Track 1: AUTOMATIC CONTROL

FEEDFORWARD STRUCTURE OF KALMAN FILTERS FOR LOW

COST NAVIGATION
T. D. Tana, L.M.Ha a, N. T. Longa, H.H.Tuea, c, N. P. Thuya, b
a

Faculty of Electronics and Telecommunication, College of Technology, VNU, Hanoi

b
International Training Institute for Materials Science (ITIMS), HUT, Hanoi
c
Laval University, Canada
ABSTRACT

Nowadays, navigation and guidance are very important problems for marine, aeronautics and space
technology. In such systems, Inertial Measurement Units (IMUs) are widely used as the core of the
Inertial Navigation Systems. However, there are existing errors in the accelerometer and gyroscope
signals that cause unacceptable drifts. To eliminate the deterministic errors, we can specify them
quantitatively by calibrating the device. It is, however, more complex in determination of the
stochastic errors. Thus, an optimal filter such as Kalman one is often used to minimize these errors on
the INS system. In this paper, we present a feed forward Kalman Filter (KF) of which the noisy INS
output is combined with the noisy GPS output in improving the quality of the navigation and guidance
systems.
1. INTRODUCTION
The demand of navigation and guidance has
been urgent for many years [1]. In fact, GPS
have been employed widely in many
applications while INS is daily used in
controlling flight dynamics. In principle, an
IMU consists of gyroscopes and accelerometers
that measure angular velocities and accelerations
in three dimensions. Recently, thank to the
development of MEMS technology, the IMUs
become smaller, cheaper and more precise.
However, there are still problems with MEMS
based the IMUs which are necessary to be
solved. The position error of an INS increases
rapidly with navigation due to the integration of
measurement errors in the gyroscopes and
accelerometers. In order to make the corrections,
there appears a new trend in navigation and
guidance domain [2]: it consists of the
integration of INS and GPS altogether.
Integrating these two methods can improve the
performance of the system and reduce
concurrently the disadvantages of both INS and
GPS.

An INS often consists of three accelerometers

and three gyroscopes in order to measure the
accelerations in three dimensions and the
rotation rates around three axes [3]. There are
two typical INS systems: gimble system and
strapdown system. The strapdown INS system
has been used more popular than gimble system.
The strapdown system is mostly based on the
MEMS technology that is relatively inexpensive
and
compact.
In
strapdown
system,
accelerometers and gyroscopes are fixed to body
frame of the aircraft. Signals from these sensors
are processed in order to obtain three Euler
angles. The results are corrected by gravity
acceleration and Earth rotation velocity.
Three accelerations (ax, ay and az) along
three axes of body frame relate to three
velocities (U, V and W) in the Earth fixed frame
by following equation:
.

U = a x +V .r W.q + g.sin
.

V = a y U .r + W.p - g.cos .sin

(2)

W = a z U .q + W.p - g.cos .cos

2. INS AND GPS

2.1. Inertial Navigation System (INS)

Where p, q, and r are velocities of roll, pitch,

and yaw, respectively.

International Symposium on Electrical & Electronics Engineering 2007 - Oct 24, 25 2007 - HCM City, Vietnam
-1-

Track 1: AUTOMATIC CONTROL

To convert the movement from body frame to

navigation frame (see Fig. 1), we could use
Direct Cosine Matrix (DCM):

. V N
U
X.
T
Y = V E = DCM V

.
W
Z V D

(3)

2.2. Global Positioning System (GPS)

Then, we can obtain the latitude, longitude and

height of the aircraft by following equations:

=

VN
Rearth

=

VE
Rearth cos

Figure 2. The discrete INS algorithm

H
= V D

(4)

The GPS consists of 24 satellites that fly above

the surface of the Earth at the height 19.200 km
in order to acquire the position of the aircraft
(latitude, longitude and height) [4]. Radio
signals often hardly transmit over solid
buildings, tunnelsTo get the correct position
of the aircraft; it requires at least four satellites.
They can separate the GPS into three segments
(see Fig. 3):
Space segment: consists of 24 satellites.
Control segment: ground control
User segment: receiver

Figure 1. Navigation frame

There are many kinds of errors in the INS. Most

of the errors in the INS are caused by sensor
imperfections that are necessary to be solved.
The position error of an INS increases rapidly
with navigation due to the integration of
measurement errors in the gyroscopes and
accelerometers. In order to make the corrections,
the errors are often classified into deterministic
errors and stochastic errors. We have succeeded
in specifying the parameters of the IMU errors,
which is a necessary step when applying errorprocessing algorithms for the INS. A discrete
INS algorithm was developed in our lab in order
to develop a completed Inertial Measurement
Unit (IMU) (see Fig. 2). The detail of this
discrete algorithm will not be presented in this
paper.

Figure 3. Structure of the GPS

Errors in the GPS mostly caused by six

following factors (not including Selective
Availability error):
Ephemeris data
Satellite clock
Multipath reflection
Atmospheric delay
Random measurement noise
Receiver (include software)

International Symposium on Electrical & Electronics Engineering 2007 - Oct 24, 25 2007 - HCM City, Vietnam
-2-

Track 1: AUTOMATIC CONTROL

3. INTEGRATION OF INS AND GPS

USING KALMAN FILTERING
The INS system has two main advantages when
comparing with other navigation system: selfcontained ability and high accuracy for shortterm navigation. In long-term navigation
applications, the INS should works with the aid
of the GPS system. The integration of INS and
GPS is considered an optimal combination. The
heart of integrated system is Kalman algorithm
[5].
Fig.
4
illustrates
a
feedforward
configuration. Its advantage is that provides a
rapid filter response. Alternatively, the
configuration in Fig. 5 is a feedback one. This
configuration is more complex than the open
loop one but it can provide better performance
in the exist of nonlinear effects. In our project,
the feedforward configuration is chosen in order
to provide fast performance [6, 7].

this is the GPS-aided INS system configuration.

Looking at the discrete model:
x k +1 = k x k + wk

(5)

Where k is the state transition matrix, wk is the

driven response at tk+1 due to the presence of
input white noise during time interval (tk, tk+1).
Because this time interval is short (it means
that the update rate of the INS is high), we can
approximate k as:
k = e ( Ft ) I + Ft

(6)

and the covariance matrix associated with wk is

Qk = E wk wkT k GQG T Tk t

(7)

Where

2
Q = diag ax

ay2 az2 2x 2x 2x

Looking at the observation equation:

zk = Hk xk + vk

(8)

where
Figure 4. Feedforward configuration

VnINS VnGPS

z k = VeINS VeGPS , H k = (I 33 0 35 ) and

Vd Vd
GPS
INS

Rk = E v k v kT = diag Vn2 Ve2

Figure 5. Feedback configuration

The Kalman filter is a multiple input, multiple

output digital filter that can optimally estimate
in real time the states of the system based on its
noisy outputs. These states are velocity errors
(eVN, eVE, eVD), drift terms (GBx, Gby, GBz) and
attitude errors (Tn, Te). The GPS output is used
as a tool to estimate the error in the INS and to
correct the error as much as possible. We call

Vd2

] is

the covariance matrix for vk.

In this system, we assume that the process
noise wk and the measurement noise vk are
uncorrelated.
Figure 6 shows the more detail of the
feedforward KF implementation.

Figure 6. The 8 states feedforward structure

International Symposium on Electrical & Electronics Engineering 2007 - Oct 24, 25 2007 - HCM City, Vietnam
-3-

Track 1: AUTOMATIC CONTROL

4. EXPERRIMETAL RESULTS
In this study, we used the MICRO-ISU BP3010
which consists of three ADXRS300 gyros and
three
heat
compensated
ADXL210E
accelerometers.
The
measurements
are
synthesized by IMUs micro-controllers and
transmitted out via RS232 interface. The unit
transmits output data as angular incremental and
velocity incremental data in serial frames of 16
bytes at 64 Hz.

Figure 7. The BP3010 A MEMS unit.

After a calibration procedure, the experiment

system is placed on precise rate table which
contains sequence of different rates for each
dimension has been made use. The IMU is
initially positioned in center of rate table and
each rate is run approximately for 10 minutes.
As mentioned above, eight states were
considered and Kalman filter was used to
estimate the INS errors. The INS can give out
correct values of the velocities by subtracting
the noisy INS to the estimation of the INS
errors. The standard deviation of the sensors
chosen was 30 mGal. The standard deviation of
the GPS was 20 m.
To point out the usefulness of the KF, look
at the Fig. 7 and Fig. 8 that are illustrate the
heading of INS while standing still in the case
without KF and with KF.

Figure 7. The drift of heading (without KF)

Figure 8. The heading with the present of KF

While the INS is in rotation at the rate 10 o/s, the
integration system with KF can provide the
exactly the heading as shown in Fig. 9.

Figure 9. The heading with the present of KF at

the rate 10 o/s

International Symposium on Electrical & Electronics Engineering 2007 - Oct 24, 25 2007 - HCM City, Vietnam
-4-

Track 1: AUTOMATIC CONTROL

Figure 10. The north velocity of the stand still

IMU in two cases: with and without KF

Come back to stand still case and look at the

north velocities in two cases: with and without
KF. The graphs for velocity computed and
corrected by the Kalman filter are given in Fig.
10. We can see that the un-aided INS deviates
from the ideal velocity by a large quantity. If the
integration system is supported by KF, the
output Vn is around 0 m/s. It is mean that our KF
could give the exact correction in feedforward
configuration.
The update from the INS was taken every
0.015625s, the GPS update was taken every 1s
and the KF was run every 0.5s to achieve better
accuracy. Every alternate 0.5s instant, when the
GPS update is not available; we have to predict
the error state by using the most recent GPS
update as the measurement, i.e. the GPS update
is taken constant for that whole one second. This
also comes in use when there are GPS outages.
For the experiment of the IMU on road [8],
GPS and the data acquisition system were
installed in a vehicle. The vehicle was driven for
12 minutes in a 5 km trajectory. Initially the
vehicle was at rest, with the engine on, for about
45 seconds. This stationary data was used for
calibration and alignment purposes.
Fig. 11 and 12 present the velocity of KF
compared with the values measured with the
GPS unit. It can be seen outputs of KF, black,
follows the GPS velocities with small error for
approximately 15 minutes.

Figure 11. Comparison of East velocities

Figure 12. Comparison of North velocities

The 2-D trajectory is presented in Fig. 13.

This figure shows the position of the vehicle
along North and East direction on the Earth
instead of the latitude and the longitude. The
reason is that we can prevent numerical
instabilities in calculating the Kalman gain. It
can be seen the INS/GPS trajectory supported by
KF follows the GPS one (dot curve) with small
error for a quite long trip.

International Symposium on Electrical & Electronics Engineering 2007 - Oct 24, 25 2007 - HCM City, Vietnam
-5-

Track 1: AUTOMATIC CONTROL

Figure 13. Comparison of trajectory

In land vehicle applications the errors in roll and

pitch can be catastrophic since they are mainly
responsible for the gravity vector compensation.
The selection of the roll and pitch gyros should
contemplate the terrain type since it will be
responsible for the maximum rate measured by
these gyros.

5. CONCLUSIONS
In this paper, an eight state Kalman filter
was proposed to be used in order to enhance the
quality of a combined GPS and INS system. The
experimental results have shown that the initial
calibration and alignment is accurate enough to
allow navigation with IMU sensors for extended
period of time with low dead reckoning errors.
In fact, the Kalman filter tremendously improves
accuracies compared to the GPS and INS when
operating alone as individual navigation
systems. The accuracy of estimated parameters
is improved by increasing the dimension of the
systems states space. Obviously, by its
simplicity, this model can be embedded easily
into a real time system. Future work will
investigate in-flight calibration and alignment
algorithms extending the error models of the
INS system.

REFERENCES
1. Collinson,
R.P.G.,
Introduction
to
Avionics, Chapman and Hall, London,
1996.
2. Omerbashich,
Integrated
INS/GPS
Navigation from a Popular Perspective",
Journal of Air Transportation, Vol. 7, No. 1
(2002), pp. 103-119.
3. Randle, S.J., Horton, M.A., Low Cost
Navigation Using Micro Machined
Technology,
IEEE
Intelligent
Transportation Systems Conference, 1997.
4. Parkinson, B.W., and Spilker, J.J.Jr.,
Global Positioning System: Theory and
Applications, Vol. 1(1996), AIAA,
Washington DC.
5. Grewal, M.S., Weill, L.R., and Andrews,
A.P., Kalman Filtering: Theory and
Practice using MATLAB, John Wiley and
Sons, New York, (2001).
6. Wolf, R., Eissfeller, B., Hein, G.W., A
Kalman Filter for the Integration of a Low
Cost INS and an attitude GPS, Institute of
Geodesy
and
Navigation,
Munich,
Germany.
7. Tran Duc Tan, Huynh Huu Tue, Nguyen
Thang Long, Nguyen Phu Thuy, Nguyen
Van Chuc, Designing Kalman Filters for
Integration of Inertial Navigation System
and Global Positioning System, The 10th
biennial Vietnam Conference on Radio &
Electronics, REV-2006, (2006), pp. 76-80.
8. Panzieri, S., Pascucci, F., Ulivi, G., An
Outdoor navigation system using GPS and
Inertial
Platform,
IEEE
ASME
Transactions on Mechatronics, Vol.
7.(2002).

Acknowledgements: This work was supported

by the QGT0509 project.

International Symposium on Electrical & Electronics Engineering 2007 - Oct 24, 25 2007 - HCM City, Vietnam
-6-