This article has been accepted for inclusion in a future issue of this journal.
Content is final as presented, with the exception of pagination.
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
Integration of GNSS Precise Point Positioning and
Reduced Inertial Sensor System for
Lane-Level Car Navigation
Mohamed Elsheikh , Member, IEEE, Aboelmagd Noureldin , Senior Member, IEEE, and Michael Korenberg
Abstract— The last decade has witnessed a growing demand for
precise positioning in many applications, including autonomous
car navigation. The safety features in autonomous driving and
Advanced Driver Assistance Systems (ADAS) require lane-level
positioning accuracy. Such accuracy can be obtained from the
Global Navigation Satellite Systems (GNSS) through either dif-
ferential techniques or Precise Point Positioning (PPP). PPP is
currently favored over differential GNSS because it provides a
global solution without the need for local reference stations. Nev-
ertheless, employing PPP for land vehicles would be challenging
due to frequent signal degradation and blockage. Integrating
PPP with an Inertial Navigation System (INS) can solve the
solution continuity problem; however, the INS solution drifts over
time, resulting in losing the desired accuracy. Implementing a
reliable PPP/INS system that can preserve the required accuracy
is not trivial, especially with financial and computational cost
constraints. This article proposes the integration of PPP with
the Reduced Inertial Sensor System (RISS) for lane-level car
navigation. The high-precision needed in lane-level positioning
can be achieved by integrating PPP with high-end INS. Since
high-end INS are expensive, this work proposes the use of RISS
instead of the traditional INS. RISS uses only one gyroscope and
two accelerometers, which can save more than half the high-end
INS cost. The proposed PPP/RISS system was tested through
three road tests that included highway driving under several
overpasses. The system was able to maintain horizontal position
errors of less than 50 cm.
Index Terms— Precise point positioning, global navigation
satellite system, inertial sensors, car navigation, lane-level, high-
way driving, autonomous driving.
Manuscript received June 9, 2019; revised March 12, 2020, August 14,
2020, and November 12, 2020; accepted November 23, 2020. This work
was supported by the Natural Sciences and Engineering Research Council of
Canada (NSERC) under Grant RGPIN-2020-03900 and Grant STPGP 521432.
The Associate Editor for this article was Z. M. Kassas. (Corresponding
author: Mohamed Elsheikh.)
Mohamed Elsheikh is with the Department of Geomatics Engineering, Uni-
versity of Calgary, Calgary, AB T2N 1N4, Canada, also with the Department
of Electrical and Computer Engineering, Queen’s University, Kingston, ON
K7L 3N6, Canada, and also with the Department of Electronics and Electrical
Communication Engineering, Tanta University, Tanta 31512, Egypt (e-mail:
mohamed.elsheikh@ucalgary.ca).
Aboelmagd Noureldin is with the Department of Electrical and Computer
Engineering, Queen’s University, Kingston, ON K7L 3N6, Canada, and
also with the Department of Electrical and Computer Engineering, Royal
Military College of Canada, Kingston, ON K7K 7B4, Canada (e-mail:
aboelmagd.noureldin@rmc.ca).
Michael Korenberg is with the Department of Electrical and Computer
Engineering, Queen’s University, Kingston, ON K7L 3N6, Canada (e-mail:
korenber@queensu.ca).
Digital Object Identifier 10.1109/TITS.2020.3040955
1524-9050 © 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.
Authorized licensed use limited to: TONGJI UNIVERSITY. Downloaded on June 11,2021 at 05:24:11 UTC from IEEE Xplore. Restrictions apply.
1
I. I NTRODUCTION
T HE emerging applications in the automotive industry,
such as autonomous driving and Advanced Driver Assis-
tance Systems (ADAS) require navigation systems with high
accuracy, reliability, and integrity [1]. Safety features such
as lane-departure warning cannot be achieved without lane-
level positioning accuracy. A common approach for lane-
level positioning is to calculate the vehicle position in the
Earth-Centered Earth-Fixed (ECEF) frame and then match this
location to a digital map [1], [2]. Therefore, the accuracy of
both the ECEF position and the road map impacts the final
solution [2]. The focus of this work is to enhance the estimated
vehicle position in the ECEF frame.
GNSS positioning is a vital component of all car navigation
systems. However, for lane-level positioning, the Standard
Point Positioning (SPP) with meter-level accuracy is no longer
sufficient. An average lane width can be approximately 3 m.
For example, the lane widths guide of the city of Toronto in
June 2017 has recommended lane widths between 3 and 4.3 m
for the main driving lanes [3]. If the lane width is 3 m and the
car width is assumed to be 2 m, as shown in Fig. 1, positioning
accuracy of half the lane width might be enough to match
the car to a specific lane; however, for safety applications,
accuracy less than 50 cm is required.
Centimeter-level GNSS accuracy can be obtained using Dif-
ferential GNSS (DGNSS) or Precise Point Positioning (PPP)
[4]. The developed system is based on PPP since it requires
less infrastructure than DGNSS and provides a globally-
consistent solution [5]. Still, GNSS positioning cannot be used
alone for vehicle navigation because the GNSS signal is prone
to interference and obstruction [6]. For applications such as
autonomous driving, multisensor fusion is typically the case
where sensors such as the Inertial Navigation System (INS),
odometer, Light Detection and Ranging (LiDAR), and cameras
can be employed. Each sensor has its pros and cons. For
example, cameras are useful in detecting the road features,
but they cannot work in adverse weather conditions such as
rain and snow. On the other hand, the integrated GNSS/INS
solution is weather independent but suffers from the effects of
multipath and obstacles on GNSS signals and the error drift of
the INS. This article focuses on an efficient GNSS/INS fusion,
which, when integrated with other sensors, will contribute to
enhancing the overall system performance.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
2 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
Fig. 1. Schematic diagram for the lane and car widths.
Although the integration of PPP with high-end inertial
sensors can provide the required high-performance navigation
solution, the high cost of such a system is a significant
challenge. This work investigates the use of the Reduced
Inertial Sensor System (RISS) [7] instead of the traditional INS
with a full Inertial Measurement Unit (IMU) to address the
INS cost problem. A full-IMU contains three accelerometers
and three gyroscopes (gyros). On the other hand, RISS uses
only one gyro and two accelerometers. Eliminating two gyros
and one accelerometer reduces the INS cost by half to two
thirds. Furthermore, reducing the number of inertial sensors
involved in the mechanization algorithm limits the inherent
inertial sensor errors from the two removed gyroscopes from
propagating to the final output position solution, thus affecting
the overall positioning accuracy. RISS also contributes to
reducing system complexity and power consumption. Our
objective is to develop a PPP/RISS integrated system that can
meet the requirements of lane-level vehicle positioning.
Since the advent of PPP, many research studies have investi-
gated the performance of PPP/INS integration for land vehicle
navigation, e.g., in [8]–[10]. The authors in [8] focused on
showing the benefit of Tightly-Coupled (TC) compared to
Loosely-Coupled (LC) PPP/INS integration using a tactical-
grade IMU. The land vehicle test included four periods where
the number of satellites dropped below four satellites. The TC
solution error remained within ±1 m in contrast to ±5 m for
the LC solution.
A PPP/INS tightly integrated system was presented in [9],
where single differencing between satellites was applied. The
results showed decimeter-level accuracy with simulated GNSS
outages of 10 s duration. However, the figures of the position
errors with simulated GNSS outages in [9] have shown instant
solution convergence after the outages. An instant convergence
of PPP after a simulated GNSS outage indicates that the
ambiguities and its covariance were kept being updated during
the outage, which does not happen in a real GNSS outage.
The research in [10] investigated the performance of a TC
integration of ambiguity-fixed PPP and INS. Two car-borne
tests were made; one in a small square and the other in a
suburban environment. The presented system has demonstrated
centimeter-level accuracy in open-sky environments after the
initial convergence. Furthermore, two simulated 10 s GNSS
Authorized licensed use limited to: TONGJI UNIVERSITY. Downloaded on June 11,2021 at 05:24:11 UTC from IEEE Xplore. Restrictions apply.
outages were introduced in the first test. The results showed
how the PPP/INS integration contributed to reducing the re-
convergence time after these outages significantly. The initial
convergence time was not affected by the integration due to
the large uncertainty in the initial INS position.
In [11], tight integration of measurements from two low-
cost GNSS receivers and an IMU was utilized for PPP-based
positioning and attitude determination. The algorithm is based
on estimating the vertical ionospheric delays and satellite
phase biases from a network of static GNSS receivers to enable
ambiguity-fixed PPP. The performance was examined through
a road test that involved passing under a bridge. The results
showed centimeter-level positioning accuracy and a heading
rms error of 0.27◦ .
Regarding lane-level positioning, early studies have used
the Differential Global Positioning System (DGPS) to obtain
a precise GNSS solution [12], [13]. In [12], DGPS data was
fused with a fiber optic gyro and a wheel sensor before the
solution is matched to a map. A test was performed on one
lane on a highway with 4 to 7 visible satellites. There was
no reference, and the authors focused on showing the results
after the map matching with no details about the performance
of DGPS or its integration with the Dead Reckoning (DR)
sensors. In [13], an integrated DGPS/INS system is used to
assist vision systems for lane-keeping. The results showed
potential in using a high-precision GNSS/INS solution to add
smoothness and robustness to the vision solution.
The problem with DGNSS is the need for reference stations,
which means added infrastructure, radio links for commu-
nication, and a local positioning solution. Employing PPP
for lane-level positioning was investigated in [14] and [15].
In [14], the focus was on designing a real-time PPP framework.
A road test was performed on an urban road, and the PPP
achieved an average of 50 cm accuracy away from down-
town and a few meters accuracy in challenging environments.
This research did not investigate the GNSS outages. On the
other hand, in [15], the test included two underpasses, which
caused solution unavailability and degradation. The authors
in this research discussed the re-convergence time problem
and suggested the integration with DR sensors, but with no
implementation or results.
From the studies mentioned above, the implementation of
PPP/INS integration for lane-level positioning has not still
been fully explored, especially in driving scenarios with real
GNSS outages. The main contribution of this work is to exploit
the benefits of RISS to enable high-precision positioning for
lane-level navigation of automated vehicles. The mathematical
model for integrating PPP with RISS in a tightly coupled
mode is developed. Results from road tests showed how RISS
contributed to maintaining the continuity of the navigation
solution within the required accuracy level even in challeng-
ing scenarios where the test vehicle goes under overpasses
and bridges. These results were achieved by using only one
gyroscope, which reduces the financial and computational cost
of such a high precision system. The developed system is
assessed based on the rms and maximum horizontal position
errors. The maximum errors are crucial in the system evalua-
tion for safety applications.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
ELSHEIKH et al.: INTEGRATION OF GNSS PPP AND RISS FOR LANE-LEVEL CAR NAVIGATION
The next section describes the utilized PPP model.
Section III is a review of RISS describing its mechaniza-
tion equations in addition to its advantages and limitations.
The developed PPP/RISS integration model is illustrated in
section IV. Section V shows the experimental setup in the per-
formed road tests. The testing results are shown in section VI,
focusing on how the developed PPP/RISS system could main-
tain lane-level accuracy on highways. Finally, section VII
concludes the presented work.
II. P RECISE P OINT P OSITIONING M ODEL
In this work, the standard PPP model is adopted, which
is based on the ionospheric-free (IF) combinations of dual-
frequency GNSS measurements with float ambiguity resolu-
tion [16]. Besides, corrections for satellite orbits, clocks, phase
and code biases from a global network are employed to achieve
the high precision of the PPP solution. The PPP algorithm also
applies corrections for the relativistic and Sagnac effects [17].
Furthermore, additional error sources have to be considered in
PPP that used to be ignored in SPP or canceled in DGNSS,
such as the phase wind-up, antenna phase center, and site
displacements [18].
The tropospheric delays consist of dry and wet components.
The dry component accounts for 90% of the delay, and it can
be accurately modeled. In contrast, the wet component, which
is responsible for 10% of the delay, is difficult to model as the
water vapor content varies locally [19]. For centimeter-level
positioning, the total tropospheric delay T for each satellite s
at the GNSS receiver r can be parametrized as [18]
Trs = m h (E s )Z r,h + m w (E s )Z r,w
where E s is the satellite elevation angle, whereas Z r,h is the
zenith hydrostatic delay, and Z r,w is the zenith wet delay at
the receiver location. The terms m h and m w represent the
elevation-dependent mapping functions of the hydrostatic and
wet components, respectively.
In our algorithm, Z r,h was calculated using [20]
0.0022768P
Z r,h =
1 − 0.00266 cos(2ϕ) − 2.8 ∗ 10−7 h
where P is the surface pressure in hPa, ϕ is the latitude in
degrees, and h is the height in meters. For centimeter-level PPP
accuracy, the pressure (in hPa) can be approximated by Berg’s
formula P = 1013.25(1 − 2.2557 ∗ 10−5 h)5.2568 [21]. On the
other hand, the wet component cannot be calculated with the
same accuracy and hence was estimated as an unknown in the
navigation filter. The mapping functions were computed using
Neill’s mapping functions [22].
After applying the corrections mentioned above, the cor-
s,G s,G
rected IF GPS pseudorange Pr,I F and carrier phase r,I F
observations can be formulated as
Pr,s,G s,G
I F = ρr + cdtr, PI F + m w Z r,w + r, PI F
s s
s,G
r, IF = ρrs + cdtr, PI F + m sw Z r,w + λG s
I F NI F
where the superscript G refers to GPS, ρrs
is the true geometric
range between the GNSS satellite s and receiver r in meters, c
Authorized licensed use limited to: TONGJI UNIVERSITY. Downloaded on June 11,2021 at 05:24:11 UTC from IEEE Xplore. Restrictions apply.
3
is the speed of light, dtr, PI F is the receiver clock bias of the IF
pseudorange observations in seconds, λ I F is the IF wavelength
in meters, N I F denotes the IF float ambiguity in cycles, and
 denotes the multipath and receiver noise in meters. It is
worth mentioning that dtr, PI F was used in both pseudorange
and carrier phase observations since the difference between
code (pseudorange) and phase biases is absorbed into the float
ambiguity parameter.
The observations of the Russian Global Navigation Satellite
System (GLONASS) were also employed to increase the
number of visible satellites. The corrected GLONASS mea-
surements can be written as
Pr,s,R s,R
I F = ρr + cdtr, PI F + I S BG−R + m w Z r,w + r, PI F
s s
(5)
s,R
r, IF = ρrs + cdtr, PI F + I S BG−R + m sw Z r,w + λs,R s
I F NI F
+ r,s,R
I F (6)
where the superscript R refers to GLONASS. GLONASS
observations have an extra unknown, which is the Inter-System
Bias (ISB) between GPS and GLONASS clock systems,
in meters. Besides, the IF wavelength for GLONASS is not
only different from GPS, but it is also different for each satel-
lite due to the employment of the Frequency Division Multi-
ple Access (FDMA) technique in GLONASS legacy signals.
This frequency difference results in an additional satellite-
dependent bias called the Inter-Frequency Bias (IFB) or Inter-
Channel Bias (ICB). Nevertheless, in float-ambiguity PPP,
the phase IFBs are absorbed in the estimated float ambiguities
[23]. On the other hand, GLONASS code IFBs are difficult
(1) to estimate, and there are no available corrections for them.
Besides, they only affect the initial PPP convergence, espe-
cially in fixed-ambiguity PPP; therefore, they were neglected
in the implemented float-ambiguity PPP.
Another GNSS useful observation is the Doppler measure-
ment. Doppler measurements can be used to calculate the
GNSS receiver velocity. Furthermore, in GNSS/INS integra-
tion, it can provide an update to the vehicle velocity, which can
contribute to a faster estimation of the INS attitude and IMU
(2)
errors [6]. The Doppler measurements D H z provided by GNSS
receivers represent the Doppler shift in Hz. The measured
pseudorange rate D in m/s is obtained by multiplying the
Doppler shift by its corresponding wavelength with a negative
sign
D = −λ × D H z (7)
s
The corrected IF Doppler observation Dr,I
F is given by
˙ r, P +  Ds
Dr,s I F = ρ˙rs + cdt (8)
IF r,I F
It should be noted that the above equation can be used for
both GPS and GLONASS since the change of the receiver
clock is the same for both constellations as they use the same
(3)
oscillator.
+ r,s,G
I F
(4) III. T HE R EDUCED I NERTIAL S ENSOR S YSTEM
There is a growing demand for lower cost, lighter, and
smaller navigation sensors in the land vehicle navigation
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

4 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS

industry. One way to achieve this goal is to rely on Micro-

Electro-Mechanical systems (MEMS) sensors rather than the
high-end ones, compromising the sensors’ quality. Another
approach to consider is to reduce the number of utilized inertial
sensors.
The motion of land vehicles has three constraints, as defined
in [24]. First, the vehicle velocity is mainly in the forward
direction, whereas the transverse and vertical velocity com-
ponents are nearly zero. Secondly, the pitch and roll angles
relative to the Earth’s surface have small values (mostly less
than 5 degrees [25]). The third constraint is that the vehicle
is always moving on the Earth’s surface. These constraints
were used in [24] to prove that three gyros and one forward Fig. 2. Block diagram of RISS mechanization (after [30]).
accelerometer may be sufficient for land applications. How-
ever, this configuration cost is still relatively high since the
gyros are more expensive than the accelerometers, especially
and roll are calculated at epoch k from the gravity components
for high-quality sensors.
in the accelerometer measurements at the same epoch by
The above constraints also show that the dominant vehicle
subtracting the effect of vehicle motion acceleration [7].
rotation is around the vertical axis, which allows the removal
of the two horizontal gyros. Furthermore, the output of the  f −a 
pk = sin−1
vertical accelerometer mainly consists of gravity and road yk odk
(9)
vibrations. Therefore, the minimum configuration of IMU g
 f +v ω 
sensors for a reduced INS is one vertical gyro and two rk = − sin−1
xk odk z k
(10)
horizontal accelerometers [25]; nevertheless, a wheel speed g cos pk
odometer may be used to replace the accelerometers in 2D
models [26]. Many researchers have studied this configuration where v od , and aod are the odometer velocity and acceleration
in the past years [25]–[28]. Some of these studies presented a respectively, f x and f y are the specific forces measured by the
simplified 2D solution ignoring the vehicle off-plane motion horizontal accelerometers, wz is the rotation rate around the
(i.e., assume zero pitch and roll angles) [26]. In contrast, others z-axis, and g is the Earth’s gravity that can be considered as
tried to model the pitch and roll angles as a stochastic process a constant or better calculated using gravity models.
[25], [28]. The azimuth angle is calculated through the numerical
Among the reduced INS configurations, RISS was initially integration of its rotation rate as follows
introduced as a 2D model in [29] using one vertical gyro and a
speed odometer. Employment of two horizontal accelerometers  v e tan ϕk−1 
was suggested to calculate the pitch and roll angles but not Ak = Ak−1 + −ωzk + ωe sin ϕk−1 + k−1 t
R N + h k−1
used in estimating the position. Later, the RISS model was (11)
updated to incorporate the calculated pitch and roll angles in
estimating the off-plane vehicle motion, and it was named
3D RISS [7]. For simplicity, in the rest of this article, the term where ωe is the Earth’s angular rotation rate, and t is the
RISS implicitly refers to the 3D RISS model. time interval between epochs.
The underlying assumption in RISS is that the measured
odometer speed represents the forward motion in the y-
Tand no other velocity components exist, i.e., v =
A. RISS Mechanization direction, b

Fig. 2 shows the block diagram of the RISS system and its 0 v od 0 where the superscript b denotes the body frame.
mechanization process. The RISS mechanization is performed For calculating the velocities in the LLF, the odometer velocity
in the Local Level Frame (LLF) since it provides the position is transformed from the body frame to the navigation frame
in terms of the latitude, longitude, and altitude parameters. using the equations
The utilized LLF shares the same origin with the vehicle
frame, and its axes point to the east, north, and up directions. v ek = v odk sin Ak cos pk (12)
The following analysis assumes that the IMU is well-aligned v nk = v odk cos Ak cos pk (13)
with the vehicle such that the y-direction of the IMU body
frame is the motion direction, the x-direction is the transverse v u k = v odk sin pk (14)
(right) direction, and the z-axis is normal to both, pointing in
the up vertical direction. Because of its simplicity, the RISS where v e , v n , v u are the east, north, and up velocities respec-
mechanization equations can be directly represented in the tively.
discrete-time form. Finally, the position states, latitude ϕ and longitude λ in
The first mechanization step is the attitude calculation, radians, and altitude h in meters can be obtained by the
i.e., the azimuth ( A), pitch ( p), and roll (r ) angles. The pitch numerical integration of their corresponding velocities using

Authorized licensed use limited to: TONGJI UNIVERSITY. Downloaded on June 11,2021 at 05:24:11 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

ELSHEIKH et al.: INTEGRATION OF GNSS PPP AND RISS FOR LANE-LEVEL CAR NAVIGATION 5

the trapezoidal rule.
h k = h k−1 + 0.5(v u k + v u k−1 )t (15)
0.5(v nk + v nk−1 )
ϕk = ϕk−1 + t (16)
RM + hk
0.5(v ek + v ek−1 )
λk = λk−1 + t (17)
(R N + h k ) cos ϕk
where R M is the meridian Earth radius, and R N is the normal
Earth radius.
B. RISS Advantages and Limitations
The use of RISS has several advantages over full IMU
and other reduced INS systems, which can be summarized
as follows:
• The elimination of two gyroscopes and one accelerometer
reduces the overall cost and minimizes the error accumu-
lation compared to the full IMU.
• The reduction of INS sensors reduces the computational
complexity and power consumption of the system.
• Using the odometer removes one mathematical integra-
tion step required when only accelerometers are used,
eliminating the error growth in the velocity.
• Using the odometer with the horizontal accelerometers
allows the calculation of the pitch and roll angles using
(9) and (10).
• Using the odometer speed as a control input rather
than an update eliminates the need to apply algorithms
such as Zero-velocity Update (ZUPT) and Nonholonomic
Constraints (NHC). Furthermore, the odometer is an
integrated part of any land vehicle, and hence, there is
no extra cost in utilizing the odometer data.
• The calculated pitch and roll angles are used in RISS to
estimate the off-plane vehicle motion.
• Calculating pitch and roll angles from the accelerometers
rather than the gyros includes no mathematical integra-
tion, which means no drift or error growth [7].
On the other hand, one limitation of RISS is that the calcu-
lated pitch and roll are noisy because they are directly calcu-
lated from the sensor measurements. Furthermore, as with all
other reduced INS, the solution degrades if the land vehicle
constraints are not fulfilled (e.g., large pitch or roll angles and
vertical motion).
IV. PPP/RISS I NTEGRATION
This section presents the developed PPP/RISS system with a
description of its system and measurement models. For the best
integration performance, the TC integration mode is chosen
such that RISS can contribute to estimating the PPP parameters
and hence reduce the solution re-convergence time after GNSS
outages. Furthermore, the TC integration is performed on the
measurement level, which means that the final PPP solution
is not necessary, and any available measurements can be used
to update the navigation filter.
The block diagram of the developed PPP/RISS system is
shown in Fig. 3. The raw pseudoranges, carrier phase, and
Doppler measurements are obtained from the GNSS receiver
and corrected using the PPP network corrections and error
models. On the other hand, the RISS mechanization output
is converted to measurement-like values. Then, the measure-
ment errors, which are the differences between the predicted
measurements and the real GNSS measurements, are fed
to an Extended Kalman Filter (EKF). The filter fuses the
measurement errors with the corresponding ones from the
system error model and generates the final estimated position,
velocity, and attitude errors. Fig. 3 also shows the closed-loop
structure of the system. The estimated navigation parameters
are fed back after each epoch to the RISS mechanization
module. Moreover, the estimated sensor biases are corrected
using the estimated values of their errors. This feedback resets
the error states every epoch and assures that the linearity
assumption of the EKF is maintained.
A. System Model
The system error model consists of the INS error states
augmented with error states related to PPP. The number of
error states in the system model is (13 + M) error states where
M is the number of satellites used in the solution. The state
vector δX can be written as
δX = [δrl δvl δA δaod δ Bz δ Br
δd Br,G−R δdr δ Zw δNIF ]T (18)
where rl and vl are the vehicle position and velocity vectors
in the LLF. The error in the azimuth angle is denoted δ A,
whereas δaod represents the odometer acceleration error. δ Bz
is the error in the vertical gyroscope bias. The term Br is
the GNSS receiver clock bias in meters, which is equal to
˙ r is the GNSS receiver
cdtr, PI F . Similarly, the term dr = cdt
clock drift in m/s. The ISB between GPS and GLONASS is
denoted d Br,G−R . The error in zenith wet tropospheric delay
is represented by δ Z w , whereas the vector δNIF represents the
errors in the float ambiguities.
The system model can be described by the formula
δ Ẋ = FδX + W (19)
where F is the system dynamic coefficient matrix, and W is
the process noise vector with covariance matrix Q [6].
Since the system states are error states, the elements of
the F matrix can be obtained from the total derivative linear
approximation of the time rate of change of the system
states. For example, the time rate of change of the latitude
is calculated by
vn
ϕ̇ = (20)
Rm + h
Applying the total derivative leads to
1 vn
δ ϕ̇ = δv n − δh (21)
Rm + h (Rm + h)2
which indicates that δ ϕ̇ depends on the changes in both v n
and h. Although values like (R v+h)
n
2 are very small, they were
m
included in the developed system model to reduce the error
accumulation and achieve the best performance in the long
term. The derivation of the other INS error states when using

Authorized licensed use limited to: TONGJI UNIVERSITY. Downloaded on June 11,2021 at 05:24:11 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

6 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS

Fig. 3. Block diagram of the developed TC PPP/RISS integrated system.

⎡ 1 ⎤
RISS is beyond the scope of this discussion and can be found 0 0
⎢ RM + h ⎥
in [31]. ⎢ ⎥
F2 = ⎢ 1
0⎥ (25)
The error in the GNSS receiver clock is represented by two ⎣ 0 ⎦
(R N + h) cos ϕ
states, the clock bias and the clock drift, similar to [32], [33].
0 0 1
The clock error model is described by ⎡   −v tan ϕ  ⎤
v e sec ϕ 
2
e
       ⎢ v n ωe cos ϕ + R + h 0 vn
(R N + h)2 ⎥
δ Ḃr 0 1 δ Br w Br ⎢  N
2   −v tan ϕ ⎥
δ d˙r
=
0 0 δdr
+
wdr
(22) F3 = ⎢
⎢−v ω cos ϕ + v e sec ϕ e
⎥
⎥
⎣ e e 0 −v e
RN + h (R N + h)2 ⎦
where w Br and wdr represent the white Gaussian process noise 0 0 0
that satisfies the assumptions of the EKF [30], for the receiver (26)
⎡ v n tan ϕ v e tan ϕ ⎤
clock bias and clock drift, respectively. −ωz +we sin ϕ + 0
⎢ RN + h R N +h ⎥
The other states that have no deterministic formulas must ⎢ ⎥
be modeled using proper stochastic models. In the developed F4 = ⎢ω −w sin ϕ − 2v e tan ϕ 0⎥
⎣ z e
R N +h
0 ⎦
system, the vertical gyro bias and odometer acceleration errors
0 0 0
were modeled by the first-order Gauss-Markov process with
correlation times β Bz and γaod , respectively. The errors in (27)
⎡ ⎤
the wet tropospheric delay and ISB terms were modeled aod cos A cos p sin A cos p vn
F5 = ⎣ −aod sin A cos p cos A cos p −v e ⎦ (28)
as random walk processes since they have slow variations
overtime. Finally, the float ambiguities were considered as 0 sin p 0
⎡ ⎤
random constants. v e sec2 ϕ −v e tan ϕ
The F matrix can now be written as ⎢ωe cos ϕ + R + h 0 (R + h)2 ⎥
F6 = ⎢ N N ⎥ (29)
⎣ 0 0 0 ⎦
⎡ ⎤
F1 F2 03×3 03×3 03×1 03×M 0 0 0
⎢ F3 F4 F5 03×3 03×1 03×M ⎥ ⎡ tan ϕ ⎤
⎢ ⎥
⎢ F6 F7 F8 03×3 03×1 03×M ⎥ 0 0
F =⎢ ⎢ 03×3
⎥
F7 =
⎢ RN + h ⎥
⎢ 03×3 03×3 F9 03×1 03×M ⎥ ⎥
⎣ 0 0 0⎦ (30)
⎣ 01×3 01×3 01×3 01×3 0 01×M ⎦ 0 0 0
0 M×3 0 M×3 0 M×3 0 M×3 0 M×1 0 M×M ⎡ ⎤
0 0 1
(23) F8 = ⎣0 −γaod 0 ⎦ (31)
0 0 −β Bz
where the matrices F1 to F9 are given by ⎡ ⎤
0 0 1
⎡ ⎤ F9 = ⎣0 0 0⎦ (32)
−v n
0 0 0 0 0
⎢ (R M + h)2 ⎥
⎢ v tan ϕ −v e ⎥
F1 = ⎢ e
0 ⎥ (24) The state transition matrix φ is approximated from F using
⎣ (R + h) cos ϕ (R N + h) cos ϕ ⎦
2
N
0 0 0 φ ≈ I + Ft (33)

Authorized licensed use limited to: TONGJI UNIVERSITY. Downloaded on June 11,2021 at 05:24:11 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
ELSHEIKH et al.: INTEGRATION OF GNSS PPP AND RISS FOR LANE-LEVEL CAR NAVIGATION
TABLE I
T HE I NITIAL STD OF THE S YSTEM E RROR S TATES AND A SSUMED
P ROCESS N OISE STD. T HE IMU-R ELATED S TATES A RE FOR IMU-
KVH1750
where I is the identity matrix, and t is the time interval
between the current and previous IMU measurement epochs.
In the prediction phase of EKF, the state transition matrix in
(33) is used to predict the error state vector δXˆ −
k and its a priori
covariance Pk− at epoch k based on their values from the
previous epoch. The EKF was implemented in the closed-loop
configuration; hence, the errors states are reset every epoch,
and the prediction equations can be written as
δXˆ −
k = 0
Pk− = +
φk−1 Pk−1 φk−1
T
+ Q k−1
+
where Pk−1 is the posterior state covariance matrix of the
previous epoch.
It is worth mentioning that the initial P matrix is diagonal,
and is set based on the uncertainty of the initial states provided
to the system. In designing the initial values of the P matrix,
we shall avoid choosing very small initial values of the state
covariance to prevent divergence of the estimator (EKF) or rel-
atively large initial values to avoid long convergence time. The
Q matrix is constant diagonal and is set based on the process
noise covariance assumed with each state. The values of the Q
matrix are determined based on the stochastic model of each
state. The datasheet of the IMU sensors can provide a good
start to help to set the Q matrix; however, further tuning is
typically required to achieve the required performance. Table I
contains the initial standard deviation (STD) of the system
error states and the assumed process noise STD.
B. Measurement Model
The EKF measurement model can be expressed as
δZ = H δX + η
where δZ is the measurement error vector, H is the measure-
ment design matrix, and η is the measurement white Gaussian
noise with covariance matrix R.
Authorized licensed use limited to: TONGJI UNIVERSITY. Downloaded on June 11,2021 at 05:24:11 UTC from IEEE Xplore. Restrictions apply.
7
The measurement error vector represents the difference
between the predicted values calculated from RISS and the
corrected PPP observations
⎡ ⎤
PR I S S − PP P P
δZ = ⎣ R I S S −  P P P ⎦ . (37)
DR I S S − DP P P
For the developed TC PPP/RISS, the design matrix H can
be formulated as (38), as shown at the bottom of the next
page. The matrix H p with size M × 3 is given by
⎡ ⎤
(−ur1 )T
⎢ (−u2 )T ⎥
⎢ r ⎥
Hp = ⎢
⎢ . ⎥L
⎥ (39)
⎣ . ⎦
(−urM )T
where uri represents the 3 × 1 unit vector from the receiver to
the i t h satellite in the ECEF frame, L is a 3 ×3 transformation
matrix that maps the position errors from the geodetic coordi-
nates (δϕ, δλ, δh) to the rectangular coordinates (δx, δy, δz)
[30] and is given by (40), as shown at the bottom of the next
page, where e is the Earth’s eccentricity.
The matrix Hv is similar to H p with replacing the matrix
L by the 3 × 3 rotation matrix Rle which rotates the velocity
errors from the LLF frame (δv e , δv n , δv u ) to the ECEF frame
(δv x , δv y , δv z ) [30], and is given by
⎡ ⎤
− sin(λ) − sin(ϕ) cos(λ) cos(ϕ) cos(λ)
Rle = ⎣ cos(λ) − sin(ϕ) sin(λ) cos(ϕ) sin(λ) ⎦ (41)
0 cos(ϕ) sin(ϕ)
(34)
The vector mw maps the tropospheric zenith wet delay to
(35)
its corresponding slant delay based on the elevation angle,
as described in section II. Finally, the vector hb contains the
coefficients for the δd Br,G−R term, which means its elements
equal one for GLONASS satellites and zero for GPS satellites.
Assuming that MG is the number of GPS satellites, M R is the
number of GLONASS satellites, and the observation vectors
are ordered to have GPS satellites first and then GLONASS
satellites, hb can be written as
 
0
hb = MG ×1 (42)
1 M R ×1
To calculate the diagonal measurement covariance matrix
R, the measurements standard deviation was assumed to be
inversely proportional to the squared sine of the satellite eleva-
tion angle. Besides, an elevation mask of 7◦ , obtained by trial
and error, was applied to reject low-elevation satellites. The
carrier phase standard deviation was scaled to be 100 times
less than the code observations because of its high precision.
Due to the differences in the signal-to-noise ratio and the
signal structure between GPS and GLONASS, the standard
deviation of GLONASS code and Doppler observations was
(36)
scaled by a factor of 3 and the phase observations by a factor
of 1.5 compared to the corresponding GPS observations [34].
Table II summarizes the adopted numerical values for the
standard deviation of the utilized GNSS observations.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
8 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
TABLE II
T HE A DOPTED N UMERICAL VALUES FOR THE S TANDARD D EVIATION OF
THE U TILIZED GNSS O BSERVATIONS
The update equations of the EKF can be written as
K k = Pk− HkT {Hk Pk− HkT + Rk }−1
δXˆ + ˆ− ˆ−
k = δXk + K k (δZk − Hk δXk )
Pk+ = (I − K k Hk )Pk−
where K k is the Kalman gain.
C. Practical Considerations
Although the description of the system and measurement
models of the EKF seems to be sufficient for implementing
the integration algorithm, other practical considerations should
be taken into account for an efficient implementation of the
system. In this section, three examples of these practical
considerations are briefly discussed.
1) Dynamic Satellite Visibility: Due to the use of carrier
phase observations, the system model has an ambiguity state
for each satellite. The total number of system states in the
developed TC PPP/RISS model is 13 + M; nevertheless,
the visible satellites vary with time due to the change in
the satellite geometry and the signal blockage. Thus, care
must be taken between different processing epochs for the
appearance or disappearance of satellites, especially when
updating the covariance of the float ambiguities in the P
matrix.
A straightforward solution to this issue is to assume that all
the satellites exist and design the system matrices based on the
maximum possible number of satellites. However, using these
matrices in the EKF equations leads to high computational cost
and requires dealing with matrices with many zero entries and
might result in singularity issues. In the developed algorithm,
the system is designed for a maximum number of satellites =
32 GPS + 26 GLONASS = 58, and hence, the total number
of states is 58+13 = 71. However, for every epoch, before
applying the EKF equations, a submatrix P  is taken from
the original P matrix such that it contains only the covariance
information for the 13+ M states in this epoch. After applying
⎡ ⎤
Hp 0 M×3
H = ⎣ Hp 0 M×3
0 M×3 Hv
⎡ ⎤
−(R N + h) sin ϕ cos λ
L = ⎣ −(R N + h) sin ϕ sin λ
(R N (1 − e2 ) + h) cos ϕ
Authorized licensed use limited to: TONGJI UNIVERSITY. Downloaded on June 11,2021 at 05:24:11 UTC from IEEE Xplore. Restrictions apply.
the EKF, the new estimated covariance values are carefully
updated in the big P matrix.
2) Cycle Slip Detection: The primary challenge of using
carrier phase observations is the problem of cycle slips.
A cycle slip means a loss of carrier tracking, which can happen
due to a signal blockage or weak signal-to-noise ratio. When
a cycle slip occurs, the ambiguity changes and the assumption
that it is constant is not valid anymore. If the cycle slips are not
detected and based on their size, the solution deteriorates, and
the PPP precision may be lost. After a cycle slip is detected,
there are two ways to deal with it. The first approach is to try
(43) repairing it; nevertheless, an incorrect reparation will affect all
the following observations. The second approach is to reset
(44) the ambiguity parameter in the navigation filter to estimate
(45) the new ambiguity, which is a safer method [35] despite the
required re-convergence time.
There are several cycle slip detection methods in the lit-
erature; however, none can achieve 100% detection success.
Each method has advantages and limitations. In the developed
algorithm, the cycle slip detection module uses two methods to
detect the cycle slips from the PPP measurements. A cycle slip
flag is raised if any of the two methods detected a cycle slip.
The two methods, initially introduced in [36], are using the
Melbourne-Wübbena (MW) function and using the Geometry-
Free (GF) phase combination.
The MW function can detect large cycle slips but might
not detect the small cycle slips and cycle slips that occur
equally on the two GNSS frequencies used to form the MW
combination. Besides, the MW function is affected by the
noise in the code observation. On the other hand, the GF phase
has low noise and can be used to detect small cycle slips,
and equal cycle slips on both frequencies. However, the GF
phase cannot detect some combinations of cycle slips, and
large changes in the ionospheric delay between the consecutive
epochs can affect the detection success.
One of the advantages of the integration with INS is that
it can also contribute to cycle slip detection, which means
an extra level of detection. In the implemented algorithm,
the carrier phase innovation sequence, which is the difference
between the PPP measurements and the predicted measure-
ments from RISS, is checked against a predefined threshold
similar to the algorithm in [37]. If the innovation sequence
was greater than the threshold, a cycle slip is declared.
3) Outlier Detection: Measurement outliers may seriously
deteriorate the system performance if not handled properly. For
INS, the measurement outliers can result from sensor failures,
large sensor noise, or unexpected environment disruption.
0 M×3 1 M×1 hb 0 M×1 mw 0 M×M
0 M×3 1 M×1 hb 0 M×1 mw λ I F I M×M ⎦ . (38)
0 M×3 0 M×1 0 M×1 1 M×1 0 M×1 0 M×M
−(R N + h) cos ϕ sin λ cos ϕ cos λ
(R N + h) cos ϕ cos λ cos ϕ sin λ ⎦ (40)
0 sin ϕ
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
ELSHEIKH et al.: INTEGRATION OF GNSS PPP AND RISS FOR LANE-LEVEL CAR NAVIGATION
In the implemented RISS, the IMU measurements are down-
sampled by averaging from 100 Hz or 200 Hz to 1 Hz, which
contributes to smoothing the measurements and reducing the
effect of the outliers.
GNSS outliers can arise from multipath effects. If passed
the cycle slip detection module, the phase cycle slips will
also look like an outlier. In the implemented algorithm, two
steps of outlier detection were applied in a coarse-to-fine
approach. The first step is done before applying the EKF,
where the measurement error vector δ Z is tested against a
relatively large threshold, e.g., 20 m for the code observations.
The threshold value can be tuned based on the available data
sets, and it can also be different for each type of GNSS
observations.
The second step is based on testing the measurement resid-
uals after applying the EKF. The corrected PPP measurements
and the predicted RISS measurements are re-calculated based
on the new estimated parameters. Then, a new measurement
error vector δY is formed that represents the solution residuals.
The normalized residual δY ¯ k, j for the j t h measurement is
calculated by dividing the residual value by its posterior
standard deviation
Ck = Hk Pk+ HkT + Rk
δY
¯ k, j =  k, j
δY
Ck,( j, j )
where Ck is the posterior measurement covariance matrix at
epoch k. Ck comprises the estimated covariance of the state
transformed into the measurement domain and the measure-
ment noise covariance.
Finally, the normalized residual is compared to a threshold
to decide if the measurement is an outlier. A threshold of 3 was
chosen in the implemented algorithm, which means that the
measurement will be rejected if the residual value exceeded
three standard deviations. If a code outlier is detected, its
corresponding phase and Doppler measurements must also be
rejected, but not vice versa [6].
V. E XPERIMENTAL S ETUP
The developed PPP/RISS system was tested using data
collected from road tests. Three trajectories have been per-
formed, one in Kingston, Ontario, and the other two in
Calgary, Alberta, Canada. The following is a description of
the equipment and software used in these tests.
A. Rover GNSS Receivers
The GNSS data were logged at a 1 Hz rate. Two GNSS
receivers were used in these tests:
• NovAtel SPAN OEMV Receiver: a dual-frequency
GNSS receiver installed inside the NovAtel SPAN-SE
unit. This receiver was used as the rover receiver in the
nKingston trajectory.
• NovAtel SPAN OEM6 Receiver: a dual-frequency
GNSS receiver installed inside the NovAtel SPAN on
ProPak6 unit. This receiver was used as the rover receiver
in Calgary trajectories.
Authorized licensed use limited to: TONGJI UNIVERSITY. Downloaded on June 11,2021 at 05:24:11 UTC from IEEE Xplore. Restrictions apply.
9
B. IMUs
Two IMUs were used in these tests:
• IMU-CPT: a tactical-grade IMU with 20◦ gyro bias
offset, 1◦ / hr gyro bias stability, and 100 Hz output rate
[38]. It has fiber-optic gyros and MEMS accelerometers,
and it was used in the Kingston trajectory.
• IMU-KVH1750: a tactical-grade IMU with 2◦ gyro bias
offset, 0.05◦ / hr gyro bias stability, and 200 Hz output
rate [39]. It has fiber-optic gyros and MEMS accelerom-
eters, and it was used in Calgary trajectories.
C. Correction Products
Since GLONASS measurements were also employed,
the final precise satellite orbit and clock corrections from the
IGS Multi-GNSS Experiment (MGEX) project were used [40],
[41]. The chosen corrections included precise satellite orbits
every 5 min, and precise satellite clock corrections every 30 s.
Although the developed algorithm used the final IGS cor-
rections, in reality, other real-time correction services such as
the IGS Real-time Service (IGS-RTS) can be used to support
the real-time realization of our method. It is worth mentioning
that real-time corrections will be less accurate and will impact
(46) the overall system accuracy. Therefore, to establish an ideal
environment that examines PPP/RISS integration benefits, we
(47)
decided to use the final IGS corrections in this work. In the
future, real-time correction services can still be used, and the
proposed method can be employed to operate in real-time
without any limitations.
D. Reference System
An accurate and precise reference solution is vital in the
evaluation of high-precision systems. The NovAtel SPAN unit
measurements were used to generate the reference solution.
However, to obtain the most accurate reference, the SPAN
data were used with data from a base station to generate a
TC DGNSS/INS solution. The reference solution is gener-
ated in post-mission mode using NovAtel’s Waypoint Inertial
Explorer Software.
In the Kingston trajectory, the base station data was col-
lected using NovAtel OEM4 ProPakG2plus GNSS receiver,
and the antenna was mounted on the roof of one of the build-
ings of the Royal Military College of Canada. Two-day static
observations were collected and post-processed to estimate the
base station accurate position. For the Calgary trajectories,
the UCAL IGS station was used as the base station. UCAL
observations and reference position are available through IGS
services.
E. Software
The TC PPP/RISS algorithm is developed using MATLAB
R2017a. The development includes processing the input data
and the corrections data.
F. Equipment Setup
To collect the test data, all the equipment except the
base station receiver was carried in the testing vehicle. The
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

10 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS

Fig. 4. Typical equipment setup on a van ready for road data collection.

equipment was mounted firmly on a testbed. This testbed was

originally designed and installed by a group of professional Fig. 5. The first test trajectory, Kingston, Ontario, Canada.
technicians. They designed the testbed to be firmly held by
a special frame that is perfectly installed inside the vehicle,
replacing one of the seat benches with special brackets to
ensure that the testbed is perfectly flat and parallel to the
vehicle surface. Also, special and precise equipment was used
to guarantee the IMU axes are perfectly aligned with the
vehicle axes. The GNSS antennas were mounted on the car
roof. The car forward speed (odometer data) was collected
from the car built-in sensors through the On-Board Diagnostics
version II (OBD II) interface. Fig. 4 shows the equipment setup
in the back of the testing van in Kingston.
The lever arm vector between the GNSS antenna and the
IMU was precisely measured in advance and then compensated
for at each processing epoch to ensure the best accuracy of the
output solution. Another lever arm component is the vector
between the IMU and the vehicle odometer; however, this
component was neglected for two reasons. First, this vector
Fig. 6. Number of visible GPS/GLONASS satellites in the first trajectory.
should only affect the solution when the car is turning, and The number ranges from 13 to 15 satellites.
its effect is expected to be minor with respect to the target
accuracy. Secondly, the exact location of the vehicle odometer
is unknown because the speed is obtained from the car through
in loops for around 40 min to guarantee a continuous open-sky
the OBDII interface. Thus, it might be counterproductive to
environment, as shown in Fig. 5. For the reference solution,
consider the IMU/odometer lever arm at this development
the baseline length was less than 1 km in this trajectory. The
stage.
IMU-CPT was employed as a tactical-grade IMU. The number
VI. R ESULTS AND D ISCUSSION of GPS/GLONASS visible satellites was at least 13 during
the whole trajectory, as shown in Fig. 6. The speed profile,
The developed PPP/RISS system is tested via three road test in Fig. 7, indicates that the test started with ≈ 4 min static,
trajectories. The first trajectory is fully open-sky to investigate then the car started to move with an average speed of around
the performance with no GNSS outages. Then, we introduced 15 km/h with a quick stop in the middle.
two simulated 10 s GNSS outages. The second and third Fig. 8 shows a comparison between the PPP-only solution
trajectories imply driving on highways and moving under and the integrated PPP/RISS solution based on the horizontal
overpasses to present examples of driving scenarios with real position error. The error trend of both solutions indicates that
GNSS outages. Although the developed system generates the the integrated solution benefited from the PPP precision in the
3D position solution, the focus of the presented results will open-sky environment. Both solutions achieved an accuracy of
be on the horizontal position because it is the component that less than 50 cm in less than two minutes, and the errors went
matters in lane-level positioning. below 20 cm in 10 min. The horizontal position rms error,
calculated after the first 10 min, was 9.8 cm for PPP/RISS
A. Road Test Trajectory 1: Fort Henry compared to 11.5 cm in the case of PPP alone. In open-
The first trajectory was carried out at the parking lot of Fort sky environments, more trust is given to the GNSS PPP over
Henry, Kingston, Ontario, Canada. The car was kept moving RISS. Nevertheless, the navigation filter continuously fuses

Authorized licensed use limited to: TONGJI UNIVERSITY. Downloaded on June 11,2021 at 05:24:11 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
ELSHEIKH et al.: INTEGRATION OF GNSS PPP AND RISS FOR LANE-LEVEL CAR NAVIGATION
Fig. 7. Speed profile of the first trajectory. The speed was on average 15 km/h.
Fig. 8. Horizontal position error comparison of the first trajectory between
the integrated PPP/RISS and PPP solutions.
the GNSS PPP and RISS information, which explains why
the integrated solution is very close to the PPP-only solution
but not the same.
To explore the real benefit of the integration with RISS, two
10 s simulated GNSS outages were introduced after 20 and
28.5 min from the beginning of the trajectory. Fig. 9 shows the
horizontal error comparison after applying these two outages.
The selected outages represent two different examples of the
performance during GNSS outages. The integrated PPP/RISS
solution did not exhibit significant drift during the first GNSS
outage, and hence continued after the outage with an error less
than 20 cm and gradually started to approach the PPP solution.
The PPP solution was worse than the integrated solution right
after the GNSS outage by several centimeters, and within 5
min went below the 20 cm level again.
On the other hand, during the second GNSS outage, the inte-
grated solution drifted to a maximum error of 40.5 cm. The
amount of drift during the GNSS outage depends mainly
on the accuracy of the estimated vertical gyro bias and the
other INS navigation error states, and also the dynamics of
the vehicle before and during the outage. Despite the drift,
the integrated solution was better than the PPP solution after
the GNSS outage. The integrated solution converged to the
Authorized licensed use limited to: TONGJI UNIVERSITY. Downloaded on June 11,2021 at 05:24:11 UTC from IEEE Xplore. Restrictions apply.
11
Fig. 9. Horizontal position error comparison of the first trajectory with two
simulated 10 s GNSS outages introduced after 20 and 28.5 minutes.
20 cm error level approximately 4 min before the PPP solution.
The horizontal position rms error with the simulated outages,
calculated after the first 10 min, was 14.1 cm for the integrated
PPP/RISS solution and 19.8 cm for the PPP solution.
As seen from this test, a significant benefit of the inte-
grated solution is continuity. The integrated PPP/RISS solution
continued to provide a navigation solution with a horizontal
error less than 50 cm during the two 10 s GNSS outages,
whereas PPP alone had no solution during these periods.
The second benefit of integrating PPP with RISS is reducing
the re-convergence time after GNSS outages. RISS could not
reduce the initial PPP convergence at the start of the trajectory
because the EKF was not converged yet. However, if a GNSS
outage occurred after the filter is converged, RISS, or generally
an INS, can help reduce the solution convergence time. This
depends on the accuracy of the INS positioning solution at
the end of the outage and how its covariance is compared to
the PPP solution covariance after the outage. The longer the
GNSS outage, the less aiding is expected from the INS.
Since the ambiguity resolution is directly affected by the
accuracy of the estimated position, the ambiguity resolution is
also affected by the INS aiding. Fig. 10 shows an example of
the ambiguity resolution using the estimated IF ambiguity of
the GPS satellite G13 during the first test trajectory with the
two simulated outages. This figure shows how the integration
with RISS affected the ambiguity resolution which is reflected
in the positioning solution convergence in Fig. 9. Nevertheless,
the effect of INS aiding on the ambiguity resolution can be
more significant when ambiguity fixing is considered.
B. Road Test Trajectory 2: Calgary Highway I
The second trajectory was meant to test the developed
system in a real highway driving scenario which lasted
74 minutes. The test trajectory, shown in Fig. 11, started
on Highway 1 (Trans-Canada Highway), then the car moved
through three rectangular loops on several highways, next to
Calgary, in open-sky conditions. In the last loop, the car moved
far south and then back through highway 201 north, passing
under two overpasses before the test ends in a suburban area.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
12
Fig. 10. Example of how the integration with RISS can affect the ambiguity
resolution using the estimated IF ambiguity of the GPS satellite G13.
Fig. 11. The second test trajectory, Calgary, Alberta, Canada.
The IMU-KVH was utilized as a tactical-grade IMU in this
test. The baseline length with respect to the UCAL reference
station was less than 20 km.
Fig. 12 shows the number of visible GPS/GLONASS satel-
lites during the test. At least ten satellites were visible most
of the test time, with a one-time drop to 5 satellites during
the open-sky driving. The car passed under the first overpass
(under Highway 1) at the end of the 65t h minute. The number
of satellites became zero for two seconds, and then four
for one second before going back to the regular visibility.
The second overpass was at the 70t h minute, and the number of
satellites became five for one second, followed by two seconds
with zero visible satellites.
Fig. 13 shows the speed of the vehicle during the test. The
car was static for only 42 s before it started to move. The
speed was more than 60 km/h for 66% of the time; however,
there were several stops at the intersections. The 5 min stop
that occurred after 15.5 min was due to a railway crossing.
Authorized licensed use limited to: TONGJI UNIVERSITY. Downloaded on June 11,2021 at 05:24:11 UTC from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
Fig. 12. Number of visible GPS/GLONASS satellites in the second trajectory.
The two drops near the end are due to two overpasses.
Fig. 13. Speed profile of the second trajectory. The speed was above 60km/h
at 66% of the time.
The horizontal position error comparison is displayed
in Fig. 14. Both solutions reached 20 cm level accuracy
within 3 min. This fast convergence may be due to the large
number of visible satellites and the low horizontal Dilution
of Precision (DOP) value, which was 0.6 at the beginning of
the test. Another factor is that the car started to move after
42 s, which resulted in different measurements every epoch
that could contribute to the faster estimation of the EKF error
states. When the number of satellites dropped to 5 in the 32nd
minute, the PPP solution had an error spike of 23 cm, whereas
the integrated PPP/RISS solution was not affected by this drop
and provided a smooth solution.
Fig. 15 zooms on the part when the car moved under the
two overpasses at the end of the trajectory; the PPP solution
had four seconds with no output solution for each overpass
followed by a high position error in the epoch right after
the GNSS outage. The position error was 48.5 cm after the
first overpass and 121.5 cm after the second overpass. On the
other hand, the integrated solution provided a continuous
solution with an error of 26.5 cm after the first GNSS outage
and 50.5 cm after the second GNSS outage. Nevertheless,
the integrated solution was generally worse than the PPP
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
ELSHEIKH et al.: INTEGRATION OF GNSS PPP AND RISS FOR LANE-LEVEL CAR NAVIGATION
Fig. 14. Horizontal position error comparison of the second trajectory
between the integrated PPP/RISS and PPP solutions.
Fig. 15. Zoomed horizontal position error comparison of the second trajectory
with a focus on the part where the car moved under the two overpasses.
solution after the 67t h minute. In this period, the solution
was re-converging after the two GNSS outages with a short
time in between, and the car was moving at speed above
80 km/hr. Since the odometer measurements are typically
provided at a low data rate (1Hz), error accumulation may
result in significant position errors when driving at relatively
high speeds. When the odometer speed was replaced by the
reference speed from the Novatel system (to validate the above
explanation), the PPP/RISS solution became better than the
PPP-only solution after the second outage. This indicates that
acquiring the speed from OBDII at a low rate and with the
odometer errors can lead to relatively large errors after GNSS
outages depending on the vehicle dynamics.
Table III compares the rms and maximum errors of the
PPP and the PPP/RISS solutions after 10 min of starting the
test. The first 10 minutes were excluded to avoid the initial
convergence time and keep consistency with the previous test.
The rms errors are almost the same for the two solutions,
which is expected since the two outages are short, and the
re-convergence time of both solutions was close. The PPP
solution had a relatively fast convergence time during this
test, and hence, there was no benefit from the integration
in this regard. However, the main benefit of the integration
is smoothing the solution and removing the error spikes,
as seen from the maximum error comparison, in addition to
Authorized licensed use limited to: TONGJI UNIVERSITY. Downloaded on June 11,2021 at 05:24:11 UTC from IEEE Xplore. Restrictions apply.
13
TABLE III
C OMPARISON OF THE RMS AND M AXIMUM H ORIZONTAL P OSITION
E RRORS IN THE S ECOND T RAJECTORY
Fig. 16. The third test trajectory, Calgary, Alberta, Canada.
the continuity advantage. The PPP/RISS solution maintained
the horizontal position error within 50 cm during this test.
C. Road Test Trajectory 3: Calgary Highway II
The third trajectory lasted 66 min and was designed to
have a more challenging highway driving scenario. Fig. 16
shows the test trajectory, which started outside Calgary on
Highway 1A. The car moved west to the city of Cochrane,
where there was a suburban environment, and then moved
south on Highway 22 and east on Highway 1 towards Calgary.
In Calgary, the car moved on Highway 201, passing under
several consecutive overpasses, and finally, the test ended in a
suburban area. The IMU-KVH was utilized as a tactical-grade
IMU in this test. The baseline length with respect to the UCAL
reference station was less than 30 km.
The number of visible GPS/GLONASS satellites in the
third trajectory is shown in Fig. 17. More than eight satellites
were visible in the first 40 minutes of the test. After that,
on Highway 201, the car passed under ten consecutive road
and pedestrian overpasses. The number of satellites dropped to
zero (full outage) eight times and was less than five satellites
two times (partial outages). The significant challenge in this
trajectory is that the duration between the outages was between
0.5 to 2.5 min, which is a short period for the PPP solution
to converge. The GNSS outage duration ranged from two to
four seconds.
The vehicle speed profile during the third trajectory is shown
in Fig. 18. The car started to move after 5 min with high speed
before the speed goes down in the suburban area in Cochrane
city. At the 22nd min, the car pulled over on the road shoulder
for 3 min before moving again at high speed. At the end of
the test, the car moved into a suburban area at a low speed.
The speed was above 60 km/h for 64% of the time and more
than 80 km/h for 46% of the time.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
14
Fig. 17. Number of visible GPS/GLONASS satellites in the third trajectory.
After 40 min, the car passed under several consecutive bridges.
Fig. 18. Speed profile of the third trajectory. The speed was more than
60 km/h at 64% of the time.
Fig. 19 shows the horizontal position error of the PPP and
PPP/INS solutions versus time. Both solutions converged to
a 20 cm error level after 5.3 min. In the first 40 min where
the environment was mostly open-sky, the two solutions have
a comparable performance. However, once the car started to
move under the overpasses, the PPP solution suffered from
momentary GNSS outages, large error spikes up to 3.6 m,
and relatively long re-convergence periods. On the other hand,
the integrated PPP/RISS solution maintained the required
accuracy with maximum errors around 40 cm.
Fig. 20 zooms on the position errors in the last 30 minutes.
While the PPP solution struggled between outages to re-
converge, the integrated PPP/RISS solution provided a con-
tinuous solution less than 20 cm most of the time with
no or shorter convergence time after outages. The amount
of re-convergence time reduction is mainly affected by the
propagated INS accuracy at the end of the outage. This is why
the integration with INS can reduce the re-convergence time in
short GNSS outages better than long ones. During the GNSS
outages in this trajectory, the INS drift was relatively small,
which indicates that the INS states were well estimated before
the GNSS outages, especially the azimuth angle and the verti-
cal gyro bias. For example, the error in the estimated azimuth
angle with respect to the reference solution was around 0.5◦
Authorized licensed use limited to: TONGJI UNIVERSITY. Downloaded on June 11,2021 at 05:24:11 UTC from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
Fig. 19. Horizontal position error comparison in the third trajectory between
PPP and PPP/RISS solutions.
Fig. 20. Zoomed horizontal position error comparison in the third trajectory
with a focus on the last 30 min that includes ten consecutive partial and full
GNSS outages.
TABLE IV
C OMPARISON OF THE RMS AND M AXIMUM H ORIZONTAL P OSITION
E RRORS IN THE T HIRD T RAJECTORY
when the vehicle was moving under the overpasses. Besides,
using a high-end INS with low bias instability was also a
primary factor in limiting the error drift.
Table IV compares the rms and maximum errors of the
PPP and PPP/RISS solutions after 10 min of starting the
third trajectory. The integrated PPP/RISS solution has better
performance in terms of both rms and maximum error results.
The horizontal position errors of the developed PPP/RISS was
less than 50 cm, not only in the rms sense but also in the
maximum errors. The consideration of the maximum error is
crucial for safety applications.
Fig. 21 shows, on Google Earth, an example of one of
the GNSS outages where the car passed under the Highway
1A bridge, and there was no PPP solution for four seconds.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
ELSHEIKH et al.: INTEGRATION OF GNSS PPP AND RISS FOR LANE-LEVEL CAR NAVIGATION
Fig. 21. Example of the developed PPP/RISS system performance under an
overpass using Google Earth. The green color is for the reference, the black
for PPP, and the red for PPP/RISS solution.
The integrated solution continued to provide a high-precision
solution very close to the reference within the lane-level.
It is worth mentioning that the position errors after the
real GNSS outages in the second and third trajectories are
higher than the errors after the simulated GNSS outages in the
first trajectory. The real outages lasted two to four seconds,
whereas the simulated outages were introduced for ten sec-
onds. In simulated outages, GNSS observations are removed
during the outage; however, the quality of observations is kept
intact before and after the outages. On the other hand, in real
GNSS outages, the GNSS signal cannot go back to its nominal
strength in no time. Based on the surrounding environment,
the GNSS signal may gradually degrade before the outage,
and typically it retrieves its strength after the outage gradually.
This explains why PPP position errors are more significant in
real GNSS outages compared to simulated outages.
VII. C ONCLUSION
In this article, the mathematical model for the tightly
coupled integration of PPP with RISS was developed targeting
lane-level car navigation. Lane-level positioning is crucial for
safe driving features such as lane-keeping and lane-departure
warning. For such applications, it is not only the rms errors
that matter, but also the maximum errors. The developed
PPP/RISS system was tested in three road tests. The first road
test was performed at low speed and open-sky environments.
The results showed that PPP dominates the performance in
open-sky conditions. The benefit of the integration with RISS
was shown by introducing two 10 s simulated GNSS outages
where the integrated system maintained the continuity of
the positioning solution with a horizontal position rms error
of 10 cm and a maximum error less than 50 cm. Because
real GNSS outages are different and more significant than the
simulated outages, the second and third tests were performed
at high speed by driving on highways and passing under
several bridges and overpasses. The integrated PPP/RISS
system could maintain rms errors less than 20 cm in these
tests and maximum errors less than 50 cm. The integration
with RISS contributed to providing a continuous solution and
reducing the re-convergence time in most outages compared
to the solution with PPP alone. The use of RISS rather than
Authorized licensed use limited to: TONGJI UNIVERSITY. Downloaded on June 11,2021 at 05:24:11 UTC from IEEE Xplore. Restrictions apply.
15
traditional INS allows the use of high-end systems with more
than half the cost reduction. Besides, RISS also contributes to
reducing the computational cost and power consumption.
The developed system employed dual-frequency
GPS/GLONASS measurements. Further improvements
to the system performance can be achieved by using more
GNSS constellations and frequencies, especially with the
continuous evolution of GNSS systems such as Galileo and
BeiDuo, and GPS modernization. More constellations and
frequencies can improve the system accuracy, increase the
availability of the PPP solution, and reduce the convergence
time after GNSS outages. This work also utilized the final
IGS-MGEX corrections to establish an ideal environment for
PPP to demonstrate the impact of the integration with inertial
sensors; nevertheless, future developments will consider using
IGS RTS products. In this article, we used a single fiber-optic
gyroscope for the proposed PPP/RISS system to provide
the performance necessary for lane-level navigation. Our
next developments will focus on high-performance MEMS
IMUs, achieving the perfect compromise between the solution
quality and the IMU cost. The developed PPP/RISS can be
a part of new emerging applications such as autonomous
driving and ADAS that require decimeter-level accuracy
targeting lane-level operation.
ACKNOWLEDGMENT
The authors would like to thank Mitacs for providing
the main author with an industrial internship through their
Mitacs Accelerate Program. Thanks also go to the team of
Profound Positioning Inc., Calgary, Canada, for their help in
data collection.
R EFERENCES
[1] I. Skog and P. Handel, “In-car positioning and navigation technologies—
A survey,” IEEE Trans. Intell. Transp. Syst., vol. 10, no. 1, pp. 4–21,
Mar. 2009.
[2] A. Chen, A. Ramanandan, and J. A. Farrell, “High-precision lane-level
road map building for vehicle navigation,” in Proc. IEEE/ION Position,
Location Navigat. Symp. Indian Wells, CA, USA: IEEE, May 2010,
pp. 1035–1042.
[3] (2017). Lane Widths Guideline, Version 2.0, City of Toronto, Transporta-
tion Services. Accessed: Nov. 1, 2020. [Online]. Available: https://www.
toronto.ca/wp-content/uploads/2017/11/921b-ecs-specs-roaddg-Lane_
Widths_Guideline_Version_2.0_Jun2017.pdf
[4] P. Misra and P. Enge, Global Positioning System: Signals, Measurements
and Performance, 2nd ed. Lincoln, MA, USA: Ganga-Jamuna Press,
2006.
[5] S. Bisnath and Y. Gao, “Current state of precise point position-
ing and future prospects and limitations,” in Observing our Chang-
ing Earth, M. G. Sideris, Ed. Berlin, Germany: Springer, 2009,
pp. 615–623.
[6] P. D. Groves, Principles of GNSS, Inertial, and Multisensor Integrated
Navigation Systems. Norwood, MA, USA: Artech House, 2013.
[7] J. Georgy, A. Noureldin, M. J. Korenberg, and M. M. Bayoumi, “Low-
cost three-dimensional navigation solution for RISS/GPS integration
using mixture particle filter,” IEEE Trans. Veh. Technol., vol. 59, no. 2,
pp. 599–615, Feb. 2010.
[8] E. Shin and B. Scherzinger, “Inertially aided precise point positioning,”
in Proc. 22nd Int. Tech. Meeting Satell. Division Inst. Navigat. (ION
GNSS), Savannah, GA, USA, Sep. 2009, pp. 1892–1897.
[9] M. Abd Rabbou and A. El-Rabbany, “Tightly coupled integration of
GPS precise point positioning and MEMS-based inertial systems,” GPS
Solutions, vol. 19, no. 4, pp. 601–609, Oct. 2015.
[10] S. Liu, F. Sun, L. Zhang, W. Li, and X. Zhu, “Tight integration of
ambiguity-fixed PPP and INS: Model description and initial results,”
GPS Solutions, vol. 20, no. 1, pp. 39–49, Jan. 2016.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
16
[11] P. S. Henkel, “Tightly coupled precise point positioning and attitude
determination,” IEEE Trans. Aerosp. Electron. Syst., vol. 51, no. 4,
pp. 3182–3197, Oct. 2015.
[12] G. Zhang and C. Wilson, “An integrated DGPS/DR/Map system for
vehicle safety applications,” in Proc. Nat. Tech. Meeting Inst. Navigat.,
Anaheim, CA, USA, Jan. 2000, pp. 253–257.
[13] J. Wang, S. Schroedl, K. Mezger, R. Ortloff, A. Joos, and T. Passegger,
“Lane keeping based on location technology,” IEEE Trans. Intell.
Transp. Syst., vol. 6, no. 3, pp. 351–356, Sep. 2005.
[14] C. Guo, W. Guo, G. Cao, and H. Dong, “A lane-level LBS system for
vehicle network with high-precision BDS/GPS positioning,” Comput.
Intell. Neurosci., vol. 2015, pp. 1–13, Jan. 2015.
[15] Y. Ohshima et al., “Application of precise point positioning (PPP)
and precise real-time ephemeris to achieve ‘where-in-lane’ positioning
accuracy for vehicles,” in Proc. Pacific PNT Meeting (ION), Honolulu,
HI, USA, Apr. 2015, pp. 583–590.
[16] J. Kouba and P. Héroux, “Precise point positioning using IGS orbit and
clock products,” GPS Solutions, vol. 5, no. 2, pp. 12–28, Oct. 2001.
[17] M. Karaim, M. Elsheikh, and A. Noureldin, “GNSS error sources,” in
Multifunctional Operation and Application of GPS, R. B. Rustamov and
A. M. Hashimov, Eds. Rijeka, Croatia: InTech, 2018, ch. 4, pp. 69–85.
[18] J. Kouba, F. Lahaye, and P. Tétreault, “Precise point position-
ing,” in Springer Handbook of Global Navigation Satellite Systems,
P. J. Teunissen and O. Montenbruck, Eds. Springer, 2017, ch. 25,
pp. 723–751.
[19] J. J. Spilker, “Tropospheric effects on GPS,” in Global Positioning
System: Theory Applications, vol. 1. 1996, pp. 517–546.
[20] J. L. Davis, T. A. Herring, I. I. Shapiro, A. E. E. Rogers, and G. Elgered,
“Geodesy by radio interferometry: Effects of atmospheric modeling
errors on estimates of baseline length,” Radio Sci., vol. 20, no. 6,
pp. 1593–1607, Nov. 1985.
[21] H. Berg, Allgemeine Meteorologie. Berlin, Germany: Dümmlers Verlag
Bonn, 1948.
[22] A. E. Niell, “Global mapping functions for the atmosphere delay at
radio wavelengths,” J. Geophys. Res., Solid Earth, vol. 101, no. B2,
pp. 3227–3246, Feb. 1996.
[23] M. Håkansson, A. B. O. Jensen, M. Horemuz, and G. Hedling, “Review
of code and phase biases in multi-GNSS positioning,” GPS Solutions,
vol. 21, no. 3, pp. 849–860, Jul. 2017.
[24] A. Brandt and J. F. Gardner, “Constrained navigation algorithms for
strapdown inertial navigation systems with reduced set of sensors,”
in Proc. Amer. Control Conf. (ACC), Philadelphia, PA, USA, vol. 3,
Jun. 1998, pp. 1848–1852.
[25] X. Niu, S. Nasser, C. Goodall, and N. El-Sheimy, “A universal approach
for processing any MEMS inertial sensor configuration for land-vehicle
navigation,” J. Navigat., vol. 60, no. 2, pp. 233–245, May 2007.
[26] B. Phuyal, “An experiment for a 2-D and 3-D GPS/INS configuration
for land vehicle applications,” in Proc. Position Location Navigat. Symp.
(PLANS), Monterey, CA, USA, Apr. 2004, pp. 148–152.
[27] P. Daum, J. Beyer, and T. F. W. Kohler, “Aided inertial LAnd NAvigation
system (ILANA) with a minimum set of inertial sensors,” in Proc. IEEE
Position, Location Navigat. Symp. (PLANS), Las Vegas, NV, USA: IEEE,
Apr. 1994, pp. 284–291.
[28] D. Sun, M. G. Petovello, and M. E. Cannon, “GPS/Reduced IMU
with a local terrain predictor in land vehicle navigation,” Int. J. Nav-
igat. Observ., vol. 2008, pp. 1–15, Nov. 2008. [Online]. Available:
https://www.hindawi.com/journals/ijno/2008/813821/
[29] U. Iqbal, A. F. Okou, and A. Noureldin, “An integrated reduced
inertial sensor system—RISS/GPS for land vehicle,” in Proc. IEEE/ION
Position, Location Navigat. Symp., Monterey, CA, USA, May 2008,
pp. 1014–1021.
[30] A. Noureldin, T. B. Karamat, and J. Georgy, Fundamentals of Inertial
Navigation, Satellite-Based Positioning and Their Integration. Berlin,
Germany: Springer-Verlag, 2013.
[31] M. M. Atia, T. Karamat, and A. Noureldin, “An enhanced 3D multi-
sensor integrated navigation system for land-vehicles,” J. Navigat.,
vol. 67, no. 4, pp. 651–671, Jul. 2014.
[32] Z. M. Kassas and T. E. Humphreys, “Observability analysis of collabo-
rative opportunistic navigation with pseudorange measurements,” IEEE
Trans. Intell. Transp. Syst., vol. 15, no. 1, pp. 260–273, Feb. 2014.
[33] Z. M. Kassas, “Navigation with cellular signals,” in Position, Navigation,
and Timing Technologies in the 21st Century: Integrated Satellite
Navigation, Sensor Systems, and Civil Applications, vol. 2, Y. J. Morton,
F. van Diggelen, J. J. Spilker Jr., and B. W. Parkinson, Eds. Hoboken,
NJ, USA: Wiley, Jan. 2021.
Authorized licensed use limited to: TONGJI UNIVERSITY. Downloaded on June 11,2021 at 05:24:11 UTC from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
[34] P. F. de Bakker and C. C. J. M. Tiberius, “Real-time multi-GNSS single-
frequency precise point positioning,” GPS Solutions, vol. 21, no. 4,
pp. 1791–1803, Oct. 2017.
[35] G. Xu, GPS: Theory, Algorithms and Applications. Berlin, Germany:
Springer-Verlag, 2007.
[36] G. Blewitt, “An automatic editing algorithm for GPS data,” Geophys.
Res. Lett., vol. 17, no. 3, pp. 199–202, Mar. 1990.
[37] T. Takasu and A. Yasuda, “Cycle slip detection and fixing by MEMS-
IMU/GPS integration for mobile environment RTK-GPS,” in Proc. 21st
Int. Tech. Meeting Satell. Division The Inst. Navigat. (ION GNSS),
Savannah, GA, USA, 2008, pp. 64–71.
[38] IMU-CPT Datasheet. Accessed: Nov. 1, 2020. [Online]. Available:
https://hexagondownloads.blob.core.windows.net/public/Novatel/assets/
Documents/Papers/IMU-CPT/IMU-CPT.pdf
[39] IMU-KVH1750 Datasheet. Accessed: Nov. 1, 2020. [Online]. Available:
https://novatel.com/products/span-gnss-inertial-navigation-systems/
span-imus/imu-kvh1750
[40] O. Montenbruck et al., “The multi-GNSS experiment (MGEX) of the
international GNSS service (IGS)–achievements, prospects and chal-
lenges,” Adv. Space Res., vol. 59, no. 7, pp. 1671–1697, Apr. 2017.
[41] O. Montenbruck et al., “IGS-MGEX: Preparing the ground for multi-
constellation GNSS science,” Inside GNSS, vol. 9, no. 1, pp. 42–49,
Jan. 2014.
Mohamed Elsheikh (Member, IEEE) was born in
Tanta, Egypt, in 1984. He received the B.Sc. and
M.Sc. degrees in electrical engineering from Tanta
University, Tanta, in 2006 and 2013, respectively,
and the Ph.D. degree in electrical and computer
engineering from Queen’s University, Kingston, ON,
Canada, in 2019.
He has been a Teaching Assistant and Assistant
Lecturer with Tanta University since 2007. From
2014 to 2019, he was a Teaching and a Research
Assistant with Queen’s University and a member
with the Navigation and Instrumentation Research Group, Royal Military
College of Canada. He is currently a Post-Doctoral Associate with the
Geomatics Department, University of Calgary, Canada. His current research
interests include high-precision GNSS positioning and integrated navigation
systems.
Aboelmagd Noureldin (Senior Member, IEEE)
received the B.Sc. degree in electrical engineering
and the M.Sc. degree in engineering physics from
Cairo University, Egypt, in 1993 and 1997, respec-
tively, and the Ph.D. degree in electrical and com-
puter engineering from the University of Calgary,
AB, Canada, in 2002.
He is currently a Professor with the Depart-
ment of Electrical and Computer Engineering, Royal
Military College of Canada (RMCC) with Cross-
Appointment at the School of Computing and the
Department of Electrical and Computer Engineering, Queen’s University. He
is also the Founder and the Director of the Navigation and Instrumentation
Research Group, RMCC. He has published two books, four book chapters,
and more than 270 papers in journals, magazines, and conference proceedings.
His research interests include global navigation satellite systems, wireless
location and navigation, indoor positioning and multi-sensor fusion targeting
applications related to autonomous systems, intelligent transportation, road
information services, crowd management, and the vehicular Internet of Things.
His research work led to 13 patents and several technologies licensed to
industry in the area of position, location, and navigation systems.
Michael Korenberg received the B.Sc. and M.Sc.
degrees in mathematics and the Ph.D. degree in elec-
trical engineering from McGill University, Montreal,
QC, Canada.
He has been a Faculty Member with Queen’s
University since 1983, where he is currently a Pro-
fessor with the Department of Electrical and Com-
puter Engineering. His research interests include
the development of practical methods for represent-
ing/identifying nonlinear systems of unknown struc-
ture and for time-series analysis. Applications relate
to communications, to the modeling of physical and physiological processes,
and to computational biology and bioinformatics. The latter applications
include protein sequence classification, gene recognition, and interpretation
of microarray data.