GPSLoc: Framework for Predicting Global Positioning
System Quality of Service
Hassan A. Karimi
1
; Xiong Liu
2
; Shuo Liu
3
; and Amin Hammad
4
Abstract: While currently numerous existing engineering applications benefit from the global positioning system 共GPS兲, it is anticipated
that operation of many new, emerging applications 共e.g., applications related to ubiquitous mobile computing兲 will rely on the information
provided by this technology. Depending on the application requirement, GPS data may be collected and post-processed or collected and
processed in real time. In either case, there are questions about availability, quality, and reliability of GPS data in engineering applications.
To date, despite available techniques for realizing, and to some extent improving, a certain level of GPS accuracy, there is no integrated,
coherent approach or technique that would provide users with solutions that combine GPS availability, quality, and reliability. To that end,
we propose quality of service 共QoS兲 assurance for GPS. With GPS QoS, users and applications would be provided with the means for
predicting GPS solutions in advance meeting the requirements in a timely and cost-effective manner. We have developed a framework for
the proposed GPS QoS called GPSLoc. In this paper, we discuss the requirements, methodologies, models, and algorithms for the GPSLoc
framework and the experimentation with one of the GPS QoS parameters 共visibility兲.
DOI: 10.1061/共ASCE兲0887-3801共2004兲18:3共196兲
CE Database subject headings: Global positioning; Data collection; Computer applications; Algorithms.
Introduction
The global positioning system 共GPS兲 has become a dominant po-
sitioning technology used in numerous applications. Example ap-
plications utilizing GPS are field tasks engineering, fleet and
freight management, workforce management, facility and data
mapping and modeling, incident/outage positioning, in-car navi-
gation systems, automatic vehicle location 共AV L 兲 systems,
location-based services, and mobile mapping systems 共MMSs兲.In
these and other civil and construction applications, data provided
by GPS play a crucial role in the operation and delivery of infor-
mation to the users. Timely and cost-effective decisions can only
be made when there is a high degree of reliability on the infor-
mation provided by GPS. However, GPS data are subject to un-
certainties, and while it may not be possible to eliminate these
uncertainties, having knowledge in advance about them helps im-
prove the timeliness, usefulness, and reliability of GPS-based ap-
plications. In this regard, there is a need for quality of service
共QoS兲 assurance for GPS. We define QoS of GPS as a set of
techniques and strategies that could assure application and users a
predictable service from GPS. Despite developments with respect
to GPS issues, such as GPS accuracy measurements and improve-
ments, currently there are no unified QoS methodologies and
models for GPS 共though disparate solutions for specific applica-
tions may exist in GPS receivers and software packages兲.
To better understand the issues of GPS QoS, we define opera-
tion mode and requested QoS. The operation mode is a reference
to the ways GPS data are collected, which could be either static or
dynamic.Inthestatic operation mode, GPS is used to compute
position data at one fixed location 共or a set of selected locations
one at a time兲 without the real-time processing constraint. In the
dynamic operation mode, GPS is used to compute position data
over a set of points in real time each at a different time; this is
also called the real-time operation mode. The requested QoS is a
reference to the required GPS solutions imposed by the user or
application and could be either passive or optimal. In either the
passive or optimal requested QoS, the assumption is that a loca-
tion and a time are given by the user, or are determined by the
application. In the passive requested QoS, a GPS solution is
sought from GPS QoS. This means that the user requires a GPS
solution no matter how good the solution is. In the optimal re-
quested QoS, the most optimal solution is sought from GPS QoS.
This means that the user requires a GPS solution that is most
optimal with respect to availability, accuracy, reliability, etc.; in
this case, the user is only interested in an optimal solution 共not
any solution兲 and that GPS QoS should only search for such
solutions that meet the requirements. We also define the following
four parameters in GPS QoS that are applicable to both passive
and optimal modes: visibility, accuracy, reliability, and flexibility.
The visibility parameter of GPS QoS is a reference to those
satellites, out of the available satellites, that are visible or have
lines of sight 共LOS兲 at the given location and time. This is needed
because there are locations where GPS signals simply are not
available due to obstruction of LOS. In other words, a LOS to a
satellite is needed in order to obtain a signal representative of the
true distance from the satellite to the receiver. Therefore, any
object in the path of the signal has the potential to interfere with
1
Dept. of Information Science and Telecommunications, Univ. of
Pittsburgh, Pittsburgh, PA 15260.
2
Dept. of Information Science and Telecommunications, Univ. of
Pittsburgh, Pittsburgh, PA 15260.
3
Dept. of Information Science and Telecommunications, Univ. of
Pittsburgh, Pittsburgh, PA 15260.
4
Concordia, Institute for Information Systems Engineering, Concordia
Univ., Montreal PQ, Canada H3G 1T7.
Note. Discussion open until December 1, 2004. Separate discussions
must be submitted for individual papers. To extend the closing date by
one month, a written request must be filed with the ASCE Managing
Editor. The manuscript for this paper was submitted for review and pos-
sible publication on December 13, 2002; approved on September 11,
2003. This paper is part of the Journal of Computing in Civil Engineer-
ing, Vol. 18, No. 3, July 1, 2004. ©ASCE, ISSN 0887-3801/2004/3-
196–206/$18.00.
196 / JOURNAL OF COMPUTING IN CIVIL ENGINEERING © ASCE / JULY 2004
the reception of that signal. Objects which can block a GPS signal
include terrain heights, tree canopies, and buildings.
The accuracy parameter of GPS QoS is a reference to the level
of accuracy the GPS receiver is able to compute at the required
location and time. There are several external sources which intro-
duce errors into a GPS position. Below are the sources of errors in
GPS and the amount of error by each:
Source Error level
Ionosphere 0–100 m
Troposphere 0–30 m
Measurement noise 0–5 m
Ephemeris data 0–5 m
Clock drift 0–1.5 m
Multipath 0–25 m
We define the reliability parameter as the ability of GPS QoS
to guarantee a solution that meets the requirements of the user or
application for the given location and time. In other words, upon
receiving information from the user or application about the re-
quired QoS at a given location and time, GPS QoS will search for
the set of solutions that meet the requirements.
We define the flexibility parameter as the ability of GPS QoS
to provide alternative solutions 共location, time, or both兲 in case
there is no possible solution for the location and time requested.
For example, when it is determined that there is no possible GPS
solution for the given location and time that meets the user re-
quirements, GPS QoS will search for a solution at the nearest
location 共at the same time兲, nearest time 共at the same location兲,or
both 共at a different location and a different time兲 that meets the
requirements.
It should be noted that the purpose of GPS QoS is not to
mitigate error or improve accuracy, rather its goal is to provide
information to predict QoS of GPS. Several benefits are expected
from GPS QoS, some of which are:
1. Evaluating the quality of GPS data collected against the QoS
required by the user;
2. Predicting the QoS that can be expected at a specific location
and time;
3. Planning GPS data collection using GPS QoS maps to avoid
undesirable solutions and maximize productivity.
Currently there are no coherent methodologies and techniques
that provide users with predictive QoS of GPS. There exist off-
the-shelf software packages 共e.g., Pathfinder 1999兲 capable of
providing the user with the satellites that produce the best solu-
tion for a given location and time. However, such solutions suffer
from two shortcomings. One shortcoming is that they do not con-
sider the actual physical environment of the user’s position, that
is, such three-dimensional 共3D兲 data as terrain heights and build-
ings are not taken into account. Detailed and accurate 3D data
共which contain terrain heights and 3D objects兲 of the geographic
area where the user is located would help determine reliable GPS
solutions. A common example of the need for 3D data on and near
the location of the user for predicting QoS is when a GPS receiver
searches for the satellites with good geometry. Without checking
for potential obstacles using 3D data, the GPS receiver may in-
clude satellites that actually do not have LOS with the receiver.
Another shortcoming is that existing solutions address mostly
GPS accuracy-related issues focusing on individual cases or ap-
plications and are not coherently integrated with other issues 共in
addition to the accuracy issue兲 to provide a comprehensive QoS
of GPS. We propose a framework called GPSLoc that includes
models and algorithms to predict GPS QoS for applications using
GPS. In GPSLoc, detailed 3D geometrical data on and near where
the user is located are used and all possible GPS-related issues are
taken into account. As is shown in Fig. 1, the LOS between each
available satellite and the receiver may be obstructed by terrain
heights, such as mountains, or by 3D objects, such as buildings.
In this paper, we focus on the visibility parameter for which we
have developed solutions in GPSLoc. Our reason for focusing on
the visibility parameter is that its result impacts all the other pa-
rameters. That is, by realizing visible satellites, GPSLoc can pro-
ceed to determine GPS QoS with respect to the other parameters,
while without knowledge about visible satellites all subsequent
computations and analyses are worthless. The two major compo-
nents of GPSLoc for predicting GPS QoS are a 3D database and
a set of algorithms for LOS intersection 共LOSI兲. The 3D database
component contains detailed data on terrain heights and other
obstacles which are used in the set of algorithms to compute the
LOS between each satellite and the receiver at a given location
and time.
This paper’s contributions are an introduction to the concept of
GPS QoS for engineering and other applications that use GPS,
development of a new methodology for predicting GPS QoS, and
development of models and algorithms for GPS QoS. The struc-
ture of the paper is as follows. First, a representative engineering
application where GPS QoS can be used is described. Second, the
3D database models and algorithms used in GPSLoc are de-
scribed. Third, the algorithms for computing LOS used in
GPSLoc are described. Fourth, the experimentation results using
GPSLoc are discussed. In the last section, conclusions and future
research are summarized.
Example Scenario
To better understand the need for GPS QoS and the benefits that
users will gain from it, a sample application scenario using
GPSLoc is described. GPSLoc can be of great value to utility
infrastructure systems. Typically in utility infrastructure systems,
data 共including GPS data兲 of a high quality are desired, for ex-
ample to identify utility lines or update existing maps based on
maintenance performed. For example, a utility maintenance field
crew that has to visit several sites during a day operation can use
GPSLoc in more than one way. In the planning mode, GPSLoc
can help the crew gain knowledge in advance about the QoS of
GPS they will be able to obtain for each site. GPSLoc will ana-
lyze the planned locations and times of the visits and will provide
the GPS solutions that meet the requirements. In doing so,
GPSLoc will check the visibility, accuracy, and reliability param-
eters. If GPSLoc realizes that there are no solutions at all or that
available solutions do not meet the requirements, it can suggest
alternatives by using the flexibility parameter. The crew can use
Fig. 1. Global positioning system signal blockage
JOURNAL OF COMPUTING IN CIVIL ENGINEERING © ASCE / JULY 2004 / 197
this information provided by GPSLoc to plan visiting each site in
an order that takes into account the priority of the maintenance
needs as well as the QoS of GPS they require. Another way
GPSLoc can assist the maintenance crew in a real-time mode is
through the use of an AVL system 共which may also have commu-
nication links with the office and other field crews兲 available in
the vehicle where the real-time location of the vehicle is com-
puted 共and may be transmitted to others at the office or at other
maintenance sites兲. GPSLoc for the AVL system will assist the
crew in navigation and routing activities optimizing their perfor-
mance while allowing the office managers to maximize the pro-
ductivity of the overall maintenance operation by using the real-
time locations of all the crews in the field. An interesting
observation in this application is that the requested QoS of GPS
for the planning mode 共e.g., an accuracy range within a few cen-
timeters, no real-time processing constraint, and flexible with re-
spect to time of data collection兲 is different from the requested
QoS of GPS for the real-time mode 共e.g., an accuracy range
within a few meters, real-time processing constraint, and fixed
with respect to location and time兲.
Three-Dimensional Database Models and
Algorithms
Terrain heights 共e.g., mountains兲 and 3D objects 共e.g., buildings兲
are the major obstacles for the GPS signal. There are two reasons
why terrain heights are needed. First, places such as cliffs or hills
are potential obstacles. Second, heights of 3D objects must be
calculated relative to elevation of their base. Therefore, the 3D
database in GPSLoc must include data on all types of obstacles
including terrain heights and 3D objects and it must cover at least
the area where GPS data are collected. The database must be
based on one or more 3D models along with appropriate data
structures and algorithms for 3D data manipulation. Currently
there exist separate data models for terrain representation, used
mostly in geospatial information system 共GIS兲 software packages,
and data models for 3D objects 共e.g., buildings兲, used mostly in
Computer-Aided Design 共CAD兲 software packages.
Miller and La Flamme 共1958兲 introduced the Digital Terrain
Model 共DTM兲 as ‘‘a statistical representation of the continuous
surface of the ground by a large number of selected points with
known X, Y, and Z coordinates in an arbitrary coordinate field.’’
DTM is a generic term for digital representation of terrain infor-
mation. There are two widely used terrain models: the digital
elevation model 共DEM兲 and the triangulated irregular network
共TIN兲. DEM is a grid 共or raster兲 model where the terrain’s eleva-
tion data are sampled at a regularly spaced interval and stored as
an array 共Lo and Yeung 2002兲. Each raster cell has an elevation
value and all values make a matrix of elevations. TIN is a vector
model 共Peucker et al. 1978兲 based on a set of irregularly spaced
points to represent the terrain. One advantage of TIN over DEM
is the possibility of adapting the irregularly spaced sample points
to the terrain’s features. For instance, in TIN more points in rough
areas and fewer points in smooth areas can be used to better
represent the terrain’s features, while in DEM the same number of
points is the only way to represent both rough and smooth areas.
Another advantage of TIN is that since it is a vector model, it is
possible to obtain highly accurate data points. TIN also results in
less storage than DEM since the number of points in TIN is ad-
justable. Because of its advantages, TIN has been adopted by
several GIS and automated mapping and contouring software
packages. Given that the accuracy in representing 3D objects is an
important requirement and that the data model should make it
possible for developing computational geometry algorithms for
computing LOS, TIN is more suitable for GPSLoc than DEM.
3D databases have been discussed in the field of 3D GISs
where geometric and semantic information has been incorporated
into one model for spatial analysis 共Maguire et al. 1991; Aronoff
1995兲. There are also 3D models such as 3D FDS 共Formal Data
Structure兲, TEN 共Tetrahedral Network兲, and the cell tuple model
共Zlatanova 2000兲 where 3D objects 共e.g., buildings兲 are treated
independent from the terrain, that is, they are located on a plane
with no elevation information. These models can handle spatial
relationships among objects existing in the database using SQL-
type queries. In computing the intersection of an arbitrary line
with a 3D object, both geometric and topological information are
needed. In a simplified 3D database model, the height of a 3D
object can be added to the Z value of the terrain. Conceivably
such a data model may be obtained either by integrating an ex-
isting vector model for representing the terrain and a model for
representing 3D objects, or by developing a new model that takes
into account the primitives of both the terrain and 3D objects in
one data structure. The main issue in the integrative approach is to
develop an algorithm to link the two different models, each as a
separate schema, together. For example, by using a TIN to repre-
sent terrain data and a TEN to represent 3D objects, there is a
need for an algorithm that links the two models together. The
main issue in the second approach, developing a new model, is to
use a single data structure for both the terrain and 3D objects. For
example, if the TIN model is considered for both types of data,
there must be a method of incorporating 3D objects into it.
The first approach requires two data structures and two sets of
retrieval procedures. This approach is inefficient in GPSLoc,
which requires frequent interactions with the database. The sec-
ond approach requires development of a new algorithm to com-
bine terrain heights and 3D objects into a single model, but since
it involves only one data structure and one set of retrieval proce-
dures, it is more efficient for GPSLoc. Because of the advantages
it offers, a new single database model, i.e., the second approach,
is adopted in GPSLoc. The single model is an extension of the
TIN model, thus we call it eXtended TIN 共XTIN兲, by adding
points that represent 3D objects to an existing TIN representing
terrain heights. The result of XTIN is a 3D model made up of
intraconnected triangles representing both the terrain and 3D ob-
jects. Since 3D objects may have regular shapes, such as a col-
umn or a cylinder, or irregular shapes, such as a football stadium,
XTIN must support methods of including both shape types 共regu-
lar and irregular兲. Therefore, the requirement of XTIN for
GPSLoc is to include both regularly and irregularly shaped ob-
jects and to preserve the shapes of objects accurately. In the fol-
lowing, the construction method of XTIN in GPSLoc is dis-
cussed.
Extended Triangulated Irregular Network
The 3D data model in GPSLoc must represent geometric and
topological information for both terrain heights and 3D objects.
XTIN is a new data model which can be used for GPSLoc, and
other applications requiring 3D databases. In GPSLoc, before
XTIN is constructed, the TIN model for the area of interest is first
generated. The algorithm for generating TIN in GPSLoc is based
on the algorithm by Garland and Heckbert 共1995兲. In this algo-
rithm, first two triangles are generated using the four corner
points of the DEM data set. Then all the remaining points in the
DEM are used to find the best point, which is defined as the point
198 / JOURNAL OF COMPUTING IN CIVIL ENGINEERING © ASCE / JULY 2004
with the largest interpolation error, to insert into the current TIN.
The interpolation errors are calculated by taking the differences
between the interpolated heights 共using the X,Y coordinates of
each DEM point in the corresponding location in the current TIN
to determine the elevation value, i.e., Z兲 and the actual heights
共the Z values in the DEM兲. The point with the largest interpola-
tion error is chosen as the next point to be inserted to minimize
the overall error in the TIN. One of the following two criteria can
be used to terminate the triangulation process: the number of total
selected points or the largest interpolation error of the TIN. In
GPSLoc, the former criterion is chosen because it allows an easier
way of controlling the total number of points and triangles in the
final TIN model. The steps of this algorithm are shown in Fig. 2.
XTIN is built by incorporating the triangles from 3D objects
into TIN. In doing so, it is expected that an object will intersect
with one or more triangles in TIN. The important part in merging
object triangles into TIN is how to update TIN 共after the object is
inserted兲 since this directly impacts the structure of the resultant
triangle set and the performance of queries on XTIN. Discussed
below are the three methods for this along with their strengths and
weaknesses.
The first method is to put the two sets of triangles together
without making any changes. This method is simple and fast to
implement. The topology integrity in the resultant triangle set,
however, is not maintained. In this method, a common edge be-
tween two triangles, one from TIN and one from object triangles,
that intersect may not exist, which is a property for two neighbor-
ing triangles and important in traversing the triangle network and
indexing. Furthermore, removing the triangles in TIN that are
covered by objects could improve performance.
Fig. 2. Triangulated irregular network 共TIN兲 generating algorithm
Fig. 3. Extended triangulated irregular network 共XTIN兲 algorithm
Fig. 4. Projecting top face of a three-dimensional object and trian-
gulated irregular network 共TIN兲 triangles
JOURNAL OF COMPUTING IN CIVIL ENGINEERING © ASCE / JULY 2004 / 199
The second method is to update TIN by inserting into it all the
vertices on the base of an object as new points in TIN. This
method seamlessly joins both triangle sets, from TIN and objects,
and maintains the Delaunay triangulation property 共Kreveld 1997兲
in TIN. But one major problem with this method is how to retain
the shape of the 3D object’s base while inserting new vertices into
TIN, which is a necessary condition for the seamless join. There
is no guarantee that any arbitrary connection between points will
be preserved after they are inserted into TIN using Delaunay tri-
angulation. Even though an algorithm may be devised to over-
come this problem, the entire TIN may have to be recomputed
each time a new point is added. This would make the computation
very expensive, especially when working on a large area with
many 3D objects.
The third method is to only update triangles in TIN around an
inserting object. The base of a 3D object is formed by projecting
its top face onto the TIN. All the edges of the adjacent triangles in
TIN that intersect, touch, or are contained by the base are re-
moved. New triangles are built in the area between the base and
the edges left from the adjacent triangles. This method guarantees
a seamless join of TIN and object triangles since the shape of the
3D object’s base is maintained. All the unnecessary triangles in
TIN covered by the base are removed to avoid redundant compu-
tation in query execution. The Delaunay triangulation property
may be no longer valid in the newly generated triangles around
the base if a polygon triangulation algorithm 共e.g., O’Rourke
1994b兲 is used. The time performance of this method is better
than the time performance of the second method since it only
requires the triangles around objects and not the entire TIN. Con-
sidering the amount of triangles generated and updated in this
method and the gain in computing time, this method is preferred
over the other two methods and was chosen in our work. Fig. 3
shows the steps of the XTIN algorithm in GPSLoc, which is
based on this method. Some of the details of the algorithm are
described below.
In order to simplify the computation, a generalization on rep-
resentation of 3D objects is adopted. Given the case of irregular
object shapes, a 3D object is considered as a solid entity com-
posed of its top horizontal face and all the vertical faces formed
by projecting the top face onto the ground. This generalization
saves time in constructing and triangulating 3D objects and is
valid in most situations which simplifies constructing XTIN by
using only the top face of the 3D objects.
After obtaining top faces, XTIN will be constructed by incor-
porating 3D objects into TIN. This is accomplished by projecting
the top faces onto the triangles of the TIN and connecting the
vertices of the top faces to the projection points in the TIN. In
doing so, the projection points are identified first, and then the
triangles in which these points fall are determined and are used to
interpolate the elevation of the projection points. The approach
taken in this work to construct XTIN first projects the top faces
and the TIN triangles onto a 2D plane and then uses a point-in-
polygon algorithm to find the triangles that contain the projection
points. A horizontal plane passing through the lowest elevation
point in the TIN was chosen as the 2D plane. A projection point
on this 2D plane should have the same horizontal coordinates
共i.e., X and Y兲 as its original vertex and the same vertical coordi-
nate 共i.e., Z兲 as the 2D plane. This projection process is illustrated
in Fig. 4, where the top face of a 3D object 共a cube兲 and the TIN
triangles are projected onto an imaginary 2D plane in the bottom.
The four circles illustrate the corners in the base of the cube,
which is the projection of the top face onto the TIN. After all the
points are on the 2D plane, the point-in-polygon algorithm by
Fig. 5. Triangulated irregular network with a top face of a three-
dimensional object
Fig. 6. Triangulated irregular network after some triangles are
removed
Fig. 7. Triangulated irregular network after polygon triangulation
200 / JOURNAL OF COMPUTING IN CIVIL ENGINEERING © ASCE / JULY 2004