cannot even begin to express how thankful I am for you. and I truly appreciate your mentoring and tutelage. Dr. Jesus Christ. analytical mind. Christopher W. Thank you to Dr.Acknowledgements I would first like to thank my Lord and Savior. and immortalizing me in one of the examples of your book. I want to thank my wife and best friend. Rakes for your insight and ideas. Cliff T. through whom all things are possible – especially this research. Dr. iv . Professor Loren Paul Rees. Dr. Carstensen and George E. William L. and great racquetball games. Finally. Thank you to my dissertation chairman. The guidance and support of my committee members are also very much appreciated. Laurence J. Morgan for your helpful guidance in the areas of wireless communications and geographic information systems. Mary.

........ 10 SOLUTION METHODOLOGIES ............ Fixed Wireless Telecommunication Networks .............................................................. 6 Scope and Limitations................................................................................................................................. 13 v ................................................................................................................................................ 3 Spatial Decision Support Systems ................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................ 6 Plan of Presentation ..................................................................................................... 1 Broadband............................................................................................ 7 Chapter 2: LITERATURE REVIEW ........................................ 12 Geographic Information Systems .....................Table of Contents Chapter 1: Introduction................ 9 Fixed Wireless Networks....................................................... 11 Mathematical Programming.................................... 6 Contributions of the Research..... 8 BROADBAND TELECOMMUNICATION NETWORKS ..................................................... 9 PLANNING ............................................................................................................ 1 Planning .............................................................................................................. 3 Statement of the Problem........ 6 Unification of Chapters.................................................... 11 Decision Support Systems .............................................................................................................................................................................................................................................................................................................................................................................. 4 Objective of the Study ............... 5 Research Methodology .......................................................... 13 Spatial Decision Support Systems .................................................................... 11 Fixed-Charge Network Flow Models ................................................................................................

..16 (WirelessMAN) at $50 and $100 per month.......................................................................................................................................................................................................................................................................................................................................................................................................................... 18 BACKGROUND .............................................................................................. 29 CONCLUSIONS............ 28 Customer Premise Equipment Cost ..Chapter 3: A MATHEMATICAL-PROGRAMMING AND GEOGRAPHIC-INFORMATIONSYSTEM FRAMEWORK FOR WIRELESS BROADBAND DEPLOYMENT IN RURAL AREAS... 31 vi .................................................. 28 Revenue Model ................................................................................................................................... 19 Costs.......................................................................................... 28 Results............................................................................................................................................. 28 LMDS at $50 per month ................................................................................ 21 Notation....................................................................................................................................................................................................... 24 Pedagogical Example............................................................................................................................................................ 20 (NEAR) LINE-OF-SIGHT OPTIMIZATION MODELS ............................................................. 19 Different Wireless Systems................................................................................................................................................................................................................................................................................................................................................................................................................................................... 30 REFERENCES ........................................................................................................................................................................................... 29 DISCUSSION .................... 29 Wi-Fi at $50 per month................ 23 Profit Model ................................................................................................................................................................... 25 Maximizing Exposure Model ....................................... 16 ABSTRACT........................................................ 27 GETWEBS Data .................................................. 28 Tower Costs .................................................................................................. 22 View Sheds ........................................................................... 26 Additional Constraints ....................... 26 A Spatial Decision Support System................ 27 Wireless Parameters and Costs ................................................................................................................................................................................................................................................................................................................................................ 29 802.................. 19 Wireless Telecommunication............................................. 17 MOTIVATION ........................................ 23 Pedagogical Mini-Example........................................................................................................................................................... 20 Differences Among Wireless Systems...................................................... 27 Range ............................................................................................. 26 REACHING THE LAST MILE ................................................................ 20 Geographic Information Systems and Decision Support Systems ........................................

.......................................................................................................................................................... 51 REFERENCES ................................................................................................................................ 44 INTRODUCTION .... WIRELESS MESH NETWORKS...................... 48 Step 2 – Artificial Directional Node Determination ....................................... NETWORK-FLOW MODEL FOR SOLVING HOP CONSTRAINED....... 47 METHODOLOGY ...................................................................................................... 49 EXAMPLE........................... 49 Network Flow Model..............Chapter 4: A CAPACITATED............................................................................ 48 Step 1 –Potential Equipment Placement and Line-of-Sight Determination ........................................................... 43 ABSTRACT............................................... 46 Fixed-Charged Network Flow Models .................................................................................................................................................... 53 v ii ..... 46 Mesh Networks ................................. 48 Preprocessing ............................................................................................................................................................................... 51 CONCLUSIONS AND FUTURE WORK .... 50 IMPLICATIONS .......................................................................................................................................................... 45 BACKGROUND ............................................................................................................................................. FIXED-CHARGE..................................................................................................................................................................................

.................................................................................................................................................................................................................... 67 Point to Point (PtP) ...................................................... 69 Step 2 – Feasible Path Enumeration ............................. 66 MMDS ................................. 69 Step 1 – Potential Equipment Placement and Line-of-Sight Determination ................................................................................................................................................................................................................................................................................................................................................................. 68 WIRELESS MESH PLANNING METHODOLOGY ........ 62 ABSTRACT.......... 70 Step 3 – Network-Flow Model Formulation............................................. 66 Network Configuration Options .................................................................................................. 67 Mesh......................................................................... 74 REFERENCES ............ 70 Random Problem Generation.......................................................................................................................................................................................... 77 APPENDIX II: PROBLEMATIC CORNERS ..................................... 74 Cost Reductions Necessary to Achieve Profitability ......................................................................................................................................................................................................... 70 Step 4 – Solution Generation ............................................................................................................................................................................................................................................................................. 74 CONCLUSIONS....................... 64 BACKGROUND ............... 72 Effect of Obstructions ........................................... 70 MANAGERIAL IMPLICATIONS ............................................................................................................................................................................................................................................................................................................................................................................. 66 Free Space Optics (FSO) ................................................................................... 63 INTRODUCTION ............................................................................. 66 Wi-Fi .................................................................. 73 Effect of Cost: Bandwidth .............................................................. 65 Technology Options..................................................................................................................................................................................................................... 78 v iii ............................................................................................................................................................. 73 Effect of Quality of Service: Bandwidth .............................................................................................................................................................................. 65 LMDS ...................................................... 68 Mesh Network Terminology.......... 72 Effect of Quality of Service: Hops....................................................... 75 APPENDIX I: THE WIRELESS MESH PROBLEM MODEL............................................................ 70 EXPLORATORY STUDY....................... 67 Point to Multipoint (PMP) ........................................................................................ 71 Effect of Customer Density .Chapter 5: ADDRESSING UNIVERSAL-BROADBAND-SERVICE IMPLICATIONS WITH WIRELESS MESH NETWORKS ..................

.............................................................................. 90 MANAGERIAL IMPLICATIONS ...................................................................... 88 GUI Base................................................................. 90 REFERENCES................................................................................................................................................................ 88 Decision models..................... 87 Database/GIS ........................................................... 87 SDSS .. 86 BACKGROUND .................................. 87 For the geographic area under consideration:.................................................................................................................................................................. 88 Possible Products/Outcomes/Outputs of the Methodology ............................................................................................................................................................................................................................................................................................................................................................. 85 INTRODUCTION ....................................................................................................................................................................................................................................................... 90 CONCLUSIONS AND FUTURE WORK ...................................... 87 Network Topologies... 89 EXAMPLES ................................. FIXED WIRELESS TELECOMMUNICATION NETWORKS .......................................................................................................................... 96 ix ........................ 88 Helper models .................................................... 87 Model Base .................................................................................................................................... 87 METHODOLOGY ........................................................................................................................................................................ 89 Charts/Graphs ................................................................................................................................................................... 88 Possible Synthesis/Flow of Above Factors in the Planning Methodology ........................ 89 Decision Trees .................................................................... 87 For each viable technology: ........................................................................................................................................................................................................................ 84 ABSTRACT................ 89 Rules ............................................................................................................................................................................. 87 Marketing models ............................ 91 VITA.....................................................................................................................................................................Chapter 6: A GENERAL FRAMEWORK FOR PLANNING BROADBAND............ 87 Technology Issues..............................................................................................................................................................

.... 57 Re-linking nodes to their artificial directional nodes................ Figure 4.. Figure 5.......................................5....... given the line-of-sight configuration of Figure 5................................. Figure 3............1.... 81 Nodes shown linked together with line-of-sight capabilities.......... Figure 3............................ .................................9............ .................... ........................................................... Figure 5.................. .............. .....List of Figures Figure 2........2b............................. Figure 5. 74 Providing broadband service to a neighborhood with a wireless mesh network.......................... 73 Bandwidth Offerings.3................. 36 The wireless spatial DSS architecture.......... Figure 3................................. 57 Customer C1 viable paths expressed in terms of artificial directional nodes.................... Figure 3....6...................5......3.........6............................... 81 Viable paths for customer C1 to communicate with the backhaul. 82 x ..................... 34 The exposure matrix E and the view shed matrix V for the fictitious town depicted in Figure 3......... 40 Providing broadband service to a neighborhood with a wireless mesh network...... Figure 3.............. 55 Nodes represent potential wireless mesh equipment locations..............4............................. Figure 3....2.............................................. given the line-of-sight configuration of Figure 4................ 33 A fictitious town.... divided into an east and west region by a railroad track that runs north/south through the center of town........ .. 39 Profit.................................. Figure 5.................... 37 Cells may be individually excluded from (red) or included (blue) for tower placement as desired.... 56 Nodes shown linked together with line-of-sight capabilities............ Virginia Tech Center for Wireless Telecommunications BTA Regions shown shaded on a map of Virginia (United States) ...................................................................................... Figure 4.............................2a............3. 38 Tower placement and profit sheds (magenta) for Wi-Fi (802........................................3.............................. ............ Figure 5.................2c....................... 80 Nodes represent potential wireless mesh equipment locations..... . 72 Visibility .............................................. Figure 5.5.......................................... Figure 4........................................................................................................ 34 A transmission tower’s signal can reach diagonally to all households two cells away.. Figure 4........................................3.................4.. Figure 5...1.......... Figure 3..................... divided into six cells.......2..................... 73 Bandwidth Cost........................................................ Figure 3. Figure 4..... Figure 3.....6.......................4............................7... and towers in Montgomery County as a function of percent coverage with Wi-Fi broadband............................................. 34 An elevation map of the fictitious town........ Figure 3.............8........................8................2.......... Figure 5................ 56 Viable paths for customer C1 to communicate with the backhaul.......... ...... Figure 4...................................7.............................. 15 A GETWEBS screen shot indicating the view shed (in magenta) of a tower placed at the green dot near the center of the top of the screen and with range indicated by the green circle.... 58 Effect of Customer Density . 35 A fictitious mini-town....1..............................1. Figure 5...............................11b) service offered at $50/month... 72 Quality of Service (latency) . households.......

....................................... Table 4......................................................List of Tables Table 2.. 59 Viable paths for node C1 with artificial directional nodes ......................2....................... 59 Viable paths for customer C1 to communicate with the backhaul . Table 4......................................... 60 List of sources and sinks for direct inclusion into the network flow model ...4............. 83 Factor Levels.................... .1............... 61 Some Results from the Example’s Preprocessing Step ............1................ Fixed-Charge... 83 xi .................1................................................... Table 4........7.. Table 3...... Table 4................. Some wireless communications examples......... Table 5...........5... Table 4.......... Table 3................................... Table 4.......................... 61 Results from the Example’s Capacitated................ Table 4............. 59 Source and immediate destination nodes with artificial directional nodes included ......2.............................3............2........ Network-Flow Problem ....... given the price per month for wireless service....................................................................................... 42 Source and neighbor nodes . 14 Some wireless applications and their allocated frequencies.......................... 61 “Typical” Wireless Assumptions............................ Table 5................................................. Table 4......6............1..... ......................... 41 The percentage participation (in Montgomery County) as a function of average household annual income. 60 Re-linking nodes to their artificial directional nodes........8.....

according to Ben Macklin. One popular alternative strategy to “going the last mile” with copper is to provide wireless services.” She points out that although investors plowed $90 billion into a crosscontinental fiber-optic broadband network. Consumer devices such as AM/FM radios. reduced costs. another wireless technology using higher frequencies and providing greater data transfer rates. whereby the signal is transmitted from a tower through various alternative mechanisms to equipment at business and residential customer sites. She then encourages both Congress and the president to subsidize sparsely populated regions of the country. Fundamentally. wireless technology has become ubiquitous in the United States and other developed countries. For example. Broadband. The problem.11 standard for local area networks (LAN) and local multipoint distribution service (LMDS). wireless computer networking has seen increasing employment. today merely 3% of that backbone is in use. p. access speeds greater than 144 kbps. 2001). Fixed Wireless Telecommunication Networks Comer (3rd edition.000 was reduced to $60. is that entrepreneurs failed to foresee the enormous cost of upgrading the “last mile – copper telephone wires that connect individual homes and small businesses to the broadband backbone” (p. for example.602).CHAPTER 1: INTRODUCTION Over the last two decades. p. 603) defines “broadband technology” as the term to describe a networking technology that uses a large part of the electromagnetic spectrum to achieve higher throughput rates. however. Advances in the technologies of wireless systems provide new opportunities for service providers. satellite televisions. broadband means: a downstream connection of 256 kilobits per second (kbps) and higher to Jupiter Communications. pagers. 2002). Kornbluh (2001. including applications meeting the Institute of Electrical and Electronic Engineers (IEEE) 802. broadband has come to mean fast connectivity whether by a wired or wireless medium (Macklin. Nokia has recently announced a national initiative to bring broadband wireless connectivity to business and residential customers via their Nokia RoofTop solution (BusinessWire. and 200 kbps in at least one direction to the US Federal Communications Commission. cordless and cellular telephones.” the term used to describe a networking technology that uses a small part of the electromagnetic spectrum and sends only one signal at a time over the underlying medium (Comer. and television channel changers are just some of the applications of wireless technology. Chapter 1: Introduction 1 . she states. 21). garage door openers.000 with a wireless system. CB radios. In an op-ed piece in the New York Times. a broadband analyst at eMarketer. p. suggesting that an ambitious broadband strategy can help revitalize the economy. simplified installation. low-income users or both. Cable TV. uses broadband transmission. car alarm signalers. 21) describes the hundreds of thousands of miles of fiber-optic cable buried in the United States as the “digital equivalent of fallow farmland. More recently. What constitutes a “large part” or a “small part” of the electromagnetic spectrum means different things to different people. A few reasons for this move toward wireless networking are improved electronics transmitters and receivers. and enhanced network expandability. Willebrand and Ghuman (2001) cite an example where a fiber cost of $400. This is as contrasted with “baseband technology.

It operates in the 28 GHz band and is LOS. every receiver must be visible to the transmitter. Some current fixed wireless technologies that are one or more of the three connection methodologies are free space optics (FSO). PTP typically offers the greatest transmission range. FSO do not currently require FCC licensing. Two general categories of wireless connectivity exist: fixed wireless and portable wireless." transmits data using low-powered infrared lasers. cellular phones. fog can corrupt the transmission as water particles act as prisms to the laser and dissipate the light beam. Multipoint Microwave Distribution Services (MMDS). therefore. The third fixed wireless method is Mesh." "optical wireless" or "infrared broadband. The other nodes in the mesh network act as receivers and as re-transmitters. and is connected to the network backbone. Of the three. and personal digital assistants (PDA). This technology uses multiple channels simultaneously and. There are three different types of connection methodologies in fixed wireless communication. consequently a shorter distance capability from the transmitter to the receivers than PTP. whether the transmission needs to be line of sight.). Instead of sending a narrow beam to a single receiver. and whether the FCC requires licensing. This method has a single transmitting tower and multiple receivers.11 standards for wireless connectivity (See Table 1.” which acts as a transmitter. MMDS is also LOS. The second method is point-to-multipoint (PMP). The focus of this research is on fixed wireless connections. Multipoint Microwave Distribution Services are also called Multi-channel Multipoint Distribution Systems and wireless cable. Furthermore. Local Multipoint Distribution Services is a PMP service. The max range of LMDS is approximately five kilometers and can transfer data up to 2 Gigabits per second (Gbps). there are several different technologies. Fixed wireless means that both the sending and receiving components of the network are situated in fixed locations and do not move without reconfiguring the network. it is thus unnecessary for every node to be visible to the insertion point. also known as "open-air photonics. and to remain in the unlicensed territory. certain power restrictions must be observed. Free space optics (FSO). the transmitter has a broader beam reaching multiple receivers and.Under the umbrella of broadband wireless communications. The data transfer rate of FSO is between 155 and 622 Megabits per second (Mbps). so long as a path exists from a node to the backbone via other nodes. where there are fixed towers but the receivers are mobile. With a mesh. Wireless communications use frequencies ranging from radio level to optical. MMDS can transfer data on the unlicensed channels up to 27 Mbps and up to 1 Gbps on licensed channels. With higher frequency wireless methods. The first is point-to-point (PTP). capable of passing signal along to other nearby nodes. This is the oldest of the three methods and is most commonly used in campus environments. where it is desirable to connect one point to another without the expense of laying cable. however. Local Multipoint Distribution Services (LMDS). the laser must not be too highly powered. Chapter 1: Introduction 2 . Varying spectral frequencies will determine the distance of transmission. A mesh network will have a socalled “insertion point. LMDS behaves more reliably when transferring data in the Mbps range. where every receiver is also a transmitter. Portable wireless refers to devices such as pagers. by aggregation creates large pathways between the sender and receiver. the maximum range of FSO is a few kilometers. however. Consequently. and the IEEE 802. MMDS uses both unlicensed and licensed channels. FSO is a line-of-site (LOS) technology. The distinguishing factor separating the technologies is frequency.

Planning One common meaning of planning used in the literature is to understand the necessity of building a system (Dennis. IEEE adopted the 802. or Turban (2001) are excellent resources for more information on DSS. Scientific. AI planning systems assume an initial (current) state. and Tegarden. and a set of allowable actions.In the 1990s. Any of these pieces of information may be critical in determining the financial success of the new store. The planning system’s purpose is to specify the actions that will take the system from its initial state to its goal state.11b. Rather. Fikes. still the general sense of the term here will be to specify those activities that will meet a service provider’s goals of maximizing profit or minimizing cost within the context of geographic. seemingly insufficient for larger scale connectivity. this work uses a definition of planning closer to that used in artificial intelligence (AI). retrieve. These bands are unlicensed and are usually used for household appliances such as microwaves to Wi-Fi routers. store. if possible (Rolston. but it is possible to connect points over greater distances with the same technology. The reported range of 802. GIS may also be used to show demographics such as residents’ income. 2002).11 standard for wireless Ethernet. also known as Wi-Fi (for wireless fidelity). Another family of bands is the Unlicensed National Information Infrastructure (UNII) band. Unstructured decisions are defined as those for which no algorithm can be written. Plans will include the choice of wireless technology plus the time sequencing of sending and receiving antenna placement. Keen and Scott-Morton (1978).4GHz. and analyze spatial data. UNII provides 300MHz of bandwidth in the 5GHz range. uses the unlicensed portion of the spectrum at 2. has been a topic of ongoing research for the past 30 years. In fact. 1978). however. the 802. models. This Wi-Fi frequency band is part of the Industrial. 1983). Hart. Although planning as used here will not be utilized in a formal way to derive a sequence of actions. and Medical (ISM) bands. For example. a branch of information systems. Wi-Fi was originally intended for local area networks (LAN). Wi-Fi networks are being implemented with distances of up to 15 kilometers (Carstensen and Morgan.11b standard was approved. 2002). GIS’ capabilities of spatial visualization often simplify difficult problems. a goal state. Semi-structured decisions fall between the other two (Keen and Scott-Morton. Spatial Decision Support Systems Decision Support Systems (DSS). Bennett (1983). taken in Chapter 1: Introduction 3 . Geographic Information Systems (GIS) are a type of relational database management system used to collect. 802. but is now also used in metropolitan area networks (MAN). Wixom. 802.11a uses the UNII bands and provides 54 Mbps data transfer rate. They used data. this dissertation will not generate long-range plans (say 10 to 20 years). 1972). and other constraints. GIS are widely used to aid decision makers in solving spatial problems. The capabilities of the GIS. fast food chains considering new locations can use GIS to determine where their nearest competitors are located. whereas algorithms can be specified for structured decisions. 1988. GIS link tabular data to graphical data by relating graphical layers to database tables. and data transfer rates up to 11 Mbps. financial. Furthermore.11b is approximately 100 meters. the GIS can show the road network so the decision maker can determine possible traffic near their store. and user interface components to help decision makers solve semi-structured or unstructured problems (Bennett. In 1999. but rather more intermediate plans in the context of three to ten years. This is not the meaning used in this research. and Nilsson. In general.

and vehicle routing (Keenan. and Paruccini. 1998. transmission may not be economical. land planning (Nehme and Simoen. The situation will possibly become similar to the days when garage door openers first became popular. GIS software required large budgets. providers will want to determine the profit sheds yielding the greatest financial return. trees. in an area of potential wireless customers under a realistic scenario that is capitally constrained. Wynne. Carstensen. they must maintain a clear path between themselves and a signal propagator. Vlachopoulou. One fundamental issue with high frequency. Because computers have become significantly cheaper and more powerful. Conversely. 2002). That means that they are unable to transfer through walls. and the majority of customers are only willing to pay a minimal amount. but they are limited to short transmission distances. Furthermore. etc. Tarantilis and Kiranoudis. mountains. 1999). and Perkins (1995) determined that SDSS enabled decision makers to complete their tasks quicker. part of this research determines that physical region providing profitable coverage from a tower. Since such equipment was very expensive. Factors affecting the optimal profit shed include the Chapter 1: Introduction 4 . 2001). For customers to be included in a wireless network. PMP technologies such as LMDS is that while it may be possible to calculate the physical coverage of a tower. Silleos. Crossland. furnish a tremendous capability in solving this semi-structured spatial problem. Bostian. However. SDSS is a relatively new research area. GIS packages can now run on desktop computers. and with greater understanding of the problem through visualization. The LOS constraint requires careful planning of the wireless network. and it is necessary to overcome or compensate for the weakness of a specific technology in order to make it a viable method of reaching customers. and it was possible to drive around town and use one remote to open other garages. This may be true if a minority of customers has the greatest propensity to pay for services provided. which creates more SDSS research possibilities. In their research. wireless solutions are being aggressively pursued. Mendes. bandwidth will fill causing other problems such as interference. memory. Technologies that use relatively lower wireless frequencies are not as restricted by LOS problems as higher frequencies. thereby aiding decision makers with semi-structured or unstructured spatial problems. Statement of the Problem Since the “last mile” of wiring has become the prohibitive factor for connectivity. and hard disk space.concert with the knowledge of the decision maker. 1988. but are limited to LOS. providers must determine where towers should be placed to maximize profit or alternatively to minimize cost. The advantage of an SDSS is it is able to integrate the model portion of the DSS with the graphical representation of the GIS. fixed wireless technologies that use relatively higher frequency bands are able to propagate data over greater distances. and labels it a profit shed. primarily because GIS software has historically needed a great amount of computing power. each of the wireless technologies introduced earlier has inherent strengths and weaknesses. The solution may or may not contain locations that reach the most customers. and Manthou. For example. and Morgan (2001) defined the physical area that is visible from a tower as a view shed. Marrying DSS and GIS creates Spatial Decision Support Systems (SDSS). Providers must still determine whether that coverage is profitable. Since different tower placements will lead to varying levels of profitability. fixed wireless. SDSS have been applied to siting problems (Maniezzo. as more people use these wireless technologies. more efficiently.

the cost of receivers. The spatial component of the DSS is the locally developed (Carstensen et al.cost of wireless towers. Factors that must be considered in planning a mesh network are the number of allowable hops from a node to the backbone. the set covering formulation must be augmented to incorporate geographic constraints discovered by the GIS in order to determine profitable configurations and regions. the terrain of the region. LOS between nodes. a classical mathematical programming problem formulation. has no ability to calculate the profitability of a specific region. however. hops. but specifies only whether a tower signal can be received. density. The overall objective of the study is to generate understanding of the planning inherent in broadband. GETWEBS. and bandwidth – are addressed in the planning of such networks. The first SDSS proposed here incorporates cost. 2001) GIS tool Geographic-Engineering Tool for Wireless: Evaluation of Broadband Systems (GETWEBS). The second portion of this research is to develop a SDSS that incorporates a capacitated. One of the functions of the GETWEBS program is the display of view sheds for specific tower locations. Chapter 1: Introduction 5 . the capacitation of arcs between nodes. as do the PMP networks. or in such a location that they cannot maintain a clear line-of-sight to the tower. In particular. It encompasses cost and range capabilities of a particular tower. As steps along the way in obtaining this understanding. and the range of a tower signal. and the density of the neighborhood. One method explored here for solving this problem is to cast it as a set-covering problem. Mesh networks do not have profit sheds. and solution models are developed that determine optimal profit sheds for PMP and optimal cost for Mesh implementations. This dissertation addresses the problem of generating overarching guidelines for a wide range of possible broadband. the propensity to pay of the customers reached. By allowing each node in the network to function as a transmitter as well as a receiver. Mesh networks are able to work around the problem of a customer being too far away from a single tower. Four factors – visibility. Moreover. fixed wireless use. The formulation requires integer solutions and is NP-hard. and signal obstructions due to local geography. the placement of additional nodes to fill in gaps on the mesh. factors are delineated that differentiate the specification of present wireless technologies. A major difficulty of the wireless tower location problem is the number of possible locations a tower can be placed. the market penetration of a given service offering. The objective of this research is to minimize the cost of the network by using as few cost incurring nodes while maintaining a desired level of service. the topology of the area in which the wireless service is proffered. the cost of receivers. there are cost incurring nodes in the network. fixed-charge network flow model to achieve a minimum cost for a Mesh network architecture. an exhaustive search over possible sites is prohibitive. signal is passed from one node to another – theoretically allowing more customers than a PMP network. the propensity to pay of customers. Instead. revenue. and performance capabilities of a wireless technology applied to a given area. Objective of the Study The objective of this study is to explore the use of an SDSS for broadband fixed wireless connectivity to solve the wireless network-planning problem. fixed wireless telecommunication networks and to implement that knowledge into a spatial decision support system.

has been written as a series of four separate journal articles all under the thematic umbrella of spatial planning of broadband wireless networks. the wireless network-planning problem may be formulated as integer programming problems (IP). It will not discuss network security issues intrinsic to wireless data transfer or wireless network protocols. The problem we are dealing with we will refer to as the general wireless planning problem. This study also derives a model embedded in an SDSS to minimize the cost associated with equipment location for broadband. Consequently. Contributions of the Research • • • This research makes three primary contributions. fixed-charge network flow problem. Loren Paul Rees. in particular. Chapters 3 through 5 are formatted as journal articles and are meant to stand on their own. Chapter 6 is another journal article. The line-of-sight restriction imposed by several extant wireless technologies necessitates the incorporation of a GIS and that a spatial DSS be built to facilitate the symbiosis of the mathematics and geography. the so-called set covering problem and capacitated. Quantitatively. fixed PMP wireless networks maximizing profitability. Chapter 1: Introduction 6 . Moreover.solutions to both fixed wireless PMP and fixed wireless mesh connectivity are developed and incorporated into the SDSS. An IP formulation is developed for both the PMP and the mesh wireless network problems. Research Methodology This research is both quantitative and qualitative in nature. References from each chapter have been alphabetically compiled at the end of the dissertation. Scope and Limitations The general wireless planning problem is largely unexplored in the literature. this research will focus mainly on planning to ensure profitability and economic feasibility of wireless connectivity. To keep the exploration manageable. fixed mesh wireless networks. under the guidance and tutelage of Dr. The qualitative aspect of this research involves the generation of planning guidelines for the general wireless planning problem. In general. in preliminary form. A case study is undertaken to illustrate the general procedure and to aid in refinement of the process and use of the SDSS. A limited case study is undertaken to support development and understanding of the guidelines. Methodologically. this requires a synthesis of the literature and insights gathered from using the SDSS above in a “what-if” mode. and is defined as providing wireless broadband services to residential and small home office customers. solution strategies such as parallel computation to solve larger problems are not addressed here. Furthermore. Unification of Chapters This dissertation. This dissertation provides an investigation into “going the last mile” in reaching customers with broadband fixed wireless networks. The research develops mathematical models embedded in an SDSS to develop profit sheds which are solutions to the equipment location problem for broadband. each has its own title page. engineering issues in the design of antennas or the propagation of radio signals at uncharted frequencies are also beyond the realm of this study. and references. abstract.

This factor gives cause for using a decision support system. there are some necessary factors for planning the wireless networks that are common to all four papers. they are closely related in their address of reaching the “last mile” using a spatial decision support system and wireless telecommunication networks. Depending upon the wireless methodology. which includes the application of the GETWEBS program. wireless telecommunication networks. A crucial one is the utilization of a GIS that accounts for wireless technological characteristics. but instead of using the GIS. whereas Chapter 5 describes the entire process. Therefore. Chapter 6 would use the methods proposed in the earlier chapters. conclusions and references. area coverage. which summarizes the research and draws conclusions by outlining an overarching planning methodology for fixed. it has identified a need for the application of SDSS in wireless technology – specifically in the areas of profitability. planning. such as which methodology(s) would be most appropriate given the characteristics of the area of a desired wireless network. Moreover. they are all thematically related. and resolution of line-of-sight issues. Chapter 4 introduces a capacitated. The method in Chapter 3 uses such a GIS. and spatial decision support systems.The overall theme of the dissertation is planning broadband wireless telecommunication networks. minimize cost. it simulates the information the GIS would provide. fixed-charge network flow problem as a means of minimizing the cost of wireless mesh networks. Future work is outlined in chapter 6. Additionally. Another common factor is the semi-structured nature of wireless network planning. Two different objective functions are considered and a detailed example using a GIS in concert with a mathematical programming model is presented. literature review. methodology. wireless broadband telecommunications. Chapter 2 is a literature review in the areas of fixed. Wireless technologies may be delivered by three different network topologies. while each chapter was written as a separate article and contains its own introduction. or guarantee coverage. to help decision makers address the larger picture of network planning. These topologies are formulated (and solved) mathematically in Chapters 3 and 4. and spatial decision support systems (including geographic information systems). Where Chapter 3 proposes the set-covering formulation to solve PMP networks. Chapter 3 presents an SDSS with an embedded mathematical programming model based on the set-covering formulation in the management science literature. some models are more appropriate than others in determining maximum profitability or minimum cost. This review establishes the need for computer support in the placement of antennas and receivers to maximize profit. Chapter 5 provides a major study developing managerial implications of mesh networks in rural / suburban settings a mechanism for addressing the “lastmile” problem. Chapter 5 develops managerial conclusions from the latter topology. The first approach is omitted as it only allows for connecting two customers. and the latter two are discussed in this dissertation. Plan of Presentation This chapter has served as an introduction to fixed wireless communications. and while each chapter either describes a different methodology in application or general description. The methodology is “delivered” to the planner in a spatial decision support system. The model presented in Chapters 4 and 5 would use the GETWEBS program for effective “real world” problem solving. Chapter 1: Introduction 7 . planning.

It has created a global community of organizations and individuals. or wireless. Moreover. satellite. only two decades ago the Internet was just a research project involving a dozen locations. owning a PC and being “on-line” is now the norm. Decisions are costly. Wireless connectivity is now a rapidly growing field with many companies racing to establish a market niche for themselves. people interact. The impact of the Internet has been far greater than just allowing people to send email. as it is used here is concerned not with implementation details. whereby the receivers move around with the customer. schools. a commonality to many. more capable. where a small part of the spectrum is used and signals are sent one at a time. fixed wireless telecommunication networks. and governmental agencies operate. there is a myriad of possible wireless solutions to the “last mile” problem. Kornbluh (2001) goes so far as to argue for governmental subsidization of connecting lower income or sparsely populated areas to revitalize the economy. for example. if not all. consumers shop. 2002). connecting from homes and companies to the backbone is a much larger problem than anybody initially seemed to realize. consumers are less satisfied with small bandwidth. This case is as opposed to cellular telephones. More specifically. 1988. The Internet has created a paradigm shift in the way companies do business. 1998). it refers to Chapter 2: Literature Review 8 . cable modems. While each alternate solution may be drastically different from another. It has become a rare thing for a company not to have at least one network that is connected to the Internet. Companies such as PSINet and Quest have spent millions in laying a transcontinental fiber optic backbone (PSINet. a fundamental problem exists in that cable companies or telephone companies cannot reach many areas because their infrastructure cannot handle the level of throughput required for broadband speeds. Additionally. In the United States the Internet connects companies. is the need for transmitters and receivers. schools educate. rather. The term fixed wireless means that neither transmission towers nor receiving towers are mobile. over the last twenty years is staggering. According to Comer (2001). However. Quest. as computers have become faster. however. government. They want faster connections. Since the wireless spectrum ranges from radio to laser. and people. 2001). this is in distinction to baseband networks. She states that the underestimation of the cost of connecting end users to the backbone of the fiber-optic network has left the majority of the capacity of the network unused. Appropriate or inappropriate tower placement can mean the difference between financial success and disappointment. This dissertation deals with the intermediate planning of broadband. and to meet the demand cable and telephone companies are racing to be broadband service providers (Wagner. The term broadband refers to the utilization of a large part of the spectrum to achieve high throughput ranges (Comer. Individuals are also connecting either by low speed methods such as modems or high speed methods such as DSL. The need for “last mile” connectivity has never been greater. nor with the electrical engineering development of improved wireless technologies. but now it has grown to reach millions of people in every populated country of the world.CHAPTER 2 LITERATURE REVIEW The growth rate of networking. Planning. and specifically of the Internet. hardware such as transmitting towers is costly and great prudence is required in determining their placement. It is in dealing with these factors that planning is essential. and less expensive.

Hasletad.11b) and LMDS systems. point to multipoint (PMP). 2001). Table 2. BROADBAND TELECOMMUNICATION NETWORKS As introduced in Chapter 1. and whether the FCC requires licensing. Simply stated. Fixed Wireless Networks Also mentioned in Chapter 1 are the three types of connection methodologies in fixed wireless communication: point to point (PTP). are near line-of-sight (LOS) operation and are license free. The main advantages of LMDS over Wi-Fi are bandwidth and coverage distance. 2002). and FCC licensing. whereas LMDS frequencies are around 28-31 GHz. and Engineering.” Practical issues. and Holm. this research examines Wi-Fi systems with a greater range. therefore. LOS. Theoretically. 2000). CWT “has succeeded in its objective of bidding and acquiring licenses in the Federal Communications Commission's (FCC's) first LMDS auction by winning licenses to provide 1150 megahertz of wireless bandwidth for the Greater Roanoke. limit the matrix of possibilities. wireless communications encompasses many different technologies differentiated by frequency. wireless telecommunication networks.4 GHz range. Instead the focus here is upon the three to ten year horizon whereby the infrastructure to reach communities or businesses. of 20 to 30 years duration. Lower frequency means shorter range. that is. Martinsville and Kingsport-Johnson City market areas” (VTCWT. Friisø. Near LOS means that signal may pass through some obstructions such as foliage. These frequencies determine the distance of transmission. licensing. LMDS has a bandwidth that is more than twice that of AM/FM Radio.11b systems operate in the 2. They range from radio level to optical. but cannot pass through mountains and such. the capability to go “the last mile. however. This research examines those frequency-connectivity choices that are being considered now or are on the horizon (Nokia. The advantages of Wi-Fi over LMDS are that no FCC licensing is required and Wi-Fi is near-LOS. VTCWT. require FCC licensing. fixed connectivity. 802.1 shows some of the radio frequencies and their applications. Furthermore. Participating colleges in the CWT are Arts and Sciences. Business. Furthermore. and. The Center for Wireless Telecommunications (CWT) at Virginia Tech is an interdisciplinary research group designed to aid client companies develop new products and services using wireless technologies. and Mesh (Willis. this research deals with the intermediate-range infrastructure planning of broadband. and transmission properties reduce the realistic combinations to a few. higher frequency means greater range. VHF/UHF Television and cellular telephones combined (VTCWT. Thus. 2001. 2001). whether the transmission needs to be line of sight. and are strongly LOS. The primary focus of this research is on the Wi-Fi (802. LMDS is also able to transmit a signal over several miles versus Wi-Fi’s range of about 300 feet (Gibbs. 2002). Radiant. 2002. Danville. Chapter 2: Literature Review 9 . Although LMDS’s transmission range is so limited as to be economically prohibitive. less LOS. one could use any of the connection methodologies at any of the frequencies to bridge the “last mile. and that FCC licensing may not be required.” is determined. is not as concerned with objects blocking the tower signal. as a means of exploring possible future near-LOS systems.determining a proper wireless infrastructure and its deployment. the word intermediate in the phrase intermediate planning stipulates that the horizon is not necessarily long-range in nature. Factors such as the frequencies available.

and time is estimated for the software system’s completion (Dennis.” or alternatively. The College of Arts and Sciences is experienced with GIS. A famous example of planning is STRIPS. as “to have as a specific aim or purpose. Each component currently exists in a somewhat isolated form.Approximately 609. Not only are the application areas in which planning is applied varied. and there is no integrated computer system available to do planning. For example. workload planning (Lewis and Slotnick. or reactive. The College of Business has financial know-how. enactment. planning may be proactive such as when producing blueprints for the construction of a house. financial. For example. 2002). The College of Engineering has technical expertise with wireless transmitters and receivers. and other constraints. Planning in an information systems context refers to the first of four phases in the information systems development life cycle. a goal state. radiation therapy planning (Hamacher and Kufer. planning is a term with many different uses. 2002). However. it is crucial in understanding why an information system should be built. and a set of allowable actions. Recent articles may also be found on network planning (Wen. 2002). 1988).1 for a map of CWT’s BTA region. Chapter 2: Literature Review 10 .” Yet. the topology of the landscape. AI planning systems assume an initial (current) state. Wu.000 people live within the basic trading area (BTA). if possible. or attainment of. See Figure 2. whereby a group of wooden blocks on a table in a particular initial orientation is transformed into an alternate configuration by a computerized robot with certain well-defined actions (Rolston. still the planning will be executed to specify those activities that meet a service provider’s goals of maximizing profit or minimizing cost within the context of geographic. This work uses a definition of planning closer to that used in artificial intelligence (AI). in the literature. Each CWT college brings specific capabilities to the table. in Chapter 6 a general planning methodology will be outlined that maps existing infrastructure and goals into a specification of technology and equipment providing wireless capability. and has created Monte Carlo simulations for various economic scenarios that determine the cost and profit of a given wireless solution. there does not exist a single unifying system that incorporates the capabilities and limitations of the towers. as when preparing a retreat from a lost battle. Although planning as used here will not be utilized in a way to derive a sequence of automated actions. but the fundamental activities included may be quite different. intend. 2002). PLANNING The American Heritage® Dictionary of the English Language (2000) defines planning as formulating “a scheme or program for the accomplishment. the feasibility of the project is assessed. Wixom and Tegarden. During this phase. and the economic analysis. a search on “planning” using ScienceDirect for the years 1999 through 2002 on titles alone (not including abstracts and keywords) results in 1863 articles. tasks are identified. For example. The planning system’s purpose is to specify the actions that will take the system from its initial state to its goal state. 2002) and vehicle planning (Horng and Li. and Shyur.

4GHz (Nokia. C. see Taha (1975). one may solve for the location of transmitters and receivers directly as integer variables in the set-covering problem. that is problems in which an objective (such to maximize profit) is pursued. The fixed-charge network flow problem (FCNFP) is one of a large class of network design problems. and the MP solution approach generates answers to the spatially constrained optimization problem. however. 1999). Lee. Some MP can be shown to have optimal solutions that can be found algorithmically regardless of the size of the problem. many MP cannot be guaranteed to have optimal solutions that can be found quickly. 1963). and heuristic solutions must be substituted. Dantzig first used the approach to solve logistic activities during World War II (Dantzig. Conversely. Mathematical programming problems (MP) are classified by the nature of the objective and constraints. 1993). 2001). 2001). the set-covering problem. For example.SOLUTION METHODOLOGIES As this research encompasses topics that range over different optimal specification of wireless transmitters and receivers the solution methodologies will be disparate. In the simplest wireless case to be examined here. and laser (Acampora and Krishnamurthy. and Taylor. for the case of line-of-sight wireless systems. In general. and Holm. For a description of the set-covering problem. Rather. LP in which the constraints form a convex region can be shown to have optimal solutions that can be found by examining the so-called corner points of the region (Moore. which have been used in many applications including Chapter 2: Literature Review 11 . Friis∅. This research will show that the wireless PMP equipment location problem may be cast as a well-known MP formulation. the number of variables in the problem. whereby the GIS determines LOS satisfiability. Mathematical programming is used to solve constrained optimization problems. In fact. this research will embed both mathematical programming solution techniques and Geographic Information Systems into a spatial decision support system. For example. each methodology will incorporate the optimization abilities of mathematical programming. the visualization facilities of GIS. A general wireless equipment location spatial DSS will require heuristic solution approaches as well. many MP can be shown to require exponential solution time in n. Hasletad. program is used in the sense of a set of activities or a schedule or a managerial activity for which a mathematical or quantitative solution approach will be advanced. solutions cannot be guaranteed. in general. Lee. Taylor. 2002). those problems in which the objectives and constraints are linear are termed linear programming problems (LP) (Moore. Fixed-Charge Network Flow Models Mesh networks are made up of nodes and arcs where multiple nodes at different locations all transmit signals to perhaps a single node (Willis. In fact. A node is a point to which a signal is transmitted or received. Mathematical Programming The term mathematical programming is a misnomer in the sense that the user of the approach does not program the computer by writing code in Basic. and the integrating capabilities of spatial decision support systems. 26 GHz to 40 GHz (Fowler. or another language. Mesh networks may use a variety of frequencies such as 2. This research uses a network flow formulation to solve the mesh network problem. However. 1993). but is limited by constraints (such as cash on hand).

2002. The dialog module generally consisted of over half of the code and served the purpose of communication with the manager. and (2) the development of easily-utilized model-base competence. Gavish. Turban and Aronson (2001). Gouveia and Requejo. whereby managers could now do dialog generation in a natural manner. 1982). a model base. A fixed cost is incurred for using arcs between nodes. 1982. perform the requested operations in a representational form consistent with the preferences of the manager. Keen and Scott Morton (1978) noted that most managerial tasks of significance (Alter. The principal reason for constraining the number of hops in a network is to maintain a level of quality of service. once developed. 1984). and whole conferences organized around such systems. The answer advanced was to encourage the manager and computer into a symbiotic relationship. From a builder’s perspective. However. 2000). The model base acted as a repository of models that managers might want or need in working their way through a particular problem. independent modules: a database. logistics and production planning (Minoux. 1999. Soni. primarily at MIT. 1991). Today. and the system would respond. the manager could ask for any model and/or data needed for any aspect of the entire decision-making process (Simon. real-time direction at the prompting of the machine. but also to support managers at their jobs. and transportation (Magnanti and Wong. Supposedly. the objective is to seek the most efficient way to move flow (in our case bandwidth) on a network in order to satisfy demand between origin and destination nodes and to minimize the overall cost. In these problems. 1998). and much research has been devoted to creating better and more efficient solutions (Kim and Pardalos. because computer programs of that time required algorithms. 1963). 1991. therefore. Another consideration for some network design problems is the number of hops from a source (transmitting node) to a sink (receiving node) (Pirkul and Soni. DSS did not reach widespread acceptance until software development introduced two fundamental capabilities: (1) the graphical user interface. This was problematic. in general. FCNFP problems are known to be NP-hard. and macros. 1983). modules. in turn. 1980) were semi-structured or unstructured. This new DSS era followed on the heels of Electronic Data Processing Systems in the 1950s and Management Information Systems in the 1960s. Girish. Bennett. with hundreds of articles in the literature.telecommunications (Balakrishnan et at. The machine would. DSS are developed routinely. hence. DSS were organized into three primary. The rationale for this new approach was that computers could be utilized not just to automate tasks. 2001. the computer was being asked to solve problems that it could not solve. Decision Support Systems Decision Support Systems were first developed four decades ago. A variety of reasons was responsible and enhancements were made (Sprague and Carlson. Early DSS failed in the sense that they were not used. 1989). Zhou and Hu. it behooves the objective to use a few arcs in which costs are associated. When the network is capacitated then upper bounds exist for the amount of flow over an arc. whereby the human provided algorithmic. Cruz et al. The reader interested in further detail may consult. for example. meaning that no algorithm could be written to specify the path to task solution. Chapter 2: Literature Review 12 . through such packages as Excel with its built-in functions. 2000. and a dialog generation component (Sprague and Carlson.

Tarantilis and Kiranoudis. minimization of cost. Chevallier. GIS have been combined with LP models to solve cropland allocation and land-use modeling (Campbell. it takes more than just layering data to extract meaningful information. this research develops a tool to address important variations of the wireless equipment location problem (in Chapters 3 and 4). and vehicle routing (Keenan. 1997. primarily because GIS software has historically needed a great amount of computing power. 1999). This future research subject is introduced in Chapter 6. memory. and with greater understanding of the problem. store. 2002). and Paruccini. Radke. and to perform multicriteria analysis (Laaribi. However. 2001) This research will develop an SDSS to solve the line-of-sight tower location problem for a PMP network. Objective functions considered include maximization of profit. land planning (Nehme and Simoen. GIS link tabular data to graphical data by relating graphical layers to database tables. they suggest using software agents to format the information in a manner most helpful to the user. and the SDSS research stream is beginning to flow. including issues such as infrastructure development and recovery. Vlachopoulou. Crossland. Chuvieco. thereby aiding decision makers with semi-structured or unstructured spatial problems. & Wirtshaffer. 1999). for example with developing countries. 2001). 1992. Wynne and Perkins (1995) determined that SDSS. through visualization. In conclusion. and both urban and rural disaster recovery. van der Merwe. a demographic layer of a town map may have sections linked to economic information such as household income or school zoning information (ESRI. but also to possess expert domain knowledge.Geographic Information Systems Geographic Information Systems are a type of relational database management system (RDBMS) used to collect. By building an SDSS to combine these symbiotically. a review of the literature indicates a plethora of knowledge on mathematical programming/set covering and GIS. Chapter 2: Literature Review 13 . more efficiently. For example. GIS software required large budgets. Mendes. Since such hardware was expensive. & Martel. SDSS have been applied to siting problems (Maniezzo. and the maximization of area coverage. Native American lands. 1998. enabled decision makers to complete their tasks quicker. As computers have become cheaper and more powerful. The dissertation will also utilize the SDSS to solve mesh-connectivity wireless problems utilizing its mathematical programming and GIS components. The advantage of an SDSS is the seamless integration of the model portion of the DSS with the graphical representation of the GIS. once the new tool is assembled. For example. Silleos. GIS are excellent tools for presenting spatial data in a format that is easy to manipulate and understand. and Manthou. However. Gless. GIS often require the user not only to be facile in manipulating the GIS. 1988. In fact. The layering of different data makes GIS extremely powerful. Spatial Decision Support Systems Spatial Decision Support Systems (SDSS) combine DSS and GIS. and hard disk space. Lohrentz. retrieve. an additional host of problems becomes amenable to solution. 1993). and analyze spatial data. West and Hess (2001) point out that it often takes someone with expert knowledge in GIS to make adequate use of the system. SDSS is a relatively new research area. but a dearth of work utilizing the techniques in concert. GIS packages can now run on desktop computers.

190 kHz Beacons AM broadcast communications 550 .24.1600 kHz Low power voice and data CB radios 27 MHz Low power voice and data Remote control 49 MHz Cordless telephones Low power voice and data FM radios 88 . (http://www.11b U-NII 5.vt.170 MHz Pager services 303 MHz garage door openers 260 .4385 GHz Army packet radio development 802.727 .2.25 GHz Radio navigation Local multipoint distribution service 28 – 31 GHz (LMDS) Source: The Virginia Tech Center for Wireless Communications.5.16 – (Tsunami equipment) 24 .470 MHz Keyless entry systems Security alarms 824 .htm).108 MHz Low power voice and data Commercial two-way voice 150 .849 MHz and 869 . Frequency Band Applications VLF band radios 170 .894 MHz Cellular phones ISM band Wireless LAN's Part 15 devices (spread spectrum cordless phones.) 902 . etc.Table 2.11a 802.cwt. Some wireless communications examples.1.928 MHz Military radiolocation systems Federal mobile communications Pager services with high transmitter 930 MHz power Amateur satellite Part 15 devices Microwave ovens and systems 2.11a 5.15-5.4 .edu/wireless_faq/default.35GHz Amateur satellite Part 15 devices Naval radar systems 5.25-5. Chapter 2: Literature Review 14 .25GHz 802.875 GHz Test range instrumentation radars 802.

1. Virginia Tech Center for Wireless Telecommunications BTA Regions shown shaded on a map of Virginia (United States) Chapter 2: Literature Review 15 .Figure 2.

CHAPTER 3

A MATHEMATICAL-PROGRAMMING AND GEOGRAPHIC-INFORMATION-SYSTEM FRAMEWORK FOR WIRELESS BROADBAND DEPLOYMENT IN RURAL AREAS

submitted to Decision Sciences July 31, 2002

Chapter 3: A Mathematical Programming…

16

A MATHEMATICAL-PROGRAMMING AND GEOGRAPHIC-INFORMATION-SYSTEM FRAMEWORK FOR WIRELESS BROADBAND DEPLOYMENT IN RURAL AREAS

ABSTRACT
Organizations and individuals are calling for the universal availability of broadband as a national imperative by 2010. Although the United States has the requisite cross-continental fiber-optic backbone infrastructure in place, only 2% to 5% of that nationwide network is used today. This is because of the prohibitive cost necessary to extend the backbone to many small businesses and homes, even for a relatively short distance. For example, nine out of ten American small businesses are less than one mile from the backbone, but still do not have service – the “last mile” problem. This research integrates a mathematical programming model and a specially developed geographic information system to examine the potential of “going the last mile” using current wireless telecommunication costs in a rural county in the Mid-Atlantic United States. The appropriateness of current wireless technologies in both for-profit and government subsidized scenarios is discovered. Implications for the national scene are discussed. Keywords: Wireless telecommunications, Broadband, Geographic information systems, Decision support systems, Set covering problem, Mathematical programming optimization

Chapter 3: A Mathematical Programming…

17

A MATHEMATICAL-PROGRAMMING AND GEOGRAPHIC-INFORMATION-SYSTEM FRAMEWORK FOR WIRELESS BROADBAND DEPLOYMENT IN RURAL AREAS

MOTIVATION
TechNet, a bipartisan group comprising more than 300 chief executive officers and senior partners of the major companies in the fields of information technology, biotechnology, venture capital, investment banking and law, claims that (1) “widespread adoption of true broadband will increase the efficiency and productivity of Americans at work and at home – with a potential $500 billion impact on the United States economy. The benefits to quality of life are immeasurable,” and that (2) true broadband is the key to the next generation of communications and Internet services (TechNet, 2002, p. 1). However, Kornbluh (2001, p.21) points out that this potential will remain unrealized without government action. She states in an op-ed piece in the New York Times, that although investors have plowed $90 billion into a cross-continental fiberoptic broadband network, today merely 3% of that backbone is in use – it is the “digital equivalent of fallow farmland.” The problem, she states, is that entrepreneurs failed to foresee the enormous cost of upgrading the “last mile – copper telephone wires that connect individual homes and small businesses to the broadband backbone” (Kornbluh, 2001, p. 21). In an article in Scientific American, Acampora (2002) elaborates by noting that for nine out of 10 American businesses with more than 100 workers, the backbone is less than a mile away. TechNet has called on the President and policymakers to make broadband a national priority and to set a goal of making an affordable 100-megabits per second (Mbps) broadband connection available to 100 million American homes and small businesses by 2010 (TechNet, 2002). They note that to achieve the 100 million homes goal will require network providers to invest hundreds of billions of dollars to upgrade infrastructures and increase bandwidth capacity in the last mile, primarily by providing new fiber connections to homes and offices. Today, virtually no homes have connections with such bandwidth (TechNet, 2002). An intermediate goal is the “availability of affordable broadband at speeds of at least 6 Mbps from 2 or more providers to at least 50% of U.S. households and small businesses by 2004” (TechNet, 2002, p. 6). TechNet developed 6 principles to address roadblocks and provide a guide to a national broadband policy. Principle 5 reads as follows (pages 2-3): Investment incentives, potentially including targeted tax incentives, should encourage broadband deployment to underserved communities and businesses. … In a market-oriented environment that encourages the deployment of broadband networks, there may still be a segment of the U.S. population that does not have broadband availability. Public policies should seek to narrow the current and future disparity in the level of high-speed access to the Internet, to ensure that all Americans can enjoy the benefits of broadband. Similarly, Kornbluh encourages both Congress and the president to subsidize sparsely populated regions of the country, or low-income users or both, suggesting that an ambitious broadband strategy can help revitalize the economy. (Kornbluh, 2001, p.21) Chapter 3: A Mathematical Programming… 18

car alarm signalers. Point-to-multipoint (PMP) networks are those where a single source (in our case an antenna or antennas on top of a tower) connected to a backbone network propagates signal to multiple customers (in our case. Table 3. the IEEE 802. approximately only 4. and television channel changers have applied wireless technology.000 was reduced to $60. many consumer devices such as AM/FM radio. BACKGROUND TechNet defines broadband as the capacity to deliver Internet access with a continuous ‘always on’ connection and the ability to both receive and transmit digital content or services at high speeds (TechNet. pagers. Willebrand and Ghuman (2001) cite an example whereby a fiber potential cost of $400. and local multipoint distribution service (LMDS).000 a mile). garage door openers. The procedure will be demonstrated for a rural county in the mid-Atlantic region of the United States. wireless radio systems may be classified along several dimensions. Many experts have defined 100 Mbps as the speed at which the web’s true potential can be achieved. As these costs are prohibitive. and several implications will be drawn for the rural “last mile” problem as a whole. For reasons that will be explained in the next section. Different Wireless Systems Operationally. receiving antennas Chapter 3: A Mathematical Programming… 19 . it is estimated that speeds of 10 Mbps will be required. cordless and cellular phones. In order to facilitate telecommuting. most agree with this one – except as to what precisely is meant by “fast. 2002). Because the wireless technologies require “line of sight” or “near line of sight” operation. including applications meeting the IEEE 802. (Nokia. this paper examines the wireless provision of broadband service to rural homes and small businesses using current costs and technologies. whereas optical fiber bridges needed to connect millions of users to the optical-fiber backbone would cost too much to install (between $100.The focus of this research is on the sparsely populated segment of this problem. satellite television. “Fixed” wireless networks are those where the transmitter and receiver locations are fixed. such as utilized with cellular telephones. 2002) Copper wires and coaxial cables connecting buildings do not possess the gigabit per second capacity necessary to carry advanced bandwidth-intensive services and applications. Although other definitions of the term abound. it will be necessary to develop a special-purpose geographic information system (GIS) and then integrate it with the mathematical programming formulation we derive. and in particular to wireless telecommunications. 2002). wireless computer networking has seen increasing employment.000 with a wireless system. More recently.000 and $500.1 indicates radio frequencies allocated for several different applications. These standards are described below.11 standard for local area networks (LAN). as opposed to mobile networks.” Today. service providers are looking to other transmission media. CB radios.4% of American households have speeds approaching 400 Kbps (TechNet. (TechNet. Solutions will be generated for both “for profit” and government subsidized scenarios. point-to-multipoint systems. 2002) Wireless Telecommunication For years.16 standard. Nokia has recently announced a national initiative to bring broadband wireless connectivity to business and residential customers via their Nokia RoofTop solution. The systems we explore in this chapter are fixed.

16. Hasletad. has been a topic of ongoing research for the past 30 years. whereas algorithms can be specified for structured decisions. 2001. Costs As this paper considers only fixed PMP systems. Willis. the transmission range from towers for cell phones is much greater than for broadband because cell phone frequencies are much lower. 2001). broadband wireless communications. To make this determination. and annual maintenance. to determine a broadband distribution network requires the ability to position towers in locations that guarantee reliable line-of-sight transmission. the receivers in the cell phone scenario are mobile. callbacks. such as mesh networks (Fowler. Geographic Information Systems and Decision Support Systems Decision Support Systems (DSS). thus allowing for a greater number of choices in equipment and. allowing users to switch from one tower to the next as they travel. Friis∅. power. Differences Among Wireless Systems It might be supposed that one could apply directly a solution methodology for cell phone networks to the wireless network determination problem.999% reliability). we turn to geographic information systems and decision support systems. They are all either line-of-sight. antennas. but is the most costly of the three. Second. Multipoint-to-multipoint networks (MPM). Schrick and Riezenman. First.16. consequently. Using data. or near line-of-sight. The three wireless technologies mentioned above (IEEE 802. but in some cases customer premise equipment is one-fourth that of the WirelessMAN. and Holm. as opposed to cell phone users accustomed to breakup. and LMDS) share similarities in characteristics and properties and are the leading contenders for fixed PMP service. connecting the tower to the backbone. Finally. Such is not the case with broadband. Tower costs may be similar to 802.11. Third. Consequently.16 technology (also known as WirelessMAN) is less expensive than LMDS. model and user interface components. At each CPE site there are receiving antenna costs and installation costs (so called “truck rolls”. DSS aid decision makers to solve semi-structured or unstructured problems (Bennett. 1983) and cover all phases of Simon’s (1960) decision-making process. LMDS offers very high-speed connectivity. 802. however. According to Keen and Scott-Morton (1978). but the customer premise equipment (CPE) costs are still high. whereby multiple sources at different locations all transmit signals to perhaps the same customers. 2002). the expectations in terms of service level are much higher for broadband (the so-called “five nines” – 99. depending on location and terrain. bringing power to the tower. Such is not the case. structure. Semi-structured decisions fall Chapter 3: A Mathematical Programming… 20 . will be discussed in Chapters 4 and 5. unstructured decisions are defined as those for which no algorithm can be written. utilizing the same frequencies at adjacent towers is not a problem. and frequent fading and/or loss of signal.located at customer premises). lower prices. a branch of information systems. The 802. because of three primary differences between the two wireless technologies.11b (Wi-Fi) service is gaining acceptance and market share. 802. there is a tower cost at the hub (the “point”) as well as customer premise equipment costs (the “multipoints”). which include antenna alignment and positioning expenses. Tower costs can vary widely. but vary in the degree of throughput. and include components such as cost of land/right of way.

Wynne and Perkins (1995) determined that SDSS enabled decision makers to complete their tasks quicker. A DEM is a digital file containing terrain elevation information that has been sampled at regularly spaced intervals. called cells. Chapter 3: A Mathematical Programming… 21 . the medium-scale category includes 2-arc-second maps with a ratio of 1:100. 7. 1999). 2002). and divide the rectangle into a grid of n equi-sized smaller rectangles. Assume for pedagogical purposes that service is to be provided to a rectangular-shaped geographic area. store. now based in Redlands. 2002). and the small-scale category includes the 1-degree case with a ratio of 1:250. 1988. large. Since such equipment was very expensive. which creates more SDSS research possibilities. now the leading global provider of GIS software (ESRI. and stored in raster (matrix of rows and columns) format. ArcView GIS is a desktop geographic information system from ESRI.5-minute DEM provides the greatest level of detail from the USGS with 30x30 meter data spacing. and 1-degree. 2-arc-second. GIS packages can run on desktop computers. GIS software required large budgets. Silleos. The 7. SDSS have been applied to (among others) siting problems (Maniezzo. Bennett (1983). and with greater understanding of the problem through visualization. and small. Mendes.between the other two. 15-minute. CA. more efficiently. consider wireless service that does not require line of sight. and hard disk space. Larger scale means greater map detail.000 and 1:63. the United States Geological Survey (USGS) produces Digital Elevation Models (DEMs). for example.000. 1999).2 to develop our own special-purpose GIS software. Keen and Scott-Morton (1978).360 respectively. was founded as Environmental Systems Research Institute in 1969 as a privately held consulting firm that specialized in land use analysis projects. and vehicle routing (Keenan. 2001). The reader interested in further detail may consult. and analyze spatial data and can provide decision support. GIS link tabular data to graphical data by relating graphical layers to database tables.5minute Alaska. land planning (Nehme and Simoen. The large-scale category includes 7. ESRI. Vlachopoulou. (NEAR) LINE-OF-SIGHT OPTIMIZATION MODELS As a simplified introduction to the problem. In this research we use ArcView version 3. retrieve. For example. The advantage of an SDSS is its ability to integrate the model portion of the DSS with the graphical representation of the GIS. As computers have become cheaper and more powerful. Marrying DSS and GIS creates Spatial Decision Support Systems (SDSS). In their research. medium. Geographic Information Systems (GIS) are used to collect. 1998. and Paruccini. a demographic layer of a town map may have sections linked to economic information such as household income or school zoning information (ESRI. SDSS is a relatively new research area. 7.5-minute. Crossland. There are five DEM products for sale by the USGS. These maps are produced either by digitizing cartographic map contour overlays or by scanning photographs from the National Aerial Photography Program (USGS. 1999). Tarantilis and Kiranoudis.000. primarily because GIS software has historically needed a great amount of computing power. or Turban and Aronson (2001). These five products fall into one of three different scale categories. memory. and Manthou. thereby aiding decision makers with semi-structured or unstructured spatial problems.5-minute and 15-minute maps with ratios of 1:24. As part of the National Mapping Program.

. where ri indicates revenue generated in cell i by a tower placed somewhere in the grid (not necessarily in cell i) that provides signal to cell i. Let r be a vector of revenues. in particular. • Let c represent a vector of costs incurred by placing towers in the grid. and/or a signal between i and j might not have line-of-sight (due to terrain). if a tower placed at the center of cell j provides a signal to cell i Eij =  0. note that Eij might equal zero for either of two basic reasons: A tower placed at j cannot transmit a strong enough signal to a distant cell i. The cost of placing towers is ct · x = ∑ ci xi . Note that this allows costs to vary from cell to cell. if a tower placed at the center of cell j does not furnish a signal to cell i. let xi =  0. which we call the exposure matrix.e. Chapter 3: A Mathematical Programming… 22 . an assumption we will relax shortly. 1. if no tower is placed anywhere in cell i of the grid . • • In order to express mathematically a tower’s ability to transmit signals successfully to cells in the grid (i. In general. we define an n x n binary matrix E. In particular.Notation Define the following four n x 1 vectors: • Let x represent a decision variable indicating tower placement in cells. “cover” the cell). 1. then the dot product Ei· · x equals the number of signals cell i will receive from all towers actually placed. If we denote the ith row of the E matrix as the 1 x n vector Ei· . This latter observation leads to the formulation of several important potential constraints: • The specification Ei· · x ≥ 1 specifies that cell i must receive at least one signal. Let e be an n by 1 vector of ones such that et = (1 1 … 1). if a tower is placed at (the center of ) cell i of the grid . But since in the simplified non-line-of-sight case being considered at the moment it does not matter whether the receiving antenna can see the tower. where et represents the transpose of e. the second reason is of no concern. namely ci equals the cost of placing a tower in cell i. With these definitions the following should be observed: • • • The number of towers placed equals the vector dot product et · x = ∑ xi .

which we term the view shed matrix.e. as will now be explained.5 2 kilometers (see Figure 3. otherwise. given the height and tower location on a DEM.2b. note that Ei· · x takes on the value zero if cell i receives no signal and equals a positive integer value if a signal or signals are received. values of one for elements Vij in the matrix V indicate which cells j actually receive signal from a tower placed at the center of cell i. then the exposure matrix E ignoring line-of-sight issues and Chapter 3: A Mathematical Programming… 23 . That is. This means that DEM maps contain the topographic detail necessary to ascertain near line-of-sight capability. if. area that can see a tower’s transmission given the tower’s location and height (40 feet) in a rural county in the mid-Atlantic United States. and the transmission signal strength. can determine the visibility over an area of any tower signal. a tower at the center of cell j does not furnish a signal to cell i.2c for the pedagogic rationale for this “unusual” choice). we define Vij such that 1. and Morgan. given the terrain. i. (1) View Sheds Introducing the line-of-sight requirement necessitates a fundamentally different analysis because terrain. if Ei• · x > 0 revenue in cell i =  i 0.1 shows (superimposed on a DEM) the view shed. trees and smaller hindrances are not a problem. Whereas the exposure matrix assumes that an antenna within the tower’s range can receive any signal sent from a tower. and planners consider tower placements using line-of-sight LMDS technology with an omni directional antenna array reaching 2. This town is chosen so we may calculate both exposure and view shed matrices without relying on GETWEBS. the exposure matrix. we may write total revenue generated in the entire grid as r . i.• The matrix stipulation E · x ≥ e mandates that each cell in the grid must receive at least one signal. This is done in practice by making a point-to-multipoint determination within the GIS of what can be seen. and (sometimes) even trees must be considered. Finally. the Geographic-Engineering Tool for Wireless: Evaluation of Broadband Systems (GETWEBS). Signals still cannot go through large obstructions such as mountains. One thing we are interested in is determining the profit shed. If the town is divided into a grid of rectangular cells 1 kilometer on a side. given the terrain. the region that can both see a tower’s signal and generate a profit for a service provider. To exploit these maps. A simple example illustrates the concept. buildings. The modified GIS... Bill Carstensen wrote a special-purpose geographic information system (Carstensen. if. To determine this. Consequently. Enhance the notion of E. “Near” line-of-sight is less restrictive. GETWEBS must make this determination. 2001) and introduced the view shed concept. we integrate the GETWEBS view shed with a mathematical programming formulation. a tower placed at the center of cell j provides a signal to cell i Vij =  0. by defining an n x n binary matrix V. Bostian.e.2a) whose elevation profile is indicated in Figure 3. Pedagogical Mini-Example Consider a fictitious town (shown in Figure 3. Figure 3.

3. cell 1 will get coverage if towers are placed in cells 1. the portion of the view shed matrix corresponding to that part of town is identical to the exposure matrix E. or 4. then cell 1 receives no signal. 0 0 1 1 0 1 0 1 1 1  0 1 0 1 1 Recall that V1· indicates the first row of the V matrix. if placed. Chapter 3: A Mathematical Programming… 24 . To fix this.) Note that since the east part of town is flat (no hills or valleys). Consider another fictitious small. The same is not true on the other side of the tracks. as can be seen in Figure 3. but not in a diagonal direction. maximized revenue cannot be determined without defining a new binary variable that indicates whether a cell is covered: 1.the view shed matrix including those considerations may be written as indicated in Figure 3. whereas we want z1=1. will reach cell 1. 2. and x4 ≠ 0.4) and assume that a tower placed at the center of any cell will cover that cell and any cell it touches horizontally and vertically. stipulate that z1 ≤ x1 + x2 + x4 (or. if cell i receives (at least ) one signal . Furthermore. This suggests an integer mathematical programming formulation similar to the set-covering problem (Taha. then z1 may equal either 0 or 1. Let zi =  0. Profit Model A wireless Internet. because the hill prevents signals from a tower on one side of it from reaching the other side. Now if x1 = x2 = x4 = 0.3. We develop the wireless formulation by way of a simple example. This may be seen in the second row of the V matrix: 1  1 0 V=  1 0  0 1 0 1 0 0  1 1 0 1 0 1 1 0 0 1 . 2. we note that the V matrix is equivalent to the E matrix. This says (as desired) that if there is no tower in cells 1. a tower in cell 2 will cover cells 1. z1 ≤ V1· · x). x2.3. 2. For example. equivalently. etc. But if one or more of x1. and 5. or 4. That is. Since the town is completely flat. 1975). then z1 = 0. flat town in which we want to consider providing wireless service.. if cell i does not receive a signal at all. (We have assumed a tower height of zero for pedagogic reasons. Although the expression V · x ≥ e guarantees coverage of all cells. we say max z1 subject to z1 ≤ x1 + x2 + x4. it specifies which towers. provider will most likely want to maximize profit subject to financial capabilities. For pedagogical purposes we divide the town into only 6 cells (see Figure 3.

we further specify the following: Assume M = $500K. zi ε [0. and all cells have the same number of households. Then. x2. and propensity to pay. fictitious town. as desired. then M is allowed to approach infinity. there is a 5-year economic horizon. The objective max rt · z subject to z ≤ V x allows us to write the vector version including all cells. and if one or more of x1.then if x1 = x2 = x4 = 0. namely: Ni = 70 households. and x4 ≠ 0. Pedagogical Example To illustrate the model for our six-celled. ccpe = 0.1} ∀i. zi ∈ {0. penetration.1] for all i. then constraint (2e) below is added to the system: V x≥e (2e) (2a) (2b) (2c) (2d) If there is no budgetary restriction. we may write: max ρt · z – ct · x st/ V x ≥ z ct · x + ctcpe · z ≤ M xi. and constraint (2c) is removed from the model. then z1 = 1. and penetration = 1/3. If everyone must receive a signal. revenue = ri = 70 * (1/3) * $3000 = $70. then in general. ci = $200K (∀i). Chapter 3: A Mathematical Programming… 25 . Equations (2) are written as (dollars are in thousands): Max st/ 70(z1 + z2 + z3 + z4 + z5 + z6) – 200(x1 + x2 + x3 + x4 + x5 + x6) 200(x1 + x2 + x3 + x4 + x5 + x6) ≤ 500 x1 + x2 x1 + x2 + x3 x2 + x3 x1 x2 x3 + x4 + x5 + x5 + x6 + x4 + x5 ≥ z1 ≥ z2 + x6 ≥ z3 ≥ z4 ≥ z6 + x4 + x5 + x6 ≥ z5 xi. propensity to pay = $50/month or $3000/horizon. then z1 still equals 0. If we let ρi = ri − ccpe (where ccpe is the customer-premise-equipment cost). with a budget of M.000 for each cell if a signal is received in that cell.

A Spatial Decision Support System We have built an SDSS that implements the mathematical modeling features described above with the GETWEBS geographic information system. zi ∈ {0. which states that all cells are covered.. which says to place towers in cells 2 and 5 only.5. and financial modeling to be intertwined in the appropriate mix. This model is interesting in examining regions where profit cannot be made. The visibility module consists of the equipment definition sub module.6. The point noted here with this SDSS is that general modules have been/can be written that allow various equipment. the view shed matrix V) given the antenna specified and the GETWEBS program.. x1 = x3 = x4 = x6 = 0. including the range. forecasting. Of the $500. and the profit is $20.000. the insistence that a tower be placed in a given cell. (3a) (3b) (3c) (3d) Additional Constraints There are other constraints not indicated above that may be written when appropriate. Chapter 3: A Mathematical Programming… 26 . The reader interested in incorporating such constraints should consult a basic mathematical programming text. then the exposure model is max ht · z st/ V x ≥ z ct · x + ctcpe · z ≤ M xi.e.000 remains unspent. z1 = z2 = z3 = z4 = z5 = z6 = 1.000. the equipment sub module contains technical capabilities. Maximizing Exposure Model In addition to the profit maximization model formulated above. circular.The solution is x2 = x5 = 1. and the evaluation module contains both the max profit and max exposure mathematical programming models as well as requisite financial sub modules. If hi is the number of households in cell i. Moreover. etc. the cost of placing towers is $400. the specification of an existing tower in a cell.000. The revenue module includes the census data access and various marketing revenue sub modules. The view shed sub module calculates exposure (i. the requirement that a specified fraction of total households be serviced. the demand that k out of r cells (where k and r are specified constants) be given service.g.). and shape of the region covered by each antenna (e.000 budget. These include: budget constraints for any given time period. and the view shed sub module. An example of a tower specification screen is shown in Figure 3. and the evaluation module.1} ∀i. the exclusion of a particular cell from tower placement. such as Taha (1975). A graphic description of the SDSS architecture we employed is shown in Figure 3. elliptical.5 shows three paths to the (output) profit-shed module: the visibility module. The total revenue generated is $420. Figure 3. angle of transmission. In addition to descriptive information on each type of equipment. etc. the revenue module. but government subsidy is considered at a specified level. an important model maximizes exposure within a prescribed budget. $100.

The parameters we used are based on published data. Wireless Parameters and Costs We examine the three fixed. The county is approximately a 40 km by 40 km rectangle. the 802. with only 10% of households and small businesses connected to it. the signal is such that any one transmitting cell can reach at most only 28 other cells. and transmitting cells on edges of the region have an even further reduced set of possible receiving cells. For example. and (3) the search over the receiving cells can be significantly limited by the range of the signal. consists of 393 square kilometers in a rural setting. We base our wireless cost data on actual systems with which we are familiar. Obviously the 50% threshold may be adjusted to meet particular circumstances. GETWEBS Data Montgomery County is a medium-sized county in the mid-Atlantic United States. (2) the software reports the percentage of the receiving cell able to get a signal. as will be noted. we assume transmission towers may be placed only in the middle of any 2-kilometer cell. We arbitrarily specify that a cell with greater than 50% coverage from a tower is covered. and has an active interest in low-cost. which we subdivide into 394 cells (arranged roughly as 20 rows with 20 columns). and then determines whether line-of-sight exists to each of the other (“receiving”) cells. and the process described below would be unchanged if analysis in greater detail were desired. As mentioned. Chapter 3: A Mathematical Programming… 27 .500 (about 29. although computational complexity would be significantly multiplied. each cell being a square of side 2 kilometers. It has a population of 77.000 households). although some cost specifications are difficult and approximate.16 standard. namely the lack of broadband infrastructure going “the last mile. we now return to the issue posed at the beginning of this paper. point to multipoint technological options currently under most consideration: local multipoint distribution systems (LMDS). in one scenario considered here. census data down to squares 30 meters on a side is available. The output from our view shed software is a 394 by 394 matrix filled with percentages. We use USGS census maps in the GETWEBS program. To demonstrate proof of concept. and enter a “1” in the corresponding cell for the view shed matrix. Three notes should be made with respect to this calculation: (1) this is a point-to-multipoint GIS calculation. GETWEBS considers a transmission tower centrally placed in a cell.” Recall that Kornbluh (2001) noted that 97% of the broadband backbone is not in use. seen earlier. and the 802.REACHING THE LAST MILE Having developed a mathematical programming model with an embedded geographic information system.1. high–speed Internet service. we develop profit and exposure models for various near line-of-sight technologies for a region approximately 40 km (25 miles) on a side. from the tower to each locale in the receiving cell.11b option. shows a screenshot of this county displayed in the GETWEBS program. Figure 3. For simplicity. partly due to the presence of a large land-grant university.

Note that in that table. the view shed matrix V is determined by GETWEBS and substituted into equation (2b).Range We use a range (i. it is a fair approximation to say the tower costs of all three technologies are approximately the same. Tower Costs We assume four 90º sectored antennas at each tower. I is the average per-household annual income (in 000s of dollars). and pm is the price per month for the wireless broadband offering. one would expect the wireless broadband service to do at least as well as the two alternatives mentioned. and a most likely value of $150.000.900 for LMDS.2. The particular functional relationships employed in the study are shown in Table 3. the percentage of households in a cell purchasing (“participating”) in the broadband offering is a function of the average household income in that cell given the monthly price of that service. depending on the wireless technology employed. I is the average annual household income for a cell. transmission signal from the tower) of 5 2 ≈ 7. Results The profit model developed above is run for each of the three technologies.000. where %P is the percent participation.000. That is.16). First. and depend primarily on the electronics costs of each system.e.. and connecting (or transmitting) to the backbone. and (2d) is solved. Chapter 3: A Mathematical Programming… 28 . $1295 for the WirelessMAN (IEEE 802. Customer Premise Equipment Cost The receiver antenna costs do vary widely by technology. we chose a triangular distribution with a minimum value of $100. providing omni-directional coverage. This value is within the scope of all three systems and is chosen because a tower placed at the center of a cell can reach diagonally to all households two cells away. Costs for each tower. and. as the wireless bandwidth would be an order of magnitude or so better. At a similar price. Talks with “experts” suggested that we model tower costs probabilistically. getting electricity to the site.1 kilometers. pm is expressed in dollars. will vary considerably depending on the location and accessibility to other infrastructure. even for Montgomery County. include the cost of land purchase and/or right-of-way. namely. a maximum of $500. Revenue Model Actual year 2000 number of households and per household annual income are available or can be calculated from United States census data for each cell in Montgomery County. These functional relationships were estimated from information on “related” services in the local area. beyond the antennas and the structure. $230 for the Wi-Fi system. The costs we used were $7. the price charged for DSL and for Adelphia Internet service. We assume the following functional form to calculate revenue: %P = f(I | pm). The tower costs. Then the system given by equations (2a). As the non-antenna costs dominate. (2b). and %P is a number between 0 and 100.

LMDS at $50 per month This technology.11b Wi-Fi alternative generates a significant profit.2M. data indicate that the profit begins to drop quite sharply beyond 90% coverage. no towers should be placed. What is happening is that. but the government wishes to increase the number of households reached. Alternatively. if the government agrees to make up for the provider’s loss in profit.000 in profit is generated.1). If the government decides it wishes service to be provided to (say) 90% of the households in the county.16 (WirelessMAN) at $50 and $100 per month The WirelessMAN reduces the customer premise equipment to $1295. for Montgomery County. further analysis with the tool can be performed to determine the incremental benefit of reaching additional homes. We first consider the case where a for-profit service provider has found a limited level of service profitable. in the Wi-Fi example given above. the maximum-profit solution provides service to 80% of the households in Montgomery County. If the government agrees to a subsidy to the provider for reaching the extra homes. Regions in the profit shed (i. the 802. then the for-profit provider will not want to service the extra households because less profit will be obtained. though touted for several years until the fairly recent past. For example. a government subsidy to provide service beyond 90% would give relatively little marginal benefit for the marginal expenditure. is substantially too expensive for deployment in a county similar to Montgomery County. generating a profit of $5M on a cost of $6. where tower costs increase dramatically. We consider only the Wi-Fi (802. towers must be placed that cover relatively few individuals. A second subsidization alternative beyond the cooperative venture between the government and a for-profit firm is one in which no for-profit wishes to provide broadband service to a region. Moreover. once again wireless service is not profitable. (See Figure 3. 802. Almost 80% of the county is projected to participate. the subsidy would be $500K. Nonetheless. hence. Moreover. So. the subsidy would be $1. We now examine Kornbluh’s (2001) and Technet’s (2002) calls for government subsidization to see at what cost Montgomery County can reach more households with broadband availability. if the government agreed to make up for the difference in total costs (not shown). only (approximately) $25. in a view shed that generates profit) are shaded. with the easiest-to-reach people having been served. universal service was not nearly achieved.. The one tower placed is positioned in the northern portion of the county (see Figure 3.11b) alternative. the amount of the subsidy could be studied using the spatial decision support system – one merely adds a constraint to the profit maximization model requiring service to the desired percentage of homes. At a price of $100/month.7). requiring 6 towers (see Figure 3. The system solution declares that no profit can be made and that. In such a case the government may believe it is necessary to subsidize the Chapter 3: A Mathematical Programming… 29 . at a price of $50/month.e.8 for some results at different levels of coverage. the optimal solution suggests that one tower be placed reaching approximately 700 households (less than three percent of the county total). For example.7M.) For example. DISCUSSION The previous section examined three wireless technologies in a “for profit” context. Wi-Fi at $50 per month With CPEs down to $230. Even with the Wi-Fi scenario.

To the extent that appropriate probabilistic models can be formulated to specify local tower costs and revenue receipts. it is safe to draw several general conclusions. But what is hard to specify are tower costs and revenue projections. It is not possible to generalize from one county to draw conclusions about broadband in the United States as a whole. 1670 households can be reached. The cost of customer premise equipment is the largest problem. For example. including the cost of land access. Tower costs also need addressing. and is still prohibitive for sparsely populated. census data maps are available with household data. CONCLUSIONS This paper has shown how mathematical programming can be integrated with an enhanced geographic information system to build a spatial decision support system for analyzing options in the deployment of wireless broadband alternatives in rural areas.project. given sufficient effort. specifying the subsidization amount in equation (3c). although significant effort may be required to generate reliable estimates. Both profitmaximization and subsidization scenarios were presented in addressing the “last-mile” problem in a particular county in the mid-Atlantic United States region. In particular. with one tower being placed. and DEMs can be analyzed to develop view sheds. the spatial decision support system developed in this paper can be used to determine the magnitude of the last-mile problem and analysis of different policy options. In this situation the maximum exposure model of equations (3) could be utilized. Chapter 3: A Mathematical Programming… 30 . although little can be done to lower several components of these costs. It is seen that at current technology levels and costs. and connecting from the tower to the backbone. Nonetheless. provision of power. one could generate answers for the entire United States. hilly terrain. for a small subsidy of $500K in Montgomery County. providing widespread broadband service to many rural regions will require significant government subsidization. However.

287(1). Spatial Decision Support System for Land Assessment. 3.html. L. M.C... (2001). Wireless broadband in a box. MA: Addison-Wesley. Building Decision Support Systems. Retrieved July 30. P. and Financial Modeling in a Software Package for Broadband Wireless Wide Area Network Design. Chapter 3: A Mathematical Programming… 31 . The new science of management decision. Bennet. Maniezzo. & Riezenman. Reading. Scientific American. Spatial Decision Support Systems: An Overview of Technology and a Test of Efficacy. Crossland. D. (1960).S. Nokia Demonstrates Wireless Mesh Broadband Market Leadership with More Than 50 Network Customers. (2001).C. (Ed. editorial desk. 47 (1). 273-284. Fowler. Press release. & Scott Morton. 219-235. Inc. MO: ACM Press. Shrick. Proceedings of the International Conference on Electromagnetics in Advanced Applications. W. J.). in ACM GIS. ESRI (1999). (2002). M. Bostian.” The New York Times..G. December 10. 2001. & Paruccini. 1. Kornbluh.. Combining Electromagnetic Propagation: Geographic Information Systems. 39(6). Keenan.J. T. P. Overview.nokia. (1978). Decision Support for Siting Problems.E. Kansas City. A.. Decision Support Systems. 3. (1998). & Simões (1999) M. 38 – 43.. “The Broadband Economy. H. 65-71. V. W. Getting to Know Arcview GIS: The Geographic Information System (GIS) for Everyone (3rd ed.. Mesh Networks for Broadband Access. & Perkins. (1998). 2002. page 21. Simon.. (2002). from the World Wide Web: http://press. MA: Addison-Wesley. B. C. Nehme. New York: Harper & Row. Keen. G.com/PR/200202/848470_5. B. Last Mile by Laser.D.B. Decision Support Systems.. & Morgan. 17-22. Wynne. (1995). M. (2002).W. K. 49-53. Nokia. (2001). IEE Review.E.C. IEEE Spectrum. Spatial Decision Support Systems for Vehicle Routing. New York: Environmental Systems Research Institute.W. Carstensen. Decision Support Systems: An Organizational Perspective. Decision Support Systems. dateline: Washington. 1-14. M. 799-810. Mendes. I. C. column 2.) (1983).REFERENCES Acampora. Reading. Section A.L.

R. http://edcwww. (1975). 205-212. & Kiranoudis. Willebrand. Information & Management. 4145. V. B. from the World Wide Web: http://www.E. C. Retrieved July 30. IEEE Spectrum. & Aronson.. Turban. Englewood Cliffs. Chapter 3: A Mathematical Programming… 32 . Decision Support Systems. C. G. Lawrence A. (1982). 2002. T.D. E.B.. Willis. Inc. and Computations..H. Hasletad. NJ: Prentice-Hall. 247-264. Decision Support Systems and Intelligent Systems.T. Geographic Information Systems in Warehouse Site Selection Decisions. (2002). Tarantilis. Using a Spatial Decision Support System for Solving the Vehicle Routing Problem. J.Theory. 359-375.. (2002).D. B.. Technical report. USGS Digital Elevation Model Data. Holm (2001). (2001). Fiber Optics without Fiber. O. T. J. (13). Journal of the IBTE. (3). E.gov/glis/hyper/guide/usgs_dem Vlachopoulou.. International Journal of Production Economics. H.Sprague.J. Silleos. & Ghuman. (2001). J..A. 867.A. Friis∅. Applications.technet. 48-53. United States Geological Survey. (5). H. (2001). West. TechNet (2002)... Building Effective Decision Support Systems.S. & Hess. New York: Academic Press. M. Integer Programming . 2(2). Exploiting Peer-to-Peer Communications – Mesh Fixed and ODMA Mobile Radio. & Carlson. & Manthou.org. Metadata as a Knowledge Management Tool: Supporting Intelligent Agent and End User Access to Spatial Data.usgs.cr. A National Imperative: Universal Availability of Broadband by 2010. (8). USGS (2002). Taha. New Jersey: Prentice-Hall. T.

A GETWEBS screen shot indicating the view shed of a tower placed at the dot near the center of the top of the screen and with range indicated by the circle.Figure 3. Chapter 3: A Mathematical Programming… 33 .1.

A fictitious town. elevation above sea level (meters) 400 300 West Figure 3. divided into an east and west region by a railroad track that runs north/south through the center of town.2b.N N 5 km West East 10 km 5 km 10 km Figure 3. A transmission tower’s signal can reach diagonally to all households two cells away. East Figure 3.2a.2c. An elevation map of the fictitious town. Chapter 3: A Mathematical Programming… 34 .

2. wheras white cells indicte a 0. Chapter 3: A Mathematical Programming… 35 . cell 1 is in the upper left corner of the town. With regard to Figure 3.1 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 Notes: 1.3. and cell 50 the bottom right corner. The exposure matrix E and the view shed matrix V for the fictitious town depicted in Figure 3. 2. Checkered cells indicate a 1 in the exposure matrix E and a 0 in the view shed matrix V.2. 3. Figure 3. Black cells indicate a 1 in cell (i. j).

1 2 3 4 5 6 Figure 3. A fictitious mini-town. Chapter 3: A Mathematical Programming… 36 . divided into six cells.4.

The wireless spatial DSS architecture Chapter 3: A Mathematical Programming… 37 . PENETRATION PROFIT SHED Census data EVALUATION CRITERIA MATH PGM MATH PGM MODEL • • max profit o budget year by year max exposure o budget in/exclude FINANCIAL MODEL • • economic life cost of money depreciation.VISIBILITY EQUIPMENT DEFINITION VIEW SHED GETWEBS REVENUE HOUSEHOLDS PROPENSITY TO PAY INCOME. ETC. etc.5. • Figure 3.

32. and 74) or included (e.Figure 3. and 88 ) for tower placement as desired. 38. 2.g. Chapter 3: A Mathematical Programming… 38 .6. Cells may be individually excluded from (e.g. 26.

Tower placement and profit sheds (magenta) for Wi-Fi (802.7.11b) service offered at $50/month.Figure 3. Chapter 3: A Mathematical Programming… 39 .

60 Profit ($00000's) Households (000's) 50 Towers 40 30 20 10 0 50% 55% 60% 65% 70% 75% 80% 85% 90% 95% Percentage of Households Figure 3. and towers in Montgomery County as a function of percent coverage with Wi-Fi broadband. Chapter 3: A Mathematical Programming… 40 . households. Profit.8.

25-5.4385 GHz 5.htm).25GHz 5. Some wireless applications and their allocated frequencies.25 GHz 28 – 31 GHz Source: The Virginia Tech Center for Wireless Communications.11b U-NII 802.470 MHz 824 .cwt. Chapter 3: A Mathematical Programming… 41 .4 .24.11a 802.11a Amateur satellite Part 15 devices Naval radar systems Test range instrumentation radars 802. (http://www.894 MHz 902 .190 kHz 550 .849 MHz and 869 .2.108 MHz 150 .170 MHz 260 .16 – (Tsunami equipment) Radio navigation Local multipoint distribution service (LMDS) Frequency Band 170 .1600 kHz 27 MHz 49 MHz 88 .727 .) Military radiolocation systems Federal mobile communications Pager services with high transmitter power Amateur satellite Part 15 devices Microwave ovens and systems Army packet radio development 802. Applications VLF band radios Beacons AM broadcast communications Low power voice and data CB radios Low power voice and data Remote control Cordless telephones Low power voice and data FM radios Low power voice and data Commercial two-way voice Pager services 303 MHz garage door openers Keyless entry systems Security alarms Cellular phones ISM band Wireless LAN's Part 15 devices (spread spectrum cordless phones.1.928 MHz 930 MHz 2.edu/wireless_faq/default.5.Table 3.15-5.35GHz 5.875 GHz 24 . etc.vt.

Table 3.05 I %P = 0.2 I %P = 0.1 I %P = 0.05 I %P = 5 Average Annual Household Income. given the price per month for wireless service.2. The percentage participation (in Montgomery County) as a function of average household annual income. Percent Participation (%P) $50/mo price %P = 0.1 I %P = 10 $100/mo price %P = 0.4 I − 15 %P = 25 $75/mo price %P = 0. in 000s (I) $0 to $75 $75 to $100 >$100 Chapter 3: A Mathematical Programming… 42 .

NETWORK-FLOW MODEL FOR SOLVING HOP CONSTRAINED. FIXED-CHARGE.CHAPTER 4 A CAPACITATED. WIRELESS MESH NETWORKS Chapter 4: A Capacitated. Fixed-Charge… 43 .

capacitated. Keywords: Network Telecommunications Flow Models. The model is capable of handling differing bandwidth requirements for each customer. then solves the resulting problem as a fixed-charge. NETWORK-FLOW MODEL FOR SOLVING HOP CONSTRAINED. Fixed-Charge… 44 . A solution procedure is developed here that generates all feasible customer paths to the backhaul in a preprocessing step. network flow problem. WIRELESS MESH NETWORKS ABSTRACT Wireless mesh networks are proposed as a possible solution to the “last-mile” problem in broadband communications. FIXED-CHARGE. Math Programming. mesh networks with hop constraints that ensure transmission quality are examined.A CAPACITATED. Broadband Wireless Chapter 4: A Capacitated. In particular. whose solution is well known.

Chapter 4: A Capacitated. In fact. and Rees. to underestimation of the cost of delivering broadband service the “last-mile. FIXED-CHARGE. to meet for the broadband requirements of the new millennium. WIRELESS MESH NETWORKS INTRODUCTION The popularity and rate of growth of the Internet over the last decade has been prodigious. mesh network configuration of a neighborhood. including mesh. as obstructions such as buildings and terrain prohibit transmission. and then by implications and limitations. primarily. 2001.A CAPACITATED. Finally. as Kornbluh states in her NY Times op-ed article. 2001). what was once a medium for government and academics to exchange ideas has now become an almost ubiquitous entity that assumes many shapes – from personal communications. the paper concludes with a summary of contributions and requisite future work. 2002) In the 1990s. to global information collection. it is not longer sufficient to connect to the Internet with low speed modems. namely the capacitated. We then enumerate every viable. Several varieties of this approach exist (see. always “on” service. fixed-charge. One approach to solving this problem is wireless telecommunications (Willebrand and Ghuman. and formulate a mixed integer programming problem. Scheibe. to determine the optimal network with minimal cost. which we will define shortly. the solution methodology (including the mixed integer formulation) is presented. thereby ensuring a quality of service (QoS) level.” where the “last-mile” is that part of the telecommunication network that connects the homes and businesses to the high speed backbone. particularly financially. whereby broadband is meant the capacity to deliver high speed. Acampora (2002) states that for nine out of ten American businesses with more than 100 workers. Fixed-Charge… 45 . Rakes. and has been sufficient for voice communications for half a century. (TechNet. As more people and companies today do business on line.. Broadband telecommunications have become a necessity. for a discussion of the point-to-multipoint (PMP) line of attack). This is due. We will utilize Geographic Information Systems (GIS) in a preprocessing analysis step to determine which neighbors can see each other electronically. they are not connected to it. companies poured billions of dollars into building a transcontinental fiber optic backbone (Kornbluh.g. feasible path from customers to the backhaul. to on-line shopping. But it will not be adequate. While most American homes and businesses exist relatively near the backbone. This last-mile link to the customer has traditionally consisted of copper wiring. only three percent of that backbone is currently in use. This discovery of viable communication links is combined with an enforcement of number of hops from each source (customer) to sink (backhaul). the need for greater bandwidth increases. NETWORK-FLOW MODEL FOR SOLVING HOP CONSTRAINED. 21). the backbone is less than a mile away. e. This is followed by an example. network flow problem (FCNFP). However. Carstensen. 2003. The purpose of this paper is to ascertain the minimum-cost. After background material on meshes as well as on the general capacitated FCNFP. p.

In Figure 4. 1992).BACKGROUND Mesh Networks As mentioned. 2001). This can reduce network costs as well as provide network survivability. and Chapter 4: A Capacitated. 2002). In a mesh network signals from a customer seeking the backhaul may be forwarded from customer to customer (possibly including repeaters). Wireless networks have traditionally been point-to-multipoint (PMP) networks (Whitehead. Although hop constraints are not new to network design (Balakrishnan and Altinkemer. but it is not economic to connect residential neighborhoods and/or businesses by wire to that backbone.1. until finally a mesh insertion point is reached. One of the key features of mesh networks is that customers are used to forward signals on to other customers. The wireless signal is not. Whitehead suggests that six to ten hops offer an acceptable level of service (Whitehead.4 GHz (Nokia. there must be no obstruction between transmitter and receiver. we define some terms as they are used in practice or in the literature. Generally. and finally to the backhaul. The backhaul point (BH in Figure 4. Latency is the time it takes for a packet to travel from source to destination (Webopedia. The former work with strategically placed towers. The signal is sent to mesh insertion points (MIPs. and then back out to customers. signals are not transmitted from customers to MIPs. Fixed-Charge… 46 . once signal is received at a MIP back from a customer. That is. the fundamental issue with the last mile problem is that capacity exists in broadband network. and laser (Acampora and Krishnamurthy. 1999). 2000). it is only forwarded to a trunk connection point and then to the backhaul. or latency. 26 GHz to 40 GHz (Fowler. 2003). Because mesh frequencies are in the gigahertz range. transmission between each transmitter and receiver must be line of sight. M2. the probability of a network meeting a given traffic contract (Wikipedia. The MIP then sends the signal to a trunk connection point. 2003). up to the capacity limits of the receiver and transmitters. connected to the backhaul. transmitted from the trunk connections directly to customers. the transmission medium changes to wireless at these trunk connection points. it is sometimes necessary to restrict the number of hops in each transmission path. However. however. sometimes it is necessary to insert repeater nodes into the network when there are not enough customers nearby to get messages to every customer. labeled M1. that is. there has been little discussion in the literature as to what should be the optimal number of links for a wireless mesh network. Thus a repeater may be thought of as a non-revenue generating customer.1) is that location along the backbone network where the mesh network will connect. Each link in the chain back to the backhaul is called a hop. redundant paths. shown as T1 and T2 in Figure 4. 2000). A traffic contract may be viewed as an agreed upon throughput rate during peak usage times. The backhaul is then generally "wired" (say with fiber optics) to trunk connection points. in general. and M3 in the figure) that are closer to customers. Too many hops in a chain of connectivity can cause inordinate delay. To ensure quality of service. To facilitate discussion of our approach to providing such capability in a wireless manner.1 an existing backbone and neighborhood without broadband service are shown. Mesh networks overcome several limitations of point-to-multipoint (PMP) networks. Mesh networks may use a variety of frequencies. not only must equipment be proximate and within range. That is. Repeater nodes are generally identical to customer equipment – there is just not a customer at the repeater location. These mesh insertion points do transmit and receive directly to/from customers. such as 2.

Mesh networks obviate most of these concerns. 2001. the backhaul as the sink. which is assumed to be nonnegative. and transportation (Magnanti and Wong. logistics and production planning (Minoux. Fixed-Charged Network Flow Models The fixed-charge network flow problem (FCNFP) is one of a large class of network design problems. A) where N is a set of n nodes and A is a set of m arcs. Girish. it is only necessary to determine a connection point location that will bring signal into one section of the mesh. upper bounds exist for the amount of flow over an arc. FCNFP problems are known to be NP-hard. 2000. and transshipment nodes. In these problems. the incremental addition of customers to the network is inexpensive. therefore. and Hu. From that point. Magnanti. That approach necessitates purchase of/permission to use land. sink nodes. fixed-charge. let (i. This occurs when. Note also that with a mesh. the design cost is denoted fk. Simply stated. Soni. it is not crucial to locate the single tower location that will best serve the entire area. and repeaters as transshipment nodes. Let dj denote the supply or demand at node j. the greater the number of hops between nodes. and arcs are constrained with an upper bound or bandwidth capacity. A fixed cost is incurred for using arcs between nodes. a connection point does not need to be optimally placed to reach every customer directly. The capacitated. This allows for greater flexibility in choosing connection node locations. and should lower the initial investment cost (Whitehead 2000). which consume flow. 1999. With mesh networks. 1991. 2001). In fact. which produce flow. Another consideration for some network design problems is the number of hops from a source to a sink (Pirkul and Soni. Every node is classified into one of three categories: Source nodes. bringing electric power to the tower. and much research has been devoted to creating better and more efficient solution procedures (Kim and Pardalos. by adding more customers. Shulman. the other nodes will pass the signal along. which has been used in many applications including telecommunications (Balakrishnan. the lower quality of service. Fixed-Charge… 47 . the only cost is the customer premise equipment. and TCN. the greater the amount of time it takes for data to be transferred between them. the objective is to seek the most efficient way to move flow (in our case bandwidth) on a network in order to satisfy demand between origin and destination nodes and to minimize the overall cost. 2002.j) denote a directed arc from node i to node j. and Mateus. Gouveia and Requejo. Gavish. The principal reason for constraining the number of hops in a network is to maintain a level of quality of service. the cost of the mesh network can decrease. and Wong. Moreover. For each node k. the longer it takes. 1991). often. We assume a balanced network where total supply equals total demand (or that we have balanced the network by introducing either slack or surplus variables to consume excess capacity or demand). it behooves the objective to use as few arcs as possible for which costs are associated. etc. 2001). Zhou. 1984). 1998). Smith. and ui > 0 be the capacity of each node i. Chapter 4: A Capacitated. The mesh network can be cast as a FCNFP with the customers acting as sources. which pass flow through. The objective is to minimize the design costs that occur when arcs are used. MIP. Cruz. the need for repeater nodes decreases because paths to the backbone may be found through revenue-paying customers (Fowler. as more customers join the network. When the network is capacitated. network-flow problem can be formulated mathematically as follows: Given a directed graph G = (N. permitting. There is a fixed cost associated with arcs that are connected to and from the transshipment nodes. 1989).within range of the maximal number of potential customers.

a minimum cost solution may be found using integer programming solution packages. where a dotted line between any two circles indicates no obstruction between the two nodes. then j is a transshipment node. yij ∈ {0.3 can then be generated. These equipment locations are primarily a matter of physical characteristics of a neighborhood. which locations are chosen is an economic decision to be determined by the model. given a hop constraint. Fixed-Charge… 48 . j )∈A (1) (2) ∑x ij − jk k ∋ ( j . the location of the single backhaul point and two trunk connection points are assumed. where right-of-way/permissions are available. etc. then j is a source node.In this formulation. we also define xij to be the flow on arc (i. either a geographic information system (GIS) or field visit is required. e. Preprocessing Step 1 –Potential Equipment Placement and Line-of-Sight Determination Preprocessing begins with the specification of the backhaul point. trunk connection points. k )∈ A ∑x k∈N = d j . METHODOLOGY The approach taken in this chapter is to introduce a preprocessing step whose main purpose is to determine. and that inflows equal outflows for transshipment nodes. It is possible to enumerate these paths. Figure 4. j ) ∈ A (3) xij ≥ 0.2. and all possible mesh insertion points. ∀i ∈ N . but all potential locations are to be specified. kj k ∋ ( k . Constraint (3) is the flow capacity constraint. ∀j ∈ N ∑x i. and yi to be a binary variable equal to one when node i is in use and zero otherwise.g. customer C3 cannot transmit or receive to/from customer C4 in Figure 4. Again. then j is a sink node. where each circle (node) indicates the (geographic) location of the equipment/customer indicated. Line-of-sight must be determined next. Once all possible paths are enumerated. (Lack of a dotted line indicates no feasible transmission. and will be zero if the node is not made operative in the design. and if dj = 0. all known customer locations are specified. (i. If dj < 0. and it ensures that flow on all arcs leaving a node will not exceed the capacity of the node at the start of that arc. If dj > 0. with the methodology developed here. that is. ∀(i. as the number of customers for any mesh insertion point is bounded by the technology and since the number of hops on each possible path is also constrained by quality of service considerations. The situation to this point is indicated in Figure 4.) At this point it may occur that a customer desirous of Chapter 4: A Capacitated. j ≤ ui yi .j). It is unknown how many MIPs will be economic. The resulting formulation is min ∑ f k y k ..1} (4) (5) Constraint (2) is the flow conservation constraint and ensures that demand/supply is satisfied at each node j. terrain is not obstructive. all viable paths from each customer back to the backhaul. j ) ∈ A.3. Next.

3. Next. Recall that signals received at MIPs from customers do not go back out to other customers. For example. the first row of Table 4. Begin with the source node (in this case. 2001). Finally in Figure 4. one per row. and from each trunk connection point to the backhaul. their addition does not violate any hop constraints. count the number of hops from the source to the BH. if the next node is not a MIP. The nodes may be viewed as a series of sources and sinks. or the backhaul. as shown in Table 4. Excel has proven to be an excellent tool for many problems of this sort (Ragsdale.4 is C2. The possible paths for customer C1 are shown in Figure 4. In such an eventuality.3) are defined at non-customer locations that will ensure an unobstructed path to the backhaul for each customer. move from the source to the subsequent hop in line to the BH. Once this information is in hand. Table 4. they are all actually node C1 themselves. ADNs are generated in this fashion for every customer and repeater node.3. either because transmission range limits are exceeded or because obstructions block all possible paths to the customer. Since ADNs are not actual physical nodes in the network. This is the last step before transforming the data into a form that may be solved directly with the network flow model.2 has node C1 reaching the backhaul in three hops.3. they move on to trunk connection points.1).3 in Table 4. Chapter 4: A Capacitated.3 and Figure 4. the network flow model may be formulated. Information equivalent to Figure 4.6.6. which becomes C2.6 shows the re-integration of the nodes into the mesh by connecting them to their artificial directional nodes. and append that number after a period. Step 2 – Artificial Directional Node Determination We use Excel to process feasible. a TCN. hop-constrained paths. so C1 becomes C1. With this approach the simplest way to generate these paths is to define what we term artificial directional nodes (ADNs).4 lists each node as a source and indicates the one-step-away nodes. arrows are shown on the links from each MIP to its closest trunk connection point.1 and a (temporary) hop limit of six that illustrates the process for customer C1. potential repeater nodes (Figure 4.3 – C1. Fixed-Charge… 49 . or the backhaul. there is no hop constraint violation. TCNs. given a limit on number of hops. the next node from C1. Even though it appears that node C1 must now also pass through C1. then append the number of hops to the BH from that point. Table 4. With this list the construction of any feasible path may be automated.2 that is based on Figure 4. The method for creating and naming the ADNs is as follows. all possible paths from each customer to the backhaul may be enumerated. as the connectivity and hop constraint information is all included. A simple example is presented in Table 4. On line two of our example.3 may be stored in tabular form (see Table 4.5 is the example with artificial directional nodes.5 presents the new mesh network with all the artificial directional nodes along with the re-integrated nodes themselves. C1). Each row in the table represents an allowable hop within the mesh network.4 and Table 4. Therefore. Figure 4. Every number with a period represents an ADN. as the number of nodes from each is constant. Shown in Table 4. Following this procedure creates a new (artificial) node for every possible number of links from the node back to the backhaul. they are not needed for MIPS. Network Flow Model Once the artificial directional nodes and the actual nodes have been integrated into the new mesh network.4.service cannot obtain a signal.

1}. the number of hops and bandwidth. M. medium. all artificial nodes should be considered transshipment nodes (with a supply of 0). and the houses are placed on the center of each lot. as follows. xij is the flow on arc (i. Constraint (8) restricts flow from trunk nodes and mesh insertion points.. seven. ∀(i.e. and high. let the cost of activating that node be fk. and the levels for the bandwidth will be 2 Mbits/s. EXAMPLE To demonstrate the model a small example was constructed. eight potential MIPs. whereas constraints (8) and (9) are flow capacity constraints. Fixed-Charge… 50 . The example assumes a visibility of 60%. Flow leaving either a customer or a repeater node (i. The model may be written as min ∑ f k y k . ∀j ∈ N ij j ∋ ( i . mesh insertion points.We modify the general FCNFP model of equations (1) – (5) as follows. There are two potential TCNs connected to the backhaul. and customer nodes respectively. any artificial nodes have a cost of 0.j). moreover. Let ui > 0 represent the capacity on the arcs emanating from node i. Let dj denote the supply or demand at node j. Let N be the set of all nodes and Na be the set of artificial nodes. their potentiality exists in whether or not they will be used. meaning that nodes that are located near enough to each other for a signal to pass between may still have a 40% chance that something obstructs line of sight. The lots are assumed to be one acre each. These factors have three levels each – low. define the set of all arcs in the network by the letter A. R. Further designate T. The example consists of three neighborhoods with a total of 25 houses requiring wireless broadband service. ∀i ∈ (T ∪ M ) ∀i ∋ {[i ∈ (C ∪ R) ∩ (i ∉ N a )] ∩ [(i. no flow may traverse any arc departing that node. We consider two independent variables. The actual location of the aforementioned nodes is set. j ) ∈ A. The objective is to minimize total cost. ∀i ∈ N Constraint (7) ensures flow conservation. if the transmitter is inoperative. and C as the sets of trunk connection nodes. the sum of flow over all arcs leaving such a node) cannot exceed the capacity of the transmitter at that node. and yi is a binary variable equal to one when node i is in use and zero otherwise. Finally. k∈N kj k ∋ ( k . The levels of the hops are five. repeater nodes. and nine. j ) ∑x ij ≤ ui yi (9) (10) (11) xij ≥ 0. j )∈A (6) (7) (8) ∑x − jk k ∋ ( j . and fourteen potential repeaters. i*) ∈ A] ∩ [i* ∈ N a ]} j ∋ ( i *. j )∈ A ∑x ≤ ui yi . yi ∈ {0. k )∈ A ∑x = d j . whereas (9) restricts flow emanating from customer or repeater nodes. For each node k. Chapter 4: A Capacitated. which is assumed to be nonnegative.

The methodology preprocesses data by first using a GIS and then manipulating customer requests and equipment locales to form a list of feasible paths from which the optimal set of paths may be found by solving a well-known network-flow problem. and 200 Mbits/s between TCNs and MIPS. The network generated is constrained by a given number of hops. Although this is a minimum cost solution procedure. but. the number of source and sink nodes and viable paths created for each of the hop levels. it is approaching an upper bound on the number of hops. many of the paths will pass through the same pairs of nodes. and 10 Mbits/s. revenue of $127 per month for the basic 2 megabit service with a hop constraint of seven proves profitable within a five-year window – a not very lucrative opportunity.7 shows pertinent data from the example’s preprocessing step. IMPLICATIONS The solution procedure successfully generates answers to the wireless mesh broadband service problem. While the number of viable paths may be high. Larger problems may need a more sophisticated integer-programming engine. the number of source and sink nodes did not. and the cost for TCNs is $25. For instance. This is due to high level of redundancy of paths from a node to the backhaul.000 each. Several simple heuristics can be attempted (such as limiting the search to a fixed number of randomly chosen paths as the algorithm steps out from each node in search of additional hops to the backhaul) to limit the number of paths and thus the execution time of the integer programming part of the problem. and consequently. or links. in particular. results may be manipulated to examine sundry revenue schemes. proves expensive to solve. the cost for MIPs is $5. Note from this example that while the number of viable paths increased dramatically with each level of hops. This issue must be addressed in future work. Table 4. Shown in Tables 4. 80 Mbits/s between MIPS and any other nodes.8 are some results of the example. thus increasing the cost as more repeaters and MIP nodes are required. the number of viable paths from a customer node to the backhaul. the overall cost may actually increase. namely the total cost for each network. 2002). However.8 shows some of the results of running the mixed integer problem code. the bandwidth is split up over multiple paths. The arcs between nodes are capacitated at 30 Mbits/s between customer nodes and repeaters. The cost for using repeaters is $2. nine hops is probably not excessive. CONCLUSIONS AND FUTURE WORK A methodology has been presented to determine minimum cost wireless mesh network configurations.000 a piece. The procedure works reasonably well at nine or fewer hops. as the number of Mbits increases. as mentioned. With each additional hop added to the allowable number. generates a very large number of viable paths in the preprocessing step. in the example given above and its assumed cost structure. Fixed-Charge… 51 . Although mesh networks should not indiscriminately route customer signals over too many hops. to ensure quality of service. While it may seem intuitive that the cost would decrease by increasing the number of hops. and consequently. Another noteworthy point is that the cost does not necessarily decrease as the number of hops increases. Since the network is capacitated. Small problems may be run using Excel combined with Frontline’s Premium Solver (Frontline. if the nodes are not located such that they may avoid repeater nodes. the problem size grows dramatically. Table 4. and Chapter 4: A Capacitated.500 each.5Mbits/s.7 and 4.

Fixed-Charge… 52 . The tool developed here may not only be used by ISPs. An example was shown demonstrating how an Internet service provider (ISP) would approach service provision constrained at seven hops in a new neighborhood. Results were also shown when quality of service was loosened to nine hops and tightened to five hops with three different bandwidth levels. Chapter 4: A Capacitated.allows differing bandwidths for each customer. Implications of customer density. and the effects of quality of service and bandwidth are possible. and by equipment manufacturers to examine the likely adoption of new transmitters with improved ranges and lower costs. as desired. cost of technology. Future work examines managerial implications of wireless mesh technology. This research is important in estimating the extent to which wireless mesh broadband solutions may be expected to mitigate the “last-mile” problem. but by policy bodies to examine the wisdom of subsidies. Bandwidth requested was ten megabit per second for all customers.

Decision Support Systems. & Perkins. Using a hop constrained model to generate alternative communications netowrk design.. Inc. and Morgan. 6(5). M. (1995). Frontline’s Premium Solver™.E. 539-552.. Proceedings of the International Conference on Electromagnetics in Advanced Applications. A new Langrangean relaxation approach for the hopconstrained minimum spanning tree problem. Mesh Networks for Broadband Access. D. W.. ESRI (1999). 4. Crossland. and Pardalos P. Kim. A. Annals of Operations Research. and Mateus G. 17-22. 67-81. L. 24. (1991). S..W. (1992). (2000). NV. Inc. New York: Environmental Systems Research Institute. and Financial Modeling in a Software Package for Broadband Wireless Wide Area Network Design. Gavish B. 219-235. (1999). Operations Research Letters. Acampora. Cruz F.L. Annals of Operations Research. G. C.K.REFERENCES Acampora. Bostian. Formulation of the Traffic Engineering Problems in MPLS Based IP Networks. 47 (1). (1991).. Gouveia. Getting to Know Arcview GIS: The Geographic Information System (GIS) for Everyone (3rd ed.E. Magnanti T.solver.T. and Hu.M. 49-53. Topological Design of Telecommunications Networks – Local Access Design Methods...D. and Krishnamurthy. Frontline Systems. (2001). 799-810. IEEE Review.. M. Computers and Operations Research. European Journal of Operational Research. Balakrishnan A. Carstensen. (2002). Smith J.R. Scientific American. 17-71. Spatial Decision Support Systems: An Overview of Technology and a Test of Efficacy. (2002). B. and Requejo C. 33. K. Shulman A. T. W. J. (1998).. A.192-205.B. IEEE Personal Communications. Solving to optimality the uncapacitated fixed-charge network flow problem. L. A broadband wireless access network based on mesh-connected free-spaced optical links.. Zhou B. Combining Electromagnetic Propagation: Geographic Information Systems. Balakrishnan..R. IEEE.. (1999). 239-284.C. (2001). 62-65. 132. Fowler.). Incline Village. A solution approach to the fixed charge network flow problem using a dynamic slope scaling procedure. From the World Wide Web: http://www. 214-219.com. Frontline. 195-203. Fixed-Charge… 53 . Last Mile by Laser. 287(1). and Wong R. Models for Planning Capacity Expansion in Local Access Telecommunication Networks. 3. 25(1). 33. ORSA Journal on Computing. Altinkemer. Girish. A. (2001). Chapter 4: A Capacitated. Wynne.

181-198. from the World Wide Web: http://press. Section A. from http://www. S. and Wong. Annals of Operations Research. K. 48-53.webopedia.Kornbluh. R.R. Network Synthesis and Optimum Network Design Problems: Models.C. Carstensen. December 10. Hasletad.” The New York Times. Nokia Demonstrates Wireless Mesh Broadband Market Leadership with More Than 50 Network Customers. (2002). Ragsdale. O. 2(2).html. M. 19. Holm (2001). European Journal of Operational Research. 2002. Pirkul. the World Wide Web: Willis. A MathematicalProgramming And Geographic-Information-System Framework For Wireless Broadband Deployment In Rural Areas. (2002).nokia.. http://www.com/TERM/l/latency. Webopedia (2003). Soni.com/PR/200202/848470_5. 18(1). New formulations and solution procedures for the hop constrained network design problem. “The Broadband Economy. Under review. Nokia. Fixed-Charge… 54 . from the World Wide Web: Whitehead. 2001. 2003. Retrieved July 30..html. RAWCON.L. Transportation Science. 106. Minoux. S. T. Third Edition. B. Cincinnati..org/wiki/Quality_of_Service. In Press. and Rees.P.W. Journal of the IBTE. page 21.wikipedia. Rakes. Spreadsheet Modeling and Decision Analysis: A Practical Introduction to Management Science. (2003). H. 313-360. T.B. D. Chapter 4: A Capacitated. Hop Constrained Network Design Problem with Partial Survivability. and Soni. K.. editorial desk. Solution Methods and Applications. T. (2001). L. (1984). 1-55. column 2. (1989).T. Magnanti.. Networks. Southwestern College Publishing. Network Design and Transportation Planning: Models and Algorithms. Retrieved February 10. Wikipedia (2003). Friis∅. Scheibe. Philip (2000). Exploiting Peer-to-Peer Communications – Mesh Fixed and ODMA Mobile Radio. Retrieved March 10. C. 2003.T. T. dateline: Washington. (2001).P. Mesh Networks: A new Architecture for Broadband Wireless Access Systems. Press release. L. 2001.

1.Figure 4. Providing broadband service to a neighborhood with a wireless mesh network. Fixed-Charge… 55 . Chapter 4: A Capacitated.

Fixed-Charge… 56 .3.BH T1 T2 C3 M1 M2 C4 C2 Figure 4. Key: BH – backhaul Ti – trunk i Mj – mesh insertion point j Rk – repeater k (not shown) Cl – customer l Nodes represent potential wireless mesh equipment locations. C1 BH T1 T2 C3 M1 M2 C4 C2 R1 C1 Figure 4. Chapter 4: A Capacitated. Nodes shown linked together with line-of-sight capabilities.2.

given the lineof-sight configuration of Figure 4.3 C1.BH T1 T2 C3 M1 M2 C4 C2 R1 C1 Figure 4.4 C1.3 Figure 4. Chapter 4: A Capacitated.6 C1. Fixed-Charge… 57 .3 C4.4 M2 R1.5 C2.5. Customer C1 viable paths expressed in terms of artificial directional nodes. BH T2 T1 C3. Viable paths for customer C1 to communicate with the backhaul.4.5 C1.3.3 M1 C2.4 C2.

5 C2.M1 C2.6 C1.6.5 C1. Fixed-Charge… 58 .3 C1 Figure 4.3 C2 C1. Re-linking nodes to their artificial directional nodes.4 C2. Chapter 4: A Capacitated.4 C1.

2.3 C1.6 M1 C2. Fixed-Charge… 59 .Table 4.4 C1.3 M2 BH T2 BH Chapter 4: A Capacitated.4 R1. Source and neighbor nodes “Reachable” Neighbors Source C1 C2 C3 C4 R1 M1 M2 T1 T2 M1 M1 M1 R1 M2 T1 T2 BH BH C2 C1 C2 C2 C4 C3 C4 Table 4.4 C2.1.5 C1.5 Viable paths T1 BH M1 T1 BH C3. Viable paths for customer C1 to communicate with the backhaul Viable paths Source C1 C1 C1 C1 M1 C2 C2 C2 T1 M1 C3 C4 BH T1 M1 R1 BH T1 M2 BH T2 BH Table 4.3.3 M1 T1 C4.3 C2. Viable paths for node C1 with artificial directional nodes Source C1.

5 C2. Fixed-Charge… 60 .3 C1. Source and immediate destination nodes with artificial directional nodes included Immediate Source Destinations C1.3 C4. Re-linking nodes to their artificial directional nodes Possible Immediate Destinations Source C1.4 C2.3 C2.3 C4.5 C1.4 C2.3 C2.4 C1.4 C1.3 C4.3 C4.3 C2.5 C3.4 C2.5 C3.5.4 M1 R1.5 M1 C3.5 M1 C3.5 C1.4 R1.3 M1 M2 T1 T2 M1 C2.6 C2.6 C2.Table 4.3 M2 T1 T2 BH BH C1.4 C2.4 R1.4 C2.4 C1.3 C3.3 C4.4 M1 R1.3 C2.4.3 M1 M2 T1 T2 C1 C2 C3 C4 M1 C2.3 M2 T1 T2 BH BH Table 4.3 C1.3 C2.4 C1.5 C1.6 Chapter 4: A Capacitated.

Fixed-Charge.3 C2.6.3 C4 C4. Fixed-Charge… 61 .4 C2.4 C2. Some Results from the Example’s Preprocessing Step Source and Sinks Five Hops Seven Hops Nine Hops 337 832 760 Viable Paths 2.928 1.202.5 C1.3 C1.000 $205.4 C1.000 $210.5 C2.6 C2 C2 C2 C2.3 M2 Table 4.6 M1 C2.4 R1.000 $190.3 C2.3 Sink BH BH T1 T2 C1.4 R1.048 Table 4.5 C1.8.000 $190.3 M1 C4.000 5 Mbits/s $255.3 C4.4 C3.4 C2.000 Chapter 4: A Capacitated.3 C2.5 C3 C3.131 18.3 C1.Table 4. Results from the Example’s Capacitated.5 M1 C3. Network-Flow Problem Five hops Seven hops Nine hops 2 Mbits/s $255.000 $285.4 C1. List of sources and sinks for direct inclusion into the network flow model Source T1 T2 M1 M2 C1 C1 C1 C1 C1.000 10 Mbits/s $300.000 $295.7.

CHAPTER 5 ADDRESSING UNIVERSAL-BROADBAND-SERVICE IMPLICATIONS WITH WIRELESS MESH NETWORKS Chapter 5: Addressing Universal Broadband Service… 62 .

A new approach to providing broadband technology is wireless communications. Spatial decision support systems. but most areas lack adequate means of delivering the desired services to residential customers and businesses – at least for the last mile of connectivity. Stated differently. and this technology has been touted as a possible savior for broadband. Planning. This paper proposes a methodology for supplying broadband services using a wireless mesh network structure. Geographic information systems. In fact. In order to examine the potential of such networks to mitigate the “lastmile” problem in rural/suburban areas. A study exploring government options is suggested as future work. costs would have to decrease (through either technological advances or government subsidy/tax breaks) by 45% just to break even. Wireless telecommunications. Broadband service ISRL Categories: Chapter 5: Addressing Universal Broadband Service… 63 . Results are discouraging for almost all cases. and then each neighborhood is evaluated for fiscal promise under nine different cases of bandwidth offerings and quality of service. Keywords: Mesh networks.ADDRESSING UNIVERSAL-BROADBAND-SERVICE IMPLICATIONS WITH WIRELESS MESH NETWORKS ABSTRACT Demand is increasing for broadband telecommunications. customer monthly charges would have to be reduced by $62/month per customer over five years for wireless broadband to achieve profitable status. 270 “neighborhoods” are randomly generated.

p. a bipartisan group comprising more than 300 chief executive officers and senior partners of the major companies in the fields of information technology. The current (2003) presidential administration has made statements concerning the importance of broadband connectivity. 1) However. Senator Tom Daschle (D . 6). there may still be a segment of the U. (TechNet. An intermediate goal is the “availability of affordable broadband at speeds of at least 6 Mbps from 2 or more providers to at least 50% of U. Acampora (2002) elaborates by noting that for nine out of ten American businesses with more than 100 workers. p. Principle 5 reads as follows (pages 2-3): Investment incentives. TechNet developed 6 principles to address roadblocks and provide a guide to a national broadband policy. In an article in Scientific American. potentially including targeted tax incentives. 2002. will have a potential affect on the United States (US) economy by $500 billion. investment banking and law. The issue of broadband connectivity has attracted the attention of the highest levels of the US government. today merely 3% of that backbone is in use. Secondly. TechNet has called on the president and policymakers to make broadband a national priority and to set a goal of making an affordable 100-megabits per second (Mbps) broadband connection available to 100 million American homes and small businesses by 2010 (TechNet. 2002. 2002). Today. She states in an op-ed piece in the New York Times that although investors have plowed $90 billion into a cross-continental fiber-optic broadband network. … In a market-oriented environment that encourages the deployment of broadband networks. primarily by providing new wired connections to homes and offices. and positively benefit quality of living. the backbone is less than a mile away.21) points out that this potential will remain unrealized without government action. in turn. p. and a joint letter to President Bush from Senator Daschle and Senator Gephardt (D .” and the problem is entrepreneurs failed to foresee the enormous cost of upgrading the “last mile – copper telephone wires that connect individual homes and small businesses to the broadband backbone” (Kornbluh. 21). although they have not been as strong as some in congress would like. 2001. venture capital. This. She says it is the “digital equivalent of fallow farmland. population that Chapter 5: Addressing Universal Broadband Service… 64 . Kornbluh (2001. network providers will need to invest hundreds of billions of dollars to upgrade infrastructures and increase bandwidth capacity in the last mile. 2002). households and small businesses by 2004” (TechNet. TechNet says true broadband is the key to the next generations of communications and Internet services. 2002).ADDRESSING UNIVERSAL-BROADBAND-SERVICE IMPLICATIONS WITH WIRELESS MESH NETWORKS INTRODUCTION Broadband is the capacity to deliver Internet access with a continuous connection and the ability to both send and receive digital content or services at high speeds (TechNet.S. Organizations such as TechNet. Moreover. claim that the productivity of Americans in their homes and at work will increase with widespread adoption of broadband.S.MO) states that deploying the technology should be a “national imperative” (Dreazen. To reach this goal. should encourage broadband deployment to underserved communities and businesses. 2002).SD) has made a call to bring broadband to every American by the end of the decade. biotechnology. p. virtually no homes have connections with such bandwidth (TechNet.

or both. and free space optics (FSO). and numerous other factors. This paper proposes a mesh network planning methodology and examines mesh networks as a possible solution to the last-mile problem. such as Japan. a mere fifteen percent of US households use them (Woolley. and transmission ranges are examined.11 standards for wireless connectivity. distance involved. Multipoint Microwave Distribution Services (MMDS). cable. is wireless networks (Willebrand and Ghuman 2001). and installation. One popular alternative. local labor rates. the cost came to $396. after discussion of some background material. suggesting that an ambitious broadband strategy can help revitalize the economy. While relative costs obviously depend on the number of locations. Fiber cost was calculated on the basis of 1220 meters required to interconnect the buildings to the hub. p. Although other options exist. In comparison. At $325 per meter for trenching.500. This is primarily due to prohibitive cost of services. details of the planning methodology are explained.21) While broadband connectivity is becoming a technological necessity. it is clear that wireless connectivity provides an alternative worth exploring in many network settings. one case encountered by LightPointe Communications involving the tradeoff between fiber cable and laser-based wireless for connecting three urban buildings to a backhaul hub clearly illustrates the cost differential (Willebrand and Ghuman 2001).does not have broadband availability. these four provide a significant starting point for this research. they have not only caught up but are surpassing them (Woolley. 2001. 2002). Public policies should seek to narrow the current and future disparity in the level of high-speed access to the Internet. America continues to fall behind other developed countries in deployment of broadband services. The paper concludes with overarching thoughts on the practicality and likely success of the wireless mesh approach to solving the last-mile problem. low-income users. Similarly. Chapter 5: Addressing Universal Broadband Service… 65 . Kornbluh encourages both Congress and the president to subsidize sparsely populated regions of the country. the cost to interconnect wirelessly was a mere $59. 2002). As a result. to ensure that all Americans can enjoy the benefits of broadband. As an example. and while the United States has a fiber-optic backbone that is severely underutilized with members of both the government and industry calling for greater broadband coverage. BACKGROUND Technology Options Some current fixed wireless technologies are Local Multipoint Distribution Services (LMDS). (Kornbluh.000 for the transmitting and receiving equipment across the four locations. In particular. equipment costs. the IEEE 802. The effects of density. which has emerged for providing this “last mile” connectivity. quality of service. Countries in which products and services have generally been more expensive. Five years after the phone and cable companies began offering broadband services. are offering broadband services at almost half of what American citizens pay. while Japan previously lagged behind the US by a factor of twenty in broadband usage. and then results of applying the approach to approximately 2500 randomly generated neighborhoods are presented. Prompted by the high cost of fixed-media connections and the problems with obtaining permission to run cable trenches within urban areas. wireless technologies have received renewed attention.

000 more across the United States by the end of 2003. this figure is projected to grow to 103. Intel is also squarely behind the push.1). This is more commonly known as Wi-Fi (short for wireless fidelity). It also cannot transmit though obstructions. In fact." transmits data using low-powered infrared lasers. however. Intel will spend over $300 million touting this advance. Wi-Fi technology is becoming popular in an application related to. (Kessler. the laser must not be too highly powered.11b standard shows.11b is approximately 100 meters. The reported range of 802. as. such as 802. Furthermore. MMDS can transfer data on the unlicensed channels up to 27 Mbps and up to 1 Gbps on licensed channels. 2003. and McDonald's in 10 stores on a trial basis in Manhattan.11g.” Companies such as Starbucks and McDonald's are putting so-called “public access points” in their stores – Starbucks in 1200 stores across the country. and though it can transfer data up to 2 gigabits per second (Gbps). 802. But the availability of Wi-Fi on chips and across the US and UK will only help promote usage of the technology and drive down prices and improve the electronics as time goes on. and Cometa Networks (financed by IBM and AT&T. Cell phone maker Ericsson is building 5.11a. the 802. its biggest marketing blitz ever. This technology’s key feature is that it uses multiple channels simultaneously and. There are currently 5. Free Space Optics (FSO) Free space optics (FSO). FSO do not currently require FCC licensing. “big companies are starting to pour big bucks into Wi-Fi technology – marking a turning point for the young technology. certain power restrictions must be observed. Other IEEE 802 wireless standards are also available. that considered in this research. 802. The Wi-Fi was originally intended for local area networks (LAN).900 public access points in the United States. among others) plans to have wireless networks installed in 50 US cities by year's end. The maximum range of LMDS is approximately five kilometers. LMDS behaves more reliably when transferring data in the Mbps range.000 wireless hot spots in the United Kingdom (UK). p. the maximum range of FSO is a few kilometers. MMDS uses both unlicensed and licensed channels. also known as "open-air photonics.800 in 2006. Wi-Fi networks are being implemented with distances of up to 15 kilometers (Carstensen and Morgan. by aggregation creates large pathways between the sender and receiver. MMDS Multipoint Microwave Distribution Services are also called Multi-channel Multipoint Distribution Systems and wireless cable. According to USA Today (Kessler. and others. seemingly insufficient for larger scale connectivity. it plans to build the Wi-Fi capability right into its chips." "optical wireless" or "infrared broadband. fog can Chapter 5: Addressing Universal Broadband Service… 66 .16 (Wireless MAN). but distinct from. Consequently. therefore. 2003) The use of Wi-Fi in coffee shops and McDonald's does not solve the last-mile problem. for example. Wi-Fi The variations of the Institute of Electrical and Electronics Engineers (IEEE) wireless standards and their usage are rapidly increasing. and to remain in the unlicensed territory.LMDS Local Multipoint Distribution Services operates in the 28 GHz band and cannot be transmitted through obstructions (such as buildings and mountains). Toshiba plans to install 10. 2002). but it is possible to connect points over greater distances with the same technology. but is now also used in metropolitan area networks (MAN). FSO is a line-ofsite (LOS) technology.

Such connections are often found in metropolitan areas when two offices for the same company are in separate buildings. point to multipoint. Some applications of PtP connections use the Free Space Optics technology.. and it is desirable to have them part of the same network. or near-line of-sight complications. trees) do not. security becomes a relative non-issue. whereas mobile wireless (cellular) does not need this imposition. but lesser obstructions (e.g. where some obstructions (buildings. where obstructions cause loss of signal. This methodology is typically implemented when there are two buildings that need to be connected as part of a LAN. moreover. and mesh (also known as multipoint to multipoint) – the subject matter of this paper. It is also worth noting that since FSO's laser beam is very high frequency and is very directed. The data transfer rate of FSO is between 155 and 622 Megabits per second (Mbps). to maintain expected quality of service (QoS) for broadband. e. PMP often must deal with either line-of-sight (LOS) issues. and everything falling under that blanket part of the network.g. or that broadcast from multiple points to multiple homes and businesses. people are very forgiving of interference (including minor static. This type of network may be envisioned with a single tower blanketing an area with a signal. This is in distinction from technologies that broadcast from a single point to multiple homes or businesses.” coverage is provided by one tower. It is not possible to “eavesdrop” on a FSO signal without blocking the beam and thereby notifying the sender and receiver. Several solutions have been introduced to address this particular problem. planning for PMP networks is significantly different from that of cell Chapter 5: Addressing Universal Broadband Service… 67 . namely the sway of the buildings involved. Second. Broadband wireless typically operates in higher frequencies and thus requires its receivers to be line-of-sight with the transmitter. this introduces another complication. The first deals with line of sight. Since the technology of FSO is a fairly narrow laser beam. This is primarily because audio communication occurs at such low data rates. The advantage of this methodology is its simplicity for connecting two locations.) cause failure. lost syllables or words) that mobile networks can operate satisfactorily where broadband networks could not. Point to Point (PtP) Point-to-point is the oldest of the three configurations and consists of transmission from one location to a second point. Cellular telephone does not have this stipulation either. Point to Multipoint (PMP) PMP networks have a single source connected to a backbone network that propagates signal to multiple receivers. point to point.corrupt the transmission as water particles act as prisms to the laser and dissipate the light beam. The disadvantages are potential security issues and possible weather disruptions. PMP networks operate somewhat similarly to those of cellular telephones in the sense that when one is within a “cell. A very common application area is campus (school) environments. Instead of laying wire or fiber optics between the edifices. Therefore. there are two major differences between cell telephone provision and wireless PMP broadband coverage. there must not be any interference with other signals. and the cost of the wireless equipment is often substantially less than that of a traditional wired network. However. Network Configuration Options Wireless technologies have been deployed in three basic network configurations. it is often more straightforward to place two small directed antennas on the tops of the buildings.

in general. 2002). increasing the number of customers can decrease the cost of the mesh network. with mesh networks. 2002). a Cambridge. With these non-mesh systems.1 shows a neighborhood for which a mesh network has been provided. The backhaul point (BH in Figure 5. VoiceStream. The incremental addition of customers to the network is often inexpensive. cheaper. purchase/lease of land. the need for and interest in wireless connectivity is increasing in the global market. however. The backhaul is generally “wired” (say with fiber optics) to trunk connection points. (Hamblen. more than 19 million mobile and remote workers will regularly use public access Wi-Fi networks (Brewin. and predicts that by 2006. based company announced plans to implement a mesh wireless network in the South Wales region. Former Soviet President Mikhail Gorbachev appealed to American companies and investors to bring information technology outsourcing to Russia to strengthen their wireless infrastructure.. … M7 in Chapter 5: Addressing Universal Broadband Service… 68 . the most recent of the network configurations and also the focus of this research. 2003). obviate many of these issues by using smaller. A small municipality in western England has taken upon itself to provide broadband wireless access to its citizens by purchasing MeshBoxes and strategically placing them throughout the area. Nokia announced that more than 50 customer were using their Nokia RoofTop™ Wireless Routing mesh network solution. Rakes. the transmission medium changes to wireless at these trunk connection points. several drawbacks of PtP and PMP are immediately overcome. it is also important to have already generated a critical mass of customers that will pay for the service. They also provide businesses with up to 3 Mbps access speeds (Wery. Kar. In fact. in Stamford.1) is that location along the backbone network where the mesh network will connect. Citizens need only to have Wi-Fi cards for their PCs to be part of the mesh (Batista. Figure 5. 2001). as long as a customer is within range and view of another customer that is within range and view of yet another customer or piece of equipment that is eventually connected to the backbone. Connecticut. transmitted from the trunk connections directly to customers. bringing electric power out to a (perhaps) remote site. The tower-based networks require permission. Mesh networks. In February 2002. As to future plans. shown as T1 and T2 in Figure 5. then the signal may be received. labeled M1. Finally. companies such as Mesh Networks “envision blanketing cities and highways with a peer network to provide continuous 802. because of the high cost of the tower. multiple transmitters. with the only additional cost being that of the customer premise equipment. England. and Rees (2003). The signal is sent to mesh insertion points (MIPs. Vista Broadband Networks has adopted Nokia’s technology and has deployed equipment providing residential customers up to 512 Kbps speeds. and Woodrum.1. Nokia claimed this was the first commercially successful deployment of a wireless mesh system (Nokia. etc. planning to offer services equivalent to asymmetric digital subscriber line (ADSL) networks (Willis et al. 2002). 2002). Radiant Networks. as will be seen later. cites figures from Gartner Inc. 2002) Mesh Network Terminology In order to define terminology common to mesh implementations.phone towers and networks. The wireless signal is not. M2. Mesh Since mesh (or multipoint to multipoint) networks do not have a tower or towers that must reach all customers. The reader interested in additional information about the planning of PMP networks is directed to Scheibe.11 presence” (Gillmor. Carstensen. For example. a company offering broadband services over public access Wi-Fi networks.

and finally to the backhaul. Next. Line-of-sight must be determined next. 2000).. that is. These numbers are an amalgamation of numbers we have heard informally or seen in various internal publications/white papers. it is only forwarded to a trunk connection point and then to the backhaul.) These equipment locations are primarily a matter of physical characteristics of a neighborhood. etc. and then back out to customers. With the methodology developed here. These mesh insertion points do transmit and receive directly to/from customers. Generally. one of the key features of mesh networks is that customers are used to forward signals on to other customers. and since cost data are confidential. Latency is the time it takes for a packet to travel from source to destination (Webopedia. all known customer locations are specified. Repeater nodes are generally identical to customer equipment – there is just not a customer at the repeater location. Each link in the chain back to the backhaul is called a hop. (We will call these neighborhoods. The first three steps are used to prepare (i. which locations are chosen is an economic decision to be determined by the model. Table 5. Since slightly different equipment and terminology are in use with different mesh providers. That is. sometimes it is necessary to insert repeater nodes into the network when there are not enough customers nearby to get messages to every customer. Figure 5. once signal is received at a MIP back from a customer. To ensure quality of service. either a geographic information system (GIS) or field visit is required. terrain is not obstructive. that is. and all possible mesh insertion points for a specified geographic region. there has been little discussion in the literature as to what the optimal number of links should be for a wireless mesh network. Although hop constraints are not new to network design (Balakrishnan and Altinkemer.e. whether residential or commercial. the probability of a network meeting a given traffic contract (Wikipedia. up to the capacity limits of the receiver and transmitters. In a mesh network signals from a customer seeking the backhaul may be forwarded from customer to customer (possibly including repeaters). trunk connection points.the figure) that are closer to customers. but all potential locations are to be specified. it is sometimes necessary to restrict the number of hops in each transmission path. or latency. etc. 1992). 2003). Thus a repeater may be thought of as a non-revenue generating customer.2. The situation to this point is indicated in Figure 5. The MIP then sends the signal to a trunk connection point. where each circle (node) indicates the (geographic) location of the equipment/customer indicated. It is unknown how many MIPs will be economic. until finally a mesh insertion point is reached. (Lack of a Chapter 5: Addressing Universal Broadband Service… 69 .1 only gives a rough profile of current mesh equipment costs. WIRELESS MESH PLANNING METHODOLOGY The methodology proposed in this research to plan wireless mesh networks is a four-step procedure. However. the locations of the single backhaul point and two trunk connection points are assumed. Step 1 – Potential Equipment Placement and Line-of-Sight Determination Preprocessing begins with the specification of the backhaul point. where a dotted line between any two circles indicates no obstruction between the two nodes. As mentioned. signals are not transmitted from customers to MIPS.3 can then be generated. where right-ofway/permissions are available. preprocess) the neighborhood or region being analyzed for solution. 2003). Whitehead suggests that somewhere from six to ten hops still offers an acceptable level of service (Whitehead. Too many hops in a chain of connectivity can cause inordinate delay.

Finally in Figure 5. and invoked the procedure above for nine different cases on each. and from each trunk connection point to the backhaul. Gavish. Step 3 – Network-Flow Model Formulation Once the feasible paths have been generated. and use the capacitated. logistics and production planning (Minoux. 1991. they move on to trunk connection points. either because transmission range limits are exceeded or because obstructions block all possible paths to the customer. The method of neighborhood generation is outlined below. customer C3 does not have line of sight with customer C4. 1989). and transportation (Magnanti and Wong.g. We have written an Excel (preprocessing) program that does this.4. arrows are shown on the links from each MIP to its closest trunk connection point. Step 4 – Solution Generation The results from our Excel preprocessing program leads to data entered in Excel cells that can be run directly in an integer programming package within Excel. network-flow model. Magnanti. two hundred seventy such neighborhoods were produced.6-acre plots. and highdensity regions (H) were created on ¼-acre lots.) and thereby portray Chapter 5: Addressing Universal Broadband Service… 70 . fixed-charge.dotted line indicates no feasible transmission. ninety at each density. (see Appendix I). Low-density neighborhoods (L) were constructed on 1. To generate these paths efficiently. Random Problem Generation We randomly placed customer homes in a square grid to achieve differently configured neighborhoods with three different rural/suburban densities.. R1 in Figure 5. fixed-charge network flow formulation described in Chapter 4. et al. and Wong.g. we randomly generated 270 such neighborhoods. Three levels of visibility were specified in order to introduce the effect of obstructions (i. EXPLORATORY STUDY To examine the potential of wireless mesh technology to reach previously unreached (and perhaps unreachable) customers in relatively low-density areas – the case of most economic uncertainty. potential repeater nodes (e.) At this point it may occur that a customer desirous of service cannot obtain a signal. discussion of these is beyond the scope of this paper. e. The fixed-charge network flow problem has been used in many applications including telecommunications (Balakrishnan. all possible paths from each customer to the backhaul may be enumerated. In such an eventuality.. Interested readers may consult Scheibe. it is helpful to define artificial directional nodes. Step 2 – Feasible Path Enumeration Once the information from step 1 is in hand. terrain. given a limit on number of hops.. medium-density areas were spawned with 1-acre lots.3) are defined at non-customer locations that will ensure an unobstructed path to the backhaul for each customer. Recall that signals received at MIPs from customers do not go back out to other customers. etc. 2002).e. buildings. they may be cast into the form of a mathematical programming problem. We used Frontline’s Premium Solver as this package (Frontline. 1984). 1991). (2003). Four possible paths to the backhaul for customer C1 may be traced in Figure 5. Shulman.3. namely the capacitated. The point was to study suburban and more rural areas as candidates for profitable wireless mesh operation.

Table 5. The data below are hindered in several cases by an inability to generate feasible neighborhoods or solutions. the use of Wi-Fi technology at current costs is economically infeasible. Each section below discusses in a similar manner one of the factors explored. Each of the 270 neighborhoods was then evaluated on two different factors at three levels. factor levels be changed to include only feasible alternatives. and nine. We then show data indicating the decline in costs necessary to “prove in” this technology for rural/suburban areas at the densities examined in the study. The details from the study are shown below for each factor. whereas a (medium) visibility level suggests that three fifths of homes have such line of sight. and 10 Mbps rates. The high level of visibility was set at 80%. A level of 40% visibility (low) infers that only two-fifths of the possible connections among homes within the broadcast range of the transmitter are viable. That is to say. bandwidth and number of hops were each evaluated at low. Appendix II discusses several of these anomalies. The levels examined of number of hops were five. This means that. Such drops in costs could be achieved either through technological advances or government incentives/subsidies. medium.2 shows each factor and its corresponding level. The particular cost and other assumptions made in the study are listed in Table 5. 5.1. and high levels. Chapter 5: Addressing Universal Broadband Service… 71 . thereby helping ensure a level of quality.realistically line-of-sight issues among customers. Plots are generated showing the monthly revenue required per customer to achieve breakeven under the conditions given. Connections were randomly assigned to the visible or obstructed categories according to their likelihoods. Development of these scenarios is beyond the scope of the present paper. Each factor reflected different aspects of quality of service (QoS). resulting in nine different cases. Bandwidth was studied at 2. When this occurred it sometimes became difficult to make comparisons across factors. no path from a customer proceeding to the backhaul is allowed to exceed seven hops.430 computer runs are economically disappointing. in particular. at the medium hop level. seven. This suggests that for the journal article to be generated from these results. this keeps latency in check. MANAGERIAL IMPLICATIONS The results from the 2. for example.

000. This is because with many buildings and/or terrain to obstruct signal paths (i. breakeven revenue is $174/month at a low density and $165/month at a high density.000.Effect of Customer Density Effect of Customer Density $265. Managers will be interested to note the magnitude of this change: for the low visibility case. low visibility).00 $255. Effect of Customer Density It can be seen from Figure 5.000.5.e.00 Total Cost $260. Visibility (The low visibility / low density point was excluded from this graph because no feasible solution was found at those factor levels.00 $250.00 Low Medium Density High Figure 5.) Not surprisingly. none of the three cases is financially feasible.00 Low Medium Density High Low Visibiltiy Med Visibility High Visibility Figure 5.000.00 $255.000..00 Total Cost $260. If a five-year payback period is specified.000. the cost is increased by 25% in going from a high customer density to a medium one. the steeper the descent from high cost across the density levels.000.000. Chapter 5: Addressing Universal Broadband Service… 72 . the lower the visibility. this necessitates more hops and often more repeaters.00 $245.000.5 that as the density of an area increases the cost of bringing a broadband wireless mesh network decreases.000. Although the drop visually appears significant. Both are beyond the ordinary reach of residential customers.00 $250. Effect of Obstructions Visibility $275. the harder it is at even medium customer densities to get from the customer’s home to the backhaul.00 $270.000.6.00 $245.00 $240.00 $240.000.00 $265.000.000.

when a five-hop maximum is enforced (thereby maintaining fairly low transmission delay).000. This indicates a hefty charge for the fastest rate of service – one that is prohibitive for almost all residential customers.00 $400.000.000.00 Total Cost Low Hops $258.000.000. This is a significant cost to pay for quality. Bandwidth Offerings Figure 5. note that there is very little effect as customer density changes for any bandwidth.00 $200.00 $253.000.8. and.000.8 illustrates the changes in cost over the three density levels for each level of bandwidth.00 Low Medium Density High Med Hops High Hops Figure 5. The 5-Mbps rates are perhaps affordable for business customers. Quality of Service (latency) The message of Figure 5.000.00 $250.00 $100.00 $350. the cost is increased by $18.00 $243. Finally. it is not clear that managers will enforce such restrictions once they become aware of these cost penalties.000.00 $268.Effect of Quality of Service: Hops Quality of Service (latency) $273.00 $150. but not for residential.000. $73/month for the 2 Mbps setting.00 Total Cost $300. For example. $152/month for the 5 Mbps scenario. Chapter 5: Addressing Universal Broadband Service… 73 . there is a large price to pay in reducing latency.7.000. at low densities.00 $Low Medium High Low BW Med BW High BW Density Figure 5.000. and the 2-Mbps is only slightly more than what many residential customers are currently paying for about 1/3 the throughput.000.000.000. monthly revenue charges for breakeven are $287/month for the 10 Mbps case. With a five-year payback.000.00 $248.00 $263.00 $50.000.000 (14%) over a more lax nine-hop maximum number of hops.7 is that. Effect of Quality of Service: Bandwidth Bandwidth Offerings $500.00 $450.

000. as the bandwidth increases.000. Chapter 5: Addressing Universal Broadband Service… 74 . While under current conditions this rate may be very reasonable. Costs and penetration could be assumed for the promising scenarios to see whether such expenditures are socially possible or desirable. density is one-acre lot sizes. consider a case of medium-level factors. That is to say.00 $Low Medium High Bandwidth Figure 5.000.Effect of Cost: Bandwidth Bandwidth Cost $450.00 Total Cost $250. This methodology was employed in a large computer study to assess wireless technology as a mechanism for solving the last-mile problem of broadband provision in rural/suburban areas.000.000.00 $50. If we assume customers are willing to pay $75/month for such service.00 $200.00 $100.00 $150.00 $400. If the desired time to recoup the investment were specified to be five years.9 that averaged over all other factors. then the total cost of equipment would need to be reduced to $112. QoS (latency) is restricted to seven hops. The total cost for providing mesh service to the average 25-customer neighborhood is $205.000. under average conditions.000.000.500 to break even – a 45% cost reduction. One way of viewing this result is to say that.000. The cost increase is almost a factor of five – the difference between a profitable service offering and an unaffordable one. costs would have to drop by 45% (or revenues be supplanted by $62/month per customer for five years) to become profitable. Bandwidth Cost It can be seen from Figure 5. we view the next area of future work to be an inquiry into various governmental options.9. The analysis also concludes that. There are those in government and industry who have called for the federal government to intervene through subsidy and/or tax breaks to make broadband service “universally” available.00 $300. The study finds that Wi-Fi service is not economically viable at current costs. Cost Reductions Necessary to Achieve Profitability To illustrate the need for cost reduction.000. and bandwidth is 5 Mbps. on average and under average/medium conditions. Now that a methodology is available for analysis of wireless mesh implementation. then each customer would be required to pay a monthly fee of approximately $137 for his or her service. the cost of bringing a broadband wireless mesh network also increases. Wi-Fi costs need to drop by 45% to achieve breakeven. CONCLUSIONS A planning methodology has been developed that configures wireless mesh networks in an optimal manner.00 $350. each of the four factors in this study takes on a medium level: visibility is 60%. we believe that very few end users would be willing to pay that much.

T. Topological Design of Telecommunications Networks – Local Access Design Methods. 2003. Models for Planning Capacity Expansion in Local Access Telecommunication Networks. A. December 10. Mesh Less Cost of Wireless. dateline: Washington. “Wireless Net Technology Taking Off. 2002. Computerworld. W. March 21..E.. (2003). “The Broadband Economy. page 1. and Wong. Retrieved March 17. Annals of Operations Research. Shulman A. NV.com/news/print/0..” Computerworld. Minoux. 33. 17-71. (2002). (2001). 18(1). and Morgan. Balakrishnan A. Incline Village. From the World Wide Web: http://www. (2003). Frontline’s Premium Solver™.S.html.com. 4. 239-284.57617.J. column 2. D. Network Design and Transportation Planning: Models and Algorithms. S.” The New York Times. 287(1). (1989). May 27.L. Annals of Operations Research. K. M. Last Mile by Laser. 19. Magnanti. Chapter 5: Addressing Universal Broadband Service… 75 . page 21. Networks. CA. A. R. ORSA Journal on Computing. Section A. Gillmor. K. Hamblen. Y. 2002. Inc. Frontline Systems. June 21.1294. and Wong R. 2002. Batista. InfoWorld. Using a hop constrained model to generate alternative communications netowrk design.C. from the World Wide Web: http://www. Private communication. (2002). (2002). 2001. 313-360. (1984). The Wall Street Journal. Altinkemer. (1991). M.solver. Solution Methods and Applications. 33. oursourcing dollars.wired. G. Transportation Science. (2002). L. (2002). Balakrishnan.T. T. Section A4.L. E. (2002). New services spur growth of public access Wi-Fi. Tech Firms Bemoan Bush Talk --. Riding the tiger. Section A. 2002.” USA Today. Kornbluh.00. March 21. (2002). editorial desk. Dreazen. 2003. Gavish B. M. Scientific American. Frontline.‘Broadband’ Policy is Viewed as Lacking Significant Details. “Russia needs U.. Kessler. Network Synthesis and Optimum Network Design Problems: Models.. Gorbachev says.192-205. Brewin.REFERENCES Acampora. (1992). dateline: San Francisco. (1991). Wired News. Magnanti T. 1-55. B. 49-53. March 12. Carstensen.

Willebrand. Carstensen. L. Wireless Mesh Networks. A MathematicalProgramming And Geographic-Information-System Framework For Wireless Broadband Deployment In Rural Areas. 4145. (2001). Retrieved February 10..org.com/TERM/l/latency. Press release.R. from the World Wide Web: Wery.A. Hasletad. Woolley.. Chapter 4 of dissertation. Holm (2001). Retrieved March 17. (2003). K. Ragsdale. B.S.W.B. (2003). from the World Wide Web: http://press. Retrieved July 30. Under review.. (2002). November 11. Nokia Demonstrates Wireless Mesh Broadband Market Leadership with More Than 50 Network Customers. 2002. from the World Wide Web: http://www. 2003. Webopedia. Rakes. and Woodrum J.nokia. A Capacitated. Network Flow Model for Solving Hop Constrained. and Rees. Technical report. R.Nokia.P. L. O. RAWCON. Forbes. FixedCharge.com/PR/200202/848470_5.P. T.org/wiki/Quality_of_Service. “When Wi-Fi Meets Mesh – The Combination Promises ‘Organic’-Like Network Growth and Low-Cost Equipment.. T.” Broadband Wireless Online.wikipedia. 3(6). K. T. Fiber Optics without Fiber. B. Chapter 5: Addressing Universal Broadband Service… 76 ... T. Whitehead. Journal of the IBTE. Willis. IEEE Spectrum.webopedia. 2003. A National Imperative: Universal Availability of Broadband by 2010. (8).. 2002. S.. 2002.shorecliffcommunications. Scheibe. Philip (2000). (2003).P.P. (2002).. (2002). from the World Wide Web: http://www. Wikipedia. 2003.com/magazine/volume. Mesh Networks: A new Architecture for Broadband Wireless Access Systems. Exploiting Peer-to-Peer Communications – Mesh Fixed and ODMA Mobile Radio. Retrieved July 30. L. & Ghuman. C. 2(2). and Rees.html. H. 48-53.asp?vol=29&story=290. Bottleneck Breakers. Rakes. Retrieved March 10. TechNet. http://www. Kar S.. Scheibe. from the World Wide Web: http://www. (2002).T.R.html. Friisø. (2003).technet.

R. and customer nodes respectively. network-flow problem (FCNFP) model as applied to the wireless mesh problem is formulated as follows. yi ∈ {0. Further designate T. Let N be the set of all nodes and Na be the set of artificial nodes in the network.j). no flow may traverse any arc departing that node. xij is the flow on arc (i. The objective is to minimize total cost. define the set of all arcs in the network by the letter A. Flow leaving either a customer or a repeater node (i. ∀(i. i*) ∈ A] ∩ [i* ∈ N a ]} ≤ ui yi xij ≥ 0. Rakes. if the transmitter is inoperative. M. Chapter 5: Addressing Universal Broadband Service… 77 . k )∈ A ∑x k∈N = d j . repeater nodes. j )∈A (A-1) (A-2) (A-3) (A-4) (A-5) (A-6) ∑x − jk k ∋ ( j . kj k ∋ ( k .e. which is assumed to be nonnegative. The model may be written as min ∑ f k y k . ∀i ∈ N Constraint (A-2) ensures flow conservation. moreover. For each node k. Let dj denote the supply or demand at node j. j ) ∑x ij j ∋ ( i . Finally. as follows. any artificial nodes have a cost of 0. ∀i ∈ (T ∪ M ) ∀i ∋ {[i ∈ (C ∪ R) ∩ (i ∉ N a )] ∩ [(i. mesh insertion points. ∀j ∈ N j ∋ ( i *.1}. let the cost of activating that node be fk. j ) ∈ A. Further details may be found in Scheibe. Constraint (A-3) restricts flow from trunk nodes and mesh insertion points.. and C as the sets of trunk connection nodes. the sum of flow over all arcs leaving such a node) cannot exceed the capacity of the transmitter at that node. Ragsdale. and yi is a binary variable equal to one when node i is in use and zero otherwise. Let ui > 0 represent the capacity on the arcs emanating from node i.APPENDIX I: THE WIRELESS MESH PROBLEM MODEL The fixed-charge. and Rees (2003). j )∈ A ij ∑x ≤ ui yi . whereas constraints (A-3) and (A-4) are flow capacity constraints. whereas (A-4) restricts flow emanating from customer or repeater nodes. all artificial nodes should be considered transshipment nodes (with a supply of 0).

while nodes may be close enough for the signal to reach. but the area is not dense. again it may not be feasible for a mesh network.000 generations. the range is conservatively held constant at 900 feet. If an area’s density is very low. It is very easy to generate a neighborhood where all the necessary criterion are met. when the visibility is low. but equally interesting condition. for the computer to generate twentythree neighborhoods with medium visibility and low density. it took another 225. that some neighborhood configurations are very rare. For example. low visibility and low density. then it just may not be feasible to create a mesh. This is especially true when the number of hops is set at the maximum level. there is something blocking the visibility. it became necessary to increase the density and visibility to a level where tests could be run. The technology used is IEEE 802. There are a few practical conclusions that may be drawn from the two corners just discussed. When this is the case. Similarly. however.000 generations just to create the twenty-three. Therefore. it was discovered. then it may be necessary to add a repeater node to bridge the gap and bring the customer node into the mesh. If the visibility is sufficient. it makes sense that some nodes in the network may be too far from others to be a part of the network. low visibility / low density and high visibility / high density / high hops. During the pre-testing of these factors. however. the algorithm used to generate every viable path may take a half of a minute to several minutes. When the density of the area in which a wireless mesh network may be desired is low.000. When the hop level was set at 5. There are many such neighborhoods in the United States.11b or Wi-Fi. To illustrate this issue consider a neighborhood with a visibility of 50%. If the visibility is too low. The computer created over 100. In its present state. Neighborhoods with high density and high visibility created an entirely different. This is exclusively due to the high level of redundancy within the mesh paths. and wireless service providers should be wary before attempting to penetrate such areas. a density of one-acre lots. When the parameters were low visibility and low density the computer could not generate a neighborhood that met the necessary criteria of each customer node being able to see at least one other node. certain “corners” of the design proved to be problematic for analysis – specifically.000 neighborhoods. low density and require a high QoS are very rare and prove difficult to serve. then it becomes virtually impossible to create a mesh network. one constant is the range of the wireless transmission. Now consider when the visibility is 80%. and the time to compute them exponentially increases. For this design. the density is ¼ acres. The range of Wi-Fi antennae can very from 300 feet to 23. then.APPENDIX II: PROBLEMATIC CORNERS In the study described in this chapter. then it may be possible to change Chapter 5: Addressing Universal Broadband Service… 78 . The implications are that neighborhoods that have medium visibility.000 feet. The first is when an area has a density and/or visibility that are very low it may be practically impossible to bring in a mesh network. it took approximately 214. When the two factors are brought together. and the hops are nine. The total number of viable paths may be in the 100 thousands to millions. that the redundancy of paths from a node to the backhaul made computation exceedingly difficult. It is worth noting. The visibility was set at 20% and the density was set at 5-acre lots. the computer would randomly generate neighborhoods based upon factor parameters. The number of viable paths for every node to the backhaul may range from 500 to more than 12. and not one neighborhood met the necessary criteria. and a hop level of seven. As initial tests were performed with these factors.

the costs change with different technologies. Another practical conclusion is when the density is high. and the hops are high. Chapter 5: Addressing Universal Broadband Service… 79 . potentially reducing network costs. and. Of course. the visibility is high. Another means of addressing the overkill issue is to reduce the number of hops from a node to the backhaul. While the mesh network will certainly work in neighborhoods with such characteristics. This may or may not be desirable as the number of viable paths decreases so does the network’s ability to simultaneously transmit multiple packets over different routes. thereby.the wireless technology to one that is capable of transmitting over greater distances. there is a level of “overkill.” It may be worth considering using a wireless technology that has a shorter signal distance.

1. Providing broadband service to a neighborhood with a wireless mesh network. Chapter 5: Addressing Universal Broadband Service… 80 .Figure 5.

BH T1 T1 C3 C3 T2 T2 M1 M1 M2 M2 C4 C4 C2 C2 R1 R1 C1 C1 Figure 5. Chapter 5: Addressing Universal Broadband Service… 81 . Nodes shown linked together with line-of-sight capabilities.3.2.BH T1 C3 T2 M1 M2 C4 C2 R1 C1 Key: BH – backhaul Ti – trunk i Mj – mesh insertion point j Rk – repeater k (not shown) Cl – customer l Figure 5. Nodes represent potential wireless mesh equipment locations.

given the lineof-sight configuration of Figure 5.4.3. Chapter 5: Addressing Universal Broadband Service… 82 .BH T1 T1 C3 C3 T2 T2 M1 M1 M2 M2 C4 C4 C2 C2 R1 R1 C1 C1 Figure 5. Viable paths for customer C1 to communicate with the backhaul.

Chapter 5: Addressing Universal Broadband Service… 83 .Table 5.2. “Typical” Wireless Assumptions Equipment Cost Assumptions Capacity Assumptions Trunk Connection Nodes Mesh Insertion Points Repeater Nodes Customer Premise Equipment Antenna Range $50.000 $10.6 Acre Lots 1 Acre Lots .000 $5.1.000 $5.000 200 Megabits/second 80Megabits/second 30 Megabits/second 30 Megabits/second 900 Feet Table 5. Factor Levels Visibility Low Medium High 40% 60% 80% Hops 5 7 9 Density 1.25 Acre Lots Bandwidth 2 Mbps 5 Mbps 10 Mbps Note: We performed 30 replications in each cell.

As such. FIXED WIRELESS TELECOMMUNICATION NETWORKS Under the recommendation of my dissertation chair and committee.CHAPTER 6 A GENERAL FRAMEWORK FOR PLANNING BROADBAND. wireless broadband networks. this future journal submission is more developed in some sections than others. Chapter 6: A General Framework… 84 . this dissertation chapter merely outlines and sketches future work to be done in developing a planning methodology for fixed. it is research in progress and a document in progress.

A GENERAL FRAMEWORK FOR PLANNING BROADBAND. Network Planning Chapter 6: A General Framework… 85 . In two previous papers. With this problem. FIXED WIRELESS TELECOMMUNICATION NETWORKS ABSTRACT Wireless broadband telecommunication networks have been touted as being the solution to the “last-mile” problem. Keywords: Broadband Wireless Telecommunications. and infrastructure. Geographic Information Systems. This paper presents a planning methodology that shows how and when to apply which technologies. The methodology varies in scope from metropolitan areas to developing countries. the authors demonstrate how to apply different wireless technologies in suburban and rural settings in a minimum-cost manner. given parameters such as customer density. terrain. A spatial decision support system is proposed as a delivery mechanism for the planning methodology. 97% of the United States’ broadband network is laying fallow because it is too expensive to go the remaining distance from the broadband backbone to American residences and businesses. Decision Support Systems.

The physical reason for the LOS requirement is the high frequency level of the signal. particularly with respect to assumptions made in our two previous papers dealing with broadband service via two different network topologies: point-to-multipoint networks (PMP). In this research we propose a spatial decision support system (SDSS) to encapsulate the methodology. i. the antennas must be visible to each other for a signal to pass successfully. Lasers provide a good example of this phenomenon. 2002) and fixed. A particular GIS tool developed locally (GETWEBS) takes topography into consideration as well as radio frequency characteristics (Carstensen. 2003). The next section provides background knowledge.e. wireless telecommunications refers to those network delivery mechanisms that depend on electromagnetic propagation to non-mobile customers. we use a GIS to aid in the determination of optimal tower placement in one case. as opposed to copper. In each of the methodologies mentioned above. The power of this GIS is that it allows the decision maker to evaluate alternatives in a straightforward. The subsequent section presents the Chapter 6: A General Framework… 86 .A GENERAL FRAMEWORK FOR PLANNING BROADBAND. et al. Each method shared a particular physical requirement due to the nature of the transmission and the physical medium for broadband wireless telecommunications. and mesh networks. and fiber-optic connectivity. (TechNet. LOS means that the path between transmitting and receiving antennas cannot be obstructed.. Geographic information systems (GIS) lend themselves well to this class of problems. and equipment visibility in the other. a second shared characteristic of the two fixed wireless methodologies is that each antenna is fixed in a specific location. so does the necessity of LOS. always “on” connections. FIXED WIRELESS TELECOMMUNICATION NETWORKS INTRODUCTION Broadband communication refers to high speed. whereas they have very little effect on FM radio. the authors demonstrated two methodologies for planning broadband. quick manner A second necessary component for a fixed wireless broadband planning methodology is a mechanism to deal with the fairly sophisticated mathematics (mixed integer programming solution) necessary for minimum-cost solution of the problem. 2002) and is recommended in this planning methodology. therefore. SDSS contain three major components: a spatial database (which can be manipulated by GETWEBS in our case). SDSS seem to fit naturally with the planning methodology we are developing as the purpose of this research. et al. Scheibe. Morgan. it is incumbent on a fixed wireless planning methodology to consider directly the terrain over which the technology is to be applied. it is generally correct to say that as frequency increases. in fact.. antennas do not move as do mobile telephones. coaxial. Note that. that is to say. which is lower in the frequency spectrum than laser. and a GUI dialog system. 2003.. by definition. In this paper we advance an SDSS for planning fixed wireless broadband networks. even fog and mist can disrupt a laser signal. and Bostian. collection of requisite mathematical models (the mixed integer programming models in our system). fixed wireless telecommunication networks (Scheibe. These two methods of point to multipoint and mesh were presented as means of bringing broadband communication to homes and small office / home offices (SOHO). Physically speaking. a model base. In previous work. namely a line-of-sight (LOS) stipulation.

and the implications then discussed. BACKGROUND SDSS (This section will provide a very brief overview of spatial decision support systems. a decision model section.actual methodology. Finally. as is understood from our previous work. and limitations. independent variables in charts. their design. Chapter 6: A General Framework… 87 .) Network Topologies (Current major topologies – such as PMP and mesh – will be outlined. whereby customer demand is developed. in which minimum-cost solutions are computed for a given technology/topology combination. and limitations such as transmission ranges and capacities. Emphasis will be placed on limitations and conditions of their applicability. conclusions are drawn and future work discussed. and helper models. etc.) Technology Issues (This section will discuss issues such as the major technologies. the equipment they demand. as well) database will include: For the geographic area under consideration: • • • • • • Developed/undeveloped Infrastructure Terrain visibility Customer density Customer clustering Disaster recovery an issue? For each viable technology: • • Transmission range (for various technologies) Transmission capacities The above issues will be incorporated into the methodology as root nodes in decision trees. Two (large-scale) examples are presented next. Model Base This portion of the SDSS will contain three major model sections: a marketing component.) METHODOLOGY Database/GIS Factors and items stored in the spatial (and non-spatial.

when planning a network. and Rees. The foremost question should be. socio-political conditions of the area. (The PMP solution requires this – see Scheibe. Rakes. etc. once those issues have been determined by the system. including telephone. individuals in a developing area/country that has no network infrastructure whatsoever may demand basic services such as telephone before requesting high speed Internet. and cable. but also which service is demanded. Therefore. • Set covering problem. some of these models would manipulate data and graphs from one format to another. location of available power. Internet. There are several sub-factors. For example. and Rees. While voice may be the only current requirement. 2003. location and condition of nearest existing infrastructure. provides support to the other SDSS components. Formats most appropriate to the task at hand are • Decision trees • Rules • Charts • Tables Possible Synthesis/Flow of Above Factors in the Planning Methodology To adequately address holistic planning of wireless broadband networks. Ragsdale. fixed-charge.) Helper models This class of models. Broadband communications encompass a variety of services. Also providing support are • Forecasting models • Tax data/models. including topography. Rakes. network-flow problem. For example. it is important to know not only the level of demand for a service. it is crucial to view the big picture. weather characteristics. what is the desired service. it would be inappropriate to fail to plan for cable or Internet connectivity as well. 2003. GUI Base The GUI base will contain presentation formats for the different aspects of the methodology.) • Capacitated. Carstensen. and is the area a new development or a disaster recovery? Chapter 6: A General Framework… 88 . However. population and/or household density of the area. in general.Marketing models • • Propensity to pay Demand models Decision models These models provide the mathematical wherewithal to solve a particular topology/technology implementation at minimum cost. future demand as well as current demand should be considered. The next factor which should be considered in this planning phase is the characteristics of the area for the desired service. is it a new network or augmentation of an existing network. (The mesh solution requires this – see Scheibe.

After the characteristics of the area under investigation have been determined. Moreover. For example. point-to-multipoint. lacks structure. Based upon customer density and clustering. Rule 1: if the area is undeveloped. the next step is to recommend a technology/topology pair. Three examples of rules are shown below. etc. regions of topology applicability can probably be derived. Given the multitude of factors that affect the planning of a wireless network and how the planning process. if the area were often very foggy. then higher frequencies such as those used with free space optics would not be appropriate. Possible Products/Outcomes/Outputs of the Methodology Charts/Graphs As an example of this potential SDSS output. A graph could be generated for any cost profile assumed for the given technology. or if the area had much rain. and then to analyze them within the SDSS. The characteristics of the area will play a large role in determining which methodology should be used. say customer density on one axis and degree/type of clustering on another. rules. and decision trees. and there is no existing infrastructure close by. then microwave frequencies would not be appropriate. the next set of facts to be examined is the wireless frequencies and their inherent capabilities and weaknesses. for instance. Decision Trees A third possible beneficial SDSS output is a decision tree. develop a knowledge base that integrates with the rest of the system). Rules Another approach to reducing the number of planning choices to only the necessary few is to employ rules in a top-down fashion. and a disaster has occurred high bandwidth PtP should be used to bridge network gaps the weather for area is foggy use lower frequency communications This scenario suggests that it may be propitious to embed intelligence in the SDSS in the form of an expert system (i. a decision support system would certainly be advantageous. This research suggests using three decision aids to narrow the possible choices for a broadband wireless network – charts/graphs. This should be done in conjunction with the potential wireless topology such as point-to-point. they will also determine which frequency may best suit the area. how the search space has been Chapter 6: A General Framework… 89 . Once the facts have been collected. and perhaps a necessary tool for a planner. These regions may be appropriately displayed in charts or graphs plotted against the key factors. consider which topology to utilize in planning wireless networks..e. then high bandwidth PtP is necessary from the backhaul Rule 2: if then Rule 3: if then the area is developed. Decision trees are visually informative. as one prunes off branches one can see. in general. or mesh.

it is believed that a way of thinking methodically about broadband such as has been suggested here would greatly benefit decision makers at higher levels. EXAMPLES (Two detailed examples will be developed and analyzed. Chapter 6: A General Framework… 90 .) CONCLUSIONS AND FUTURE WORK While the research in this paper is yet untested.reduced. The current plan is for these examples to possess widely different characteristics.) MANAGERIAL IMPLICATIONS (Implications will be drawn from the examples as to the utility and limitations of the fixed wireless broadband planning methodology. Decision trees also help enforce organization and complete consideration of all factors to be considered in planning.

(2002). E. 6(5). NY: John Wiley & Sons. Computer Networks and Internets with Internet Applications. Brewin.wired. Decision Support Systems. Building Decision Support Systems. Comer. Annals of Operations Research. Batista. Alter. Inc.. Shulman A. 239-284. NJ: Prentice Hall.. Bennet. 2003. March 21. (2000).E. (1992). Computers and Operations Research. Houghton Mifflin Company. G. Bostian.. Carstensen. L. Solving to optimality the uncapacitated fixed-charge network flow problem.L.com/news/print/0. A. (1998).. Reading.. New York. (2002). and Morgan. Crossland.REFERENCES Acampora. (1991). Fourth Edition.D. W. A. A broadband wireless access network based on mesh-connected free-spaced optical links. 33.T. C. (2001). “Mesh Less Cost of Wireless. Upper Saddle River.R. (1995). 2002. (Ed. ORSA Journal on Computing. Balakrishnan. Altinkemer. MA: Addison-Wesley. and Krishnamurthy. Dennis A. (2001). W. 799-810. 67-81. and Mateus G. and Wong R. B.B. D. (1999). (1980). 49-53.E. Balakrishnan A. Spatial Decision Support Systems: An Overview of Technology and a Test of Efficacy.) (1983). Reading.. K. Systems Analysis & Design: An ObjectOriented Approach with UML. H. S. Models for Planning Capacity Expansion in Local Access Telecommunication Networks.” Computerworld. J.C. Using a hop constrained model to generate alternative communications network design. 192-205. Third ed. Wynne. References 91 . Magnanti T. Acampora. Wixom.57617. Retrieved March 17. Overview. MA: Addison-Wesley. 25(1). Last Mile by Laser.00... Smith J. 1-14.M.W. B... Scientific American. D.L. S.E.. Combining Electromagnetic Propagation: Geographic Information Systems.1294. from the World Wide Web: http://www. Decision Support Systems: Current Practice and Continuing Challenges. Proceedings of the International Conference on Electromagnetics in Advanced Applications. (2003). IEEE Personal Communications. and Financial Modeling in a Software Package for Broadband Wireless Wide Area Network Design.” Wired News.. 287(1). M. (2002). 62-65 The American Heritage® Dictionary of the English Language.R. Cruz F. B. (3).html. “New services spur growth of public access Wi-Fi. 219-235.L. & Tegarden. & Perkins. A. 4.

Dreazen, Y.J. (2002). “Tech Firms Bemoan Bush Talk --- ‘Broadband’ Policy is Viewed as Lacking Significant Details,” The Wall Street Journal, June 21, 2002, Section A4. ESRI (1999). Getting to Know Arcview GIS: The Geographic Information System (GIS) for Everyone. Third ed. New York: Environmental Systems Research Institute, Inc. Fikes, R.E., Hart, P.E., and Nilsson, N.J. (1972). STRIPS: learning and executing generalized robot plans. Artificial Intelligence, 3, 1-3, 251-288. Fowler, T. (2001). Mesh Networks for Broadband Access. IEE Review, 47 (1), 17-22. Frontline. (2002). Frontline’s Premium Solver™, Frontline Systems, Inc., Incline Village, NV. From the World Wide Web: http://www.solver.com. Gavish B. (1991). Topological Design of Telecommunications Networks – Local Access Design Methods. Annals of Operations Research. 33, 17-71. Gillmor, S., (2002). Riding the tiger. InfoWorld, May 27, 2002. Girish, M.K., Zhou B., and Hu, J. (2000). Formulation of the Traffic Engineering Problems in MPLS Based IP Networks. IEEE. 214-219. Gouveia, L. and Requejo C. (2001). A new Langrangean relaxation approach for the hopconstrained minimum spanning tree problem. European Journal of Operational Research. 132, 539-552. Hamblen, M. (2002). “Russia needs U.S. oursourcing dollars, Gorbachev says,” Computerworld, March 21, 2002. Hamacher, H.W., & Kufer, K.-H. (2002). Inverse Radiation Therapy Planning -- a Multiple Objective Optimization Approach. Discrete Applied Mathematics, (1-2), 145-161. Horng, J.-H., & Tienyi Li, J. (2002). Vehicle Path Planning by Using Adaptive Constrained Distance Transformation. Pattern Recognition, (6), 1327-1337. Keen, P.G.W., & Scott Morton, M.S. (1978). Decision Support Systems: An Organizational Perspective. Reading, MA: Addison-Wesley. Keenan, P.B. (1998). Spatial Decision Support Systems for Vehicle Routing. Decision Support Systems, (1), 65-71. Kessler, M. (2003). “Wireless Net Technology Taking Off,” USA Today, March 12, 2003, Section A, page 1, dateline: San Francisco, CA. Kim, D. and Pardalos P. (1999). A solution approach to the fixed charge network flow problem using a dynamic slope scaling procedure. Operations Research Letters. 24, 195-203.

References

92

Kornbluh, K. (2001). “The Broadband Economy,” The New York Times, December 10, 2001, Section A, page 21, column 2, editorial desk, dateline: Washington, D.C. Lewis, H.F., & Slotnick, S.A. (2002). Multi-Period Job Selection: Planning Work Loads to Maximize Profit. Computers & Operations Research, (8), 1081-1098. Macklin, Ben (2001). The Broadband Revolution: You Say You Want a Definition, eMarketer, http://www.emarketer.com/analysis/broadband/20010306_bband.html. Magnanti, T.L. and Wong, R.T. (1984). Network Design and Transportation Planning: Models and Algorithms. Transportation Science. 18(1), 1-55. Maniezzo, V., Mendes, I., & Paruccini, M. (1998). Decision Support for Siting Problems. Decision Support Systems, (3), 273-284. Minoux, M. (1989). Network Synthesis and Optimum Network Design Problems: Models, Solution Methods and Applications. Networks. 19, 313-360. Nehme, C.C., & Simões (1999) M. Spatial Decision Support System for Land Assessment. In ACM GIS. Kansas City, MO: ACM Press. Nokia, (2002). Nokia Demonstrates Wireless Mesh Broadband Market Leadership with More Than 50 Network Customers. Press release. Retrieved July 30, 2002, from the World Wide Web: http://press.nokia.com/PR/200202/848470_5.html. Pirkul, H. and Soni, S. (2002). New formulations and solution procedures for the hop constrained network design problem. European Journal of Operational Research, In Press. PSINet. (1998). PSINet Begins Operating North American Transcontinental Fiber-Optic Network. PSINet Press Release. http://www.psi.net/news/pr/98/aug1198.html Quest. (1998). Qwest Awarded $20 Million Contract to Provide High Speed Internet Access to Centurion Telecommunications. Quest Press Release. http://www.qwest.com/about/media/pressroom/1,1720,121_archive,00.html Radiant Networks Plc. (2001). http://www.radiantnetworks.com. Ragsdale, C.T., Spreadsheet Modeling and Decision Analysis: A Practical Introduction to Management Science, Third Edition, Southwestern College Publishing, Cincinnati, 2001. Rolston, D. (1988). Principles of Artificial Intelligence and Expert Systems Development. New York, NY: McGraw-Hill Book Company. Scheibe, K.P., Carstensen, L.W., Rakes, T.R., and Rees, L.P. (2003). A MathematicalProgramming And Geographic-Information-System Framework For Wireless Broadband Deployment In Rural Areas. Under review.

References

93

Scheibe, K.P., Ragsdale. C.T., Rakes, T.R., and Rees, L.P. (2003). A Capacitated, FixedCharge, Network Flow Model for Solving Hop Constrained, Wireless Mesh Networks. Chapter 4 of dissertation. Shrick, B., & Riezenman, M.J. (2002). Wireless broadband in a box. IEEE Spectrum, 39(6), 38 – 43. Simon, H. (1960). The new science of management decision. New York: Harper & Row. Soni, S. (2001). Hop Constrained Network Design Problem with Partial Survivability. Annals of Operations Research. 106, 181-198. Sprague, J.R.H., & Carlson, E.D. (1982). Building Effective Decision Support Systems. Englewood Cliffs, NJ: Prentice-Hall, Inc. Taha, H.A. (1975). Integer Programming - Theory, Applications, and Computations. New York: Academic Press. Tarantilis, C.D., & Kiranoudis, C.T. (2002). Using a Spatial Decision Support System for Solving the Vehicle Routing Problem. Information & Management, (5), 359-375. TechNet (2002). A National Imperative: Universal Availability of Broadband by 2010. Technical report. Retrieved July 30, 2002, from the World Wide Web: http://www.technet.org. Turban, E., & Aronson, J.E. (2001). Decision Support Systems and Intelligent Systems. New Jersey: Prentice-Hall. 867. USGS (2002). USGS Digital Elevation Model Data. United States Geological Survey. http://edcwww.cr.usgs.gov/glis/hyper/guide/usgs_dem Vlachopoulou, M., Silleos, G., & Manthou, V. (2001). Geographic Information Systems in Warehouse Site Selection Decisions. International Journal of Production Economics, (1-3), 205-212. Virginia Tech Center for Wirless Telecommunications. (2002). http://www.cwt.vt.edu/. Wagner, Jim. (2002). Teaming Up to Beat Cable. internetnews.com. http://www.internetnews.com/isp-news/article/0,,8_1011151,00.html. Webopedia (2003). Retrieved March 10, 2003, from the World Wide Web: http://www.webopedia.com/TERM/l/latency.html. Wen, U.-P., Wu, T.-L., & Shyur, C.-C. (2002). Bi-Directional Self-Healing Ring Network Planning. Computers & Operations Research, (12), 1719-1737.

References

94

.S. Willis. Exploiting Peer-to-Peer Communications – Mesh Fixed and ODMA Mobile Radio..A. (2001). T. Bottleneck Breakers. B.” Broadband Wireless Online. 2003. Whitehead.. T.J. (8). Willebrand. Friis∅.. Mesh Networks: A new Architecture for Broadband Wireless Access Systems. & Hess. 3(6). (2002).org/wiki/Quality_of_Service. “When Wi-Fi Meets Mesh – The Combination Promises ‘Organic’-Like Network Growth and Low-Cost Equipment. from the World Wide Web: http://www. H.Wery. IEEE Spectrum. 2003. Hasletad. Fiber Optics without Fiber. November 11. Forbes. Decision Support Systems. J. West.. 4145. Holm (2001). T. Wikipedia (2003). 2002. Kar S.asp?vol=29&story=290. from the World Wide Web: http://www. 2(2). Woolley. 48-53. O. and Woodrum J. Journal of the IBTE. 247-264. & Ghuman.shorecliffcommunications.B. Retrieved February 10.com/magazine/volume. Lawrence A. (2002). R. B. Metadata as a Knowledge Management Tool: Supporting Intelligent Agent and End User Access to Spatial Data.. Retrieved March 17. References 95 . RAWCON. Philip (2000). S.wikipedia. (3). (2002).

Liu. 2001. May 2003 Dissertation Title: A Spatial Decision Support System for Planning of Broadband.. in Computer Science.P. pp. Carstensen. LoPinto. Charlotte. San Diego. at Pamplin College of Business.B.. “A Coverage Model for Topology Design in Wireless Mesh Networks. Virginia Polytechnic Institute and State University. Scheibe. and Rees. S. SC.P. October 4-5. Vita 96 . F.A. pp. 1991 REFEREED CONFERENCE PROCEEDINGS: 1. T. (2003). 2002. February 21-23.P. Dong. Biola University.” Thirty-First Annual Meeting of Southeast Decision Sciences Institute. Rakes.. 2... Scheibe.” Proceedings of the Thirty-First Annual Meeting of Southeast Decision Sciences Institute. K. K.” Proceedings of the Decision Sciences Institute. “Non-Standard Methods for Simulation Optimization Including Nonparametric Regression. Scheibe. CA.P. Metamodeling and Knowledge-Based Approaches. Rees. California State University.. 3. L. La Mirada.D. New York.W. “Information Model for Power Equipment Diagnosis and Maintenance.A.. Under review at Decision Sciences.VITA KEVIN P. 1056-1063..D. Fixed Wireless Telecommunication Networks • • M. San Marcos. pp.P.R. Rakes. 1998 B. CA. A MathematicalProgramming And Geographic-Information-System Framework For Wireless Broadband Deployment In Rural Areas. CA. K. 2001. L..701-706. 334-336.” Proceedings of the Thirty-Seventh Annual Meeting of Southeastern INFORMS. “From Decision Models to Decision Support: Excel with VBA. “A Comparative Analysis of Heuristic Techniques for Estimating the Parameters for a Nonlinear Time Series Model. Blacksburg. T. 4. L. Myrtle Beach.” Proceedings of Power Engineering Society Winter Meeting. K. Scheibe. NC. Scheibe. PAPERS UNDER REVIEW: 1. X.. K. Candidate in Business Information Technology. SCHEIBE EDUCATION: • Ph. NC. Sheetz.P. 2002.S. CONFERENCE PRESENTATIONS: 1. February 21-23.P. VA. Y. 2001. Charlotte.

CA 1991 – 1992. Graduate Assistant. The Lockman Foundation. DOS. System Administrator.P. La Mirada. CA TEACHING EXPERIENCE: • • BIT 4444: Computer-based Decision Support Systems (Virginia Tech) BIT 5495: Decision Support Systems Design and Implementation (Virginia Tech) SPECIAL AWARDS and HONORS: • • R. C.D. Britton Capital Corporation. Java.. T. San Diego..” WORK EXPERIENCE: • • • • • • • 1999 – Present.. San Diego. Rees. ASP. T. Rakes.T. VB. Assembly. VA 1996 – 1999. Director of Operations. VAX/VMS Vita 97 . Macintosh. Software Engineer.T. Fisk Communications Inc. Student Representative. CA 1990 – 1991. System Administrator / Software Engineer. CA 1989 – 1991.. Fixed-Charge. Wireless Mesh Networks. HTML. Pamplin Doctoral Fellowship 2001 Southeastern INFORMS Ph. L. Ragsdale. LINUX.P.WORKING PAPERS: 1. La Habra. PASCAL. K. Britton Capital Corporation.” 2.. ADA. Rakes. Blacksburg. Computer Programmer..R... Basic Research Corporation. Rees.. BASIC. L.NET. B. Department of Business Information Technology.P. UNIX. Network Flow Model for Hop Constrained. Apple Computer. Ragsdale. “A Capacitated. “Addressing Implications for Universal Broadband Service with Wireless Mesh Networks. Inc. Student Paper Competition award winner AFFILIATIONS: • • • • • • Association for Computing Machinery (ACM) Association for Information Systems (AIS) Decision Sciences Institute (DSI) Institute of Electrical and Electronics Engineers (IEEE) The IEEE Computer Society Institute for Operations Research and the Management Sciences (INFORMS) COMPUTER SKILLS: • • Languages: C/C++. Scheibe. Virginia Tech. San Diego. COBOL. CA 1992 – 1994.1.. C.R. CA 1994 – 1996. San Diego.P. SQL Systems: Windows XP/2000/NT/98/95/3. K. Scheibe.