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 1

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 I. I NTRODUCTION
precise positioning in many applications, including autonomous
car navigation. The safety features in autonomous driving and
Advanced Driver Assistance Systems (ADAS) require lane-level
T HE emerging applications in the automotive industry,
such as autonomous driving and Advanced Driver Assis-
tance Systems (ADAS) require navigation systems with high
positioning accuracy. Such accuracy can be obtained from the
Global Navigation Satellite Systems (GNSS) through either dif- accuracy, reliability, and integrity [1]. Safety features such
ferential techniques or Precise Point Positioning (PPP). PPP is as lane-departure warning cannot be achieved without lane-
currently favored over differential GNSS because it provides a level positioning accuracy. A common approach for lane-
global solution without the need for local reference stations. Nev- level positioning is to calculate the vehicle position in the
ertheless, employing PPP for land vehicles would be challenging
due to frequent signal degradation and blockage. Integrating Earth-Centered Earth-Fixed (ECEF) frame and then match this
PPP with an Inertial Navigation System (INS) can solve the location to a digital map [1], [2]. Therefore, the accuracy of
solution continuity problem; however, the INS solution drifts over both the ECEF position and the road map impacts the final
time, resulting in losing the desired accuracy. Implementing a solution [2]. The focus of this work is to enhance the estimated
reliable PPP/INS system that can preserve the required accuracy vehicle position in the ECEF frame.
is not trivial, especially with financial and computational cost
constraints. This article proposes the integration of PPP with GNSS positioning is a vital component of all car navigation
the Reduced Inertial Sensor System (RISS) for lane-level car systems. However, for lane-level positioning, the Standard
navigation. The high-precision needed in lane-level positioning Point Positioning (SPP) with meter-level accuracy is no longer
can be achieved by integrating PPP with high-end INS. Since sufficient. An average lane width can be approximately 3 m.
high-end INS are expensive, this work proposes the use of RISS For example, the lane widths guide of the city of Toronto in
instead of the traditional INS. RISS uses only one gyroscope and
two accelerometers, which can save more than half the high-end June 2017 has recommended lane widths between 3 and 4.3 m
INS cost. The proposed PPP/RISS system was tested through for the main driving lanes [3]. If the lane width is 3 m and the
three road tests that included highway driving under several car width is assumed to be 2 m, as shown in Fig. 1, positioning
overpasses. The system was able to maintain horizontal position accuracy of half the lane width might be enough to match
errors of less than 50 cm. the car to a specific lane; however, for safety applications,
Index Terms— Precise point positioning, global navigation accuracy less than 50 cm is required.
satellite system, inertial sensors, car navigation, lane-level, high- Centimeter-level GNSS accuracy can be obtained using Dif-
way driving, autonomous driving. ferential GNSS (DGNSS) or Precise Point Positioning (PPP)
[4]. The developed system is based on PPP since it requires
Manuscript received June 9, 2019; revised March 12, 2020, August 14, less infrastructure than DGNSS and provides a globally-
2020, and November 12, 2020; accepted November 23, 2020. This work consistent solution [5]. Still, GNSS positioning cannot be used
was supported by the Natural Sciences and Engineering Research Council of
Canada (NSERC) under Grant RGPIN-2020-03900 and Grant STPGP 521432. alone for vehicle navigation because the GNSS signal is prone
The Associate Editor for this article was Z. M. Kassas. (Corresponding to interference and obstruction [6]. For applications such as
author: Mohamed Elsheikh.) autonomous driving, multisensor fusion is typically the case
Mohamed Elsheikh is with the Department of Geomatics Engineering, Uni-
versity of Calgary, Calgary, AB T2N 1N4, Canada, also with the Department where sensors such as the Inertial Navigation System (INS),
of Electrical and Computer Engineering, Queen’s University, Kingston, ON odometer, Light Detection and Ranging (LiDAR), and cameras
K7L 3N6, Canada, and also with the Department of Electronics and Electrical can be employed. Each sensor has its pros and cons. For
Communication Engineering, Tanta University, Tanta 31512, Egypt (e-mail:
mohamed.elsheikh@ucalgary.ca). example, cameras are useful in detecting the road features,
Aboelmagd Noureldin is with the Department of Electrical and Computer but they cannot work in adverse weather conditions such as
Engineering, Queen’s University, Kingston, ON K7L 3N6, Canada, and rain and snow. On the other hand, the integrated GNSS/INS
also with the Department of Electrical and Computer Engineering, Royal
Military College of Canada, Kingston, ON K7K 7B4, Canada (e-mail: solution is weather independent but suffers from the effects of
aboelmagd.noureldin@rmc.ca). multipath and obstacles on GNSS signals and the error drift of
Michael Korenberg is with the Department of Electrical and Computer the INS. This article focuses on an efficient GNSS/INS fusion,
Engineering, Queen’s University, Kingston, ON K7L 3N6, Canada (e-mail:
korenber@queensu.ca). which, when integrated with other sensors, will contribute to
Digital Object Identifier 10.1109/TITS.2020.3040955 enhancing the overall system performance.
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.
 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

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◦ .
Fig. 1. Schematic diagram for the lane and car widths. Regarding lane-level positioning, early studies have used
the Differential Global Positioning System (DGPS) to obtain
Although the integration of PPP with high-end inertial a precise GNSS solution [12], [13]. In [12], DGPS data was
sensors can provide the required high-performance navigation fused with a fiber optic gyro and a wheel sensor before the
solution, the high cost of such a system is a significant solution is matched to a map. A test was performed on one
challenge. This work investigates the use of the Reduced lane on a highway with 4 to 7 visible satellites. There was
Inertial Sensor System (RISS) [7] instead of the traditional INS no reference, and the authors focused on showing the results
with a full Inertial Measurement Unit (IMU) to address the after the map matching with no details about the performance
INS cost problem. A full-IMU contains three accelerometers of DGPS or its integration with the Dead Reckoning (DR)
and three gyroscopes (gyros). On the other hand, RISS uses sensors. In [13], an integrated DGPS/INS system is used to
only one gyro and two accelerometers. Eliminating two gyros assist vision systems for lane-keeping. The results showed
and one accelerometer reduces the INS cost by half to two potential in using a high-precision GNSS/INS solution to add
thirds. Furthermore, reducing the number of inertial sensors smoothness and robustness to the vision solution.
involved in the mechanization algorithm limits the inherent The problem with DGNSS is the need for reference stations,
inertial sensor errors from the two removed gyroscopes from which means added infrastructure, radio links for commu-
propagating to the final output position solution, thus affecting nication, and a local positioning solution. Employing PPP
the overall positioning accuracy. RISS also contributes to for lane-level positioning was investigated in [14] and [15].
reducing system complexity and power consumption. Our In [14], the focus was on designing a real-time PPP framework.
objective is to develop a PPP/RISS integrated system that can A road test was performed on an urban road, and the PPP
meet the requirements of lane-level vehicle positioning. achieved an average of 50 cm accuracy away from down-
Since the advent of PPP, many research studies have investi- town and a few meters accuracy in challenging environments.
gated the performance of PPP/INS integration for land vehicle This research did not investigate the GNSS outages. On the
navigation, e.g., in [8]–[10]. The authors in [8] focused on other hand, in [15], the test included two underpasses, which
showing the benefit of Tightly-Coupled (TC) compared to caused solution unavailability and degradation. The authors
Loosely-Coupled (LC) PPP/INS integration using a tactical- in this research discussed the re-convergence time problem
grade IMU. The land vehicle test included four periods where and suggested the integration with DR sensors, but with no
the number of satellites dropped below four satellites. The TC implementation or results.
solution error remained within ±1 m in contrast to ±5 m for From the studies mentioned above, the implementation of
the LC solution. PPP/INS integration for lane-level positioning has not still
A PPP/INS tightly integrated system was presented in [9], been fully explored, especially in driving scenarios with real
where single differencing between satellites was applied. The GNSS outages. The main contribution of this work is to exploit
results showed decimeter-level accuracy with simulated GNSS the benefits of RISS to enable high-precision positioning for
outages of 10 s duration. However, the figures of the position lane-level navigation of automated vehicles. The mathematical
errors with simulated GNSS outages in [9] have shown instant model for integrating PPP with RISS in a tightly coupled
solution convergence after the outages. An instant convergence mode is developed. Results from road tests showed how RISS
of PPP after a simulated GNSS outage indicates that the contributed to maintaining the continuity of the navigation
ambiguities and its covariance were kept being updated during solution within the required accuracy level even in challeng-
the outage, which does not happen in a real GNSS outage. ing scenarios where the test vehicle goes under overpasses
The research in [10] investigated the performance of a TC and bridges. These results were achieved by using only one
integration of ambiguity-fixed PPP and INS. Two car-borne gyroscope, which reduces the financial and computational cost
tests were made; one in a small square and the other in a of such a high precision system. The developed system is
suburban environment. The presented system has demonstrated assessed based on the rms and maximum horizontal position
centimeter-level accuracy in open-sky environments after the errors. The maximum errors are crucial in the system evalua-
initial convergence. Furthermore, two simulated 10 s GNSS tion for safety applications.

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 3

The next section describes the utilized PPP model. is the speed of light, dtr, PI F is the receiver clock bias of the IF
Section III is a review of RISS describing its mechaniza- pseudorange observations in seconds, λ I F is the IF wavelength
tion equations in addition to its advantages and limitations. in meters, N I F denotes the IF float ambiguity in cycles, and
The developed PPP/RISS integration model is illustrated in  denotes the multipath and receiver noise in meters. It is
section IV. Section V shows the experimental setup in the per- worth mentioning that dtr, PI F was used in both pseudorange
formed road tests. The testing results are shown in section VI, and carrier phase observations since the difference between
focusing on how the developed PPP/RISS system could main- code (pseudorange) and phase biases is absorbed into the float
tain lane-level accuracy on highways. Finally, section VII ambiguity parameter.
concludes the presented work. The observations of the Russian Global Navigation Satellite
System (GLONASS) were also employed to increase the
II. P RECISE P OINT P OSITIONING M ODEL number of visible satellites. The corrected GLONASS mea-
In this work, the standard PPP model is adopted, which surements can be written as
is based on the ionospheric-free (IF) combinations of dual- 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
frequency GNSS measurements with float ambiguity resolu-
(5)
tion [16]. Besides, corrections for satellite orbits, clocks, phase s,R
and code biases from a global network are employed to achieve r, IF = ρrs + cdtr, PI F + I S BG−R + m sw Z r,w + λs,R s
I F NI F
the high precision of the PPP solution. The PPP algorithm also + r,s,R
I F (6)
applies corrections for the relativistic and Sagnac effects [17].
Furthermore, additional error sources have to be considered in where the superscript R refers to GLONASS. GLONASS
PPP that used to be ignored in SPP or canceled in DGNSS, observations have an extra unknown, which is the Inter-System
such as the phase wind-up, antenna phase center, and site Bias (ISB) between GPS and GLONASS clock systems,
displacements [18]. in meters. Besides, the IF wavelength for GLONASS is not
The tropospheric delays consist of dry and wet components. only different from GPS, but it is also different for each satel-
The dry component accounts for 90% of the delay, and it can lite due to the employment of the Frequency Division Multi-
be accurately modeled. In contrast, the wet component, which ple Access (FDMA) technique in GLONASS legacy signals.
is responsible for 10% of the delay, is difficult to model as the This frequency difference results in an additional satellite-
water vapor content varies locally [19]. For centimeter-level dependent bias called the Inter-Frequency Bias (IFB) or Inter-
positioning, the total tropospheric delay T for each satellite s Channel Bias (ICB). Nevertheless, in float-ambiguity PPP,
at the GNSS receiver r can be parametrized as [18] the phase IFBs are absorbed in the estimated float ambiguities
[23]. On the other hand, GLONASS code IFBs are difficult
Trs = m h (E s )Z r,h + m w (E s )Z r,w (1) to estimate, and there are no available corrections for them.
Besides, they only affect the initial PPP convergence, espe-
where E s is the satellite elevation angle, whereas Z r,h is the
cially in fixed-ambiguity PPP; therefore, they were neglected
zenith hydrostatic delay, and Z r,w is the zenith wet delay at
in the implemented float-ambiguity PPP.
the receiver location. The terms m h and m w represent the
Another GNSS useful observation is the Doppler measure-
elevation-dependent mapping functions of the hydrostatic and
ment. Doppler measurements can be used to calculate the
wet components, respectively.
GNSS receiver velocity. Furthermore, in GNSS/INS integra-
In our algorithm, Z r,h was calculated using [20]
tion, it can provide an update to the vehicle velocity, which can
0.0022768P contribute to a faster estimation of the INS attitude and IMU
Z r,h = (2)
1 − 0.00266 cos(2ϕ) − 2.8 ∗ 10−7 h errors [6]. The Doppler measurements D H z provided by GNSS
where P is the surface pressure in hPa, ϕ is the latitude in receivers represent the Doppler shift in Hz. The measured
degrees, and h is the height in meters. For centimeter-level PPP pseudorange rate D in m/s is obtained by multiplying the
accuracy, the pressure (in hPa) can be approximated by Berg’s Doppler shift by its corresponding wavelength with a negative
formula P = 1013.25(1 − 2.2557 ∗ 10−5 h)5.2568 [21]. On the sign
other hand, the wet component cannot be calculated with the D = −λ × D H z (7)
same accuracy and hence was estimated as an unknown in the
s
The corrected IF Doppler observation Dr,I
navigation filter. The mapping functions were computed using F is given by
Neill’s mapping functions [22].
˙ r, P +  Ds
Dr,s I F = ρ˙rs + cdt (8)
After applying the corrections mentioned above, the cor- IF r,I F
s,G s,G
rected IF GPS pseudorange Pr,I F and carrier phase r,I F It should be noted that the above equation can be used for
observations can be formulated as both GPS and GLONASS since the change of the receiver
Pr,s,G s,G clock is the same for both constellations as they use the same
I F = ρr + cdtr, PI F + m w Z r,w + r, PI F
s s
(3)
s,G oscillator.
r, IF = ρrs + cdtr, PI F + m sw Z r,w + λG s
I F NI F + r,s,G
I F

(4) III. T HE R EDUCED I NERTIAL S ENSOR S YSTEM

where the superscript G refers to GPS, ρrs
is the true geometric There is a growing demand for lower cost, lighter, and
range between the GNSS satellite s and receiver r in meters, c smaller navigation sensors in the land vehicle navigation

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.

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

TABLE I The measurement error vector represents the difference

T HE I NITIAL STD OF THE S YSTEM E RROR S TATES AND A SSUMED between the predicted values calculated from RISS and the
P ROCESS N OISE STD. T HE IMU-R ELATED S TATES A RE FOR IMU-
KVH1750
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
where I is the identity matrix, and t is the time interval page, where e is the Earth’s eccentricity.
between the current and previous IMU measurement epochs. The matrix Hv is similar to H p with replacing the matrix
In the prediction phase of EKF, the state transition matrix in L by the 3 × 3 rotation matrix Rle which rotates the velocity
(33) is used to predict the error state vector δXˆ −
k and its a priori errors from the LLF frame (δv e , δv n , δv u ) to the ECEF frame
covariance Pk− at epoch k based on their values from the (δv x , δv y , δv z ) [30], and is given by
previous epoch. The EKF was implemented in the closed-loop ⎡ ⎤
configuration; hence, the errors states are reset every epoch, − sin(λ) − sin(ϕ) cos(λ) cos(ϕ) cos(λ)
and the prediction equations can be written as Rle = ⎣ cos(λ) − sin(ϕ) sin(λ) cos(ϕ) sin(λ) ⎦ (41)
0 cos(ϕ) sin(ϕ)
δXˆ −
k = 0 (34)
The vector mw maps the tropospheric zenith wet delay to
Pk− = +
φk−1 Pk−1 φk−1
T
+ Q k−1 (35)
its corresponding slant delay based on the elevation angle,
+
where Pk−1 is the posterior state covariance matrix of the as described in section II. Finally, the vector hb contains the
previous epoch. coefficients for the δd Br,G−R term, which means its elements
It is worth mentioning that the initial P matrix is diagonal, equal one for GLONASS satellites and zero for GPS satellites.
and is set based on the uncertainty of the initial states provided Assuming that MG is the number of GPS satellites, M R is the
to the system. In designing the initial values of the P matrix, number of GLONASS satellites, and the observation vectors
we shall avoid choosing very small initial values of the state are ordered to have GPS satellites first and then GLONASS
covariance to prevent divergence of the estimator (EKF) or rel- satellites, hb can be written as
atively large initial values to avoid long convergence time. The  
0
Q matrix is constant diagonal and is set based on the process hb = MG ×1 (42)
noise covariance assumed with each state. The values of the Q 1 M R ×1
matrix are determined based on the stochastic model of each To calculate the diagonal measurement covariance matrix
state. The datasheet of the IMU sensors can provide a good R, the measurements standard deviation was assumed to be
start to help to set the Q matrix; however, further tuning is inversely proportional to the squared sine of the satellite eleva-
typically required to achieve the required performance. Table I tion angle. Besides, an elevation mask of 7◦ , obtained by trial
contains the initial standard deviation (STD) of the system and error, was applied to reject low-elevation satellites. The
error states and the assumed process noise STD. carrier phase standard deviation was scaled to be 100 times
less than the code observations because of its high precision.
B. Measurement Model Due to the differences in the signal-to-noise ratio and the
The EKF measurement model can be expressed as signal structure between GPS and GLONASS, the standard
deviation of GLONASS code and Doppler observations was
δZ = H δX + η (36)
scaled by a factor of 3 and the phase observations by a factor
where δZ is the measurement error vector, H is the measure- of 1.5 compared to the corresponding GPS observations [34].
ment design matrix, and η is the measurement white Gaussian Table II summarizes the adopted numerical values for the
noise with covariance matrix R. standard deviation of the utilized GNSS observations.

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.

8 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS

TABLE II the EKF, the new estimated covariance values are carefully
T HE A DOPTED N UMERICAL VALUES FOR THE S TANDARD D EVIATION OF updated in the big P matrix.
THE U TILIZED GNSS O BSERVATIONS
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,
The update equations of the EKF can be written as there are two ways to deal with it. The first approach is to try
K k = Pk− HkT {Hk Pk− HkT + Rk }−1 (43) repairing it; nevertheless, an incorrect reparation will affect all
the following observations. The second approach is to reset
δXˆ + ˆ− ˆ−
k = δXk + K k (δZk − Hk δXk ) (44) the ambiguity parameter in the navigation filter to estimate
Pk+ = (I − K k Hk )Pk− (45) the new ambiguity, which is a safer method [35] despite the
required re-convergence time.
where K k is the Kalman gain.
There are several cycle slip detection methods in the lit-
erature; however, none can achieve 100% detection success.
C. Practical Considerations Each method has advantages and limitations. In the developed
Although the description of the system and measurement algorithm, the cycle slip detection module uses two methods to
models of the EKF seems to be sufficient for implementing detect the cycle slips from the PPP measurements. A cycle slip
the integration algorithm, other practical considerations should flag is raised if any of the two methods detected a cycle slip.
be taken into account for an efficient implementation of the The two methods, initially introduced in [36], are using the
system. In this section, three examples of these practical Melbourne-Wübbena (MW) function and using the Geometry-
considerations are briefly discussed. Free (GF) phase combination.
1) Dynamic Satellite Visibility: Due to the use of carrier The MW function can detect large cycle slips but might
phase observations, the system model has an ambiguity state not detect the small cycle slips and cycle slips that occur
for each satellite. The total number of system states in the equally on the two GNSS frequencies used to form the MW
developed TC PPP/RISS model is 13 + M; nevertheless, combination. Besides, the MW function is affected by the
the visible satellites vary with time due to the change in noise in the code observation. On the other hand, the GF phase
the satellite geometry and the signal blockage. Thus, care has low noise and can be used to detect small cycle slips,
must be taken between different processing epochs for the and equal cycle slips on both frequencies. However, the GF
appearance or disappearance of satellites, especially when phase cannot detect some combinations of cycle slips, and
updating the covariance of the float ambiguities in the P large changes in the ionospheric delay between the consecutive
matrix. epochs can affect the detection success.
A straightforward solution to this issue is to assume that all One of the advantages of the integration with INS is that
the satellites exist and design the system matrices based on the it can also contribute to cycle slip detection, which means
maximum possible number of satellites. However, using these an extra level of detection. In the implemented algorithm,
matrices in the EKF equations leads to high computational cost the carrier phase innovation sequence, which is the difference
and requires dealing with matrices with many zero entries and between the PPP measurements and the predicted measure-
might result in singularity issues. In the developed algorithm, ments from RISS, is checked against a predefined threshold
the system is designed for a maximum number of satellites = similar to the algorithm in [37]. If the innovation sequence
32 GPS + 26 GLONASS = 58, and hence, the total number was greater than the threshold, a cycle slip is declared.
of states is 58+13 = 71. However, for every epoch, before 3) Outlier Detection: Measurement outliers may seriously
applying the EKF equations, a submatrix P  is taken from deteriorate the system performance if not handled properly. For
the original P matrix such that it contains only the covariance INS, the measurement outliers can result from sensor failures,
information for the 13+ M states in this epoch. After applying large sensor noise, or unexpected environment disruption.

⎡ ⎤
Hp 0 M×3 0 M×3 1 M×1 hb 0 M×1 mw 0 M×M
H = ⎣ Hp 0 M×3 0 M×3 1 M×1 hb 0 M×1 mw λ I F I M×M ⎦ . (38)
0 M×3 Hv 0 M×3 0 M×1 0 M×1 1 M×1 0 M×1 0 M×M

⎡ ⎤
−(R N + h) sin ϕ cos λ −(R N + h) cos ϕ sin λ cos ϕ cos λ
L = ⎣ −(R N + h) sin ϕ sin λ (R N + h) cos ϕ cos λ cos ϕ sin λ ⎦ (40)
(R N (1 − e2 ) + h) cos ϕ 0 sin ϕ

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 9

In the implemented RISS, the IMU measurements are down- B. IMUs

sampled by averaging from 100 Hz or 200 Hz to 1 Hz, which Two IMUs were used in these tests:
contributes to smoothing the measurements and reducing the • IMU-CPT: a tactical-grade IMU with 20◦ gyro bias
effect of the outliers. offset, 1◦ / hr gyro bias stability, and 100 Hz output rate
GNSS outliers can arise from multipath effects. If passed [38]. It has fiber-optic gyros and MEMS accelerometers,
the cycle slip detection module, the phase cycle slips will and it was used in the Kingston trajectory.
also look like an outlier. In the implemented algorithm, two • IMU-KVH1750: a tactical-grade IMU with 2◦ gyro bias
steps of outlier detection were applied in a coarse-to-fine offset, 0.05◦ / hr gyro bias stability, and 200 Hz output
approach. The first step is done before applying the EKF, rate [39]. It has fiber-optic gyros and MEMS accelerom-
where the measurement error vector δ Z is tested against a eters, and it was used in Calgary trajectories.
relatively large threshold, e.g., 20 m for the code observations.
The threshold value can be tuned based on the available data
C. Correction Products
sets, and it can also be different for each type of GNSS
observations. Since GLONASS measurements were also employed,
The second step is based on testing the measurement resid- the final precise satellite orbit and clock corrections from the
uals after applying the EKF. The corrected PPP measurements IGS Multi-GNSS Experiment (MGEX) project were used [40],
and the predicted RISS measurements are re-calculated based [41]. The chosen corrections included precise satellite orbits
on the new estimated parameters. Then, a new measurement every 5 min, and precise satellite clock corrections every 30 s.
error vector δY is formed that represents the solution residuals. Although the developed algorithm used the final IGS cor-
The normalized residual δY ¯ k, j for the j t h measurement is rections, in reality, other real-time correction services such as
calculated by dividing the residual value by its posterior the IGS Real-time Service (IGS-RTS) can be used to support
standard deviation the real-time realization of our method. It is worth mentioning
that real-time corrections will be less accurate and will impact
Ck = Hk Pk+ HkT + Rk (46) the overall system accuracy. Therefore, to establish an ideal
δY environment that examines PPP/RISS integration benefits, we
¯ k, j =  k, j
δY (47)
Ck,( j, j ) decided to use the final IGS corrections in this work. In the
future, real-time correction services can still be used, and the
where Ck is the posterior measurement covariance matrix at proposed method can be employed to operate in real-time
epoch k. Ck comprises the estimated covariance of the state without any limitations.
transformed into the measurement domain and the measure-
ment noise covariance.
D. Reference System
Finally, the normalized residual is compared to a threshold
to decide if the measurement is an outlier. A threshold of 3 was An accurate and precise reference solution is vital in the
chosen in the implemented algorithm, which means that the evaluation of high-precision systems. The NovAtel SPAN unit
measurement will be rejected if the residual value exceeded measurements were used to generate the reference solution.
three standard deviations. If a code outlier is detected, its However, to obtain the most accurate reference, the SPAN
corresponding phase and Doppler measurements must also be data were used with data from a base station to generate a
rejected, but not vice versa [6]. TC DGNSS/INS solution. The reference solution is gener-
ated in post-mission mode using NovAtel’s Waypoint Inertial
Explorer Software.
V. E XPERIMENTAL S ETUP
In the Kingston trajectory, the base station data was col-
The developed PPP/RISS system was tested using data lected using NovAtel OEM4 ProPakG2plus GNSS receiver,
collected from road tests. Three trajectories have been per- and the antenna was mounted on the roof of one of the build-
formed, one in Kingston, Ontario, and the other two in ings of the Royal Military College of Canada. Two-day static
Calgary, Alberta, Canada. The following is a description of observations were collected and post-processed to estimate the
the equipment and software used in these tests. base station accurate position. For the Calgary trajectories,
the UCAL IGS station was used as the base station. UCAL
A. Rover GNSS Receivers observations and reference position are available through IGS
services.
The GNSS data were logged at a 1 Hz rate. Two GNSS
receivers were used in these tests:
E. Software
• NovAtel SPAN OEMV Receiver: a dual-frequency
GNSS receiver installed inside the NovAtel SPAN-SE The TC PPP/RISS algorithm is developed using MATLAB
unit. This receiver was used as the rover receiver in the R2017a. The development includes processing the input data
nKingston trajectory. and the corrections data.
• NovAtel SPAN OEM6 Receiver: a dual-frequency
GNSS receiver installed inside the NovAtel SPAN on F. Equipment Setup
ProPak6 unit. This receiver was used as the rover receiver To collect the test data, all the equipment except the
in Calgary trajectories. base station receiver was carried in the testing vehicle. The

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.

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 11

Fig. 7. Speed profile of the first trajectory. The speed was on average 15 km/h. 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
Fig. 8. Horizontal position error comparison of the first trajectory between outage occurred after the filter is converged, RISS, or generally
the integrated PPP/RISS and PPP solutions. 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 GNSS PPP and RISS information, which explains why the PPP solution covariance after the outage. The longer the
the integrated solution is very close to the PPP-only solution GNSS outage, the less aiding is expected from the INS.
but not the same. Since the ambiguity resolution is directly affected by the
To explore the real benefit of the integration with RISS, two accuracy of the estimated position, the ambiguity resolution is
10 s simulated GNSS outages were introduced after 20 and also affected by the INS aiding. Fig. 10 shows an example of
28.5 min from the beginning of the trajectory. Fig. 9 shows the the ambiguity resolution using the estimated IF ambiguity of
horizontal error comparison after applying these two outages. the GPS satellite G13 during the first test trajectory with the
The selected outages represent two different examples of the two simulated outages. This figure shows how the integration
performance during GNSS outages. The integrated PPP/RISS with RISS affected the ambiguity resolution which is reflected
solution did not exhibit significant drift during the first GNSS in the positioning solution convergence in Fig. 9. Nevertheless,
outage, and hence continued after the outage with an error less the effect of INS aiding on the ambiguity resolution can be
than 20 cm and gradually started to approach the PPP solution. more significant when ambiguity fixing is considered.
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. B. Road Test Trajectory 2: Calgary Highway I
On the other hand, during the second GNSS outage, the inte- The second trajectory was meant to test the developed
grated solution drifted to a maximum error of 40.5 cm. The system in a real highway driving scenario which lasted
amount of drift during the GNSS outage depends mainly 74 minutes. The test trajectory, shown in Fig. 11, started
on the accuracy of the estimated vertical gyro bias and the on Highway 1 (Trans-Canada Highway), then the car moved
other INS navigation error states, and also the dynamics of through three rectangular loops on several highways, next to
the vehicle before and during the outage. Despite the drift, Calgary, in open-sky conditions. In the last loop, the car moved
the integrated solution was better than the PPP solution after far south and then back through highway 201 north, passing
the GNSS outage. The integrated solution converged to the under two overpasses before the test ends in a suburban area.

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.

12 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS

Fig. 10. Example of how the integration with RISS can affect the ambiguity Fig. 12. Number of visible GPS/GLONASS satellites in the second trajectory.
resolution using the estimated IF ambiguity of the GPS satellite G13. 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

Fig. 11. The second test trajectory, Calgary, Alberta, Canada. 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
The IMU-KVH was utilized as a tactical-grade IMU in this of Precision (DOP) value, which was 0.6 at the beginning of
test. The baseline length with respect to the UCAL reference the test. Another factor is that the car started to move after
station was less than 20 km. 42 s, which resulted in different measurements every epoch
Fig. 12 shows the number of visible GPS/GLONASS satel- that could contribute to the faster estimation of the EKF error
lites during the test. At least ten satellites were visible most states. When the number of satellites dropped to 5 in the 32nd
of the test time, with a one-time drop to 5 satellites during minute, the PPP solution had an error spike of 23 cm, whereas
the open-sky driving. The car passed under the first overpass the integrated PPP/RISS solution was not affected by this drop
(under Highway 1) at the end of the 65t h minute. The number and provided a smooth solution.
of satellites became zero for two seconds, and then four Fig. 15 zooms on the part when the car moved under the
for one second before going back to the regular visibility. two overpasses at the end of the trajectory; the PPP solution
The second overpass was at the 70t h minute, and the number of had four seconds with no output solution for each overpass
satellites became five for one second, followed by two seconds followed by a high position error in the epoch right after
with zero visible satellites. the GNSS outage. The position error was 48.5 cm after the
Fig. 13 shows the speed of the vehicle during the test. The first overpass and 121.5 cm after the second overpass. On the
car was static for only 42 s before it started to move. The other hand, the integrated solution provided a continuous
speed was more than 60 km/h for 66% of the time; however, solution with an error of 26.5 cm after the first GNSS outage
there were several stops at the intersections. The 5 min stop and 50.5 cm after the second GNSS outage. Nevertheless,
that occurred after 15.5 min was due to a railway crossing. the integrated solution was generally worse than the PPP

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 13

TABLE III
C OMPARISON OF THE RMS AND M AXIMUM H ORIZONTAL P OSITION
E RRORS IN THE S ECOND T RAJECTORY

Fig. 14. Horizontal position error comparison of the second trajectory

between the integrated PPP/RISS and PPP solutions.

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
Fig. 15. Zoomed horizontal position error comparison of the second trajectory shows the test trajectory, which started outside Calgary on
with a focus on the part where the car moved under the two overpasses. Highway 1A. The car moved west to the city of Cochrane,
where there was a suburban environment, and then moved
solution after the 67t h minute. In this period, the solution south on Highway 22 and east on Highway 1 towards Calgary.
was re-converging after the two GNSS outages with a short In Calgary, the car moved on Highway 201, passing under
time in between, and the car was moving at speed above several consecutive overpasses, and finally, the test ended in a
80 km/hr. Since the odometer measurements are typically suburban area. The IMU-KVH was utilized as a tactical-grade
provided at a low data rate (1Hz), error accumulation may IMU in this test. The baseline length with respect to the UCAL
result in significant position errors when driving at relatively reference station was less than 30 km.
high speeds. When the odometer speed was replaced by the The number of visible GPS/GLONASS satellites in the
reference speed from the Novatel system (to validate the above third trajectory is shown in Fig. 17. More than eight satellites
explanation), the PPP/RISS solution became better than the were visible in the first 40 minutes of the test. After that,
PPP-only solution after the second outage. This indicates that on Highway 201, the car passed under ten consecutive road
acquiring the speed from OBDII at a low rate and with the and pedestrian overpasses. The number of satellites dropped to
odometer errors can lead to relatively large errors after GNSS zero (full outage) eight times and was less than five satellites
outages depending on the vehicle dynamics. two times (partial outages). The significant challenge in this
Table III compares the rms and maximum errors of the trajectory is that the duration between the outages was between
PPP and the PPP/RISS solutions after 10 min of starting the 0.5 to 2.5 min, which is a short period for the PPP solution
test. The first 10 minutes were excluded to avoid the initial to converge. The GNSS outage duration ranged from two to
convergence time and keep consistency with the previous test. four seconds.
The rms errors are almost the same for the two solutions, The vehicle speed profile during the third trajectory is shown
which is expected since the two outages are short, and the in Fig. 18. The car started to move after 5 min with high speed
re-convergence time of both solutions was close. The PPP before the speed goes down in the suburban area in Cochrane
solution had a relatively fast convergence time during this city. At the 22nd min, the car pulled over on the road shoulder
test, and hence, there was no benefit from the integration for 3 min before moving again at high speed. At the end of
in this regard. However, the main benefit of the integration the test, the car moved into a suburban area at a low speed.
is smoothing the solution and removing the error spikes, The speed was above 60 km/h for 64% of the time and more
as seen from the maximum error comparison, in addition to than 80 km/h for 46% of the time.

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.

14 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS

Fig. 17. Number of visible GPS/GLONASS satellites in the third trajectory. Fig. 19. Horizontal position error comparison in the third trajectory between
After 40 min, the car passed under several consecutive bridges. PPP and PPP/RISS solutions.

Fig. 18. Speed profile of the third trajectory. The speed was more than Fig. 20. Zoomed horizontal position error comparison in the third trajectory
60 km/h at 64% of the time. with a focus on the last 30 min that includes ten consecutive partial and full
GNSS outages.
Fig. 19 shows the horizontal position error of the PPP and
PPP/INS solutions versus time. Both solutions converged to TABLE IV
a 20 cm error level after 5.3 min. In the first 40 min where C OMPARISON OF THE RMS AND M AXIMUM H ORIZONTAL P OSITION
E RRORS IN THE T HIRD T RAJECTORY
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. when the vehicle was moving under the overpasses. Besides,
While the PPP solution struggled between outages to re- using a high-end INS with low bias instability was also a
converge, the integrated PPP/RISS solution provided a con- primary factor in limiting the error drift.
tinuous solution less than 20 cm most of the time with Table IV compares the rms and maximum errors of the
no or shorter convergence time after outages. The amount PPP and PPP/RISS solutions after 10 min of starting the
of re-convergence time reduction is mainly affected by the third trajectory. The integrated PPP/RISS solution has better
propagated INS accuracy at the end of the outage. This is why performance in terms of both rms and maximum error results.
the integration with INS can reduce the re-convergence time in The horizontal position errors of the developed PPP/RISS was
short GNSS outages better than long ones. During the GNSS less than 50 cm, not only in the rms sense but also in the
outages in this trajectory, the INS drift was relatively small, maximum errors. The consideration of the maximum error is
which indicates that the INS states were well estimated before crucial for safety applications.
the GNSS outages, especially the azimuth angle and the verti- Fig. 21 shows, on Google Earth, an example of one of
cal gyro bias. For example, the error in the estimated azimuth the GNSS outages where the car passed under the Highway
angle with respect to the reference solution was around 0.5◦ 1A bridge, and there was no PPP solution for four seconds.

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 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
Fig. 21. Example of the developed PPP/RISS system performance under an PPP to demonstrate the impact of the integration with inertial
overpass using Google Earth. The green color is for the reference, the black
for PPP, and the red for PPP/RISS solution. sensors; nevertheless, future developments will consider using
IGS RTS products. In this article, we used a single fiber-optic
The integrated solution continued to provide a high-precision gyroscope for the proposed PPP/RISS system to provide
solution very close to the reference within the lane-level. the performance necessary for lane-level navigation. Our
It is worth mentioning that the position errors after the next developments will focus on high-performance MEMS
real GNSS outages in the second and third trajectories are IMUs, achieving the perfect compromise between the solution
higher than the errors after the simulated GNSS outages in the quality and the IMU cost. The developed PPP/RISS can be
first trajectory. The real outages lasted two to four seconds, a part of new emerging applications such as autonomous
whereas the simulated outages were introduced for ten sec- driving and ADAS that require decimeter-level accuracy
onds. In simulated outages, GNSS observations are removed targeting lane-level operation.
during the outage; however, the quality of observations is kept
ACKNOWLEDGMENT
intact before and after the outages. On the other hand, in real
GNSS outages, the GNSS signal cannot go back to its nominal The authors would like to thank Mitacs for providing
strength in no time. Based on the surrounding environment, the main author with an industrial internship through their
the GNSS signal may gradually degrade before the outage, Mitacs Accelerate Program. Thanks also go to the team of
and typically it retrieves its strength after the outage gradually. Profound Positioning Inc., Calgary, Canada, for their help in
This explains why PPP position errors are more significant in data collection.
real GNSS outages compared to simulated outages.
R EFERENCES
[1] I. Skog and P. Handel, “In-car positioning and navigation technologies—
VII. C ONCLUSION A survey,” IEEE Trans. Intell. Transp. Syst., vol. 10, no. 1, pp. 4–21,
Mar. 2009.
In this article, the mathematical model for the tightly [2] A. Chen, A. Ramanandan, and J. A. Farrell, “High-precision lane-level
coupled integration of PPP with RISS was developed targeting road map building for vehicle navigation,” in Proc. IEEE/ION Position,
lane-level car navigation. Lane-level positioning is crucial for Location Navigat. Symp. Indian Wells, CA, USA: IEEE, May 2010,
pp. 1035–1042.
safe driving features such as lane-keeping and lane-departure [3] (2017). Lane Widths Guideline, Version 2.0, City of Toronto, Transporta-
warning. For such applications, it is not only the rms errors tion Services. Accessed: Nov. 1, 2020. [Online]. Available: https://www.
that matter, but also the maximum errors. The developed toronto.ca/wp-content/uploads/2017/11/921b-ecs-specs-roaddg-Lane_
Widths_Guideline_Version_2.0_Jun2017.pdf
PPP/RISS system was tested in three road tests. The first road [4] P. Misra and P. Enge, Global Positioning System: Signals, Measurements
test was performed at low speed and open-sky environments. and Performance, 2nd ed. Lincoln, MA, USA: Ganga-Jamuna Press,
The results showed that PPP dominates the performance in 2006.
[5] S. Bisnath and Y. Gao, “Current state of precise point position-
open-sky conditions. The benefit of the integration with RISS ing and future prospects and limitations,” in Observing our Chang-
was shown by introducing two 10 s simulated GNSS outages ing Earth, M. G. Sideris, Ed. Berlin, Germany: Springer, 2009,
where the integrated system maintained the continuity of pp. 615–623.
[6] P. D. Groves, Principles of GNSS, Inertial, and Multisensor Integrated
the positioning solution with a horizontal position rms error Navigation Systems. Norwood, MA, USA: Artech House, 2013.
of 10 cm and a maximum error less than 50 cm. Because [7] J. Georgy, A. Noureldin, M. J. Korenberg, and M. M. Bayoumi, “Low-
real GNSS outages are different and more significant than the cost three-dimensional navigation solution for RISS/GPS integration
using mixture particle filter,” IEEE Trans. Veh. Technol., vol. 59, no. 2,
simulated outages, the second and third tests were performed pp. 599–615, Feb. 2010.
at high speed by driving on highways and passing under [8] E. Shin and B. Scherzinger, “Inertially aided precise point positioning,”
several bridges and overpasses. The integrated PPP/RISS in Proc. 22nd Int. Tech. Meeting Satell. Division Inst. Navigat. (ION
GNSS), Savannah, GA, USA, Sep. 2009, pp. 1892–1897.
system could maintain rms errors less than 20 cm in these [9] M. Abd Rabbou and A. El-Rabbany, “Tightly coupled integration of
tests and maximum errors less than 50 cm. The integration GPS precise point positioning and MEMS-based inertial systems,” GPS
with RISS contributed to providing a continuous solution and 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
reducing the re-convergence time in most outages compared ambiguity-fixed PPP and INS: Model description and initial results,”
to the solution with PPP alone. The use of RISS rather than GPS Solutions, vol. 20, no. 1, pp. 39–49, Jan. 2016.

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.

16 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS

[11] P. S. Henkel, “Tightly coupled precise point positioning and attitude [34] P. F. de Bakker and C. C. J. M. Tiberius, “Real-time multi-GNSS single-
determination,” IEEE Trans. Aerosp. Electron. Syst., vol. 51, no. 4, frequency precise point positioning,” GPS Solutions, vol. 21, no. 4,
pp. 3182–3197, Oct. 2015. pp. 1791–1803, Oct. 2017.
[12] G. Zhang and C. Wilson, “An integrated DGPS/DR/Map system for [35] G. Xu, GPS: Theory, Algorithms and Applications. Berlin, Germany:
vehicle safety applications,” in Proc. Nat. Tech. Meeting Inst. Navigat., Springer-Verlag, 2007.
Anaheim, CA, USA, Jan. 2000, pp. 253–257. [36] G. Blewitt, “An automatic editing algorithm for GPS data,” Geophys.
[13] J. Wang, S. Schroedl, K. Mezger, R. Ortloff, A. Joos, and T. Passegger, Res. Lett., vol. 17, no. 3, pp. 199–202, Mar. 1990.
[37] T. Takasu and A. Yasuda, “Cycle slip detection and fixing by MEMS-
“Lane keeping based on location technology,” IEEE Trans. Intell.
IMU/GPS integration for mobile environment RTK-GPS,” in Proc. 21st
Transp. Syst., vol. 6, no. 3, pp. 351–356, Sep. 2005.
Int. Tech. Meeting Satell. Division The Inst. Navigat. (ION GNSS),
[14] C. Guo, W. Guo, G. Cao, and H. Dong, “A lane-level LBS system for
Savannah, GA, USA, 2008, pp. 64–71.
vehicle network with high-precision BDS/GPS positioning,” Comput. [38] IMU-CPT Datasheet. Accessed: Nov. 1, 2020. [Online]. Available:
Intell. Neurosci., vol. 2015, pp. 1–13, Jan. 2015. https://hexagondownloads.blob.core.windows.net/public/Novatel/assets/
[15] Y. Ohshima et al., “Application of precise point positioning (PPP) Documents/Papers/IMU-CPT/IMU-CPT.pdf
and precise real-time ephemeris to achieve ‘where-in-lane’ positioning [39] IMU-KVH1750 Datasheet. Accessed: Nov. 1, 2020. [Online]. Available:
accuracy for vehicles,” in Proc. Pacific PNT Meeting (ION), Honolulu, https://novatel.com/products/span-gnss-inertial-navigation-systems/
HI, USA, Apr. 2015, pp. 583–590. span-imus/imu-kvh1750
[16] J. Kouba and P. Héroux, “Precise point positioning using IGS orbit and [40] O. Montenbruck et al., “The multi-GNSS experiment (MGEX) of the
clock products,” GPS Solutions, vol. 5, no. 2, pp. 12–28, Oct. 2001. international GNSS service (IGS)–achievements, prospects and chal-
[17] M. Karaim, M. Elsheikh, and A. Noureldin, “GNSS error sources,” in lenges,” Adv. Space Res., vol. 59, no. 7, pp. 1671–1697, Apr. 2017.
Multifunctional Operation and Application of GPS, R. B. Rustamov and [41] O. Montenbruck et al., “IGS-MGEX: Preparing the ground for multi-
A. M. Hashimov, Eds. Rijeka, Croatia: InTech, 2018, ch. 4, pp. 69–85. constellation GNSS science,” Inside GNSS, vol. 9, no. 1, pp. 42–49,
[18] J. Kouba, F. Lahaye, and P. Tétreault, “Precise point position- Jan. 2014.
ing,” in Springer Handbook of Global Navigation Satellite Systems,
Mohamed Elsheikh (Member, IEEE) was born in
P. J. Teunissen and O. Montenbruck, Eds. Springer, 2017, ch. 25,
Tanta, Egypt, in 1984. He received the B.Sc. and
pp. 723–751.
M.Sc. degrees in electrical engineering from Tanta
[19] J. J. Spilker, “Tropospheric effects on GPS,” in Global Positioning University, Tanta, in 2006 and 2013, respectively,
System: Theory Applications, vol. 1. 1996, pp. 517–546.
and the Ph.D. degree in electrical and computer
[20] J. L. Davis, T. A. Herring, I. I. Shapiro, A. E. E. Rogers, and G. Elgered, engineering from Queen’s University, Kingston, ON,
“Geodesy by radio interferometry: Effects of atmospheric modeling Canada, in 2019.
errors on estimates of baseline length,” Radio Sci., vol. 20, no. 6, He has been a Teaching Assistant and Assistant
pp. 1593–1607, Nov. 1985. Lecturer with Tanta University since 2007. From
[21] H. Berg, Allgemeine Meteorologie. Berlin, Germany: Dümmlers Verlag 2014 to 2019, he was a Teaching and a Research
Bonn, 1948. Assistant with Queen’s University and a member
[22] A. E. Niell, “Global mapping functions for the atmosphere delay at with the Navigation and Instrumentation Research Group, Royal Military
radio wavelengths,” J. Geophys. Res., Solid Earth, vol. 101, no. B2, College of Canada. He is currently a Post-Doctoral Associate with the
pp. 3227–3246, Feb. 1996. Geomatics Department, University of Calgary, Canada. His current research
[23] M. Håkansson, A. B. O. Jensen, M. Horemuz, and G. Hedling, “Review interests include high-precision GNSS positioning and integrated navigation
of code and phase biases in multi-GNSS positioning,” GPS Solutions, systems.
vol. 21, no. 3, pp. 849–860, Jul. 2017.
[24] A. Brandt and J. F. Gardner, “Constrained navigation algorithms for Aboelmagd Noureldin (Senior Member, IEEE)
strapdown inertial navigation systems with reduced set of sensors,” received the B.Sc. degree in electrical engineering
in Proc. Amer. Control Conf. (ACC), Philadelphia, PA, USA, vol. 3, and the M.Sc. degree in engineering physics from
Jun. 1998, pp. 1848–1852. Cairo University, Egypt, in 1993 and 1997, respec-
[25] X. Niu, S. Nasser, C. Goodall, and N. El-Sheimy, “A universal approach tively, and the Ph.D. degree in electrical and com-
for processing any MEMS inertial sensor configuration for land-vehicle puter engineering from the University of Calgary,
navigation,” J. Navigat., vol. 60, no. 2, pp. 233–245, May 2007. AB, Canada, in 2002.
[26] B. Phuyal, “An experiment for a 2-D and 3-D GPS/INS configuration He is currently a Professor with the Depart-
for land vehicle applications,” in Proc. Position Location Navigat. Symp. ment of Electrical and Computer Engineering, Royal
(PLANS), Monterey, CA, USA, Apr. 2004, pp. 148–152. Military College of Canada (RMCC) with Cross-
[27] P. Daum, J. Beyer, and T. F. W. Kohler, “Aided inertial LAnd NAvigation Appointment at the School of Computing and the
system (ILANA) with a minimum set of inertial sensors,” in Proc. IEEE Department of Electrical and Computer Engineering, Queen’s University. He
Position, Location Navigat. Symp. (PLANS), Las Vegas, NV, USA: IEEE, is also the Founder and the Director of the Navigation and Instrumentation
Apr. 1994, pp. 284–291. Research Group, RMCC. He has published two books, four book chapters,
and more than 270 papers in journals, magazines, and conference proceedings.
[28] D. Sun, M. G. Petovello, and M. E. Cannon, “GPS/Reduced IMU
His research interests include global navigation satellite systems, wireless
with a local terrain predictor in land vehicle navigation,” Int. J. Nav-
location and navigation, indoor positioning and multi-sensor fusion targeting
igat. Observ., vol. 2008, pp. 1–15, Nov. 2008. [Online]. Available:
applications related to autonomous systems, intelligent transportation, road
https://www.hindawi.com/journals/ijno/2008/813821/
information services, crowd management, and the vehicular Internet of Things.
[29] U. Iqbal, A. F. Okou, and A. Noureldin, “An integrated reduced His research work led to 13 patents and several technologies licensed to
inertial sensor system—RISS/GPS for land vehicle,” in Proc. IEEE/ION industry in the area of position, location, and navigation systems.
Position, Location Navigat. Symp., Monterey, CA, USA, May 2008,
pp. 1014–1021. Michael Korenberg received the B.Sc. and M.Sc.
[30] A. Noureldin, T. B. Karamat, and J. Georgy, Fundamentals of Inertial degrees in mathematics and the Ph.D. degree in elec-
Navigation, Satellite-Based Positioning and Their Integration. Berlin, trical engineering from McGill University, Montreal,
Germany: Springer-Verlag, 2013. QC, Canada.
[31] M. M. Atia, T. Karamat, and A. Noureldin, “An enhanced 3D multi- He has been a Faculty Member with Queen’s
sensor integrated navigation system for land-vehicles,” J. Navigat., University since 1983, where he is currently a Pro-
vol. 67, no. 4, pp. 651–671, Jul. 2014. fessor with the Department of Electrical and Com-
[32] Z. M. Kassas and T. E. Humphreys, “Observability analysis of collabo- puter Engineering. His research interests include
rative opportunistic navigation with pseudorange measurements,” IEEE the development of practical methods for represent-
Trans. Intell. Transp. Syst., vol. 15, no. 1, pp. 260–273, Feb. 2014. ing/identifying nonlinear systems of unknown struc-
[33] Z. M. Kassas, “Navigation with cellular signals,” in Position, Navigation, ture and for time-series analysis. Applications relate
and Timing Technologies in the 21st Century: Integrated Satellite to communications, to the modeling of physical and physiological processes,
Navigation, Sensor Systems, and Civil Applications, vol. 2, Y. J. Morton, and to computational biology and bioinformatics. The latter applications
F. van Diggelen, J. J. Spilker Jr., and B. W. Parkinson, Eds. Hoboken, include protein sequence classification, gene recognition, and interpretation
NJ, USA: Wiley, Jan. 2021. of microarray data.

Authorized licensed use limited to: TONGJI UNIVERSITY. Downloaded on June 11,2021 at 05:24:11 UTC from IEEE Xplore. Restrictions apply.