See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/237322666

The Role of Kalman Filter in the Modelling of GPS Errors

Article · January 2009

CITATIONS READS
7 733

5 authors, including:

B. L. Malleswari Iyyanki Muralikrishna

sridevi women;s Engg college Defence Research and Development Organisation
17 PUBLICATIONS   64 CITATIONS    98 PUBLICATIONS   340 CITATIONS   

SEE PROFILE SEE PROFILE

M. Seetha
G. Narayanamma Institute of Technology and Science
59 PUBLICATIONS   216 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Geospatial Public health data management system View project

Image Fusion View project

All content following this page was uploaded by Iyyanki Muralikrishna on 04 June 2014.

The user has requested enhancement of the downloaded file.

 Journal of Theoretical and Applied Information Technology

© 2005 - 2009 JATIT. All rights reserved.

www.jatit.org

THE ROLE OF KALMAN FILTER IN THE MODELLING OF

GPS ERRORS
*B.L.Malleswari, **I.V.MuraliKrishna, ***K.Lalkishore,*M.Seetha, *Nagaratna, P. Hegde

*Associate Professor, Gandipet, Hyderabad, Andhra Pradesh, India

** Director, R&D, JNTU, Kukatpalli, Hyderabad, Andhra Pradesh, India,
*** Registrar , JNTU, Kukatpalli, Hyderabad, Andhra Pradesh , India.
E-Mail: , [email protected], [email protected]

ABSTRACT

This paper describes about the modeling of errors (like ionospheric delays, atmospheric delays,
Troposphericelays, Multipath effects and dilution of precision etc.,) affecting the GPS signals as they
travel from satellite to user on Earth. These errors degrade the accuracy of GPS position. An attempt is
made to improve the accuracy in locating the GPS receiver by filtering the range measurements with the
datum conversion between Universal Transverse Mercator (UTM) and World Geodetic System (WGS -
84) using single frequency ML-250 hand held GPS receiver and smoothening of these coordinates using
Kalman filter. A linear recursive filtering technique, Kalman filter is used for greater accuracy in
estimating the position of user by considering the initial state of the system, statistics of system noise and
measurement errors from sensor noise measurements. The results of proposed Kalman filter technique
give better accuracy with more consistency and are found superior to the standard one

Key words: GPS, Datum conversion, Kalman filter, Recursive algorithm, Error covariance, Measurement
noise.

1. INTRODUCTION OF GLOBAL
POSITIONING SYSTEM
A Satellite-based system Global Positioning A key advantage of the UTM projection is its
System uses a constellation of 24 satellites to preservation of the shape of the small areas on a
give an accurate position of user. GPS receivers map and its grid coordinates permit easy
have been developed to observe signals calculations using plane trigonometry (Langely
transmitted by the satellites and achieve sub- B Richard, 2000). In UTM, the ellipsoid is
meter accuracy in point positioning and a few portioned in to 60 zones with a width of six
centimeters in relative positioning. GPS can be degrees longitude each. A scale factor of 0.996
operated in all weather, day and night with out is applied to the central meridian (Leick, 1995).
any requirement of Inter visibility between The scale factor is to avoid fairly large
points. GPS provides a global absolute distortions in the outer areas of zone.
positioning capability with respect to a There are several sources of error that degrade
consistent terrestrial reference frame and the GPS position from few meters to tens of
considered as an absolute global geodetic meters (Pratap Misra, 2001). These error sources
positioning system. The GPS satellites are are Ionospheric, Atmospheric delays, Satellite
positioned in such a way that at least five to and Receiver Clock Errors, Multipath, Dilution
eight satellites are accessible at any point on of Precision, Selective Availability (S/A) and
earth and at any time (Hoffmann-Wellonhaff et Anti Spoofing (A-S) as described by Hoffmann -
al, 1998). GPS is based on a system of Wellenhof et al (1998). The errors could be
coordinates called the World Geodetic System transmitted via VHF/UHF links and the users
1984 (WGS-84 whose coordinates are the can make use of the corrections to fix their
latitude, longitude, and height) (Kaplan.E.D, positions more accurately. These errors can be
1996and Parkinson, 1996). GPS data is observed reduced to arrive at a more accurate estimate of
in WGS 84 and Universal Transverse Mercator coordinates of user by means of a recursive
(UTM). The most common map projection and algorithm- KALMAN FILTER. The emphasis is
grid system used for land navigation is the given on the above errors to analyze the Kalman
Universal Transverse Mercator (UTM) system. filter (Grewel.M. S et al, 2001).

95
 Journal of Theoretical and Applied Information Technology

© 2005 - 2009 JATIT. All rights reserved.

www.jatit.org

Beran et al (2001) could try to develop Kalman extended their work further and recent precise
filter based functional model for single point positioning results show decimeter
frequency positioning which is suitable for low accuracy
dynamic platforms. Bisnath and Langely (2002)
with the use of dual frequency geodetic-quality proposed which can efficiently address the
receiver. Kevin Milton et al (2006) have also above problems.
adapted a technique, which utilizes semi-
codeless technique to with the use of dual 2. THE KALMAN FILTER
frequency geodetic quality receiver. Kevin
Milton et al (2006) have adopted a technique A significant mathematical toolbox used for
which utilizes a semi-codeless technique to stochastic estimation from noisy sensor
provide high accuracy GPS measurements and measurements is Kalman filter. Kalman filtering
L1-L2 carrier phase measurement for wide-lane is based on linear mean square error filtering
applications. The delay of the P code is varied (estimation) and it is essentially a set of
and the signals are cross correlated until a mathematical equations that implement a
maximum value is reached. When the cross Predictor-corrector type estimator which is
correlation of the L1 and L2 signals is at a peak optimal. It minimizes the estimated error
value, the relative delay between the L1 and L2 covariance —when some presumed conditions
signals is proportional to the ionospheric delay. are met. For the given spectral characteristics of
The derived ionospheric delay may be accounted an additive combination of signal and noise, the
for in measurement analysis of the L1 signal to linear operation on this input combination yields
provide for a high precision position solution. the best (meaning minimum square error)
(http:// www.google.com /patents? separation of the signal from the noise is to be
hl=en&lr=&vid=USPAT5903654&id=TkcXAA known.
AAEBAJ&oi=fnd&dq=ionospheric+delays; The distinctive feature of Kalman filter is about
29TH DEC 2006). the description of its mathematical formulation
The GPS receiver is giving both WGS-84 data in terms of state space analysis as per (Bozic,
observed in latitude, longitude & altitude and 1999) and its solution is computed recursively.
UTM data observed in Northing & Easting As each update estimate is computed from the
(Langely.R.B, 2000). In the conversion process previous estimate and the input data, only
of UTM to WGS 84, accuracy must be obtained previous estimate requires storage. The filter is a
without distortions .In the work attempted by computational algorithm that processes
Ravindhra .et al (Feb, 2002), only the datum measurements to deduce a minimum error
conversion from WGS- 84 to UTM and estimate of the system by utilizing knowledge of
inaccuracy were discussed. The role of the noise the system, measurement dynamics, and
in GPS is only at satellite and receiver segments. statistics of the system, noises measurement
The modeling of the noise in satellites and errors and initial condition information. It is to
receiver segments are discussed by Langely improve the quality of datum conversion using
(March 1997). this smoothening technique. It also reduces the
Y. Yuan and J. Ou (2001) developed one robust error while converting from UTM to WGS-84
recurrence technique, based on the efficient GPS data. Here by employing this smoothening
combination of single-frequency GPS technique, Kalman filter puts up better UTM to
observations by users and the high-precision WGS-84 conversion efficiency. The effects of
differential ionospheric delay corrections from ionospheric delays have already been discussed
WAAS. For the commonly used GPS wide-area by Klobuchar (May 1987). Smoothening of
augmentation systems (WAAS) with a grid WGS- 84 with the help of Kalman filter has
ionospheric model, the efficient modeling of been discussed by Malleswari et al (2005). But
ionospheric delays in real time, for single- in this proposed technique, the Kalman filter is
frequency GPS users, is still a crucial issue used to smoothen the UTM coordinates. Since
which needs further research. This is particularly the time of its introduction, the Kalman filter has
necessary when differential ionospheric delay been the subject of extensive research and
corrections cannot be broadcast, when users application, particularly in the area of
cannot receive them or when there are autonomous or assisted navigation. This is
ionospheric anomalies. Ionospheric delays have likely due in large part to advances in digital
a severe effect on navigation performance of computing, relative simplicity and robust nature
single-frequency receivers. A new scheme is of the filter itself. Rarely do the conditions

96
 Journal of Theoretical and Applied Information Technology

© 2005 - 2009 JATIT. All rights reserved.

www.jatit.org

necessary for optimality actually exist, and yet equations as shown in figure 1. Discrete Kalman
the filter apparently works well for many filter time update equations (5 & 6) are given as
applications in spite of this situation as xˆ -k = Axˆk −1 + Buk
emphasized by (Peter S Maybeck, 2001). ---------- 5 & 6
Langely (October, 2000) discussed about the Pk − = APk −1 AT + Q
GPS receivers tracking 8 or more satellites Time update equations project the state and
simultaneously and determining the positions covariance estimates forward from time step k-1
using all of the available observations with a to step k. A and B are from equation (1), while
least – squared – error algorithm. Q is from is from equation (.3). Initial conditions
for the filter are discussed in the earlier
3 MATHEMATICAL FORMULATION OF references. Discrete Kalman filter measurement
KALMAN FILTER
update equations (7, 8 & 9) are given below.
The Kalman filter addresses the general K k = Pk − H T ( HPk − H T + R ) −1
problem of trying to estimate the state
xˆk = xk − + K k ( Z k − Hxˆk − ) -------7, 8 & 9
x ∈ ℜ of a discrete-time controlled process
n
−
that is governed by the linear stochastic Pk = ( I − K k H ) Pk
difference equation as in equation 1. The time update equations are responsible for
XK = A X K -1 + B U K + W K -1 ------------- (1) projecting forward (in time) the current state and
error covariance estimates to obtain the a priori
With a measurement x ∈ ℜ
m
that is (as estimates for the next time step. The
stated in equation 2). measurement update equations are responsible
ZK = HX K +V K ----------------------------- (2) for the feedback--i.e. for incorporating a new
measurement into the a priori estimate to obtain
The random variables WK and VK represent the an improved a posteriori estimate. The time
process and measurement noise (respectively). update equations can also be thought of as
They are assumed to be independent (of each predictor equations, while the measurement
other), white, and with normal probability update equations can be thought of as corrector
distributions equations. Indeed the final estimation algorithm
P (W) – N (0, Q) --------------------- (3) resembles that of a predictor-corrector algorithm
for solving numerical problems.
P (V) – N (0, R) -------------------- (4) The first task during the measurement update is
The process noise covariance Q and to compute the Kalman gain, Kk. The next step
measurement noise covariance R matrices as in is to actually measure the process to obtain Zk,
equations 3 & 4 might change with each time and then to generate an a posteriori state
step or measurement, however here we assume estimate by incorporating the measurement as in
they are constant (Peter S Maybeck (2001). equation (8). The final step is to obtain an a
The n×n matrix A in the difference equation (1) posteriori error covariance estimate via equation
relates the state at the previous time step K-1 to (9). After each time and measurement update
the state at the current step K, in the absence of pair, the process is repeated with the previous a
either a driving function or process noise. Note posteriori estimates used to project or predict
that in practice A might change with each time the new a priori estimates. This recursive nature
step, but here we assume it is constant. The n×l is one of the very appealing features of the
matrix B relates the optional control input Kalman filter—it makes practical
u ∈ ℜl to the state x. The m×n matrix H in implementations much more feasible than (for
the measurement equation (2) relates the state to example) an implementation of a Wiener filter
the measurement ZK. In practice H might change (Brown and Hwang 1992) which is designed to
with each time step or measurement, but here we operate on all of the data directly for each
assume it is constant. The Kalman filter estimate. The Kalman filter instead recursively
estimates a process by using a form of feedback conditions the current estimate on all of the past
control: the filter estimates the process state at measurements. Figure 1 below offers a complete
some time and then obtains feedback in the form picture of the operation of the filter, combining
of (noisy) measurements. As such, the equations the high-level equations 5 & 6.
for the Kalman filter fall into two groups: time
update equations and measurement update

97
 Journal of Theoretical and Applied Information Technology

© 2005 - 2009 JATIT. All rights reserved.

www.jatit.org

4. FILTER PARAMETERS AND TUNING process model can produce acceptable results if
one "injects" enough uncertainty into the process
In the actual implementation of the filter, the via the selection of Q.
measurement noise covariance R is usually Certainly in this case one would hope that the
measured prior to operation of the filter. process measurements are reliable. In either
Measuring the measurement error covariance R case, whether or not we have a rational basis for
is generally practical (possible) because we need choosing the parameters, often times superior
to be able to measure the process anyway (while filter performance (statistically speaking) can be
operating the filter) so we should generally be obtained by tuning the filter parameters Q and
able to take some off-line sample measurements R. The tuning is usually performed off-line,
in order to determine the covariance of the frequently with the help of another (distinct)
measurement noise The determination of the Kalman filter in a process generally referred to
process noise covariance Q is generally more as system identification, which is clearly stated
difficult as we typically do not have the ability in (Bozic, 1999).
to directly observe the process we are
estimating. Sometimes a relatively simple (poor)

Figure1. A complete picture of the operation of the Kalman filter

In closing we note that under conditions where changed dynamically during filter operation –
Q and R are in fact constant, both the estimation becoming Qk - in order to adjust to different
error covariance PK and the Kalman gain KK will dynamics. For example, in the case of tracking
stabilize quickly and then remain constant. If the head of a user of a 3D virtual environment
this is the case, these parameters can be pre- we might reduce the magnitude of Qk if the user
computed by either running the filter off-line, or seems to be moving slowly, and increase the
for example by determining the steady-state magnitude if the dynamics start changing
value of PK rapidly. In such cases Qk might be chosen to
It is frequently the case however that the account for both uncertainties about the user's
measurement error (in particular) does not intentions and uncertainty in the model.
remain constant. For example, when sighting
beacons in our optoelectronic tracker ceiling 5. RESULTS AND DISCUSSIONS
panels, the noise in measurements of nearby
beacons will be smaller than that in far-away Using single frequency ML 250 GPS hand held
beacons. Also, the process noise Q is sometimes receiver, the field data is collected at different

98
 Journal of Theoretical and Applied Information Technology

© 2005 - 2009 JATIT. All rights reserved.

www.jatit.org

locations in around Hussain Sagar and Gandipet, filter is 0.03084715 for Gandipet and
Hyderabad. The receiver is giving WGS – 84 0.0006331302 for Hussain Sagar. So, lesser the
data (Ørx and λrx) and UTM data (Nrx & Erx). variance more will be the consistency. Again,
The variance for WGS 84 of Gandipet is: λrx- the same GPS receiver’s UTM data (Nrx & Erx)
Longitude of Receiver is 27.0337204 in degrees and is fed to Web Software “Coordinate. Transform”
Φrx– Latitude of Receiver is 6.89938593 in degrees. to validate WGS 84 data, i.e., (Øs/w & λs/w).
The variance for WGS 84 of Hussain Sagar is: This converted data is also again analyzed as
λrx- Longitude of Receiver is 13.39910013 in degrees Φs/w- Latitude in degrees after applying websoft is
and Φrx– Latitude of Receiver is 0.020659696 in 0.011725697 for Gandipet and0.000198005 for
degrees. Now the UTM coordinates are Hussain Sagar. After comparing all the variances,
converted into WGS- 84 coordinates. It is quite that is ( Øs/w, Ørx, Økf & Øprg and λs/w, λprg,
evident that the converted values of latitude and λkf & λrx), it is found that the smoothened
longitude (Øprg and λprg) with out applying converted data, which is developed from
Kalman filter are giving poor resolution as Kalman filtering technique (λkf & Økf), is having
shown in tables 1and 2. And they are more better consistency. A comparative analysis is
inconsistent as the variance is more for the emphasized in the figures 2, 3, 4 and 5. The
converted data, i.e., Φprg - Latitude in degrees accuracy of results obtained from program
before applying kalman filter: 7.3488 (Gandipet) conversions has been validated. Table 1a and 1b
and 0.001264126 (Hussain Sagar) Hence the indicate the comparison of all latitudes and
UTM data to WGS 84 converted data now longitudes of Gandipet with respect to source
smoothened by Kalman filtering algorithm to data (receiver), before and after applying
test the accuracy and consistency (Økf and λkf) Kalman filter and after applying web soft. Table
which is giving a very small variance. It is 2a and 2b indicate the comparison of all
shown from the tables 1 and 2 that, Φkf – the latitudes and longitudes of Hussain Sagar with
Latitude in degrees after applying kalman filter respect to source data (receiver), before and after
is 0.004221766 for Gandipet and applying Kalman filter and after applying web
0.00003667424 for Hussain Sagar.. Similarly, soft.
λkf - Longitude in degrees after applying kalman
Table1a: Comparison of variance for Table2a: Comparison of variance for different
different longitudes of Gandipet longitudes of Hussain Sagar

Longitude variance Longitude variance

λrx- Longitude in λrx- Longitude in degrees
degrees (Receiver) 27.0337204 (Receiver) 13.39910013
λprg - Longitude in λprg - Longitude in
degrees before applying degrees before applying
kalman filter 27.0337194 kalman filter 0.001264126
λkf - Longitude in λkf - Longitude in
degrees after applying degrees after applying
Kalman filter 0.03084715 Kalman filter 0.0006331302
λs/w- Longitude in λs/w- Longitude in
degrees after applying degrees after applying web
web soft 6.904284042 soft 0.00355667
Table1b: Comparison of variance for Table2b: Comparison of variance for
different latitudes of Gandipet different Latitudes of Hussain Sagar
Latitude variance Latitude variance
Φrx– Latitude in degrees Φrx– Latitude in degrees
(Receiver) 6.89938593 (Receiver) 0.020659696
Φprg - Latitude in degrees Φprg - Latitude in degrees
before applying kalman 7.3488 before applying kalman
filter filter 0.000292464
Φkf - Latitude in degrees Φkf - Latitude in degrees
after applying Kalman after applying Kalman
filter 0.004221766 filter 0.00003667424
Φs/w- Latitude in degrees Φs/w- Latitude in degrees
after applying web soft after applying web soft
0.011725697 0.000198005

99
 Journal of Theoretical and Applied Information Technology

© 2005 - 2009 JATIT. All rights reserved.

www.jatit.org

g
17.5
latitude gandipet
17.45
17.45

17.4
17.4

17.35
17.35

WGS 84 latitude rx

d e g re e s
17.3
17.3 WGS 84 latitude prog
d e g re e s

WGS 84 latitude KF
17.25
17.25 WGS84 latitude rx WGS 84 latitude s/w
WGS84 latitude prg
17.2
WGS84 latitude KF 17.2
WGS84 latitude s/w

17.15 17.15

17.1 17.1

17.05 17.05
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49

17 number of readings
1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58 61 64 67 70 73 76 79

number of readings

Figure 4. Plot of Latitudes for the field data of

Figure2. Plot of Latitudes for the field data of
Hussain Sagar
Gandipet

longitude gandipet longitude hussain sagar

78.55 78.56

78.5 78.54

78.45
78.52

78.4
d e g re e s

78.5
d e g re e s

WGS 84 longitude rx
WGS 84 longitude rx WGS 84 longitude prg
78.35 WGS 84 longit ude prog WGS 84 longitude KF
WGS 84 longit ude KF 78.48 WGS 84 longitude s/w
WGS 84 longit ude s/w
78.3

78.46

78.25

78.44

78.2

78.42
78.15 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58 61 64 67 70 73 76 79 number of readings
number of readings
Figure 3.Plot of Longitudes for the field data of Figure 5. Plot of Longititudes for the field data of
Gandipet Hussain Sagar

100
 Journal of Theoretical and Applied Information Technology

© 2005 - 2009 JATIT. All rights reserved.

www.jatit.org

6. CONCLUSIONS [8]. Kevin Milton et al (Dec, 2006), “Method

and apparatus for eliminating ionospheric
A comparison of accuracy is suggesting, delay error in global positioning system
accuracy through Kalman filter application is signals”
certainly yielding better results. The present http://www.google.com/patents?hl=en&lr=&v
study & data analysis methodology showed that id=USPAT5903654&id=TkcXAAAAEBAJ&
the variations in the signal related to WGS- 84 oi=fnd&dq=ionospheric+delays.
data can be smoothened using Kalman filter [9]. Klobuchur J.A (May 1987), “Ionosphere
with in the range of studies made and the time delay model for single frequency GPS
analysis is found to yield better accuracies as, it users” ,IEEE Trans Aerospace Electron
is shown from the tables 1 and 2 that, Φkf – the system (USA).
Latitude in degrees after applying [10]. Langely R.B (March 1997), “The GPS
kalman filter is 0.004221766 for Gandipet error budget” in GPS world.
and 0.00003667424 for Hussain Sagar. [11]. Langely.R.B, (2000), “The UTM Grid
System, 2nd Ed, (New York: John Wiley &
Similarly, λkf - Longitude in degrees after
Sons).
applying kalman filter is 0.03084715 for [12]. Langely R.B (October 2000) “Basic
Gandipet and 0.0006331302 for Hussain Navigation with a GPS Receiver” GPS World.
Sagar. So, lesser the variance more will be the [13]. Leick.A, (1995), GPS Satellite Surveying,
consistency. However the extensive application 2nd edition (New York: John Wiley &
of the methodology for the data in bringing out Sons).
the limitations of the smoothening of the signal, [14]. Malleswari B.L, MuraliKrishna I.V and
a statistical evaluation LalKishore K (Jan 2007) “Kalman filter for
atleast in robust domain could throw some light GPS Datum conversion”, Mapworld Forum,
on actual information content & loss of the Hyderabad.
information through Kalman filtering. [15]. Parkinson, B.W and spilker, J.J (eds),
(1996), Global Positioning System: Theory
REFERENCES and applications, Volume .1(694pp), Volume
2(632pp), I edition (American Institute of
[1]. Bozic.S, M, (1999), Digital and Kalman Aeronautics & Astronautics, Inc).
Filtering: An Introduction to Discrete-Time [16]. Peter S Maybeck, (2001), Stochastic Models
Filtering and Optimum Linear Estimation, 2nd – estimation and control, 1st Ed, (New York:
edition (New York: John Wiley& Sons). PTH).
[2]. Beran T, S.B. Bisnath and Lagely (2001), [17]. Pratap Misra and Per Enge, (2001), Global
Single receiver GPS positioning in support of Positioning System Signals, Measurements,
Airborne gravity for exploration and mapping, and Performance, 1st edition (Massachusetts:
Geoide, Fredenton, June 20-22, 2001. Ganga-Jamuna Press).
[3]. Bisnath S.B and Lagely (2002), High [18]. Ravindhra K, Malleswari B.L. and Sarma
Precision Kinematic Positioning with a A.D (Feb, 2002), “Development of conversion
single GPS Receiver, Navigation; Journal of Parameters for GPS Datum”, Proceedings of
the institute Navigation ,vol 49, No 3, Fall National conference on GIS and their
2002 161 – 169. applications in Civil Engg, Deccan College of
[4]. Brown. R. G. and P. Y. C. Hwang. Engg, Hyderabad.
Introduction to Random Signals and Applied [19]. Yuan and J.Ou (August, 2001), An
Kalman Filtering, 2nd Edition, John Wiley & improvement to ionospheric delay correction
Sons, Inc, 1992. for single-frequency GPS users - the APR-I
[5]. Grewel.M. M S, Lawrence R. Weill, Angus P. scheme, Journal of Geodesy, Volume 75,
Andrews, (2001), Global Positioning System, Numbers 5-6 / , 331-336.
Inertial Navigation and Integration, 1st [20]. www.franson.com
edition (New York: John Wiley & Sons).
[6]. Hofmann-Wellenhof, H.Lichteneegger,
J.Collins, (1998), Global Positioning Systems:
Theory &Practice, 4th edition, (New
York/Berlin: Springer-Verlag Wien).
[7]. Kaplan E (ed), (1996), Understanding GPS:
Principles and Applications, 1st edition
(Boston: Artech House).

101

View publication stats