Journal of Global Positioning Systems (2005)
Vol. 4, No. 1-2: 2-11
Using Multiple Reference Station GPS Networks for Aircraft Precision
Approach and Airport Surf a c e N a v ig a tion
Ahmed El-Mowafy
Department of Civil a n d Environmental Engineering, UAE University, P.O.Box 17555, Al Ain, UAE
e-mail:Ahmed.Mowafy@ u ae u .ac.ae; Tel: + 97150 7535840; Fax:[phone]
Received: 6 December 2004 / Accepted: 16 October 2005
Abstract. The use of multiple real-time reference stations
(RTK Networks) for positioning during the aircraft’s
precision approach and airport surface navigation is
investigated. These existing networks can replace the
proposed airport LAAS systems and have the advantage
of improving coverage area. Real-time testing of the
proposed technique was carried out in Dubai, UAE, with
a helicopter and a small fixed-wing aircraft using a
network known as the Dubai Virtual Reference System
(DVRS). Results proved the feasibility of the proposed
approach as they showed that cm to sub-meter
positioning accuracy was achieved most of the time. For
some periods, only meter-level positioning accuracy was
available due to temporary breaks in reception of the
network carrier-phase corrections. Some solutions to
improve availability o f the corrections are discussed. It is
also proposed to integrate the GPS with an IMU. The
inertial system aids positioning during periods when the
corrections are lost, as well as providing attitude
information. The GPS and IMU systems were integrated
using a decentralized adaptive Kalman filtering
technique. The measurement noise covariance matrix and
the system noise matrix are adaptively estimated, taking
the aircraft dynamics changes into account. Tests of the
integrated system show that it has a good overall
performance, and navigation at categories III and II can
be achieved during short outages of RTK-GPS network
corrections.
Key words: Global Positioning, Airborne Navigation,
Wide Area Networks, Adoptive Systems, INS
1 Introduction
Interest in the use of Global Navigation Satellite Systems
(GNSS) as a main source of navigation reference is
increasing. The system employed for such a purpose
should be capable of meeting the strict requirements of
air navigation in terms of accuracy, availability, integrity,
and reliability. At present, the accuracy requirements for
all flight categories up to precision approach are
summarized in Table 1 (Whelan, 2001). The accuracy
requirement for Category I can be achieved most of the
time using wide area differential systems such as the
American “WAAS”, the European “EGNOS”, and the
Japanese “MSAS”. To meet the most demanding
accuracy for categories II and III, which involve the final
and precision approach phases of flight, more accurate
systems are needed. The American Federal Aviation
Authority (FAA) has undertaken the development of a
navigation augmentation system based on GPS in the
form of a Local Area Augmentation System (LAAS).
This LAAS is designed to enable precision approach
navigation within the airp ort area. It includes at least four
reference GPS receivers located at each airport. GPS
measurements are collected from the four reference
stations and processed in real time in a control computer.
Next, GPS differential corrections are sent to aircraft to
compute their locations for navigation at the sub-meter
level of accuracy. Corrections are provided via a Very
High Frequency (VHF) radio link from a ground-based
transmitter. LAAS preliminary test results have generally
demonstrated accuracy of less than 1 meter in the
horizontal and vertical axes. However, the percentage of
system availability is still under evaluation to see if it can
meet the FAA requirements. The cost of establishing
LAAS for major airports is also expected to be
significant.

El-Mowafy: Using Multiple Reference Station GPS Networks for Aircraft Precision Approach 3
Tab. 1 Positioning accuracy requirements for all flight categories
horizontal vertical
Category I 17.1 m 4.1 m
Category II 5.2 m 1.7 m
Category III
(precision approach) 4.1 m 0.6 m
This study proposes the use of Real-Time Kinematic
(RTK) multi-station reference networks, as an alternative
to the airport LAAS, to aid accurate positioning of
aircraft during precision approach, takeoff and airport
surface navigation. The feasibility of this approach, the
problems experienced in practice and possible methods
for improving the overall performance are investigated in
this paper. The investigation is carried out on a typical
RTK network, bearing in mind that the presented results
are dependent, to some extent, on the network design and
operational features.
2 Using the multi-station RTK networks for airborne
navigation
GPS RTK multi-station reference systems were originally
developed for surveying applications. The basics of this
type of reference systems are discussed in Enge et al.
(2000), Raquet & Lachapelle (2001), Hu et al. (2003 ) and
El-Mowafy et al. (2003). In principle, observations from
multiple reference stations covering a large area are
gathered and processed in a common network adjustment
at a central processing facility and measurement
corrections are computed. The corrections are optimized
for the coverage area to account for distance dependent
errors. A single rover GPS receiver receives these
measurement corrections from the control centre of the
network and uses the corrections to estimate its position
in real-time accurate to the cm-level with fixed integer
carrier-phase ambiguity resolution, or to the sub-meter
level with a float solution. Currently, the RTK network
approach is mostly used in static or kinematic ground
applications. In this study, the use of these networks in
airborne navigation is considered, where the rover
receiving the network corrections is mounted on the
aircraft to determine its positions during flight.
2.1 Advantages of Using RTK Networks in Airborne
Navigatio n
The main advantages of using multi-station reference
RTK networks for precise airborne navigation can be
summarized as follows:
- Due to the fact that multi-station reference networks
usually have an area of coverage that extends to several
tens or hundreds of kilometres, each network can cover
more than one airport, including small airports, unlike the
airport LAAS. In addition to airport navigation, the
system can be used in search and rescue operations,
emergency landing, road traffic control from the air, as
well as emergency response.
- RTK networks provide cm to decimetre positioning
accuracy even in the case of malfunctioning of some
stations, particularly for dense networks. This situatio n is
however more critical in airport LAAS due to the low
number of rece ivers used.
- Compared to LAAS, no significant additional
infrastructure cost is involved as the hardware and
software of the GPS-RTK networks are available in most
developed countries and the establishment of new
networks is currently underway (or planned) in different
regions worldwide.
- RTK networks can give better runway utilization by
improving airport surface navigation. They can also
enhance air traffic management by increasing dynamic
flight planni n g.
2.2 The DVRS Network as an Example
The feasibility of using real-time reference networks to
provide precise positioning navigation information for
aircrafts is examined in this study. A network known as
the Dubai Virtual Reference System (DVRS), located in
Duabi, UAE, was used for this investigation. The focus
was on various aspects of aircraft navigation including
precision approach, takeoff and airport surface
navigation. For accurate determination of aircraft heights
from the ground using GPS-derived ellipsoidal heights, a
recently established accurate geoid model for Dubai was
utilized. The Dubai geoid model was developed from
varying data sources, mainly: gravity measurements, a
digital elevation model (DEM), orthometric heights and
GPS-observations at levelling benchmarks.
The DVRS network consists of five active reference
stations, with baseline lengths varying between 23.4 km
and 90.8 km. The main software used in the processing of
the DVRS data utilizes the area parameter method (FKP)
to estimate and represent the state of individual GPS
errors in real time. All stations of the network are
processed simultaneously using un-differenced
observables. Therefore, all error components including
clock errors are estimated. The state vector (X
G
) used in
the Kalman filtering process can be given as:
X
G
= (i x
G
, s
i
N , δti , δts, δs O
G
, δs
i
T, δs
i
I, δs
i
M)Τ (1)

4 Journal of Global Positioning Systems
where i
x
G is the position vector, δti and δts denote the
receiver and clock errors,s
i
N is the ambiguity, δs
O
G
,
δs
i
Tand δs
i
I represent the distance dependent errors (the
orbital, tropospheric, and ionos pheric errors respectively),
and δs
i
M is the multipath error. To compute its position,
the rover receiver sends its approximate position via a
cellular modem to the network control centre where
computations are carried out for each user. The estimated
network measurement corrections, mainly the distance
dependent errors, are interpolated for a virtual reference
station (VRS) close to the rover position and instantly
sent to it. The predicted distance dependent error term (δi)
at the VRS position (i) from the reference station (j) with
respect to the satellite (s) can be expressed in the
functional form:
δi = f(s
j
FKP , ∆φji, ∆λji, ∆hji) (2)
where ∆ is the differential operator, φ, λ, and h denote the
latitude, longitude and height respectively, and
s
j
FKP represents the FKP computed error. The
corrections at the VRS station (s
ji
VRS ), which are used
to correct the observations at the rover receiver, can be
expressed as follows:
s
ji
VRS = s
ji
CR + δi + ∆Tji (3)
where s
ji
CR denotes the corrected carrier phase
observations of the reference station computed from the
network solution, and ∆Tji represents the difference in
tropospheric modeling between processing of the network
at the reference station and processing of the virtual
reference station. For in-depth mathematical formulation
of this method, interested readers may refer to Wübbena
et al. (2001). Previous testing of the DVRS system for
kinematic ground surveying showed that system
positioning accuracy was typically 1-2 cm in planimetry
and 3-5 cm in altimetry (El-Mowafy et al., 2003).
2.3 Concerns and Recommendations in Using the
DVRS Network for Airborne Navigation
When applying the VRS technique to airborne
navigation, the aircraft rover receiver uses a ground VRS
station. The drawback is that continuously updated
approximate coordinates have to be used for the VRS
computation. This is similar to having a moving reference
station. A system reset should thus be frequently
performed when the VRS coordinates are changing,
which will result in frequent initialization of the carrier-
phase ambiguities. Therefore, it is preferable to keep the
VRS location for the longest possible range and apply
appropriate extrapolation. This can, however, affect the
performance of the system. In addition, the duplex
communication approach used for the DVRS network
puts a restriction on the number of users, as this number
is limited by the ability of the control centre to
simultaneously perform calculations for different users.
As this number grows, extended latency in receiving the
corrections may result.
These problems can however be alleviated in the
implementation phase of the system in aviation by using
a one-directional communication method. In this case,
one or two ground transmitters (repeaters) at the airport
will be established; they will receive the reference-station
measurement corrections from the control centre on-line
and send them to the aircraft by means of, for instance,
VHF modems. The receiver on board the aircraft will
then be responsible for interpolating the corrections at its
location and processing the measurements to estimate its
position. Thus, the rover can independently use its own
interpolation and processing models, and no restrictions
exist on the number of users. For faster and continuous
prediction of the corrections at the rover location, it is
recommended that the software computes a particular set
of aviation corrections sent to the airport transmitters,
emphasizing the airport area with a preset radius (e.g. 30-
40 km). The current architecture of the DVRS
communication with the user can, however, be kept to
serve ground-based surveying applications. Hence, both
types of communication can be used to serve different
applications (aviation and surveying), using the same
infrastructure of the real-time reference stations.
The establishment of ground transmitters at the airport
can also improve the current availability of the
corrections to the rover receiver, as breaks in receiving
such corrections frequently take place. Another
recommendation is to integrate GPS with an inertial
system. More details will be given in a following section.
Since the proposed system is based on the use of satellite
measurements, the integrity of the system and continuo us
reception of the corrections are primary concerns. These
items should be continuously monitored, and methods
such as RAIM should be implemented to warn the pilot
against any deficiency in the system. Other concerns in
the use of RTK networks in airborne navigation include:
- Due to the high dynamics involved in airborne
navigation, a high update rate of sending the corrections
is needed compared with that implemented for land
applications, which is usually 5-70 seconds for current
networks. This rate has a direct impact on the Time-To-
First-Fix of phase ambiguities, and thus on the overall
positioning feasibility and accuracy (El-Mowafy, 2004).
- The format of GPS measurement corrections should be
standardized to ensure that the system is independent of

El-Mowafy: Using Multiple Reference Station GPS Networks for Aircraft Precision Approach 5
any single receiver manufacturer. The use of the RTCM
standard for RTK multiple reference stations v3.0 is thus
recommended, see Euler et al. (2004).
- The need to ensure the security of the reference station
locations: these stations should be safe and unreachable
by the public in order to prevent possible tampering.
- The possibility that the airport authorities share control
of the system with surveying authorities is recommended.
3. Testing the DVRS system for Airborne Navigation
3.1 Test Description
The use of the DVRS network for aircraft navigation in
the airport area was evaluated by conducting several
flight tests. Two types of aircraft were used for this
purpose, a helicopter and a small fixed-wing airplane. In
these tests, aircraft positions (planimetric + height) were
determined using a dual-frequency GPS rover receiver
(Leica SR530) equipped with a DVRS GSM modem
capable of receiving the DVRS corrections. The
GPS/DVRS rover receiver collected both the GPS and the
correction data during the aircraft takeoff, enroute flying,
landing and airport surface navigation. The data were
processed in real time at one-second intervals. On the
other hand, the DVRS reference stations collected data at
five-second intervals. Processing of their RTK network
corrections was also carried out at this interval. Thus, the
corrections were interpolated in time for the rover
receiver to compute the measurement corrections at the
one-second interval. The satellite window during testing
was generally normal, and 6 to 8 satellites were observed
at any moment, giving Dilution of Precision (PDOP)
values ranging from 1.4 to 3.7 at the rover receiver
locations. The GPS data were also stored in the receiver’s
internal memory for further post processing testing and
analysis after being integrated with the data from the
DVRS reference stations.
In the helicopter test, the GPS and GSM antennae were
rigidly mounted on an arm approximately 0.9 m long
extending outside th e helicopter. The arm was attached to
a frame rigidly fixed inside the helicopter. No arm
vibration was experienced during flight testing. For better
GPS as well as GSM signal reception, the GPS antenna
was mounted high on the arm for better visibility of the
sky, while the GSM antenna faced down. This
architecture was designed only for testing purposes.
Figures 1 and 2 show the system installation on the
helicopter, and the test trajectory. For the fixed-wing
aircraft test, the GSM antenna was installed inside the
aircraft, which is acceptable for GSM communication.
Figures 3 and 4 show the fixed-wing aircraft and the test
trajectory respectively. Both tests were carried out over
the city of Dubai.
Fig. 1 System instillation for the helicopter test
25.236
25.237
25.238
25.239
25.24
25.241
25.242
25.243
25.244
25.245
25.246
55.3655.365 55.3755.375 55.38
Longitude (deg. )
Lattitude (deg .)
Fig. 2 Trajectory of the helicopter test
Fig. 3 Testing using t h e f ix e d-wing aircr aft
25.18
25.19
25.2
25.21
25.22
25.23
25.24
25.25
25.26
25.27
55.34 55.36 55.3855.4
Longitude (deg. )
Latitude (deg.)
Fig. 4 Trajectory of the fi xe d -wing aircraft test
GPS Antenna
DVRS GSM
Antenna

6 Journal of Global Positioning Systems
3.2 Test Results
Figure 5 shows the helicopter test results, illustrating the
flying height and the 2-D and height positioning
accuracies achieved during testing. At the beginning of
the test and after an initial warming up period of less th an
20 seconds, the phase ambiguities were successfully
fixed. Thus, positioning accuracy was feasible at the cm
level before starting the engine. The DVRS corrections
were continuously received during takeoff until reaching
the required height (first dashed region in Figure 5),
which was approximately 145m. During the major part of
the enroute flying time, the DVRS corrections were
continuously received. However, during most of the
landing phase the DVRS corrections were lost, but were
regained after the helicopter landed (second dashed
region in Figure 5). This can be mainly attributed to the
use of GSM signals in sending the DVRS data, and
partially to changes in the helicopter dynamics. In
addition, due to changes in the VRS positions,
initialization of the phase-ambiguities was often carried
out. The change in error values from the decimetre to the
cm level, which can be observed at some instances in the
figure, can be ascribed to reaching a fixed ambiguity
solution after a float solution. In general, during the two
marked periods, the ambiguities were resolved as integers
and the average positioning accuracy, represented by
coordinate standard deviations, was 0.022 m in the
planimetric 2-D positions and 0.034 m in height.
One can, however, note that the highest accuracies
needed in airborne navigation corresponding to category
III, which are 4.1 m for 2-D positioning and 0.6m for
height determination, can generally only be achieved with
a float ambiguity resolution. For instance, for the
helicopter test, during the periods where the DVRS
corrections were received but the ambiguities were
resolved in a float solution, the positioning accuracy was
at the sub-meter level, as shown in Table 2. This accuracy
was on average 0.322 m for the 2-D positioning and
0.539 m for height determination. However, when the
DVRS signals were not received, the 2-D positioning and
the height errors were increased to more than 3.5 m,
which are only suitable for navigation at category I, i.e.
during the enroute flying.
Tab. 2 Average positioning accuracies (m)
Helicopter test Fixed-wing
aircraft test
2D
(E&N) Height 2D
(E&N) Height
fixed solution 0.022 0.034 0.016 0.028
float solution 0.322 0.539 0.263 0.525
all test periods 0.484 0.642 1.107 0.831
Figure 6 shows the results of the fixed-wing aircraft test.
In this test, the DVRS corrections were available during
airport surface navigation and manoeuvring to the
runway, in addition to the periods of takeoff, reaching the
designated height, landing and parking, which are shown
in the dashed areas in the figure. The DVRS corrections
were, however, lost during flying for some periods. This
can also be attributed to the use of GSM signals as the
means of communication with the DVRS centre, which
might result from changes in the aircraft dynamics as
clearly seen from the top figure. This result, together with
the outcome of the helicopter test, shows that accurate
positioning usin g RTK network correction s in real-time is
feasible during the critical phases of takeoff, landing, and
airport surface navigation. However, the use of GSM
signals for sending the RTK network corrections is not
efficient and other methods are needed. The average 2-D
and height positioning accuracies achieved during the
fixed-wing aircraft test are given in Table 2. As with the
helicopter test, when receiving the DVRS corrections and
initializing fixed phase measurement ambiguities, the
average 2-D and height positioning accuracies were at the
cm level, as they were 0.016 m and 0.028 m respectively.
When phase ambiguities were solved in a float solution,
these accuracies were 0.263 m and 0.525 m. Both cases
are sufficient for positioning in all phases of flight,
including category III. When the DVRS corrections were
lost, the 2-D and height positioning errors were more than
4 m, which can be only used for category I of navigation.
To compare results of the RTK network approach for this
particular application with the standard double
differencing technique, the phase data of the rover
receiver stored in its internal memory were processed in a
post-mission mode referenced to one of the DVRS
network reference stations. This was possible since the
tests were carried out at a distance of approximately 6 km
from this station and the flight paths were within a range
of a few kilometres. Precise IGS orbits were used. When
comparing the results of positioning obtained by the
DVRS real-time multi-station reference network with the
post-mission positions, the 3D differences were within
the range of a few millimetres to a few centimetres when
the phase ambiguities were fixed. In general, the
discrepancies were less than 7 cm.

El-Mowafy: Using Multiple Reference Station GPS Networks for Aircraft Precision Approach 7
Fig. 5 Helicopter test results
Fig. 6 Fixed-wing aircraft test results
4 Integration with the INertial system
4.1 Integration and Estimation Methodology
One method to increase the availability of the positioning
accuracy at the required level is to integrate GPS with an
Inertial Measuring Unit (IMU). Thus, for the same tests
given above, the GPS/DVRS system was integrated with
an IMU running simultaneously, with the purpose of
bridging positioning outage by the Inertial Navigation
System (INS) of the IMU during short breaks in reception
of network corrections. The data of both systems were
recorded for post mission processing and analysis. For
testing purposes and due to hardware availability, a
Honeywell tactical-grade (medium accuracy) IMU
system of approximately 1-10 degrees/hour gyro drift was
used. For simplicity, the GPS/INS integration was carried
out in a decentralized loose coupling scheme. In this
approach, the GPS and IMU (INS) filters ran
independently in parallel. The GPS filter used the rover
data and the corrections received from the multiple-
reference station network as an input to the filter. The
states are given in Equation (1), which includes the
rover's position, phase ambiguities and measurement
errors. The state vector of the INS comprises the
misalignment, position, velocity, gyro drifts and
accelerator biases. Positions determined from the GPS
filter and velocity estimates were used as an update to the
INS filter. Since real-time processing was required, no
bridging algorithms such as backward smoothing were
applied.
For the purposes of the test, positioning by the INS was
mainly investigated during bridging of the GPS
positioning outages for the short breaks in reception of
network corrections. Figure 7 shows a flowchart of the
integration scheme of the GPS/INS adopted during
0
1
2
3
4
5
6
7
8
9
10
0250 500 75010001250
Hz. E rror (m )
0
50
100
150
200
250
02505007501000 1250
Fl ying Hei ght (m
)
0
1
2
3
4
5
6
7
8
9
10
02505007501000 1250
T ime (Sec.)
Height Err or (m
)
0
1
2
3
4
5
6
7
8
9
10
0100200 300 400
Hz. Er ro r (m)
0
50
100
150
200
0100200300400
Fl ying Hei ght (m
)
0
1
2
3
4
5
6
7
8
9
10
0100200 300 400
T ime (Sec.)
Hei ght Er ror (m
)

8 Journal of Global Positioning Systems
testing. To externally evaluate the performance of the
INS positioning during the network correction outages,
the rover receiver kept on collecting GPS observations,
which were processed in a post-mission mode referenced
to one of the DVRS network reference stations to
compute position data that were compared with the INS
results. Apart from this testing purpose, positioning
information can be generally acquired from the IMU in
an integrated GPS/INS system to benefit from its high
frequency output. In addition, the INS is useful for
determination of the attitude information of the aircraft,
as well as cycle slip detection and repair, and ambiguity
resolution, if a centralized filtering scheme is used.
Fig. 7 Flowchart of GPS/INS integration fo r testing purposes
The mathematical models used in the filtering estimation
approach can be written in matrix form as:
Dynamics model: i
S
 = Fi,i-1 Si + G ui (4)
taking, for simplicity, Φi,i-1= Fi,i-1 dt + I (5)
Observation model: Mi = h(Si ) + ei (6)
where Si denotes the state vector, Mi is the measurement
vector, Φi,i-1 is the transition matrix, F represents the
dynamics matrix, dt is the prediction time interval
(ti – ti-1), and ei represents the measurement noise. G is the
design matrix and ui denotes a forcing vector function,
such that the term (G ui) represents the noise of the
dynamics model. This model for the INS is described
using a first order Gauss-Markov process.
An extended Kalman filtering approach was used to
represent the non-linear observation equations, where the
filter states become estimated corrections (δ) to an
approximate state (So) represented as a nominal time
varying state updated using filter estimation, such that:
δ = S

- So (7)
The time update (prediction) equations take the form:
δi,i-1 = Φi,i-1 δi-1 (8)
Pi,i-1 = Φi,i-1 Pi-1 ΦΤi,i-1 + Qi-1 (9)
Ki = Pi,i-1 T
i
H (Hi Pi,i-1 T
i
H+ Ri-1)-1 (10)
The measurement update (information) equations can be
formulated as:
δi = δi,i-1 + Ki ( ωi – Hi δi,i-1 ) (11)
Pi = ( I – Ki Hi) Pi,i-1 (12)
where Q and R denote the covariance matrices of the
dynamics model and the measurement model
respectively. P is the covariance matrix of the filter states,
Hi represents the partial derivatives (linearized design
matrix) derived from the observation equation, K is the
Kalman gain matrix, and ω symbolizes the measurement
misclosure.
For the INS filter, the measurement equations can be
formulated as follows:
Mi =
⎪
⎪
⎪
⎪
⎭
⎪
⎪
⎪
⎪
⎬
⎫
⎪
⎪
⎪
⎪
⎩
⎪
⎪
⎪
⎪
⎨
⎧
−
−
−
−
λ−λφ+ξ
φ−φ+ς
d
GPS
d
IMU
e
GPS
e
IMU
n
GPS
n
IMU
GPSIMU
GPSIMU
GPSIMU
vv
vv
vv
hh
)(cos)h(
))(h(
+ ei (13)
where ζ and ξ are the radii of curvature for the meridian
and prime vertical. vn, ve and vd are the velocity
DGPS
INS
bridging
RTK networ k
corrections are
available
Rover Data
Position,
velocity
accuracy
achieved Position from
GPS/DVRS
Yes
No Position
from INS
accuracy
achieved
Aircraft
position
No
Y
es
Yes
No
GPS point
positioning
high
accuracy
low accuracy
(Warning sent
to
p
ilo
t
)

El-Mowafy: Using Multiple Reference Station GPS Networks for Aircraft Precision Approach 9
components in the navigation frame axes (north, east,
down).
The measurement noise matrix can be estimated from:
R = diag ( 2
φ
σ 2
λ
σ2
h
σ 2
vn
σ 2
ve
σ2
vd
σ ) (14)
where n
v
σ, e
v
σ and d
v
σdenote the standard deviations
of velocity. The initial position and velocity standard
deviations are taken from the GPS solution.
The Q matrix can be calculated from (Shin, 2001):
Q = Φ G q GT ΦT dt (15)
where q is the spectral density matrix computed as:
q = diag ( 2
ax
σ 2
ay
σ2
az
σ 2
xψ
σ 2
yψ
σ2
zψ
σ ) (16)
the σa and σψ are the standard deviations of the
accelerometers and gyroscopes, respectively.
Both the Q and R matrices play a main role in
determining the quality of the estimated states owing to
the fact that the predicted states covariance is affected by
the Q matrix, while the update measurements covariance
is R. The change of these covariance matrices reflects
changes in the system dynamics, which represent a major
factor affecting the performance of the tactical-grade
IMU system. Thus, for medium accuracy IMU, tuning of
the Q matrix is crucial to achieve filter stability. Hence,
arrangement of the Q and R matrices in an adaptive
manner can improve estimation, as they would
dynamically reflect the actual situation. Prior field-testing
results for a kinematic ground survey showed that the
adaptive Kalman filter approach outperformed the
conventional approach, both on internal and external
bases (El-Mowafy and Mohamed, 2005). It was also
shown that the track ability of the adaptive filter for the
filter states was much better than that of the convention al
filter.
For the above reasons, an adaptive Kalman filtering
approach was employed in the processing of the test data.
In this approach, the residual sequence ηi was first
computed as:
ηi = Mi – h(S i) (17)
Then, the adaptive formulation of the R and Q matrices
followed the following formulations (Mohammed and
Schwarz, 1999):
Cη = N
1 ∑
=
i
kk0
ηk T
k
η (18)
Ri = Cη + Hi Pi T
i
H (19)
Qi = Ki Cη T
i
K (20)
where Cη is the covariance matrix of the residual
sequence, and using ko = i – N +1 as the first epoch inside
the estimation of a moving time window of the size (N),
which can be taken as 20-30 epochs.
4.2 GPS/INS Integration Results
When breaks in reception of network corrections take
place, an extrapolation of these corrections continues for
a few seconds; after that GPS solution accuracy
deteriorates. As a result, the GPS positions and velocity
input are de-weighted in the filter, and the INS works in a
stand-alone mode. Thus, the acceptable period of outage
in reception of the network corrections is the summation
of the extrapolation period of the network corrections,
during which the GPS still provides positioning accuracy
at the cm to decimetre level, and the time through which
the INS positioning accuracy in a stand-alone mode
without GPS updates is within the accuracy required for
navigation. In the case of regaining GPS observations
with network corrections, the time needed to resolve the
ambiguities should be included in the GPS positioning
outage period. For the tests in hand, an outage in
receiving the network corrections occurred after the
dashed areas, illustrated in Figures 5 and 6. The INS
stand-alone positioning errors grew very rapidly with
time in a non-linear form. However, unlike GPS, the
tested IMU system has a better height determination
accuracy than its horizontal accuracy. This is
advantageous for airborne navigation, which is more
restricted by the height accuracy. For instance, the
maximum allowable height error for category III is 0.6m,
while it is 4.1m for the horizontal error.
The test results showed that the accuracy requirements
for precision approach (category III) were generally
achieved within 25-31 seconds of the GPS data outages.
This was dependent to some extent on the aircraft
dynamics. During enroute flying, the aircraft generally
had uniform dynamics, which resulted in a longer
positioning outage bridging, while during takeoff and
landing, more changes in dynamics took place, which
resulted in shorter coverage of outages. In addition,
during curved parts of the course, the INS performed less
well than during straight flying. Thus, shorter position
bridging periods were recorded during curved flying.
Overall, for data outages up to 43 seconds, the
positioning accuracy achieved was suitable for category
II. After that, the vertical positioning error was several
meters, which is only suitable for category I of airborne
navigation. These results, however, correspond to the
used system and may differ for other IMU systems. The
performance of the tested INS in the stand-alone mode
during positioning bridging of the GPS data outages is
shown in Table 3 for the helicopter and the fixed-wing
aircraft tests. The average and maximum standard

10 Journal of Global Positioning Systems
deviations of the planimetric (horizontal) and height
components are given for GPS network data outages of 20 seconds and 40 seconds.
Tab. 3 Standard deviatio ns of the INS posi t io n i n g results dur in g G PS d a t a o u t a ge s ( m)
Helicopter test Fixed-wing aircraft test
2D (E&N) Height 2D (E&N) Height
Period of GPS
data outages Avg. Max. Avg. Max. Avg. Max. Avg. Max.
20 seconds 2.635 3.725 0.354 0.582 2.140 3.971 0.310 0.534
40 seconds 4.161 5.103 0.915 1.662 3.651 4.837 0.832 1.388
Although the system hardware used and their integration
processing schemes still have room for improvement, this
configuration was tested to investigate the feasibility of
the presented concepts, namely: using the multiple
reference station RTK GPS networks for precision
airborne navigation, and the ability of an integrated
GPS/INS system to bridge positioning during short
breaks in reception of network corrections. Other
GPS/INS integration types, including tight coupling with
centralized filtering, are currently under investigation.
Tight coupling, as compared with loose coupling, is
expected to provide a solution for longer data outage
periods. Partial GPS data of less than 4 satellites, which
only give under-determined solution, can also be used. In
addition, the INS can help in detecting and correcting
cycle slips, and aiding ambiguity resolution, see for
instance Wu (2003). Other studies (e.g. Petovello, 2003)
have also shown that the tight integration approach
outperforms loose integration approaches in terms of the
overall system accuracy, due to the reduced amount of
process noise in the tight integration. The duration range
of the INS positioning bridging under different
operational conditions is also under investigation.
However, this will vary according to the quality of the
IMU used. For instance, better results can be achieved
with higher accuracy systems (navigation grade IMU)
compared with the medium accuracy IMU system used in
this test.
5 Conclusions and Re co m mendations
The test results show that the use of RTK multi-station
reference networks (e.g. the DVRS network) in precise
aircraft navigation is feasible, particularly for the airport
area. This new technology can increase the coverage area
compared with other GPS-navigation systems, such as
airport LAAS, with significant cost reduction. Small
airports can thus benefit from this service. For the test
flights conducted, the DVRS GSM signals were received
during most of the testing periods. The loss of the signals,
which took place for some periods, was expected due to
the use of GSM signals and changes in aircraft dynamics.
During the majority of periods of receiving the DVRS
corrections, the phase measurement ambiguities were
fixed and the average positioning accuracies were less
than 4 cm. The accuracy needed for category III was
achieved even with a float ambiguity resolutio n.
One can see from these results that to achieve the
accuracy requirement of all phases of flight using the
DVRS system, it is necessary to guarantee continuous
transmission of the DVRS corrections in a suitable form
for civil aviation. One suggestion to achieve this goal is
by establishing ground transmitters at the airport. These
transmitters will receive the corrections from the network
control centre on-line and send them to the aircraft using,
for instance, VHF modems instead of the currently used
GSM modems. This can be implemented in the update
phase of the DVRS network. A one-way direction of
communication from the ground tran smitter to th e aircraft
is recommended. In addition, it is advisable to add one
reference station in the vicinity of the airport to enhance
correction estimation in this area and act as a backup for
the system, such that its corrections can be readily
applied in case of any interruption in reception of the
signals from the network control centre.
In the final implementation phase, the integrity of the
system should be fully ascertained, with systems that can
warn the pilot in case of system accuracy and availability
deficiency being added as necessary. The security of the
reference station locations should also be maintained at
the highest levels. In addition, the format of the GPS
corrections sent should be standardized so as to be
independent of any single receiver manufacturer. This
can be achieved by adopting the upcoming RTCM
version 3.0 multi-station reference RTK standards.
One way to increase the availability of the positioning
data is to integrate GPS with an Inertial Measuring Unit
(IMU). Test results with medium accuracy IMU
integrated u sing an adap tiv e Kalman filtering show ed that
positioning bridging can give acceptable results for
category III and II if breaks in GPS solution availability

El-Mowafy: Using Multiple Reference Station GPS Networks for Aircraft Precision Approach 11
are less than 40 seconds on average. After this period,
without having new accurate GPS position updates,
positioning errors grow and reach several meters. This
accuracy is only suitable for category I of airborne
navigation. However, better results can be achieved if
navigation-grade IMU systems are used.
Acknowledgements
Dr. H. Fashir, Mr. A. Al Habbai, and Mr. Y. Al Marzooqi
from the Survey Section, Planning & Survey Department,
Dubai Municipality are sincerely acknowledged for
providing the DVRS system, the logistics support and
help in testing the system. The Dubai Air Wing is
gratefully acknowledged for providing the test airplanes,
airport facilities, and air flight crew, with special thanks
to Captain G. Waddington. This study was partially
funded by a grant from the Research Affairs Sector, the
UAE University.
References
El-Mowafy A., Fashir H.; Al Marzooqi Y.; Al Habbai A.;
Babiker T. (2003): Testing of the DVRS National GPS-
RTK Network, In: Proceedings of the 8th ISU International
Symposium, Strasbourg, France, 25-28 May, 2003.
El-Mowafy A. (2004): Analysis of the Design Parameters of
Multi-Station Reference RTK GPS Networks, Journal of
Satellite and Land Information Science, ACSM, Vol. 65,
No. 1, 17-26.
El-Mowafy A.; Mohamed A. (2005): Attitude Determination
from GNSS Using Adaptive Kalman Filtering, Journal of
Navigation, Cambridge Press, UK, Vol. 58, No. 1, 135-
148.
Enge P.; Fan R.; Tiwari A. (2000): GPS Reference Network -
New Role Providing Continuity and Coverage, GPS
World, Vol. 11, No. 1, 56-62.
Euler H-J.; Seeger S.; Zelzer O.; Takac F.; and Zebhauser B.E.
(2004): Improvement of Positioning Performance Using
Standardized Network RTK Messages, In: Proceedings of
ION NTM, San Diego, CA, January 26-28, 2004.
Hu G.R.; Khoo H.S.; Goh P.C.; and Law C.L. (2003):
Development and Assessment of GPS Virtual Reference
Stations for RTK Positioning, Journal of Geodesy, Vol.
77, 292–302.
Petovello M. (2003): Real-Time Integration of a Tactical-
Grade IMU and GPS for High-Accuracy Positioning and
Navigation, UCGE Report No. 20173, Dept. of Geomatics
Engineering, University of Calgary, Canada.
Raquet J.; Lachapelle G. (2001): RTK Positioning with
Multiple Reference Stations, GPS World, Vol. 12, No. 4,
48-53.
Shin E-H. (2001): Accuracy Improvement of Low Cost
INS/GPS for Land Applications, UCGE Report No.
20156, Dept. of Geomatics Engineering, University of
Calgary, Canada.
Whelan C. (2001): Evolution of Air Navigation Services
during this Decade, Journal of Navigation, Vol. 54, No. 3,
447-462.
Wübbena G.; Bagge A.; Schmitz M. (2001): Network-Based
Techniques for RTK Applications, Presented at the GPS
JIN 2001 Symposium, GPS Society, Japan Institute of
Navigation, Tokyo, Japan, November 14-16.
Wu H. (2003): On-The-Fly GPS Ambiguity Resolution with
Inertial Aiding, UCGE Report No. 20186, Dept. of
Geomatics Engineering, University of Calgary, Canada.