Development and Test Results of a Cost Effective

Inverse DGPS System

J.K. Ray, K. V. Kalligudd
Accord Software and Systems, Bangalore India

BIOGRAPHIES Therefore, DGPS was the only way to achieve acceptable

accuracy for various Intelligent Transportation System (ITS).
Jayanta Kumar Ray is a Manager of Research and
Development at Accord Software and Systems Private After the SA was turned off on May 1-2, 2000, the
Limited, Bangalore, India. He has been working in GPS- standalone GPS receiver accuracy improved dramatically
related areas since 1992. He has a Masters degree in obviating the need for differential GPS for many
Electronics Engineering from Indian Institute of Science, applications which were earlier using the differential
Bangalore and a PhD in Geomatics Engineering from the technique for improved position accuracy.
University of Calgary.
However, even without SA, the accuracy from the
Vagish Kalligudd is a Software Engineer at Accord standalone GPS receiver falls short compared to the
Software & Systems Private Limited, Banglaore, India. differentially corrected GPS receiver. The major
He has a Bachelor of Engineering degree from K.L.S remaining error sources in stand alone receiver is
Gogte College of Engineering, Belgaum, India. atomospheric delay error and residual ephemeris error.
These errors together with receiver noise and multipath
ABSTRACT makes the position error bad enough for some ITS
applications. Also, there is no guaranty that SA will not
An Inverse Differential GPS (IDGPS) system is be turned on in a future date.
developed to achieve better GPS accuracy and significant
cost savings on the air radio traffic charges by not This paper addresses a differential GPS system used for
broadcasting direct RTCM messages to the vehicles. The fleet management. The use of standalone GPS was
system consists of a Reference Station and up to two- overruled due to the accuracy requirements of the system,
hundred and fifty In-Vehicle Equipment (IVE) fitted in which could not be fulfilled by standalone GPS. An
cars. Each IVE has a low cost GPS receiver and a inverse DGPS based network was developed instead of a
communication link to send data to a Control Station, standard DGPS based network due to the cost implication
which also receives RTCM data from the Reference and suitability. The improvement in accuracy of the
Station through a leased line. The IDGPS is implemented Inverse DGPS system with respect to a stand alone
in the measurement domain. The accuracy of the IVE system is analyzed and demonstrated through static and
position is calculated before and after the range correction mobile field trials.
to evaluate the performance achievable by this system.
Initial results show that the corrected solution is INVERSE DGPS THEORY
significantly better compared to the uncorrected case,
reducing the standard deviation of the position error from Differental Global Positioning System (DGPS) is a
more than 10 m to less than 2 m. technique whereby a GPS receiver uses information from
another GPS receiver to improve the accuracy and
INTRODUCTION integrity of its solution. A conventional DGPS has
reference stations placed at surveyed locations, which
Prior to May 2, 2000, Selective Availability (SA) was the simultanously track the GPS satellites and generate range,
major source of measurement error in standalone GPS phase and/or range rate corrections. These corrections are
receiver. By using differential technique the Selective broadcast to the DGPS users using a convenient
Availability error could be removed almost completely communication link. The DGPS users are often mobile
except for the spatially decorrelated part of the error.

ION GPS 2001, 11-14 September 2001, Salt Lake City, UT 2618
receivers, which apply those corrections to obtain more The errors that are correlated in measurements from two
accurate position, velocity, time solutions. receivers simultaneously tracking a satellite, can be
reduced by taking the single difference of the range
As the name suggests, the Inverse DGPS concept is the observation equations for a single satellite and two
opposite of DGPS, whereby the mobile receivers send receivers. This is shown in Figure 1 and is given by:
raw range, phase and/or range rate measurements or the
computed position to the reference stations. The reference ∆P = ∆ρ + ∆dρ + c∆dT + ∆d ion + ∆d trop + ∆d hw + ∆ε p + ∆ε Mp
staions, which simultanously track the satellites apply (2)
correction to the data from the mobile receivers to obtain
high accuracy position and velocity of the mobile where ∆ represents a between-receiver single difference.
receiver. Unlike DGPS system, accurate position of the
mobile receiver is known only at the reference station. In Equation 2 the satellite clock error term has
The mobile receiver does not have the accurate position disappeared, as it is the same for the two receivers at a
information. Therefore, this has applications in vehicle particular time epoch. Other errors have now become the
tracking, fleet management, emergency messaging difference of errors in the two receivers. As a result, a
applications. high degree of correlation of errors in the two receivers
results in cancellation of the error in the differenced
The following sections breifly describe GPS raw equation. For a short baseline, the orbital error,
measurements and various error sources in the ionospheric delay error and the tropospheric delay error
meaurements. It then give a brief description of using are highly correlated, so the residual error can be assumed
code measurements and position information from a to be very small. The receiver clock bias, multipath and
mobile receiver to compute Inverse DGPS position at the receiver noise, however, do not cancel. The hardware
reference station. delay error, which is completely receiver dependent, is
likely to be cancelled if both the receivers are of the same
GPS Measurements type, and from the same manufacturer. Under these
circumstances, multipath error is the most dominant
The range measurement from a receiver contains various source of error in the single differenced measurements.
other small error components and is given by (Wells, 1987): For a long baseline, however, the residual orbital,
ionospheric, and tropospheric errors become more
P = ρ + dρ + c(dt − dT ) + dion + d trop + d hw + ε p + ε Mp (1) significant compared to multipath errors.
where
P is the measured code range (m)
ρ is the geometric range between the satellite
and receiver antennas (m)
dρ is the orbital error, nominal and SA (m)
c is the velocity of light (m/s)
dt is the satellite clock error with respect to
GPS time, nominal and SA (s)
dT is the receiver clock error with respect to
GPS time (s)
dion is the ionospheric delay error (m)
dtrop is the troptospheric delay error (m)
dhw is the hardware delay in the satellite and in
the receiver (m) Monitor Remote
εMp is the code range multipath error (m), and Receiver
Receiver
εP is the receiver code noise (m).
Figure 1: Between-receiver single differencing
Inverse DGPS Using Pseudorange
If the mobile receiver sends raw pseudorange
Many of the errors in the observation Equation are measurements to the reference station, the reference
spatially correlated between receivers tracking a satellite station uses Equation 2 to generate differential
simultaneously. This is because those errors are satellite observations. It may then determines the differential
dependent, or caused by atmospheric propagation and position of the mobile receiver using weighted least
therefore common for two receivers on earth separated by square as shown in the following expression:
a short distance. Often the degree of correlation between
errors at two receivers is a function of the baseline length.
(
∆x = H TP R P H P )H
T T −1
P R P ∆P (3)

ION GPS 2001, 11-14 September 2001, Salt Lake City, UT 2619
where, the position at the mobile receiver, then it is better not to
∆x is the position and clock bias error vector correct the position at that particular epoch.
HP is the dynamic matrix or direction cosine
matrix Due to the above mentioned constraints, Pseudorange
RP is the measurement error covariance matrix based Inverse DGPS was the preferred technique used for
this Inverse DGPS System.
The position and clock bias vector has elements
corresponding to 2D or 3D position error, either in INVERSE DGPS ARCHITECTURE
Cartesian co-ordinate system or curvilinear co-ordinate
system. In the curvilinear co-ordinate system, it is given by: CET Technologies Pte Ltd and Accord have jointly
developed this IDGPS System for vehicle tracking in
∆x = [∆ϕ ∆λ ∆h ∆T ]T (4) Singapore. The system consists of a Reference Station and
up to two-hundred and fifty In-Vehicle Equipment (IVE)
fitted in cars. In addition to that a Control Station was set up
where, where the Inverse GPS and fleet management software was
∆ϕ is the latitude error (m) installed. If the Control Station was on a high rise building
∆λ is the longitude error (m) with open sky, then the reference station could have been
∆h is the height error (m), and installed on the roof top of the same building. But, here, this
∆T is the clock bias error (m). was not the case. Therefore, the Reference Station was set up
on a different building which is moderately high rise and at
These position error when added to the reference station around 10 miles from the Control Station. A Leica MX9250
position, gives the position of the mobile receiver. receiver was set up as the Reference Station. The antenna
with choke ring was placed at the highest point of the
Then the design matrix for i number of common satellites building at the top of a shaft to maximize visibility and mini-
between the mobile and the reference receiver is given by mize multipath effects. The data from the Reference Station
the following expression: was connected to a Modem which was sending the data
through a dedicated telephone line at a fixed 9.6K baud rate.
 δ∆P0 δ∆P0 δ∆P0 δ∆P0  The other side of the dedicated telephone line was connected
 δ∆ϕ δ∆λ δ∆h δ∆T  to another Modem in the Control Station. The output of the
 Modem was connected to the PC hosting the Inverse DGPS
HP =   (5)
  software. Figure 2 shows a block diagram of the set up.
 δ∆Pi δ∆Pi δ∆Pi δ∆Pi 
 δ∆ϕ δ∆λ δ∆h δ∆T  The mobile IVE has a low cost GPS receiver and a
communication link to send data to a Control Station. In
the current configuration no dead reckoning sensor was
The measurement covariance matrix (RP) is generally a
used in the IVE. The IVE sends the data to the Control
function of the satellite elevation angle.
Station using Mobitex NetworkTM. Mobitex is a packet
Inverse DGPS Using Raw Position switched network for mobile data communication. Data
transmission over the Mobitex is both secure and
The raw position of the mobile receiver may be sent to the efficient compared to a circuit switched network. A
reference station, instead of raw measurements. It is then Mobitex radio modem is used at both the mobile receiver
necessary to send the satellite numbers that were used to and the Control Station to communicate over the network.
compute that raw position. If a position filter was used at
the mobile receiver, then the weight or coeffecient of the The Reference Station sends corrections for the satellite
filter also needs to be sent to the reference station. In measurements in RTCM format. The IVE sends raw pseu-
addition, if the mobile receiver uses a ionosphereic or dorange measurements and receiver derived position in a
tropospheric model for correction, then the reference predefined format at a nominal adjustable rate of once in 30
station needs to have the knowledge of which model was sec. The IDGPS is implemented in the measurement
used at the mobile receiver. domain, wherein in the corrections for satellite range
measurements are extracted from the Reference Station
As the reference station is generally established in a data and applied to the range measurements from the IVEs.
location with good satellite visibility, generally the As a result the common errors between the Reference
reference station has measurements from all the satellites Receiver measurements and the Remote Receiver
that were used in the position computation at the mobile measurements are cancelled out to give more accurate
receiver. If, however, the reference receiver does not have measurements. The accuracy of the IVE position is cal-
measurement from a satellite that was used to determine culated before and after the range correction to evaluate the
performance achievable by this system.

ION GPS 2001, 11-14 September 2001, Salt Lake City, UT 2620
 Mobitex Network

Mobitex
In-Vehicle

Modem
Equipment

Central
DGPS Station
Reference Leased with
Inverse

Mobitex

Mobitex
Station
Modem
line

Modem
In-Vehicle
Modem

Modem
(Leica DGPS
module Equipment
MX9250)

Mobitex
In-Vehicle

Modem
Equipment

Figure 2: Inverse DGPS System architecture

INVERSE DGPS DATA FLOW t0 is the time of measurement of reference station

data (sec).
The Leica MX9250 is configured in Reference Station
mode to transmit RTCM Type 1 messages and satellite Figure 3 shows the data flow diagram of the Inverse
ephemeris data at the baud rate of 9600 bps. RTCM Type DGPS system.
1 message contains PRC (pseudo-range), RRC (range-
rate corrections), IODE (issue of data ephemeris) used to
generate the corrections, for all the tracked satellites.
D ata fro m
The data from the mobile receiver is received in a format R ef
called Vehicle Movement Information (VMI). VMI S tatio n:
includes the following information: PRN
PRC
• Vehicle ID (identified by the MAN ID of the RRC
Mobitex modem) IO D E and C o rrected
• Vehicle position (latitude, longitude and altitude and ephem eris A p ply DG PS
GPS time of fix) D ifferential p ositio n
• Satellite ID and pseudorange for all tracking satellites C o rrectio ns and q uality
ind icators
The differential corrections from the reference station are o f p ositio n
D ata fro m
applied to the pseudorange data from the IVEs using the
Waypoint Inverse DGPS DLL. The data from the
M o bile
mobile unit and the reference station is aligned in time receiver:
before it is fed to the DLL using the following. PRN
p seud orange
PRC(t) = PRC(t0) + RRC*(t - t0) (6) p ositio n
fix tim e
Where,
t is the time of measurement of mobile data
(sec), and Figure 3: Inverse DGPS data flow diagram

ION GPS 2001, 11-14 September 2001, Salt Lake City, UT 2621
The IDGPS system is implemented using Microsoft Position Error Spread Before IDGPS
120
Visual C++ 5.0. The system consists of three threads: the
first, to handle user interface events, the second reads raw 100
bytes from the serial port in real time and stores into a
80
circular buffer, and the third to extract the useful data
from the raw stream of bytes stored in the circular buffer. 60
The data from the mobile receiver is received, through a

Longitude Error (m)

socket connection, from a different module, which is a 40

part of the fleet management system. The user interface 20

displays the status of the corrections made by the IDGPS
system to the GPS data from the mobile units. The 0
number of corrections made hourly and the reasons for
-20
non availability of IDGPS system are also displayed.
-40
TEST DESCRIPTION
-60

Static and mobile tests were conducted to assess the -80

performance of the system. The following describe various -40 -20 0 20 40 60 80 100 120 140
Latitude Error (m)
results and statistics that were obtained from the tests.

Static Tests Figure 4: Latitude vs. Longitude errors before IDGPS

correction (September 14)
During the Static test, the IVE was placed on the top of
the CET building and data was continuously sent through
Position Error Spread Before IDGPS
the network. The position of the IVE was known with 60
fairly good accuracy by averaging the GPS derived
position for approximately 1 day duration. The Vehicle
Movement Information (VMI) data from a IVE was sent 40
at an approximate rate of 2 solution /minute. Inverse
differential correction was applied at the Control Station
using the data from the Reference Station. The test was 20
Longitude Error (m)

carried out on September 14 and 15, 2000.

Figures 4 shows the x-y plot of latitude vs. longitude error 0

before IDGPS correction for the data collected on
September 14. The latitude error (ϕe) and longitude error
-20
(λe) were computed by subtracting the Mean of latitude
and longitude from each sample of the incoming data. The
plot shows that the latitude and longitude errors are -40
generally clustered between –15 m and + 15 m. However,
there are some large position error spikes which are due to
large measurement errors probably caused by mulitpath -60
-20 0 20 40 60 80 100 120
and/or bad geometries. Similar observations can be made
Latitude Error (m)
for the uncorrected input data collected on September 15
as shown in Figure 5.
Figure 5: Latitude vs. Longitude errors before IDGPS
Figure 6 shows the x-y plot of latitude vs. longitude error correction (September 15)
after IDGPS correction for the data collected on September
14. Figure 7 shows similar plots for the data collected on
September 15. From both these plot it can be seen that the
position errors are much smaller compared to the
uncorrected data and restricted to –5 m and +5 m most of
the cases. The spreads of the latitude and longitude errors
as well as the maximum horizontal error are well below the
uncorrected case with all the large spikes removed.

ION GPS 2001, 11-14 September 2001, Salt Lake City, UT 2622
 Position Error Spread After IDGPS errors can be as large as over 100 m. It is also clear that
15 the largest number of samples have an error of around
5 m. The statistics computed from these histograms
10 show that the number of samples having less than 5 m
of horizontal position error is less than 50 percent. The
5 standard deviation of the position error is above 10 m.
Longitude Error (m)

-5

-10

-15

-20
-10 -8 -6 -4 -2 0 2 4 6 8 10
Latitude Error (m)

Figure 6: Latitude vs. Longitude errors after IDGPS

correction (September 14)

Position Error Spread After IDGPS

10

Figure 8: Hor. position error histogram before IDGPS

correction (September 14)
Longitude Error (m)

-5

-10

-15
-8 -6 -4 -2 0 2 4 6 8 10 12
Latitude Error (m)

Figure 7: Latitude vs. Longitude errors after IDGPS

correction (September 15)

Figures 8 and 9 show the histogram of the horizontal

position error before IDGPS correction for the data
collected on September 14 and 15 respectively. The
horizontal position error is computed using the following:

Horizontal position error = ϕ e2 + λ2e (7) Figure 9: Hor. position error histogram before IDGPS
correction (September 15)
where,
ϕe is the latitude error (m), and
Figure 10 and 11 show the histogram of the horizontal
λe is the longitude error (m) position error after IDGPS correction for the data
collected on September 14 and 15 respectively. From the
The X axis represents the horizontal position error and figures it is clear that the largest number of samples have
the Y axis represents the number of samples for a an error of around 2 m. The statistics computed from
particular error. From the figures it is evident that the these histograms show that over 90 percent of the samples

ION GPS 2001, 11-14 September 2001, Salt Lake City, UT 2623
have horizontal position error less than 5 m. The standard delay i.e. ionospheric and tropospheric delays and have
deviation of the position error is below 2 m. These slow variations. The high frequency components of the
statistics are well within the desirable error limits errors, which are very small, are due to ionospheric
expected out of the IDGPS system. This clearly indicates scintillation, multipath at the reference site and receiver
the effectiveness of the IDGPS system in correcting the noise. These atmospheric errors are common to the
errors from the raw measurements. reference receiver and any receiver within a short distance
(say up to 50-100 KM). Such large errors in range would
introduce large position errors, if remained uncorrected.
The IDGPS system effectively corrects these common
errors and improves the performance of the system
significantly as evident from the above analysis.

IDGPS Correction to Satellites 15, 20, 25 and 29 Ranges

5

Satellite 25
-5

Range Correction (m)

Satellite 20

-10
Satellite 29

-15

Figure 10: Hor. position error histogram after IDGPS -20 Satellite 15

correction (September 14)

-25
3.81 3.82 3.83 3.84 3.85 3.86 3.87 3.88 3.89
GPS Time (sec) 5
x 10
Figure 12: Range error corrections from the reference
station (September 14)

Table 1 shows the improvement that was achieved by

applying IDGPS correction to the uncorrected GPS data.
It is evident from the Table that Inverse DGPS has
improved the accuracy of the solution substantially.

The availability of the IDGPS solution is less than 100

percent due to large DOP in some cases. The large DOP is
due to less number of common satellites between the
reference station and remote receivers. The IDGPS applies
correction only to those satellites, which are common in the
reference and remote receivers. The reference station is set
to have a cut-off elevation angle of 10 degrees to avoid
large errors in the range measurements from the low
elevation satellites. The remote receiver however, tracks
Figure 11: Hor. position error histogram after IDGPS some of those low elevation satellites. As a result
correction (September 15) sometimes the number of common satellites between the
reference and remote receivers remains less, even when the
Figure 12 shows the corrections sent by the reference remote tracks more satellites. To overcome this problem,
station for various satellites based on the data collected on the cut-off angle of the reference receiver is to be set to 5
September 14. degree or lower. This would improve the availability. But
the IDGPS accuracy may deteriorate to some extent in that
It can be seen that the range errors can be as large as 20 m case.
in some cases. These errors are mainly due to atmospheric

ION GPS 2001, 11-14 September 2001, Salt Lake City, UT 2624
 Table 1: Performance improvement due to the use of IDGPS correction in static tests

September 14, 2000 September 15, 2000

Item Before IDGPS After IDGPS Before IDGPS After IDGPS
Duration 15 hours 15 hours 15.8 hours 15.8 hours
No. of samples 1744 (100%) 1681 (96.4%) 1819 (100%) 1801 (99.0%)
Sample frequency 30 sec/sample 30 sec/sample 30 sec/sample 30 sec/sample
Standard Deviation of Horizontal 14.3 m 1.9 m 10.7 m 1.6 m
Position Error
Samples With Horizontal Position 38% 90% 49% 92%
Error Less Than 5 m
Spread in Latitude 175.8 m 18.2 m 119.4 m 18.5 m
Spread in Longitude 172.3 m 29.9 m 106.6 m 20.5 m
Maximum horizontal error 174.8 m 19.0 m 117.6 m 12.2 m

Further static tests were carried out with the IDGPS as the true position of the IVE was not known. The
system after tuning various thresholds. The performance positions were eventually matched on to a map, but the
of the IDGPS system repeated by reducing the spreads in accuracy of the map itself is approximately 7 m. In
latitude, longitude and the maximum horizontal error addition, the map may correspond to the center of the road,
before correction. The availability of the IDGPS was also whereas the car might have been driven on the side-lane of
improved. the road. Therefore, true position of the IVE at the instant
of GPS and IDGPS solution time was not available.
Mobile Test Therefore the analysis is done based upon the level of
consistency of the solution before and after IDGPS.
A mobile test was carried out in the downtown of
Singapore on September 20, 2000. A car, fitted with an From the table it can be seen that IDGPS was able to correct
IVE was driven in the city. Data was collected for 84 percent of the solutions sent by the mobile receiver. The
approximately half an hour at a rate of 10 sec per sample. cause of not correcting in the remaining cases is mainly due
The IDGPS corrected the raw data from the mobile IVE. to poor DOP. The standard deviation of the position solution
The map matching software used the corrected solution, before and after IDGPS correction for only those samples
whenever it was available and uncorrected solution, when for which correction was made is about 13 m.
corrected solution was not available. Table 2 shows the
results of the mobile test. Figure 13 shows the trace of the field trial. The blue trace
(dark shade) is before IDGPS correction and the green
Table 2: Mobile test results trace (light shade) is after IDGPS correction. The '*'
symbol indicates the GPS or IDGPS derived position. The
September 20, 2000 starting point of the field trial is assumed to be origin of
the X-Y plot. The X axis of the plot shows the East-West
Item Before IDGPS After IDGPS spread of the trace the Y axis of the plot shows the North-
Duration 1/2 hour 1/2 hour South spread of the trace.

No. of samples 70 (100%) 59 (84.3%) From the figure it could be seen that the trace is quite well
defined by the GPS and IDGPS solutions. The IDGPS
Sample 10 sec/sample 10 sec/sample
solutions are close to the GPS solutions most, but a few
frequency cases. As during the static cases it was proven that the
Position 12.89 m IDGPS position is quite accurate, the difference between
difference before the GPS and IDGPS position can be considered to be the
and after IDGPS error in the GPS position.

The table does not contain statistics about the absolute

accuracy of the solution before and after IDGPS correction,

ION GPS 2001, 11-14 September 2001, Salt Lake City, UT 2625
 Position Trace Before (Blue) and After (Green) IDGPS Difference of Position Before and After IDGPS
400 70

200
60

0
50
-200

Position Difference (m)

East-Weast (m)

-400 40

-600
30

-800
20
-1000

-1200 10

-1400
0
2.848 2.85 2.852 2.854 2.856 2.858 2.86 2.862 2.864 2.866
-1600 GPS Time (s) 5
-200 0 200 400 600 800 1000 1200 1400 x 10
North-South (m)
Figure 14: Position error differences between solutions
Figure 13: Position trace before and after IDGPS before and after IDGPS correction during the mobile
correction during the mobile test on September 20. test
Figure 14 shows the difference between the GPS and the ACKNOWLEDGEMENTS
IDGPS position during the test.
The authors would like to thank CET Technologies Pte
The X-axis of the plot shows the GPS time, whereas the Y Ltd (www.cet.st.com.sg) and the Land Transport
axis shows the difference. It can be seen that the Authority of Singapore (www.lta.gov.sg) for making this
difference between the two solution is less than 20 m paper possible.
most of the cases. However, in some cases the difference
could be as large as 60 m. This is also evident in the REFERENCES
traces in Figure 14. This indicates that the GPS position
has such large errors in some cases and the IDGPS is able 1. Parkinson, B.W and J.J. Spilker Jr. (1996),
to correct them. This confirms the effectiveness of the Global Positioning Systems: Theory and
IDGPS system in mobile condition as well. Applications, Vol. I and Vol. II, American
Institute of Aeronautics and Astronautics,
CONCLUSIONS Washington DC.
An Inverse DGPS system was developed which showed 2. Wells, D.E., N. Beck, D. Delikaraoglou, A.
signification improvement in accuracy compared to stand Kkeusberg, E.J. Krakiwsky, G. Lachapelle, R.B.
alone GPS receiver. The IDGPS solution was found to be Langley, M. Nakiboglu, K.P. Schwarz, J.M.
less than 5 m in over 90% of the times compared to less Tranquilla, P. Vanicek (1987), Guide to GPS
than 50% times in case of stand alone GPS. This has also Positioning, Canadian GPS Associates,
reduced the standard deviation of the horizontal position Fredericton, N.B.
error from over 10 m before correction to less
approximately 2 m after correction. These test results 3. Test Report on Inverse Differential Global
suggest that there is still a need for differential correction Positioning System (2000), CET Technologies
even in the absence of SA error, if the required accuracy Pte Ltd., Singapore, September 22.
from the system is to be good.
4. Manual of Inverse DGPS DLL from Waypoint
Consulting Inc.

ION GPS 2001, 11-14 September 2001, Salt Lake City, UT 2626