Journal of Global Positioning Systems (2004)
Vol. 3, No. 1-2: 132-142
Process for improving GPS acquisition assistance data and server-side
location determination for cellular networks
Neil Harper
Nortel, Nortel Building, Northfields Ave, University of Wollongong, NSW 2500
e-mail: [e-mail]; Tel: [phone]; Fax: [phone]
Peter Nicholson
Nortel, Nortel Building, Northfields Ave, University of Wollongong, NSW 2500
e-mail [e-mail]; Tel: [phone]; Fax: [phone]
Peter Mumford
School of Surveying & Spatial Information Systems, University of New South Wales, Sydney, NSW 2052
e-mail: [e-mail]; Tel: [phone]; Fax: [phone]
Eric Poon
Nortel, Nortel Building, Northfields Ave, University of Wollongong, NSW 2500
e-mail: [e-mail]; Tel: [phone]; Fax: [phone]
Received: 15 Nov 2004 / Accepted: 3 Feb 2005
Abstract. This paper introduces a process for analysing
and improving the quality of the acquisition assistance
data produced for Assisted GPS (A-GPS) positioning in
cellular networks. An experimental test bed is introduced
and a series of experiments and results are provided. The
experiments validate the GPS acquisition assistance data
in open-sky conditions. Accuracy results for initial testing
of our server location determination engine in a range of
different environments are also given with results of a
long-term run. Acquisition assistance data provides the
GPS handset with information that allows it to detect the
GPS signals more quickly and allows detection of much
weaker signals. It does this by providing information to
the handset about where to look for the signals. The A-
GPS server is a mobile location server determining the
location of devices within a cellular network. In order to
measure the quality of the acquisition assistance data
produced by the A-GPS server a piece of hardware has
been developed called the “A-GPS Trainer”. This takes
assistance data and uses it to lock on to the satellites. It
then provides code phase measurements back to the
server for it to do a location calculation. The trainer also
reports on the amount of time it takes to lock on to the
signals of the individual satellites. It can run multiple
calculations over a period of time and report the results.
Key Words: A-GPS, Assistance Data, GPS acquisition
assistance data.
1. Introduction
Assisted-GPS is considered by many the best solution to
meet the accuracy requirements of the U.S. Federal
Communications Commission’s E-911 mandate. The E-
911 mandate requires telecommunications carriers to give
emergency services providers accurate location
information in a timely manner. Additionally, the
accuracy of GPS facilitates Location Based Services that
have good economic potential.
The two different paradigms in cellular networks are MS-
Based Assisted-GPS and MS-Assisted Assisted-GPS. In
MS-Based A-GPS, the network provides the handset with
detailed assistance data to help the handset to lock on to
the satellites and perform the position calculation.
In the case of MS-Assisted A-GPS, the network provides
a smaller amount of assistance data to help the handset to
lock on to the satellites. The handset then provides the
satellite measurements to the network for the network to
perform the position calculation.
There are advantages and disadvantages of both MS-
Assisted A-GPS and MS-Based A-GPS. Some of the
advantages of MS-Assisted A-GPS are that less

Harper et al: Process for improving GPS acquisition assistance data and server-side location determination 133
information is transmitted over the air interface, and there
is potential on the server to use other information such as
network timing information from base stations or a
Digital Elevation Model. Another benefit is that the
position calculation is centralised so that any software
problem can be fixed in one place on the server instead of
millions of customer handsets.
GPS acquisition assistance data is used by an MS-
Assisted A-GPS handset to lock on to satellite signals.
This paper describes a test bed and a process for tuning
the server-side algorithm which produces the acquisition
assistance data.
In an emergency situation, the time it takes to lock on to
the satellites is very important and in this paper the time
to lock on to the satellites in open sky conditions from a
cold start is optimised. The results show that acquisition
assistance data speeds up the acquisition by about 200
times with our hardware.
A measure called the ‘acquisition assistance usefulness
factor’ is introduced. This factor is the ratio of the relative
speed of the GPS fixes scaled by the proportion of the
satellites in view that the device locks on to. It defines
how good the acquisition assistance data is for a
particular GPS fix.
The other major part of an MS-Assisted A-GPS solution
is the position calculation function. In this paper some
results of testing at the University of New South Wales
(UNSW) and the results of a long-term run (18 months)
are given.
2. Background
Nortel Networks manufactures various 3GPP-compliant
cellular network elements. These elements include
Serving Mobile Location Centers (SMLC), Radio
Network Controllers (RNC) and Standalone A-GPS
SMLCs (SAS). These elements generate GPS acquisition
assistance data for cellular phones with embedded A-GPS
capability. The GPS assistance data speeds up acquisition
of GPS signals and increases GPS yield in weak signal
environments.
It was necessary to develop a process and a test
environment that allowed the effectiveness of the
supplied GPS assistance data to be measured. The
ultimate measure of the value of the assistance data is to
make empirical observations on the impact to sensitivity
and time-to-fix with an A-GPS receiver. The testing
process and environment developed facilitates many such
observations being made, ensuring a valid set of
statistical measurements.
The test environment provides a means of measuring the
effects of optimising the assistance data. By measuring
the results of changes to the supplied assistance data, a
validation can be performed to ensure that the data
provides optimal results in practice. In addition, the test
environment provides a practical regression test suite to
ensure future versions of the assistance data calculation
software produce correct assistance data.
Assistance data helps to significantly reduce the search
space required by the GPS enabled handset (Zhao 2002,
Djuknic and Richton 2001). Fast, accurate fixes can be
made using a combination of good assistance data and
high sensitivity receivers (van Diggelen and Abraham
2001, and Heinrich et al 2004).
Much of the literature concentrates on improvements to
the handset. This paper is focused on improvements that
can be made on the server.
3. GPS acquisition assistance data
The cellular network can provide assistance data to the A-
GPS cellular handset through the air when it is in contact
with a base station. Assistance data types that can be sent
to the handset includes the ephemeris, reference time,
ionosphere model, UTC model, real time integrity and
acquisition assistance.
The format of the GPS acquisition assistance data is
defined by the relevant standards body; the specifics of
the encoding and field ranges depend on the access
technology. The protocol for the GSM system for
example is defined in 3GPP TS 04.21 and is referred to as
RRLP (Radio Resource LCS protocol). Other protocols
include Positioning Calculation Application Part (PCAP)
for UMTS networks (between the SAS and the RNC) and
Position Determination Service Standard for Dual Mode
Spread Spectrum Systems (PDDM) for CDMA networks.
A high-level diagram of an SMLC network is shown in
Fig. 1. In this scenario, the SMLC receives a request for
assistance data for a particular handset. The request
effectively includes the initial location estimate and the
uncertainty of the estimate, which is based on the size of
the serving cell.
This initial location estimate and the uncertainty are used
to calculate the acquisition assistance data using
information from the Wide Area Reference Network
(WARN). This acquisition assistance data is used to
populate the message that is sent back to the network and
on to the handset.
The fields that the server provides are summarised in
Tab. 1. They are named similarly and have similar ranges
across the different network access technologies

134 Journal of Global Positioning Systems
Fig. 1. A-GPS in a cellular network
Tab. 1. The fields for GPS Acquisition assistance data
Field Approximate Range Notes
GPS Time 0 to 604800.0 The GPS time of week in seconds at which the assistance data is
relevant
Satellite Information The following fields appear for each satellite in the message:
Satellite ID The satellite ID
Doppler -5120 to 5117.5 Hz The predicted doppler at the specified time
Doppler (1st order) -1 to 0.5 Hz The rate of change of doppler
Doppler Uncertainty 12.5 to 200 Hz The doppler uncertainty value
Code Phase 0 to 1022 chips The centre of the code phase search window (in chips)
Integer Code Phase 0 to 19 The number of 1023 chip segments
GPS Bit number 0 to 3 The GPS bit number (20 1023-chip segments)
Code Phase Search Window 1 to 1023 The width of the search window for code phase
Azimuth 0 to 360 The horizontal distance in degrees of the satellite from North
from the estimated location of the handset on the ground
Elevation 0 to 90 The angular distance of the satellite in degrees above the horizon
from the estimated location of the handset on the ground
The server estimates the information including the search
windows in the acquisition assistance data (Doppler
uncertainty and code phase search window) using the
initial location estimate and its uncertainty only. The
SMLC
GPS Reference feed
WARN
GPS assistance data
GPS capable
cellular handset
GPS satellites

Harper et al: Process for improving GPS acquisition assistance data and server-side location determination 135
server makes no assumptions about the handset hardware
since there may be many different types of handset in a
network. When the handset receives the acquisition
assistance data, it adjusts the windows based on its
hardware.
For example, the measured Doppler is affected by the
local handset TXCO (Temperature Controlled Crystal
Oscillator). The size of the Doppler search window needs
to be interpreted in relation to the handset’s estimate of
its local oscillator error. If a handset that has a local
oscillator error of up to 100 Hz uses the predicted search
window of 12.5 Hz, it is unlikely to find the satellite.
Similarly, the code phase values are highly dependent on
the clock error of the handset at the millisecond level
because the PRN replica is driven by the local handset
clock. The handset needs to know its time accurately in
order to make use of the code phase predictions as is. In a
synchronised network, such as CDMA, the handsets are
synchronised with the base stations, which are
synchronised to GPS time, so the code phase can be used.
In a non-synchronised network, such as GSM, the
handset needs to treat supplied code phase measurements
as relative code phase offsets. To do this, it locks on to
the first satellite using its Doppler information and a
potentially large search in the time (code phase) domain.
Once it locks on to this first satellite, it calculates the
difference between the code phase measurement for this
satellite and that supplied in the assistance data. This
offset is then applied to all of the other code phase
estimates in order to determine where to search for those
satellites. In a non-synchronised network, it will take a
longer time to lock on to the first satellite but once it gets
that one, the narrow code phase search using the
assistance data can be applied to the other satellites.
4. GPS Acquisition Results from logged GPS
measurement data
This section reports on the results of calculating the GPS
acquisition assistance data and comparing some of the
fields in the acquisition assistance data against
measurements that have been logged from a GPS receiver
over a long period of time. The results show that the
acquisition assistance data calculated is close to the
measured values of the fields.
GPS measurements have been logged from a reference
receiver for more than 18 months and approximately 130
million measurements of individual satellites have been
made. The GPS reference receiver tracks in open-sky
conditions and operates continuously.
Software reads through each of these measurements and,
for each observation, compares the calculated GPS
acquisition assistance data with the measurements. The
calculated Doppler is compared with the measured
Doppler and the predicted range (used to estimate the
code phase for acquisition assistance data) with the
measured pseudorange. The predicted range is compared
with the measured pseudorange because the reference
receiver does not supply code phase. Before they can be
compared, the predicted range is converted into a
predicted pseudorange by applying the clock offset of the
receiver calculated by the position calculation function.
This means that both ranges are offset by the same clock
error and can be compared.
The results in Tab. 2 show that there is a very small
difference in the calculated Doppler and range compared
to the measured values. This gives us confidence that the
basic algorithm used to calculate the Doppler and range
(from which code phase is derived) is correctly coded and
that the assistance data can be tested using the A-GPS
Trainer described in the next section.
The first column Tab. 2 shows the difference between the
predicted Doppler and the measured Doppler for over 130
million individual satellite observations. The average
error is 0.46 Hz with a standard deviation of 0.35 Hz.
There were 15 outliers that had more than 10 Hz of error
and they have been split out into a separate statistic
shown in the column titled ‘Doppler error over 10 Hz’.
There is presently no real-time integrity information such
as provided by the Wide Area Augmentation System
(WAAS) to determine whether there was a problem with
the calculation, measurement or satellite for these
outliers.
The predicted range error is shown as 15.21 metres with a
standard deviation of 12.47 metres. There were 3 records
with an error over 150 metres and they are shown in the
last column of the table.
Tab. 2. The difference between calculated and measured values
Doppler
error
Doppler
error
over 10
Hz
Predicted
range error
Predicted
range
error
over 150
metres
Number of
Records 130667033 15 130667045 3
Average 0.46 393.43 15.21 4363.20
Standard
Deviation 0.35 1096.96 12.47 7286.02
Minimum 1.4E-09 10.08 3.97 155.97
Maximum 9.96 4341.67 149.82 12776.37

136 Journal of Global Positioning Systems
5. A-GPS Trainer
The A-GPS Trainer is a custom device designed and
developed by the U NSW Satellite Navigation and
Positioning group within the School of Surveying &
Spatial Information Systems in consultation with Nortel
Networks. It is a device for testing GPS acquisition
assistance data.
The device consists of a Sigtec MG5001 GPS receiver
running a modified version of the Mitel GPS Architect
software.
The Mitel software has been extensively modified to
facilitate control and reporting of signal acquisition. A
custom interface has been developed that allows the
device to run in autonomous GPS mode or in A-GPS
mode. This alows a comparison of the times to make
different types of fixes. A model of the device is shown
in Fig. 2.
Sigtec
MG5001
connection
Modified Mitel
GPS Architect
Custom Interface
Fig. 2. A-GPS Trainer model
During a typical measurement session, the A-GPS
Trainer is sent acquisition assistance data and then
attempts to acquire and track those satellites and report
back. The Doppler search begins at the given Doppler
value and searches each side of this value until the
maximum width is reached. The size of this Doppler
search window depends on the estimated receiver clock
error as well as the Doppler window size given in the
assistance data. If a signal is not detected, the search
starts again.
The code search uses the standard Mitel GPS Architect
method, and starts at one end of the window and slides
across to the other. An improvement could be made on
this method by starting the code search at the given code
phase value and then searching more widely on each side
until the search window is reached. This improvement is
possible because the true location of the handset is more
likely to be at the centre of the search window than near
the boundary so the lock should be made faster. This is a
device improvement however and since the focus of this
paper is measuring server-side improvements it is not
described further.
The information reported at the end of a measurement
session includes (for each satellite allocated)
pseudorange, the time taken to acquire locks on the
signal, Doppler, code phase and SNR.
The A-GPS Trainer treats the code phase values as
absolute code phase values. This means that the code
phase search uses the code phase values in the assistance
data as absolute values in order to centre the search. The
problem is that the generation of the code phase replica in
the handset is driven by the handset clock which is
subject to error. If the clock is out by just half of a
millisecond then the code phase will be out by half of its
range of 511 chips.
The handset needs to treat the supplied code phase values
as relative values. It will do a complete search in the time
domain and when it locks on to the first satellite, the code
phase for the centre of the search windows for the other
satellites can be set relative to that satellite using the
relative differences in the supplied acquisition assistance
data.
The A-GPS trainer presently does not have this facility
which means that the code phase is of less value and the
lock is slightly slower than it could be.
6. A-GPS Train Driver tool
The A-GPS Train Driver is a software tool that calculates
the acquisition assistance data using the ephemeris for the
satellites in view extracted from an independent GPS
receiver that has already locked on to the satellites. The
A-GPS Train Driver then provides the GPS acquisition
assistance data to the A-GPS Trainer.
The A-GPS Train Driver parses the information returned
by the A-GPS Trainer and reports on the Carrier and
Code (CC) and Carrier, Code, Frame and Bit (CCFB)
lock times of individual satellites in view. The Train
Driver can also invoke the position calculation function
using the GPS code phase measurements made by the A-
GPS Trainer.
Fig. 3 shows the connection between the various
components of the test system. On the right hand side is
an image of the system in the lab.
In order for the A-GPS Trainer to make use of the code
phase supplied in the assistance data, a synchronised
network is crudely simulated. In a synchronised network,
the handset has accurate time. In order to simulate this
using the A-GPS Trainer, two pieces of information are
used.

Harper et al: Process for improving GPS acquisition assistance data and server-side location determination 137
A-GPS Train Driver
A-GPS Trainer
TCP/IP
serial
GPS antenna
GPS
referen ce
receiver
A-GPS
Trainer
Term inal
Server
GPS
Referenc e
Receiver
Terminal server
serial
Fig. 3. A-GPS Train Driver tool and its connections
Firstly, a statistical measure is made of the time that it
takes to generate the acquisition assistance data up until
the A-GPS Trainer acknowledges the receipt of the
assistance data. The time for which the acquisition
assistance data is calculated is advanced using this
information.
Secondly, information about the clock in the handset is
used to artificially adjust the code phase. To do that, a
autonomous position calculation is performed, which
allows the A-GPS Trainer to update its clock. This is
followed by a series of GPS fixes where assistance data is
supplied. After each fix using the assistance data, the
position calculation function is invoked to calculate the
clock offset of the A-GPS Trainer. This offset is applied
to the code phase measurements sent to the A-GPS
Trainer for the next assistance data based request.
Fig. 4 shows graphically the simulated synchronised
network process. Note that the A-GPS Trainer is run on a
machine that has a Network Time Protocol (NTP) server
attached in order for it to have accurate time.
7. Quality Measurement of Assistance Data
The model for analysing the acquisition assistance data
and collecting data is shown in Fig. 5. Time increases
along the axis from left to right.
Initially, the A-GPS Trainer is issued with a “clear”
command so that the next GPS fix will be a cold start.
The A-GPS Trainer then performs an autonomous GPS
fix. This gives a cold-start baseline for the amount of time
that the A-GPS Trainer takes to lock on to the satellites
without assistance data.
Another “clear” command is then issued and acquisition
assistance data is calculated before supplying it for a fix
that starts at time T1. The A-GPS Trainer locks on to as
many satellites as possible and the process completes at
time T2. The A-GPS Trainer then reports on its results.
Get current GPS
time
Start
Apply receiver clock offset from
previous calculation to code phase
A-GPS Trainer
clock offset
from GPS time
Advance GPS time by time it takes to
calculate AA and get to hardware
Calculate GPS
acquisition
assistance data
Send data to A-GPS Trainer
Receive results and calculate position
and clock error using the code phase
information
receiver clock
offset
next calculation
Fig. 4. Synchronised network simulation using the A-GPS Train Driver
The acquisition assistance data is calculated again but this
time for time T2. This is so that the measured satellite
information can be compared with the assistance data for
the time when the measurement was actually made.
During testing, there is limited value in comparing the
measured values against the assistance data at T1 because
there are cases when the A-GPS Trainer is not able to
lock on to one or more satellites and the difference
between T1 and T2 can be up to 30 seconds. This large
elapsed time is due to the fact that the A-GPS Trainer is
configured to lock on to all of the satellites supplied and
for some of the test cases satellites that are not in view
are supplied.
When analysing the results, T1 is only looked at for
sanity and a comparison between the calculations at T2
and the measured values are made.
From this data, there are two measures of the quality of
the acquisition assistance data. One is the relative amount
of time that it takes to lock on to the first four satellites
between a fix with assistance data and a fix without
assistance data. The other is the difference between the
assistance data calculated at time T2 and the
measurements reported by the A-GPS Trainer.

138 Journal of Global Positioning Systems
time
T1
Calc ulat e
acquisition
assistance for
time T1
T2
Calc ulat e
acquisition
assistance for
time T2
A-GPS Trainer performs cold
start, locks onto satellites
using assistance data and
reports results
Stand
alone
GPS fixcomplete
A-GPS Trainer performs cold
start, locks onto satellites and
reports results
Fig. 5. Data Capture Cycle
From the relative times between the autonomous fix and
the assisted fix, the speed-up factor is calculated using
Equation 1. This is a measure of the number of times
faster a fix with assistance data is.
Note that the time for the first four satellites to get CCBF
lock is used for an autonomous fix. An autonomous fix
needs to get frame lock in order to get the ephemeris data.
This contrasts to a fix with assistance data which only
needs CC lock since it only needs to measure the code
phase and Doppler information and return this to the
network to do the position calculation.
TimeCCassisted
oneTimeCCBFstndAl
torSpeedUpFac = (1)
A new measure called the Acquisition Assistance
Usefulness Factor is defined. This measure weights the
speed up factor by the number of satellites that it
manages to lock on to when provided with assistance
data. This then gives a measure of how good the
assistance data is across all the satellites in view of the
antenna. The Acquisition Assistance Usefulness Factor is
calculated using Equation 2.
tsInViewnumberOfSa
tsLockedOnnumberOfSa
torSpeedUpFacseFactorAcqAssistU *= (2)
The Acquisition Assistance Usefulness Factor is the same
as the speed up factor when the device locks on to all the
satellites in view. The scale factor part of Equation 2
reduces the speed up factor relative to the number of
satellites that it locks on to.
Due to the constantly changing satellite constellation, the
time that a GPS fix takes can vary significantly from one
period to another, especially in the case of a cold-start
autonomous fix. So all measurements are made multiple
times and the averages are reported.
8. Improving assistance data
The process used is a feedback loop where results are
analysed, and then algorithms and code are modified
based on this analysis.
The parameters are varied and the results are used for an
analysis phase before making improvements to the
algorithms.
The general model is shown in Fig. 6. The full set of
scenarios are run, results are analysed over all the
scenarios and areas of potential improvement identified.
Areas of potential improvement include:
Satellites that don’t achieve lock
Satellites that take a long time to lock on
Satellites where the assistance data at T2 does not match
what the A-GPS Trainer measures
Run scenarios
Start
Analyse
results
improve
algorithm and
implementation
needs
improvement
check results
Stop
Figure 6. Process
The experiments described in the rest of this paper deals
with how the initial location affects the GPS fix. The
experiments are focused in the following areas with
different sizes of the circle surrounding the initial
location estimate:
Testing with the centre of the initial location estimate co-

Harper et al: Process for improving GPS acquisition assistance data and server-side location determination 139
located with the antenna
Testing individual pieces of assistance data by zeroing
the Doppler or the code phase values
Testing with the antenna within the initial location
estimate but in different horizontal locations and altitude
relative to the antenna
Testing with the centre of the initial location estimate a
fixed distance from the antenna where the antenna is
outside of the initial location estimate
9. Results
The A-GPS Trainer tests shown in this section were
performed using a simulated synchronised network and
the results are an average of 100 measurements. The
results shown in this section have already been through
the process discussed in Section 8.
During the experiments the improvement process
described in Section 8 was found to be useful for
identifying and tracking down problems with the custom
developed firmware of the A-GPS trainer. Once the
problems with the A-GPS Trainer were identified using
the process and subsequently fixed, then acquisition of
the satellites in open-sky conditions “just worked”. That
is, the A-GPS Trainer was able to lock on to the satellites
very quickly.
Further work needs to be done in weak signal
environments in order to find areas of improvement to the
process and the GPS acquisition assistance data.
9.1 Tests with the GPS antenna at centre of initial
location estimate
During this test, the GPS antenna is at the centre of the
initial location estimate ellipse. The radius of the circle
for the initial location estimate is 5000 metres. The
results in Tab. 3 show that it is 185.6 times faster to lock
on to the satellite using the GPS acquisition assistance
data compared to an autonomous GPS fix. The data
shows that over the 100 tests, it never misses any of the
satellites that are in view of the antenna.
The difference between the Doppler calculated at T2 and
the measured Doppler is out by 198 Hz with a small
standard deviation of 7 Hz. This is due to the local
oscillator error of the TXCO and is quite consistent over
time as shown by the small standard deviation. The code
phase difference on the other hand varies significantly
because it is affected more by the clock error of the
device, which is quite variable.
Tab. 3. The GPS lock information for GPS antenna at the centre of the location estimate
Avg autonomous
time seconds
Avg time
using acquis
seconds
Speed-up
Factor
Avg
number of
Sats in
view
Avg
number of
Sats
missed
Acquisition
Assistance
Usefulness
Factor
Avg
Doppler
diff
(stdev)
Avg code phase diff
(stdev)
185.8 1.0 185.6 10.0 0.0 185.6 198 (7) 171 (245)
9.2 Tests for relative contribution of Doppler and code
phase
During this test, information is provided to the A-GPS
Trainer that contains zero for either code phase or
Doppler fields of the acquisition assistance data. It can be
seen that code phase has relatively low importance to the
acquisition of the signal for the acquisition algorithm
used in the A-GPS Trainer. The A-GPS Trainer uses a
code sweeping technique which is driven by the Doppler
search. The A-GPS Trainer searches each Doppler and
then within each Doppler searches for the code. A more
sophisticated GPS receiver may employ a matched-filter
based implementation which searches all possible code
phases simultaneously (Syrjarinne 2000).
The acquisition assistance usefulness factor is slightly
lower because a larger search in the code phase domain is
required and very occasionally the A-GPS Trainer is
unable to lock on to one of the satellites.
Doppler on the other hand is very important for the A-
GPS Trainer because if it is set to zero then no fix is
possible because it does not lock on to the satellites. This
is because the searching algorithm in the A-GPS Trainer
is driven by the Doppler.
The results for zero Doppler show that the average
number of satellites in view is 9.3 and it misses 8.1 of
them. Further analysis of the data shows that the satellites
that the A-GPS Trainer could acquire were the ones with

140 Journal of Global Positioning Systems
Doppler around 0 Hz. A potential improvement to the test
tool is to improve the algorithm to supply incorrect
Doppler, for example, setting the Doppler in the
assistance data to 5000.0 when the doppler is around 0
results in the A-GPS Trainer being unable to lock on to
any satellite.
Tab. 4. The GPS lock information for zero code phase and Doppler
Avg
autonomo
us time
secs
Avg time
using
acquis
secs
Speed-
up
factor
Avg no
of sats in
view
Avg no
of Sats
missed
Acquisition
Assistance
Usefulness
Factor
Avg
Doppler
diff
(stdev)
Avg Code
phase diff
(stdev)
Zero code phase 144.3 1.2 122.7 10.1 0.1 121.5 216 (3) 343 (204)
Zero Doppler 206.3 No Fix N/A 9.3 8.1 N/A N/A N/A
9.3 Tests for offset in the initial location
In this test, the A-GPS Trainer is provided with
acquisition assistance data calculated at different distance
offsets from the true location of the antenna. The results
show that there is very little difference in the A-GPS
Trainer’s ability to lock on when the acquisition
assistance data is calculated for a location away from the
receiver until the distance error is more than 100
kilometres. It is at this point that the acquisition
assistance usefulness factor drops dramatically because
the device can’t lock on to all the satellites. Even though
the A-GPS Trainer only needs to lock on to a minimum
of 4 satellites, this is still a problem because if any of the
useful satellites are obscured by obstacles then the
handset won’t be able to lock on to them and hence
enable a location fix.
Tab. 5. The GPS lock information for GPS antenna different distances from the location estimate
Initial location
offset distance
(direction)
Avg
autonomo
us time
secs
Avg
time
using
acquis
secs
Speed-
up
factor
Avg
no of
sats in
view
Avg no of
sats
missed
Acquisition
Assistance
Usefulness
Factor
Avg Dopp
diff
(stdev)
Avg code
phase diff
(stdev)
5 Km (altitude) 215.7 1.2 173.2 10.1 0.5 165.2 220 (17) 295 (265)
10 Km (horizontal) 183.2 1.1 169.9 10.6 0.3 165.3 209 (18) 193 (239)
50 Km (horizontal) 113.4 1.4 83.3 10.4 0.5 79.2 198 (168) 354 (239)
100 Km (horiz) 151.5 1.1 135.0 10.0 0.1 133.4 219 (45) 272 (237)
500 Km (horiz) 241.4 No Fix N/A 9.3 8.7 N/A 251 (136) 327 (264)
1000 Km (horiz) 169.5 4.7 36.3 10.8 5.6 17.5 278 (155) 348 (234)
10. Position Calculation Accuracy Results
The other major component of an MS-Assisted A-GPS
solution is the position calculation function. The position
calculation function is a combined least squares
implementation consisting of a mathematical model and
stochastic model as described in Parkinson et al (1996)
and Harvey (1998).
10.1 Testing at UNSW
The position calculation function was tested against
ground truth and a post-processing software package
called “P4” at points around the UNSW Campus. The

Harper et al: Process for improving GPS acquisition assistance data and server-side location determination 141
School of Surveying and SIS has established a network of
ground points with known coordinates (to centimetre
accuracy) over a wide range of environments.
The position calculation function determined coordinates
over a subset of these points with reasonable GPS signal
conditions complied with the E-911 regulation statistical
method (as described in OET BULLETIN No. 71
(2000)). The comparison between the position calculation
function and “P4” established the position calculation
function as a good performer on real world data as it
outperformed P4 on 76% of epochs in which both had a
solution. The position calculation function had an average
position error of 12 metres with a standard deviation of
34 metres over a total of 2500 GPS measurements on 42
ground points.
10.2 Long-term run
The position calculation has been running since
September 2003. A test tool gets the ephemeris and
ionosphere data at startup and every 10 minutes while it
is running from a NovAtel GPS receiver on the roof at
Nortel Networks Wollongong. Once the test tool has the
ephemeris and ionosphere data, it requests pseudoranges
and invokes the position calculation function to calculate
the location once per second.
The current statistics of this process are as follows. The
position calculation uses the pseudorange from all the
satellites in view, which is generally 8 to 12 satellites:
• Total number of position calculations: 30717770
• Average error in metres from ground truth: 4.08
• Standard deviation of the error in metres from
ground truth: 2.38
• RMS error: 4.72
• Minimum number of metres of error from the
ground truth: 0.001
• Maximum number of metres of error from the
ground truth: 62.8
11. Conclusion
In this paper, a simple process has been described for
improving the quality of GPS acquisition assistance data
supplied to a handset. A test bed to facilitate the testing
has been introduced and a series of experiments to show
how acquisition assistance data can enable a fast location
fix in open-sky conditions have been described. Some
results of our position calculation are also given.
With GPS acquisition assistance data, the hardware can
lock on to satellites up to 200 times faster than without
GPS acquisition assistance data. The usefulness of
acquisition assistance data degrades when the location of
the handset is not known to within 100 kilometres.
The process for improving the quality of the GPS
acquisition assistance data proved useful in identifying
problems with the A-GPS Trainer firmware. Once these
problems were fixed, the GPS acquisition assistance
provided to the A-GPS enabled it to lock on to all of the
satellites in view very quickly. This demonstrates that the
GPS acquisition assistance that is supplied to the A-GPS
Trainer is valid. It also validates our interpretation of the
3GPP specification against an independent
implementation. This will reduce the chance of
interoperability problems with A-GPS enabled cellular
handsets.
A new test bed has been recently acquired which includes
a Motorola FS Oncore development kit. The Motorola
hardware is much more typical of the hardware in a
cellular phone. The hardware is being interfaced to our
test bed and will be the basis for further tuning of our
GPS acquisition assistance data generation algorithms. It
is also more portable, so weak signal conditions will be
analysed. We will also have the opportunity to compare
response times between MS-Based A-GPS and MS-
Assisted A-GPS.
Acknowledgements
Chris Rizos (UNSW) for assisting with numerous
technical issues, especially when the position calculation
function was being developed. Martin Dawson (Nortel)
for his suggestions for the focus of this paper and
management support. Stewart Needham (Nortel) for
valuable suggestions about the draft and management
support. David Evans (Nortel) for management support.
John Walsh (Nortel) for editing the final version of the
paper.
References
3GPP TS 04.31 3rd Generation Edge Project; Technical
Specification Group GSM/EDGE Radio Access Network;
Location Services (LCS); Mobile Station (MS) – Serving
Mobile Location Centre (LCS) Radio Resources LCS
Protocol (RRLP)
3GPP TS 25.453 3rd Generation Edge Project; Technical
Specification Group Radio Access Network; UTRAN Iupc
interface Positioning Calculation Application Part
(PCAP) signalling
3GPP2 C.S0022-0-1 3rd Generation Partnership Project 2;
Position Determination Service Standard for Dual Mode
Spread Spectrum Systems (PDDM)

142 Journal of Global Positioning Systems
Djuknic G, Richton R (2001), Geolocation and Assisted GPS,
IEEE Comminications, February 2001
Harvey BR (1998), Practical Least Squares for Statistics and
Surveyors, The University of New South Wales School of
Geomatic Engineering Monograph 13
Heinrich G, Eissfeller B, Pany T, Weigel R, Ehm H, Schmid A
, Neubauer A , Rohmer G , Ávila-Rodríguez J, HIGAPS A
Highly Integrated Galileo/GPS Chipset for Consumer
Applications, GPS World September 2001
Syrjarinne J, Possibiliies for GPS Time Recovery with GSM
Network Assistance, ION GPS 2000, 19-22 September
2000, Salt Lake City, UT
Parkinson and Spilker, Eds., Axelrand and Enge, Assoc. Eds
(1996) Global Positioning System: Theory and
Applications Volume I,American Institute of Aeronautics
and Astronautics
van Diggelen F and Abraham C (2001) Indoor GPS
Technology, CTIA Wireless-Agenda, Dallas, May 2001
Zhao Y (2002), Standardization of Mobile Phone Positioning
for 3G Systems, IEEE Communications Magazine, July
2002