Journal of Global Positioning Syste ms (2008)
Vol. 7, No. 1 : 46-61
A Novel Arc hitecture for Ultra-Tight HSGPS-INS Integration
Guojiang Gao and Gérard Lachapelle
Positioning, Location and Navigation Group (PLAN)
Department of Geomatics Engineering, University of Calgary, Alberta, Canada
ABSTRACT
Global Positioning System (GPS) currently fulfills the
positioning requirements of many applications under
Line-Of-Sight (LOS) environments. However, many
Location-Based Services (LBS) and navigation
applications such as vehicular navigation and personal
location require positioning capabilities in environments
where LOS is not readily available, e.g., urban areas,
indoors and dense forests. Such environments either
block the signals completely or attenuate them to a power
level that is 10-30 dB lower than the nominal signal
power. This renders it impractical for a standard GPS
receiver to acquire and maintain signal tracking, which
causes discontinuous positioning in such environments.
In order to address the issue of GPS tracking and
positioning in degraded signal environments, a novel
architecture for ultra-tight integration of a High
Sensitivity GPS (HSGPS) receiver with an inertial
navigation system (INS) is proposed herein. By
enhancing receiver signal tracking loops through the use
of optimal estimators and with external aiding, the
capabilities of the receiver can be substantially improved.
The proposed approach is distinct from the commonly
used ultra-tightly coupled GPS/INS approaches and
makes use of different tracking enhancement
technologies used in typical HSGPS receivers, multi-
channel cooperated receivers and the current ultra-tightly
coupled GPS/INS methods. Furthermore, the effects of
inertial measurement unit (IMU) quality, receiver
oscillator noise and coherent integration time on weak
signal tracking are also analyzed.
Simulated test results in both static and dynamic testes
show that, the designed INS-aided GPS receiver can
track the incoming weak GPS signals down to 15 dB-Hz
without carrier phase locked, or 25 dB-Hz with carrier
phase locked. When there are multiple strong GPS
signals in view, the other weak signals can be tracked
down to 15 dB-Hz with carrier phase locked.
KEY WORDS: ULTRA-TIGHT INTEGRATION,
HSGPS, INS
INTRODUCTION
Standard Global Positioning System (GPS) technologies
fulfill the positioning requirements of many applications
intended for environments with clear Line-Of-Sight
(LOS) to satellites. However, many Location Based
Services (LBS) and applications such as vehicular
navigation and personal location require positioning
capabilities in environments where LOS to satellites is
not readily available, e.g., urban areas, indoors and dense
forest areas (e.g., Lachapelle et al., 2003). Such
environments either completely block the GPS signals or
attenuate them to a power level that is 10-30 dB lower
than the nominal signal power (van Diggelen and
Abraham, 2001). This makes it impractical for a standard
receiver to acquire and maintain signal tracking, which
causes discontinuous positioning in such environments.
In degraded signal environments, e.g., urban canyons and
indoors, positioning availability and accuracy are
affected by weak signal power, strong multipath/echo-
only signals and receiver dynamics. The characteristics
of degraded GPS signal environments are summarized in
Table 1. In these environments, signal attenuation and
strong specular reflectivity are primary sources of signal
degradation. For vehicle navigation in urban areas,
multipath/echo-only signals constitute interference
sources that change quickly and behave randomly due to
vehicle motion (MacGougan, 2003). The signal intensity
for personal positioning in indoor environments (e.g.,
fireman positioning in buildings) is commonly 20-30 dB
lower than that found outdoors (Lachapelle, 2007)
To address this issue, High Sensitivity GPS (HSGPS)
technologies (Watson et al., 2006), Assisted GPS
(AGPS) systems (van Diggelen and Abraham, 2001),

Gao et al: A Novel Architecture for Ultra-Tight HSGPS-INS Integration 47
multi-channel co-operated receivers (Zhodzishsky et al.,
1998) and cellular network-based solutions (Ma et al.,
2007) have been developed. However, these technologies
and systems still fail to maintain continuity of positioning
with acceptable accuracies, specifically in the indoors.
Thus, new receiver technologies have to be explored for
enhanced signal acquisition and tracking performance.
Recently, ultra-tight integration of GPS and inertial
navigation systems has received considerable attention
for this purpose. In an INS-assisted GPS receiver, which
is also called ultra-tightly coupled or deeply integrated
GPS/INS, an external INS is used to provide receiver
dynamics information to allow GPS receiver to do long
coherent integration to track weak signals in sight
(Soloviev et al., 2004). Measuring receiver dynamics
through INS aiding enables the INS-assisted GPS
receiver to track an incoming weak signal which is 20-30
dB lower or more than normal and therefore projects a
strong light beam into the “indoor darkness” (Beser et al.,
2002; Soloviev et al., 2004b; Kreye et al., 2000; Sennott,
1997).
Table 1 Characteristics in Different Operating
Environments and Applications
Table 2 (Gao, 2007) summarizes the performance of
HSGPS, multi-channel co-operated receivers (also called
COOP tracking receivers) and INS-assisted GPS
receivers. It shows that an INS-assisted GPS receiver is
far superior to the other positioning technologies
mentioned above and offers the greatest potential for
meeting navigation and positioning requirements under
attenuated signals. In INS-assisted GPS receivers,
velocity aiding from INS enhances the GPS phase lock
loops (PLL), which are the weakest loops in the receiver.
Furthermore, full navigation capability, including carrier
phase output under attenuated signals, is preserved in
INS-assisted GPS receivers. This availability of accurate
carrier phase measurements is deemed necessary for
many high-accuracy applications.
The paper continues with the discussion on the current
ultra-tightly coupled GPS/INS systems and other weak
signal tracking technologies, such as HSGPS and multi-
channel co-operated receivers. Then, the design of a
novel INS-assisted GPS receiver for degraded GPS
signal tracking is introduced. The proposed architecture
is distinct from current ultra-tightly coupled GPS/INS
systems and uses a combination of different tracking
technologies like HSGPS or multi-channel co-operated
GPS receivers or traditional ultra-tightly coupled
GPS/INS approaches. Some system design issues, such
as IMU quality requirements, the limits of very long
coherent integration time and receiver clock error
compensation, are also addressed. Information about the
testing tools utilized herein, including an INS simulator
developed for this purpose, is provided. Test results and
analysis are then presented using simulated data sets to
assess the performance of the INS-assisted GPS receiver,
followed by conclusions.
CURRENT ULTRA-TIGHTLY COUPLED
GPS/INS SYSTEMS
Based on the type of Kalman filter used, ultra-tight
integration can be implemented in three different ways
(Gao, 2007), namely: (1) loosely coupled Kalman filter-
based ultra-tight integration, (2) tightly coupled Kalman
filter-based ultra-tight integration, and (3) ultra-tightly
coupled Kalman filter-based ultra-tight integration.
Figure 1 summarizes the different architectures. Figure 2
shows two different types of architectures for INS-
assisted GPS receivers, as proposed by Gautier and
Parkinson (2003), Alban et al. (2003) and Gustafson et
al. (2000).
The first architecture shown in Figure 2(a) is based on a
loosely or tightly coupled integration scheme. All
individual DLL (delay lock loops) and PLL are inside the
receiver. The Kalman filter utilizes either raw
measurements or processed positions and velocities from
the GPS receiver to update the INS periodically. The
updated INS information is then used to predict the phase
and Doppler used as aiding to the receiver. Thus, based
on the type of measurements used for updating the INS,
these strategies can be classified as loosely coupled
Kalman filter-based ultra-tight integration or tightly
coupled Kalman filter-based ultra-tight integration.
However, in the second architecture shown in Figure
2(b), an ultra-tightly coupled Kalman filter is used in
place of conventional in-receiver PLLs and, in some
cases, even DLLs. This filter operates on in-phase (I) and
quadra-phase (Q) components of the signal directly. This
Features Urban Canyon Vehicle
Navigation
Indoor
Personal
Positioning
Signal Fading 10-30 dB 20-30 dB
Multipath
Signal Strong, high
frequency
Strong, low
frequency
Platform
Dynamics Moderate Low
Map Matching Easy to
implement
Maps not readily
available
Desired
Receiver Size Moderate Small

Gao et al: A Novel Architecture for Ultra-Tight HSGPS-INS Integration 48
integration strategy is referred to as ultra-tightly coupled
Kalman filter-based ultra-tight integration.
Most of the work done in ultra-tight integration of GPS
and INS has focused on the above three integration
strategies. Although these new architectures offer
flexibility, from the point of view of information theory,
it is suggested herein that simply adopting different kinds
of Kalman filters will not improve positioning
performance of GPS/INS integrated systems
significantly. This concept is best illustrated by an
analogy of water in a river. Although water looks very
different at higher, medium, and lowest points of the
river, the volume of water is the same at all these points.
This is the case for line-of-sight environments, where a
tightly coupled Kalman filter will not provide significant
improvements in an integrated system performance as
compared to a loosely coupled filter.
There are also some specific limitations in present
architectures. The first limitation is that the receiver
tracking capability is sensitive to IMU quality. For
reliable aiding from INS, a velocity accuracy of 1 cm/s
along the LOS direction is required from the INS
solution (Soloviev et al., 2004a), which requires a high
quality IMU.
Table 2 Performances of Different Positioning Methods under Attenuated Signals
Items HSGPS CO-OP
Tracking
GPS/INS
Ultra-tight
integration Notes
Tracking
Sensitivity Good Good Excellent
15-25 dB lower than regular signals for HSGPS and CO-
OP tracking; 20-30 dB for ultra-tight integration
Acquisition
Sensitivity Poor Good Excellent
Because of long integration time, long Time-To-First Fix
(TTFF) for HSGPS; INS measurements and/or information
from other tracking channels can be used to speed up the
acquisition process in ultra-tight integration and CO-OP
tracking, especially in hot start.
Re-Acquisition
Capability Poor Good Excellent
Due to long pre-detection integration time (PIT), re-
acquisition is time-consuming in HSGPS; Aiding from INS
measurements facilitates rapid re-acquisition in ultra-tight
integration.
Position Data
Up Rate Low Low High
In ultra-tight integration, the data rate can be increased to
above 100 Hz using INS aiding, with a Kalman filter
running at a low recursive rate.
Positioning
Accuracy Poor Good Excellent
In HSGPS, positioning accuracy is degraded by multipath
signal and frequency/phase tracking error; In ultra-tight
integration, INS solution can help in blunder detection and
noise compression (by using long time integration).
Carrier Phase
Output Poor Good Excellent
In HSGPS, limited benefit for PLL tracking, and thus
difficult to output carrier phase observation; Ultra-tight
integration method can output precise phase observation
and avoid/reduce cycle slips.
Dynamic
Response Poor Poor Excellent
HSGPS is used mainly for low dynamic users. Ultra-tight
integration can be used for both low and high dynamic
users and thus in both commercial and military applications
Receiver Size Small Small Moderate/Big For HSGPS, no need for any other hardware; a good size
under Ultra-tight integration for MEMS IMU
Power Cost Low Low Moderate/High
For HSGPS, no other hardware required so no additional
power cost. For Ultra-tight integration, additional external
sensor needed, so more power is required.
Multipath
Mitigation Poor Good Excellent
Ultra-tight integrated navigator can detect multipath signals
and track weak LOS signals directly in urban areas and
indoor environments.

Gao et al: A Novel Architecture for Ultra-Tight HSGPS-INS Integration 49
Fig. . 1 Ultra-tightly Coupled GPS/INS with Loosely,
Tightly or Ultra-tightly Coupled Kalman Filter
The second limitation is that before the integrated system
moves to attenuated signal environments, it has to be
initialized under LOS environments, which includes
GPS-only receiver initial acquisition and INS initial
alignment.
Furthermore, the accuracy estimation of a Kalman filter
relies on the assumption of correct stochastic modeling of
both system measurement errors, which may not be
possible in severe urban canyon environments. For
vehicle navigation in urban canyons, GPS measurement
faults such as those caused by multipath or echo-only
signals are very significant. Also, IMU measurement
errors may not be compensated effectively, which
ultimately degrades the receiver tracking performance.
Small INS velocity errors usually remain after the INS is
well aligned with GPS measurements and the vehicle’s
motion contains sufficient dynamics to make all the INS
error states observable. However, in the static case or
where the vehicle motion does not contain enough
dynamics to ensure observability, the accuracy of the
Doppler information from the INS may be of poor
quality. A particular concern is the poorly observable
states of the azimuth and gyro bias. If the estimates of
these states are of poor quality, the velocity component
errors provided by a low cost INS in static situations can
be very large. It should also be noted that the dynamics
that needs to be estimated and compensated by the INS
are not limited to the LOS velocity component but to all
other components. For example, an erroneous azimuth
estimate can result in dynamics errors due to “phase
wind-up” effects (Tetewsky and Mullen, 1996; Don et
al., 2005).
Fig. . 2 Two Different Architectures of Current Ultra-
tightly Coupled GPS/INS
Furthermore, when a low cost MEMS-based INS is
involved in an INS-assisted GPS receiver, because of the
very low accuracy and very poor stability of IMU
sensors, the INS velocity errors might contain jumps or
blunders. These large errors may cause major problems
in an INS-assisted GPS receiver. Given the GPS/INS
Kalman filter observability issue discussed above,
MEMS-based INS/GPS ultra-tight integration is still a
major challenge.
DESIGN OF NOVEL INS-ASSISTED GPS
RECEIVER
To overcome the limitations of current INS-assisted GPS
receivers, a technique based on multi-channel co-
operated tracking (COOP tracking) is proposed herein to
estimate and track the Doppler prediction errors caused
by INS errors. This technique is fully described in Gao
(2007). Initial results were presented by Gao and
Lachapelle (2006).
The architecture of the proposed integration system is
shown in Figure 3. The ultra-tightly integrated system
used the software receiver GNSS_SoftRxTM (Ma et al.,
2004) as a starting point. The proposed strategy includes
three loops. The first loop includes the conventional
loosely or tightly coupled Kalman filter, which predicts

Gao et al: A Novel Architecture for Ultra-Tight HSGPS-INS Integration 50
the user Doppler based on information from the INS. In
this study, a loosely coupled Kalman filter is suggested.
This loop provides external Doppler aiding and clock
error correction from the INS. In order to decrease the
effects of INS positioning errors on receiver Doppler
prediction and receiver clock error compensation, a
COOP loop is used as the second loop to estimate the
carrier Doppler error and receiver clock error caused by
INS positioning errors. Thus, the INS and COOP loops
operate together in order to provide a nearly perfect
reference for receiver dynamics and receiver clock errors.
Since receiver dynamics is removed from the signal, long
coherent/non-coherent integration time for individual
PLLs and DLLs becomes possible, which constitutes the
third loop of the proposed architecture. Thus, the
PLL/DLL loops track the differences between the
incoming and the local signals, which have been
compensated by INS and COOP loops.
Fig. . 3 Proposed Architecture of Ultra-tightly Coupled
GPS/INS
The characteristics and advantages of the proposed
system can be summarized as follows:
1 The HSGPS receiver optimizes parameterization and
employs data wipe off technology for long coherent
integration. This structure enables initialization in a
weak signal environment.
2 To provide INS aiding to the GPS receiver, a
loosely/tightly coupled GPS/INS approach such as
that used by Petovello (2003) in the SAINT™
software is suggested to handle INS measurements
and provide a corrected INS solution for receiver
aiding.
3 The multi-channel cooperated tracking loops track
the weak signals and eliminate the effects of INS
errors. The design of this estimator is discussed later
in this section.
4 The individual DLL/PLLs are different from those of
traditional ultra-tightly coupled GPS/INS systems,
which adopt ultra-tightly coupled Kalman filters for
signal tracking. In this design, traditional
sophisticated DLL/PLLs are still located in every
individual signal tracking channels, as shown in
Figure 3. Furthermore, the individual DLL/PLLs are
enhanced using a data wipe off technology
mentioned in item 1 to perform very-long coherent
integration. The individual DLL/PLLs are combined
with INS aiding loops and COOP loops to track both
strong and weak signals.
5 Since INS is used in this approach to mainly provide
receiver dynamics information, any other sensors
such as odometers or radars can replace INS sensors
to provide Doppler measurements for the enhanced
receiver. This provides an effective capability to re-
configure the software receiver to suit various aiding
hardware.
A. Multi-Channel CO-OP Tracking Loop
Figure 4 shows the modular design of the multi-channel
co-operated tracking loops, i.e., the COOP tracking
loops. The basic strategy of COOP tracking is to project
signals from the channel domain to the position domain
and then try to track/estimate the signals in the position
domain. Then the signals are projected back to the
channel domain for Doppler removal (Zhodzishsky et al.,
1998).
Fig. 4 Architecture of co-operated (CO-OP) Tracking
Module
In Figure 4,
e
n
en
en
e
n
CAAA
TCTT
==
== (1)
where the superscripts n and e represent the Local-Level
frame (LLF) and Earth-Centered Earth-Fixed (ECEF)
frame, respectively. C is the rotation matrix and A is the
direction cosine matrix (also called geometry matrix). T
is the transfer matrix and defined as follows:

Gao et al: A Novel Architecture for Ultra-Tight HSGPS-INS Integration 51
() () () ()
1
1
1−
−
−
=
φφ
CAACAT T
nn
T
nn (2) (2)
1−
φ
C is a weighted matrix determined by FLL lock
detectors in individual PLLs.
The COOP loop can be based on either a least-squares
estimation method or on a Kalman filter. In this study,
the least-squares method is used. Least-squares
estimation is an effective optimal estimation method,
especially when measurement redundancy is high. Many
fault detection algorithms such as Receiver Autonomous
Integrity Monitoring (RAIM) are in fact based on least-
squares estimation, provided redundant measurements
are available. Herein, the principle of “using an optimal
estimator in GPS positioning to fully utilize measurement
redundancy”, is put forward from the measurement
processing domain into the GPS baseband signal
processing domain, which leads to the COOP tracking
method. COOP loops allow a GPS receiver to do signal
tracking based on the multi-channel vector tracking
approach. Furthermore, based on measurement
redundancy, blunder detection algorithms and adaptive
estimation methods now can be realized at the signal
processing stage rather than at the measurement
processing stage.
B. Effect of IMU Quality
In a loosely or tightly coupled GPS/INS for use in urban
canyons, the receiver frequently loses phase lock on
incoming signals. Therefore the INS in this kind of
integrated system must be able to accommodate GPS
outages for relatively long periods, e.g., up to 30 s. In an
ultra-tightly coupled GPS/INS however, the receiver can
track weak GPS signals continuously because of INS
aiding. Thus, the INS is constantly corrected by GPS
measurements which are typically available every 1 s.
Consequently, the maximum INS prediction duration
may be limited to 1 s in many common scenarios. This
implies that a low-cost Micro Electro-Mechanical System
(MEMS) IMU might be acceptable for ultra-tightly
coupled GPS/INS for certain applications.
INS velocity errors caused by sensor errors over one
second can be estimated using velocity error signatures
as (Scherzinger, 2004)
gg
aa
gbtgbV
btbV
6
1
6
12−=−=
==
δ
δ
(3) (3)
where V
δ
is the velocity error, a bis the accelerometer
bias and g bis the gyro bias, g is the gravity, and t is the
time interval.
With COOP loops, the effects of an INS positioning error
on the receiver Doppler prediction and clock error
compensation will be limited. Therefore, the accuracy for
aiding velocity need not be accurate to 1 cm/s level any
more, as pointed out by Soloviev et al. (2004a). In this
research, a velocity error of 0.1 m/s is used for the
Doppler aiding accuracy, which makes it feasible to use a
lower grade IMU quality with either an accelerometer
bias of 10 mg or a gyro bias of 3.4˚/s.
In actual applications, a velocity error of 0.1 m/s in 1 s
can be achieved with many MEMS IMUs available
today, such as the Crista IMU from Cloud Cap
Technology, as described by Godha and Cannon (2005).
The in-run gyro biases and accelerometer biases of this
unit are about 0.3 ˚/s and 2.5 mg, respectively.
C. Effect of Allan Oscillator Phase Noise on Carrier
Phase Tracking
The measurement characteristics of a low-cost
Temperature-Compensated Crystal Oscillator (TCXO)
are similar to those of a low-cost gyro. Oscillator clock
errors can be divided into turn-on bias, in-run drift and
the remaining colored noise components, the latter being
characterized by the Allan Variance. In most GPS/INS
systems, the clock bias and drift are estimated and then
compensated using a Kalman filter.
In theory, since the clock noise described by the Allan
variance is characteristically colored, it can be modeled
and thus partly estimated by GPS/INS Kalman filters. If
the colored noise is modeled perfectly, the remaining part
will be limited to white noise, which can be regarded as
thermal noise and easily handled by receiver tracking
loops. To simplify the filter design, Allan clock noise is
regularly assumed to be white noise (Brown and Hwang,
1992) and thus will not be estimated. Most of the energy
of the colored noise is located in the low frequency band
in the frequency domain. For this reason, when low-pass
loop filters in a receiver try to eliminate the thermal noise
from the signal, most of the clock noise passes through
these low-pass filters, since it is mixed with the GPS
signal in the low frequency band of the spectrum.
Therefore, Allan oscillator phase noise must be
considered in receiver design, especially in weak signal
environments. The effect of the correlated clock noise on
signal tracking is discussed in Kaplan (1996).

Gao et al: A Novel Architecture for Ultra-Tight HSGPS-INS Integration 52
D. Determination of Coherent Integration Time for
Internal Individual DLLs/PLLs
The tracked GPS signal power after accumulators can be
expressed as (Raquet, 2004)
)sin()(
)sin(
2
)cos()(
)sin(
2
0
0
φπτ
π
π
φπτ
π
π
+Δ
Δ
Δ
=
+Δ
Δ
Δ
=
fTDR
fT
fTAM
Q
fTDR
fT
fTAM
I
E
E
(4) (4)
where E
M is the number of samples accumulated in one
sample period, fΔ is the frequency error over the
integration interval, R is the self-correlation function of
PRN code, and D is the navigation data bit modulated on
the signal. In Equation (4), the tracked signal undergoes a
power loss due to the aiding velocity errors, with the loss
characterized by the functionfT
fT
Δ
Δ
π
π
)sin( . Since the L1
carrier wavelength is about 19 cm, a velocity error of 0.1
m/s will lead to a maximum Doppler error of 0.5 Hz
along a given LOS vector. The resulting signal power
loss over the total integration time due to this
phenomenon is shown in Figure 5. A coherent integration
time of 1 s would result in a power loss of 4 dB.
The design of receiver tracking loops using continuous
update approximation is discussed in details in Kaplan
(1996). Stephens and Thomas (1995) have shown that,
when the product of loop bandwidth (n B) and coherent
integration time (coh
T) is much greater than 0.1 or close
to 1, the continuous update approximation does not hold.
Gao (2007) and Ilir et al. (2007) give the expressions for
the signal tracking errors of both FLL and PLL in a
digital GPS receiver. Since the Doppler uncertainty from
INS aiding is 0.5 Hz, as shown in Figure 5, coherent
integration time should be shorter than 0.2 s to satisfy the
condition 11.0 <<×<× cohncohn TBorTB .
Fig. 5 Signal Power Loss over Integration Time
Another factor that limits the choice of very long
coherent integration time is the navigation data bit
transition. Soloviev et al. (2004a) presents an energy-
based bit estimation algorithm to account for possible bit
transitions during signal integration. The algorithm
searches for the bit combination every 20 ms to
maximize signal energy over the tracking integration
interval. This approach presumes that the right bit
combination is the one that maximizes the signal energy
over the tracking integration interval, namely, the Signal-
to-Noise Ratio (SNR) after 20 ms of integration should
be above zero. This yields a limitation of -157 dBm for
this approach (Gao, 2007), as given by:
dBm
SNRBNC I
157
0
20
1000
log10204
200
−=
+
⎟
⎠
⎞
⎜
⎝
⎛
+−=
=
(5)
where 20I SNR is the signal-to-noise ratio after 20 ms
coherent integration,0 Nis the environmental thermal
noise and dBmN 204
0
−
=
.
B
is the bandwidth of
the phase tracking loop using 20 ms coherent integration
time and HzB )
20
1000
log(10=.
Equation (5) illustrates that an incoming signal lower
than -157 dBm , which equals to 17 dB-Hz, will fail the
assumption implied in this algorithm and, hence, the
energy-based bit detection approach. In this case,
increasing the length of coherent integration time cannot
help to track this very-weak signal.

Gao et al: A Novel Architecture for Ultra-Tight HSGPS-INS Integration 53
In consideration of the above factors, a maximum
coherent integration time of 100 ms is used in this paper.
TEST SETUP AND TEST RESULTS
Several static and dynamic tests were conducted to assess
the performance of the above INS-assisted HSGPS
receiver. In the static case, a scenario where a GPS
receiver is tracking strong and weak signals at the same
time is simulated to examine the performance of COOP
loops. In the dynamic case, a scenario where a GPS
receiver is tracking all weak signals at the same time is
simulated to assess the tracking sensitivity of the ultra-
tightly coupled GPS/INS system.
A. Simulation Test Tools
For simulating the incoming GPS signal and IMU
measurements, the GPS simulator GPS_GenTM,
developed previously by Dong et al (2004), and an INS
simulator INS_Sim developed by Gao (2007) were used.
Figure 6 shows the architecture of the INS simulator.
Fig. 6 The Architecture of INS Simulator
B. Static Test Scenario
The test summary is shown in Figure 7. The total
duration of the data collection was 50 s. During the first
14 s, the system initialization, including GPS signal
acquisition and ephemeris collection, was performed.
Then, from 14 to 50 s, the INS output, i.e., the user
velocity and position from the INS, was fed to the GPS
receiver to assist with the GPS signal tracking. To
simplify the test analysis, INS operated in stand-alone
mode for this 36 s aiding period, such that it kept
accumulating errors. The GPS simulator scenario was
designed to output strong GPS signals (45 dB-Hz) for all
satellites, except for PRN 07, as illustrated in Figure 7.
For PRN 07, the signal power was 45 dB-Hz during the
first 20 s period and then reduced gradually to 15 dB-Hz
by the GPS simulator over the next 20 s period. During
the last 10 s period, the signal power of PRN 07 was kept
at 15 dB-Hz.
The error characteristics of the IMU data simulated were
similar to a tactical grade HG1700 IMU, as discussed by
Petovello (2003). Figure 8 shows the interface of
INS_Sim and
Fig. 8 The Interface of INS Simulator
Table 3 lists the parameters used for simulating the IMU
data. The simulation assumes that the INS has been
initialized with GPS measurements. So the accelerometer
bias residual and gyro bias residual were limited to
around 20 μg and 0.3 ˚/hr respectively. The IMU noise
bandwidth was kept at about 10 Hz so that the gyro noise
was equal to 16.5 ˚/hr. The pitch and roll alignment
errors were 0.01˚ and the heading alignment error was
0.05˚.

Gao et al: A Novel Architecture for Ultra-Tight HSGPS-INS Integration 54
Fig. 7 Static Test Setup for INS-assisted HSGPS
Receiver
Fig. 8 The Interface of INS Simulator
Table 3 Simulated-INS Parameters
Accelerometer Gyro
Scale Factor 300 ppm 150 ppm
Bias
Residual 20
g
μ
0.3 ˚/hr
Random
Noise 1000
g
μ
5.5 Hzhr ⋅°
In this test, the ephemeris at 19:27 on July 06, 2000 in
UTC time was chosen for the GPS signal simulation. At
that time, seven satellites were visible from the test
position (Latitude 51˚ North, longitude -114˚ West), such
that the GDOP and HDOP were both less than 2 during
the test period. The PRNs visible above a 15˚ masking
angle were 04, 05, 07, 09, 17, 24 and 30. The receiver
parameters used in this static test were 3 Hz bandwidth
and 20 ms coherent integration time for COOP, and 0.2
Hz bandwidth and 100 ms coherent integration time for
individual PLL. To assess the improvement provided by
INS aiding, the standard version of the software GPS
receiver without INS aiding was also used in all tests to
measure normal GPS tracking performance. In this
standard GPS receiver, 10 Hz bandwidth and 10 ms
coherent integration time were used for signal tracking.
In the test, previous experience was used to empirically
select these parameters for this comparison. The
simulated INS velocity error is shown in Figure 9. It is
noted that, at the 50-s point, INS velocity errors reach
0.1 m/s, which is similar to the velocity accuracy
available from a MEMS IMU, during 1 s in the
integrated GPS/INS system. As shown in Figure 9, the
INS errors have low frequency content. In contrast to
GPS errors, their time growth is somewhat smooth.
Fig. 9 Simulated INS Velocity Error
C. Static Test Results
Figure 10 shows the carrier-to-noise (C/No) density of
satellite PRN 07 over the entire test period both with and
without INS aiding. Figure 11 shows the carrier phase
tracking error of individual PLLs and Figure 12 shows
the total carrier phase tracking errors with and without
INS aiding. In Figure 12, for computing the errors, the
“true” reference carrier phase was determined by another
test where satellite PRN 07 was kept at 45 dB-Hz and all
other scenario parameters kept the same as the present
test. The figures show that, although the signal power of
satellite PRN 07 is attenuated from 45 to 15 dB-Hz
during the last 30-s period, the INS-assisted GPS receiver
can track the satellite with carrier phase locked.
However, a standard GPS receiver without INS aiding
cannot track satellite PRN 07 any longer when the signal
power is lower than 30 dB-Hz.
As stated previously, the first 14-s period of the test is
used to finish frame synchronization and receive code-
time-delay (τ) from the navigation data; when the latter is
received, the receiver continues to output positioning
solutions from the 14 s point onward. Since INS is
mainly used to aid carrier phase tracking loops in this
paper, only the velocity solution is shown in Figure 13.
For the standard receiver without INS aiding, because of
the large tracking errors on satellite PRN 07 around
epoch 30 s, there are two large faults at epoch 31 s and
32 s, which are about 4 m/s and 11.5 m/s, respectively.
The reason for these two faults in the GPS solution is that
an epoch-by-epoch least-squares positioning approach is
used here to calculate receiver velocity. In this basic
least-squares estimator, no fault testing is performed.

Gao et al: A Novel Architecture for Ultra-Tight HSGPS-INS Integration 55
After the 32-s epoch, the standard receiver loses lock on
satellite PRN 07 so that velocity error returns to a normal
level.
Fig. 10 CN0 of PRN 07 Tracked in Static Test
Fig. 11 PLL Carrier Phase Error in Static Test
Fig. 12 Total Carrier Phase Error in Static Test
Fig. 13 Horizontal Velocity Error in Static Test
Figure 12 and Figure 13 show that the signal tracking
sensitivity of the INS-assisted GPS receiver is improved
for at least 15 dB as compared to that of the standard
GPS receiver when there are multiple strong signals
available. This improvement can be attributed to two
aspects:
1. Since INS aiding removes the signal Doppler and
thus decreases the receiver dynamics uncertainty, a
long coherent integration (here is 100 ms) is used for
weak signal tracking.
2. Since the COOP tracking method is used, the strong
signals aid the weak signal tracking. In this test, the
tracking of other strong satellites is of benefit to the
tracking of satellite PRN 07, as will be explained
later in this section.
Figure 14 shows the carrier phase tracking error of
satellite PRN 07 for the INS-assisted GPS receiver, with
and without the use of COOP estimators. In the case
when COOP is not used, the individual PLL parameters
are kept the same as for the case with COOP estimators,
i.e. 0.2 Hz bandwidth and 100 ms coherent integration
time. From Figure 14, it can be seen that, although the
coherent integration time is 100 ms, without the COOP
loop, the tracking performance is even worse than that of
a standard GPS receiver (without INS aiding) where
coherent integration time is 10 ms. At epoch 26 s, the
PLL loses lock and the carrier phase error drifts away
rapidly. The reason for the loss of phase lock and poor
performance of the INS-aided receiver without COOP in
Figure 14 is shown in Figure 15, which gives the carrier
Doppler of satellite PRN 07 tracked by COOP and
individual PLLs separately. Because of strong signals in
view, COOP can perfectly track Doppler residuals caused
by the INS aiding errors (shown in Figure 9). Since
COOP tracking compensates for the INS aiding errors,
combined COOP and INS aiding provides a nearly

Gao et al: A Novel Architecture for Ultra-Tight HSGPS-INS Integration 56
perfect reference for receiver dynamics. Therefore, even
though the power of PRN 07 drops to 15 dB-Hz during
the last 30 s, the carrier Doppler tracked by individual
PLLs is close to zero. If COOP were not used, the
individual PLL would have to track the INS aiding
Doppler error, as shown in Figure 14. When the coherent
integration time is very long, the INS aiding Doppler
error will fail the pure PLL tracking.
Fig. 14 Total Carrier Phase Error of PLL-Only Receiver
with INS Aiding in Static Test
Fig. 15 Carrier Doppler Tracked by PLL and CO-OP
Separately in Static Test
To investigate the signal tracking stability of the INS-
assisted HSGPS receiver while the receiver parameters
vary, different combinations of receiver tracking
parameters were examined when all incoming GPS
signals were kept at 15 dB-Hz. In this static test, the
parameters adopted for the three test receivers are listed
in Table 4: Receiver one used very narrow noise
bandwidths for both COOP and FPLL, with a very long
FPLL integration time, namely, 2 s. In receiver two,
wider noise bandwidth and shorter integration time were
adopted. However, compared to those used in a standard
receiver, these parameters were still very stringent.
Receiver three used a set of parameters that can also be
used in a standard GPS-only receiver in static situations.
The test results statistics are summarized in Table 5,
including the receiver PLL discriminator output and the
carrier Doppler tracked by the COOP, respectively. The
standard derivations of these two observations are used to
assess carrier phase tracking performance of the three
receivers. When the receiver parameters become more
stringent from receiver three to receiver one, the standard
deviation of the carrier Doppler tracked by COOP
decreases, which means that COOP can track the
incoming signals with increasing accuracy. One can also
see that the 15 dB-Hz signal is locked in the entire test
and all three receivers can output reasonable velocity
solutions. Based on the test results, it is evident that,
while the adopted receiver parameters vary in a large
range, the INS-assisted HSGPS receiver presents very
stable performance in both phase and code tracking,
although there are cycle slips present in all three
receivers. Finally, it should be noted that due to different
receiver dynamics, different levels of signal power, etc,
the range of suitable parameters for the INS-assisted
HSGPS receiver may change from one case to another. In
dynamic situations, a narrower range of suitable receiver
parameters is expected.
Table 4 Parameters Adopted in Three Receivers
Adopted
Receiver
Parameters
Receiver
One Receiver Two Receiver Three
Phase Noise
Bandwidth of
COOP (Hz)
0.2 1.2 3
Phase Noise
Bandwidth of
FPLL (Hz)
0.1 0.2 0.2
Coherent
Integration Time
of COOP (ms)
20 20 20
Non-Coherent
Integration Times
of COOP
1 1 1
Coherent
Integration Time
of FPLL (ms)
100 100 100
Non-Coherent
Integration Times
of FPLL
20 10 10

Gao et al: A Novel Architecture for Ultra-Tight HSGPS-INS Integration 57
Table 5 Tracking Result Statistics of Different Receivers
When the Incoming Signal is 15 dB-Hz
Observation Name Receiver
One Receiver Two Receiver Three
Estimated Carrier
Phase Error Std on
Satellite PRN 07
(cycle)
0.029 0.030 0.030
COOP Tracking
Doppler Std on
Satellite PRN 07
(Hz)
0.4 1.3 1.9
Estimated C/No
Mean on Satellite
PRN 07 (dB-Hz)
17.9 17.8 17.8
Estimated C/No Std
on Satellite PRN 07
(dB-Hz)
3.1 3.2 3.1
Horizontal Velocity
Error Mean (m/s) 0.21 0.38 0.40
Horizontal Velocity
Error Std (m/s) 0.16 0.23 0.30
Vertical Velocity
Error Mean (m/s) -0.10 0.23 0.14
Vertical Velocity
Error Std (m/s) 0.24 0.50 0.79
D. Dynamic Test Scenario
The receiver trajectory and the velocities simulated in the
dynamic test are shown in Figure 16 and Figure 17.
During the first 20 s, the vehicle moved east with a
velocity of 100 m/s. In the next 30 s, the vehicle made an
“S” shaped trajectory, with an angular rate of 6 ˚/s.
Fig. 16 Vehicle Trajectory in Dynamic Test
Fig. 17 Vehicle Velocity in Dynamic Test
The change of signal power in the dynamic test is shown
in Figure 18. In contrast to the static test, the power of all
signals was degraded simultaneously from 45 dB-Hz to
25 dB-Hz during the period of 20 s to 30 s. Then the
power level for all signals was kept at 25 dB-Hz from
30 s to 40 s, and then increased back to 45 dB-Hz during
the last 10 s of the test. The parameters of the INS
simulator were the same as those in the static test. The
INS velocity errors are shown in Figure 19. GPS
parameters used in the standard GPS software receiver
were the same as those in the static test, namely 10 Hz
bandwidth and 10 ms coherent integration time. For INS-
assisted HSGPS, the bandwidth was 0.4 Hz for individual
PLL and 3 Hz for COOP. The coherent integration time
of 100 ms for individual PLL and 20 ms for COOP were
used. The other parameters were kept same as those of
the static test.
Fig. 18 Simulated Change of Signal Power

Gao et al: A Novel Architecture for Ultra-Tight HSGPS-INS Integration 58
Fig. 19 INS Velocity Error in Dynamic Test
E. Dynamic Test Res ults
Figure 20 shows the C/No of the received PRN 07
satellite signal during the test period both with and
without INS aiding. Figure 21 shows the PLL carrier
phase tracking errors, Figure 22 shows the total carrier
phase errors, and Figure 23 shows the horizontal velocity
errors for the INS-assisted HSGPS receiver and the
standard GPS receiver. It can be seen from these figures
that the tracking performances of the standard receiver is
very poor under dynamic conditions as compared to
those of the INS-assisted HSGPS receiver. Figure 21 and
Figure 22 show clearly that the standard GPS receiver
cannot lock on the incoming carrier phase when the
vehicle starts to make the “S” shaped trajectory. During
the period between 20 s and 28 s, although the individual
PLLs in the standard receiver show lock on the incoming
carrier phase in Figure 21, there are cycle slips due
vehicle dynamics. From epoch 28 s onward, the standard
receiver stops to output the GPS solution. In contrast, the
INS-assisted HSGPS receiver can track the incoming
weak signals down to 25 dB-Hz with carrier phase
locked during the entire test.
Fig. 20 CN0 of PRN 07 in Dynamic Test
Fig. 21 PLL Carrier Phase Error in Dynamic Test
Fig. 22 Total Carrier Phase Error in Dynamic Test
Fig. 23 Horizontal Velocity Error in Dynamic Test

Gao et al: A Novel Architecture for Ultra-Tight HSGPS-INS Integration 59
CONCLUSIONS
This paper proposes a novel design for an INS-assisted
GPS receiver to improve GPS tracking performance for
navigation in degraded signal environments. The effects
of IMU quality and receiver parameters such as coherent
integration time on the designed system are analyzed.
Compared to a standard GPS receiver without INS
aiding, the INS-assisted GPS receiver proposed here
yields much better performance under attenuated signal
environments, based on the tests carried out. An analysis
of the results leads to the following conclusions:
1. INS aiding can effectively reduce the receiver
dynamics uncertainty and improve tracking
performance of a standard GPS receiver significantly
under both weak signal and high dynamic signal
environments.
2. When an INS solution is available, an effective
signal tracking strategy can be summarized in three
steps. First, the INS solution is implemented in order
to remove most of receiver dynamics uncertainty; as
a consequence, the residual Doppler signal left for
the COOP and FPLLs to track is close to zero. Next,
a vector tracking-based COOP loop is designed to
track the residual carrier Doppler effectively; since
six to 10 satellites are usually in view, COOP
3. tracking yields much better performance than
conventional FPLLs, especially under weak signal
environments. Finally, FLL-assisted PLLs can be
used to track the carrier. Since the Doppler signal is
compensated by the combined INS and COOP
aiding, the residual Doppler error is close to zero.
This significantly decreases carrier phase tracking
errors and, therefore, increases the FLL/PLL tracking
sensitivity.
4. Although INS error increases rapidly with time
during a GPS outage, the INS solution errors change
smoothly. These errors can be easily tracked by the
COOP method and thus, will not affect signal
tracking significantly. Therefore, even if the INS
solution error is as large as 0.1 m/s, an INS-assisted
GPS receiver can track a GPS signal that is 30 dB
lower than LOS signals with relatively good
positioning accuracy.
5. The combined tracking of the FPLL and COOP
loops presented herein have been shown to track
signals as low as 15 dB-Hz. When the signal power
is above 22-23 dB-Hz, this method can lock on the
incoming carrier and provide accurate carrier phase
measurements. When the signal is lower than 22 dB-
Hz but higher than 15 dB-Hz, the method can track
the incoming carrier most of the time, although cycle
slips may occur. When there are several strong
signals in view, the receiver can lock the other weak
carrier signals as low as 15 dB-Hz due to the
assistance from the strong signals.
6. Because INS aiding provides most of the Doppler
measurements, high receiver dynamics do not affect
signal tracking significantly in INS-assisted GPS
receivers. With INS aiding and by adopting COOP
tracking, long coherent integration can be
implemented safely when necessary.
REFERENCES
Alban S., D. M. Akos and S. M. Rock (2003),
Performance Analysis and Architectures for
INS-Aided GPS Tracking Loops, in Proceedings
of ION National Technical Meeting (NTM), 22-
24 January, Anaheim CA, pp. 611-622, U.S.
Institute of Navigation, Fairfax VA.
Beser J., S. Alexander, R. Crane, S. Rounds and J.
Wyman (2002), Trunavtm: A Low-Cost
Guidance/Navigation Unit Integrating A
SAASM-Based GPS And MEMS IMU In A
Deeply coupled Mechanization, in proceedings of
ION GPS/GNSS, 24-27 September, Portland OR,
pp. 545-555, U.S. Institute of Navigation, Fairfax
VA.
Brown, R. G. and P. Y. C. Hwang (1992), Introduction
To Random Signals and Applied Kalman
Filtering, John Wiley & Sons, Toronto ON,
second edition.
Don K., S. Luis and R. B. Langley (2005),
Compensation of the Effects of Phase Wind-up
for Improving the Performance of a GPS RTK-
Based Vehicle Navigation System, in Proceedings
of ION GNSS 18th International Technical
Meeting of the Satellite Division, 13-16
September, Long Beach, CA, pp. 346-354,
Institute of Navigation, Fairfax VA.
Dong L., C. Ma and G. Lachapelle (2004),
Implementation and Verification of a Software-
Based IF GPS Signal Simulator, in proceedings
of ION National Technical Meeting (NTM), 26-28
January, San Diego CA, pp. 378-389, Institute of
Navigation, Fairfax VA.

Gao et al: A Novel Architecture for Ultra-Tight HSGPS-INS Integration 60
Gao G. (2007), INS-Assisted High Sensitivity GPS
Receivers for Degraded Signal Navigation, PhD
Thesis, Published as Reports Number 20252,
Department of Geomatics Engineering, University
of Calgary, Canada (Available at
http:\plan.geomatics.ucalgary.ca).
Gao G. and G. Lacahpelle (2006), INS-Assisted High
Sensitivity GPS Receivers for Degraded Signal
Navigation, in proceedings of ION GNSS 2006,
Fort Worth TX, 26-29 September 2006.
Gautier J. D. and B. W. Parkinson (2003), Using the
GPS/INS Generalized evaluation Tool (GIGET)
for the Comparison of Loosely Coupled, Tightly
Coupled and Ultra-Tightly Coupled Integrated
Navigation Systems, in proceedings of ION 59th
Annual Meeting/CIGTF 22nd Guidance Test
Symposium, 23-25 June, Albuquerque, NM, pp.
65-76, Institute of Navigation, Fairfax VA.
Godha S. and M. E. Cannon (2005), Integration of
DGPS with a Low Cost MEMS – Based Inertial
Measurement Unit (IMU) for Land Vehicle
Navigation Application, in proceedings of ION
GPS/GNSS, September13-16, Long Beach CA,
pp. 333-345, Institute of Navigation, Fairfax VA.
Gustafson D., J. Dowdle and K. Flueckiger (2000), A
High Anti-Jam GPS based Navigator, in
proceedings of ION National Technical Meeting
(NTM), 26-28 January, Anaheim CA, pp. 495-
503, Institute of Navigation, Fairfax VA.
Ilir F. P., W. K. Clifford, G. Gao, R. M. William, J.
Wang and J. Lavrakas (2007), Discrete vs.
Continuous Carrier Tracking Loop Theory,
Implementation, and Testing with Large BnT, in
proceedings of ION GNSS 2007, 25-28
September, Fort Worth TX, Institute of
Navigation, Fairfax VA.
Kaplan E. D., ed (1996), Understanding GPS: Principles
and Application, Artech House, Boston MA.
Kreye C., B. Eissfeller and J. Ó. Winkel (2000),
Improvements of GNSS Receiver Performance
Using Ultra-tightly coupled INS Measurements,
in proceedings of ION GPS/GNSS, 19-22
September, Salt Lake City UT, pp. 844-854,
Institute of Navigation, Fairfax VA.
Lachapelle, G. (2007), Pedestrian Navigation With High
Sensitivity GPS Receivers and MEMS, Journal of
Personal and Ubiquitous Computing, Springer, 11,
6, 481-488.
Lachapelle G., H. Kuusniemi and D. T. H. Dao (2003),
HSGPS Signal Analysis and Performance under
Various Indoor Conditions, in proceedings of
ION GPS/GNSS, 9-12 September, Portland OR,
pp. 1171-1184, Institute of Navigation, Fairfax
VA.
Ma C., R. Klukas and G. Lachapelle (2007), A Non-
Line-of-Sight Error Mitigation Method for TOA
Measurements. IEEE Transactions on Vehicular
Technology, 56, 2 (March), 641-651.
Ma C., G. Lachapelle and M. E. Cannon (2004),
Implementation of a Software GPS Receiver, in
proceedings of ION GPS/GNSS, September 21-
24, Long Beach CA, pp. 956-970, Institute of
Navigation, Fairfax VA.
MacGougan G. D. (2003), High Sensitivity GPS
Performance Analysis in Degraded Signal
Environments, Master thesis, UCGE Reports
Number 20176, Department of Geomatics
Engineering, University of Calgary, Canada
(Available at http://plan.geomatics.ucalgary.ca).
Petovello M. G. (2003), Real-Time Integration of a
Tactical-Grade IMU and GPS for High-
Accuracy Positioning and Navigation, PhD
thesis, UCGE Reports Number 20173, Department
of Geomatics Engineering, University of Calgary,
Canada (Available at
http://plan.geomatics.ucalgary.ca).
Raquet J. F. (2004), GPS Receiver Design, ENGO
699.10 Course Notes, Department of Geomatics
Engineering, University of Calgary, Canada.
Scherzinger B. (2004), Estimation with Applications to
Navigation, ENGO 699.11 Course Notes,
Department of Geomatics Engineering, University
of Calgary, Canada.
Sennott J. (1997), Robustness of Tightly Coupled
Integrations for Real-Time Centimeter GPS
Positioning, in proceedings of ION GPS, 16-19
September, Kansas City KS, pp. 655-663, Institute
of Navigation, Fairfax VA.

Gao et al: A Novel Architecture for Ultra-Tight HSGPS-INS Integration 61
Soloviev A., F. van Graas and S. Gunawardena (2004a),
Implementation of Deeply Integrated GPS/Low-
Cost IMU for Acquisition and Tracking of Low
CNR GPS Signals, in proceedings of ION
National Technical Meeting (NTM), 26-28
January, San Diego CA, pp. 923-935, Institute of
Navigation, Fairfax VA.
Soloviev A., S. Gunawardena and F. van Graas (2004b),
Deeply Integrated GPS/Low-Cost IMU for Low
CNR Signal Processing: Flight Test Results and
Real Time Implementation, in proceedings of
ION GNSS 17th International Technical Meeting
of the Satellite Division, 21-24 September, Long
Beach CA, pp. 1598-1608, Institute of Navigation,
Fairfax VA.
Stephens S. A. and J. B. Thomas(1995), Controlled-Root
Formulation for Digital Phase-Locked Loops,
IEEE Transactions on Aerospace and Electronic
Systems, Vol. 31, No. 1, pp. 78-95.
Tetewsky, A. K. and F. E. Mullen (1996), Carrier phase
wrap-up induced by rotating GPS antennas, in
Proceedings of the Institute of Navigation ION
AM-96, Cambridge, MA, pp. 21-28, , Institute of
Navigation, Fairfax VA.
Van Diggelen F. and C. Abraham (2001), Indoor GPS
Technology, CTIA Wireless-Agenda, Dallas,
May.
Watson, R., G. Lachapelle, R. Klukas, S. Turunen, S.
Pietilä and I. Halivaara (2006), Investigating GPS
Signals Indoors with Extreme High-Sensitivity
Detection Techniques, Navigation, Institute of
Navigation, 52, 4, 199-213.
Zhodzishsky M., S. Yudanov, V. Veitsel and J. Ashjaee
(1998), COOP Tracking for Carrier Phase, in
proceedings of ION GPS, 15-18 september,
Nashville, Tennessee, pp. 653-664, Institute of
Navigation, Fairfax VA.