Beruflich Dokumente
Kultur Dokumente
Shovan Bhaumik
I. Introduction
The satellites used in communications are of three types
in practice, namely Low earth orbit (LEO), Medium earth
orbit (MEO) and Geostationary earth orbit (GEO). With
respect to any fixed point at earth, the LEO and MEO
satellites are moving while the GEO satellites are stationary. A particular satellite can cover only a part of total
earth area at a time. Hence, we need a number of satellites
to cover whole earth, so that continuous service can be
provided. The number of satellites used is proportional
to its altitude. Hence, the number of satellite required to
cover whole earth is highest for LEO and lowest for GEO.
As the LEO and MEO satellites are not stationary, the
relative position between any specific satellite and earth
or between different satellites itself changes with time.
Hence, the dynamic property of the satellites should must
be known. Consequently, we need to track the position and
velocity of the satellite at every moment.
The LEO satellites have several advantages over MEO
satellites like high bandwidth, small power requirement
to transmit the data, small delay in time between transmission and receiving of data etc. Subsequently, the LEO
satellites are preferred for most of the practical uses.
Hence, our work is concentrated around the LEO satellites.
The LEO satellites rotate in an orbit having altitude
in between 160 to 2000 kilometers with an orbital period
90 to 127 minutes depending on altitude. The altitude
of these satellites lies in thermosphere (80-500 km) or
Observer
meridian
plane
Spacecraft
(1)
.
yk = (xk ) + vk
(2)
Observer
position
(3)
2) Update step: In this step, the posterior probability
density function is evaluated using Bayes rule,
p(xk |y1:k ) =
(4)
Equatorial
plane
A. Process Model
The process model includes the inertial position and
velocity components of satellite in I, J and K directions
respectively. Let us consider, the position components in
respective directions are x, y and z, then the process model
is X = [x, y, z, x,
y,
z]
T
Let us consider, re and rs are the position vectors of
earth and satellite respectively in inertial frame. Then, the
relative position vector between earth and satellite is r=rs es . If Me and ms are the masses of earth and satellite
respectively, then using the Newtons Law we get
ms rs =
GMe ms r
r2
r
(6)
and,
GMe ms r
(7)
r2
r
where G is the gravitational constant and r =
p
x2 + y 2 + z 2 is the magnitude of r provided x, y and
z are the position vectors in I, J and K directions.
The equations (6) and (7) can further be written as
Me re =
rs =
GMe r
r2 r
(8)
and,
Gms r
(9)
r2 r
Subtracting (9) from (8) and neglecting the mass of
satellite with respect to earth, we get
re =
r =
GMe r
= 3r
2
r r
r
(10)
r = 3 r + aD + w.
(11)
r
The atmospheric drag can be given as
1 Cd A
d rel vrel
(12)
2 ms
where Cd is drag coefficient, A is cross-sectional area of
satellite, d is atmospheric density, vrel is relative velocity
between earth atmosphere and satellite and rel is the
magnitude of vrel .
1) Calculation of relative velocity: To calculate vrel , it is
assumed that the earth atmosphere rotates with the same
angular velocity as earth. Also, the earth is centered at
origin of inertial coordinate system, hence r can be used
to represent the inertial velocity vector of satellite. Under
these assumptions
aD =
vrel = r r
(13)
u
cos() 0 sin()
e =
0
1
0
n
sin() 0 cos()
cos() sin() 0
x ER cos()cos()
sin() cos() 0 y ER cos()sin() ,
0
0
1
z ER sin()
where is latitude and is longitude angle of the observer.
IV. Cubature quadrature Kalman filter
A brief discussion about the cubature quadrature
Kalman filter (CQKF) is provided here, while for more
elaborated formulation readers are requested to read [8].
We are not describing the UKF in this paper because
now a days the algorithm is extensively used and easily
available in literature [11].
For estimating the states, we need to approximate
the integrals (3) and (4). Unavailability of closed form
solutions of the above mentioned integral forces us to
adopt several numerical integration techniques [12]. The
estimation accuracy is proportional to the accuracy of
approximation of the integrals and hence depends on the
numerical method adopted for approximation.
A. Cubature rule of integration
The intractable integral which arises while estimating
the nonlinear dynamic system is decomposed into the
surface integral and the radial integral. The third-degree
spherical cubature rule fulfill the purpose of approximating
the surface integral, while the arbitrary order GaussLaguerre quadrature rule fulfills the same purpose for the
radial integral.
Let us assume an arbitrary function f (X) i.e. X <n ,
then the integral
Z
T 1
1
f (X)e(1/2)(X) (X) dX
I(f ) = p
| | (2)n <n
(15)
can be decomposed to two integrals [8], [13] as given below
Z Z
2
1
I(f ) = p
[f (CrZ + )d(Z)] rn1 er /2 dr
n
(2) r=0 Un
(16)
where X = CrZ + with and = CC T denoting the
mean and covariance matrix, k Z k= 1 and Un stands for
the surface of unit hyper-sphere.
For calculating the integral given in equation (16), one
needs to calculate the integral
Z
f (CrZ + )d(Z)
(17)
Un
(18)
Un
2n
2 n X
f (r[u]i )
f (rZ)ds(Z)
2n(n/2) i=1
Un
(19)
where [u]i (i = 1, 2, ..., 2n) are the points where a hypersphere with unity radius meets with the axes. It is also
called as cubature points. For a system with mean and
covariance = CC T , the cubature points are transformed
as (C[u]i + ).
B. Gauss-Laguerre quadrature rule
It states that the integral
Z
f () e d
(20)
=0
V. Simulation results
In this section, the LEO satellite tracking problem
formulated in section-III is simulated using CQKF, CKF
and UKF. During simulation, the CQKF uses second order
Chebyshev Laguerre polynomial equation (n0 = 2) and the
tuning parameter kappa is fixed at 3 for UKF.
As the mass body is earth in our context, hence the
gravitational parameter () and atmospheric density (d )
are defined accordingly. For earth these parameters are
given as = 398600.4418 106 km3 sec2 and d =
1.2922kgm3 . For the satellite considered in this paper,
the drag coefficient Cd = .0022 and A/ms = 0.02 m2 kg 1 .
The observer position on earth could be defined with
latitude angle 10.749 and longitude angle 70.5983 .
The
process
and
measurement
noises
are
considered
to
be
normally
distributed
with
zero
mean
and
covariances
Q
=
diag([0 0 0 1017 (km/s)2 1017 (km/s)2 1017 (km/s)2 ])
and R = diag([0.0415 , 0.0415 0.07252 km2 ]) respectively.
The true initial state is considered as x = [rT0 r T0 ]T ,
where
r0 = [6949.6 1045.73 164.92] km
(25)
and, r 0 = [0.9 5.69 4.84] km/s
0 =
The filter is initialized with the initial estimate x
[rT0 r T0 ]T , where
r0 = [7252 1358.4 383.9] km
and, r 0 = [0.613 6 5.13] km/s
(26)
L
n0 ()
0
dn
e n0 +n e = 0
d
n0
= (1)
i0 =
n0 !( + n0 + 1)
2
i0 [L
n0 (i0 )]
(21)
(22)
f () e
d
=0
n
X
i0 f (i0 )
(23)
i0 =1
2n X
n0
X
i0 f ( 2i0 [u]i ) .
(24)
I(f )
2n(n/2) i=1 0
i =1
RMSE of positio
position(km)
References
0.9
20
60
100
140
180
220
260
300
Time(sec)
0.01
UKF
CKF
CQKF
0.001
0.0001
40
80
120
160
200
240
280
Time(sec)