Professional Documents
Culture Documents
An Angle-Only Tracking Filter in Modified Spherical Coordinates by David V. Stallard. Sc.D.
An Angle-Only Tracking Filter in Modified Spherical Coordinates by David V. Stallard. Sc.D.
1.0 Abstract
Among previous approaches to tracking with angle measure- For this application, most papers1-' have utilized a four-
ments only, the IBM filter for planar tracking appears to be superior state Cartesian formulation with two relative position and two rela-
because of its use of modified polar coordinates (MPC), which re- tive velocity components, sometimes with a "pseudolinearization"
duce problems with observability, range bias and covariance ill- of the bearing measurement. Typically, difficulty was experienced
conditioning. Here, this approach is extended to three dimensions with instability andtor bias in the filter, apparently because of the
by the use of modified spherical coordinates (MSC). In the MSC fil- choice of the coordinate system.
ter, the six state variables are two angles, their derivatives, inverse
range and range-rate over range, which are transformable into Carte- The discovery of Modified Polar Coordinates (MPC) and
sian position and velocity. This filter successfully estimates the state their application to planar underwater tracking with sonar bearing
of a nonmaneuvering aircraft with noisy angle measurements from measurements appear to have been made at IBMB. Figure 1 shows
a moderately weaving interceptor aircraft. Questions remain as to the planar relative kinematics in a slightly different coordinate sys-
design refinements, desirable trajectories of the measuring vehicle, tem, in which only $ and the motion of the "ownship" is measured.
extension to a maneuvering target, and limitations of the MSC ap- The state vector in MPC is:
proach.
2.0 Introduction
The problem of determining the position of an object with-
out using measurements of range to it is familiar to surveyors and
astronomers. In the surveying problem, measurements of azimuth It is noteworthy that MPC uses l/r rather than r. The four scalar
and elevation angles to a point on the object can be made from each plant equations may be found from Equations (A-3), (A-18), (A-21)
end of a baseline of known length, and the position of the object can and (A-22) herein, with 8 and the target accelerations equal to 0:
be found by trigonometric calculations.
2.1 Genesis of Problem
There are military situations in which it is desirable to esti-
mate the position, velocity and perhaps acceleration of a target from
measurements of angle but not range. A well-known example is the
determination by a submarine of planar position and velocity of a
ship from passive sonar measurements, because the submarine com-
mander does not want to reveal his presence by pinging.
In air warfare, a fighter defending against a raid may wish
to launch a missile against a jammer at unknown range, but should
not do so unless the jammer's position and velocity can be estimat-
ed.
The simplest guidance of interceptor missiles relies on the
proportional navigation (PN) guidance law, in which the accelera-
tion command to the missile pitch autopilot is:
where A& and A& are ownship accelerations along and perpendicu-
lar to the range vector.
Reference 8 showed that the first three elements of x in
Equation (2) are observable from the ownship while it is on a
straight course with a nonzero LOS angular rate. See Figure 2, from
where is the estimated angular rate of the line of sight (LOS), V, which it is apparent that both the true target and the false target at
is the closing velocity, N' is a dimensionless effective navigation ra- half the true range have the same angular rate when viewed by the
tio (typically 3.5) and OH is the gimbal ande. The LOS angular rate constant-velocity ownship, which therefore cannot determine the
is easy to estimate approximately, and V, can be measured by a range.
doppler radar. Here, information on closing velocity but not relative The ownship must maneuver in order to determine llr,
range is necessary. which is observable8 by reason of being multiplied by A& in Equa-
A more recent problem is estimation of target position, ve- tion (4). It seems at least intuitively clear from Figure 2 that a see
locity and acceleration in three dimensions from angle measure- ond straight leg for the ownship would remove the range uncertain-
ments only, either with a passive IR receiver or a jammed radar re- ty. For some reason, Reference 8 used two straight legs of ownship
ceiver on a missile, in order to utilize optimal guidance. motion meeting at a comer, but it appears that a more realistic join-
ing circular arc would also have served in estimating l/r. The latter
2.2 quantity instead of r is used as the fourth state because Equations
There appear to have been two main categories of effort in (4) through (6) are linear in l/r and so the incremental plant matrix
angle-only tracking filters: (1) Work on the planar tracking of a non- of the vector plant equation, the transition matrix and the covari-
maneuvering ship; and (2) work on three-dimensional target track- ance propagation are less dependent on the initially unknown range.
ing in Cartesian coordinates by a highly maneuvering interceptor A major virtue of the MPC approach is that "it decouples
missile. the relatively accurate crossrange states from the downrange states,
* Associate Fellow, AIAA
Copyright D American Institute of Aeronautics and
Astronautics, Inc., 1987. All rights resewed.
which prevents covariance matrix ill-~onditioning."~ Also, "unlike Reference 13 examined various formulations of the nonlin-
Cartesian or conventional polar coordinates, it prevents false rang- ear Kalman filter and chose the simplest Extended Kalman Filter
ing and premature covariance collapse on a nonmaneuve~gown- (EKF)formulation. The state variables were relative X, Y, Z, their
ship leg." Indeed, this writer has experienced difficulty in estimating derivatives and absolute target acceleration in X, Y, Z, which recalls
range in a three-dimensional, missile-borne, tracking filter in Carte- somewhat the Bryson-HoZ0formulation of the interceptor-missile
sian coordinates. problem in two dimensions.
Despite its apparent superiority for this application, MPC Reference 14 took the interesting approach of deriving a
appears to have been used in relatively few papersg-lo on planar, "maximum information trajectory" with zero expected miss distance
bearings-only tracking by the underwater tracking community. The and a performance index which is the trace of the observability ma-
References list two other planar tracking papers 11-12 for complete- trix. The target had zero acceleration. The optimal missile trajectory
ness. was roughly sinusoidal in the horizontal and vertical planes, with
peak accelerations of about 24 g and 59 g respectively, based on cal-
culations from the hard-to-read figures.
Presumably these accelerations were within the constraints
of the postulated, high-performance, bank-to-turn missile. These ac-
celeration levels may explain while ill-conditioning of the covariance
matrix was not reported in References 13-19. Unfortunately, the
foregoing acceleration levels are not available in the majority of in-
terceptor missile applications.
2.3 Statement of the Problem
The problem is to develop a filter to estimate the three-
dimensional position and velocity of a constant-velocity aircraft tar-
get, using only angle measurements at 2.5 sec intervals in Track-
While-Scan (TWS) mode from an interceptor aircraft that maneu-
vers with an acceleration not exceeding 2 g.
2.4 Contribution of this P a ~ e r
.4 solution has been found for the problem, with a six-state
filter in MSC coordinates which has good accuracy in the presence
of noisy measurements and an acceptable initial transient.
An approach to the further problem of estimating three-
dimensional position, velocity cmd acceleration is presented, with a
nine-state filter in MSC coordinates.
3.0 Recursive Tracking Filters in Modified S~hericd
Coordinates
-
Figwe 1 Geometry of Planar Tracking in MPC The approach here is to extend IBM's work8 in planar track-
ing of a nonmaneuvering target in MPC to the three-dimensional
tracking of an essentially nonmaneuvering airborne target in MSC,
so as to solve the problem in Subsection 2.3.
Figure 3 shows the missile at the origin of a moving, non-
rotating Cartesian coordinate system, with X positive North, Y posi-
tive West and Z positive up. The target is at range r, with a compo-
nent rh in the X-Y plane making an angle $ with the X axis. The de-
pression angle 8 of the target is measured downward from rh to r.
The Cartesian velocities and accelerations of the missile target are
omitted from Figure 3 for simplicity.
The angles $ and 8 are chosen as two of the state variables.
Keeping in mind the imporlance of LOS angular rates in guidance,
8, shown as a vector perpendicular to the rh-r plane and w, perpen-
dicular to both the 8 vector and r, are also chosen as states. The iner-
tial target acceleration A$ is along r, while A$ and & are respec-
tively parallel to the 0 and w vectors.
3.1 Six-State Filter
3.1.1 State Vector. Plant Eauation and Observability
A six-state filter has been developed with the following state
OWNSH IP
vector in MSC:
-
Figure 2 Relative Come of Target for Constant Ownship Velocity
Z (VERTICAL)
t cd (1- r PLANE) 4
r =
4 ~4 A
X c o s $ + Y sin$
0 4 A 4 A
r = r cost9-Z sine
I1
( X - Y PLANE)
4 A h
i= - ( Z cost9 + E
11
sin01
" A 3
+ = ( Y c o s $ + X sin$)
Figure 3 - Basic Coordinates for MSC Filters where the measurement matrix H is:
where the predicted angles were found from Equations (16) and
(17).
It is important that the covariance propagation and Kalman-
gain computations be done in MSC, as previously indicated. Appen-
dix B summarizes these computations. In an actual application,
these would be done before the measurement and residual computa-
tion.
The updated state estimate in MSC is found as the sum of
the predicted estimate plus the Kalman gain matrix times the resid-
ual:
This completes the operations for the i-th estimate. Prediction of the
(i+l)th estimate then proceeds as described at the beginning of this
subsection.
3.1.3 Simulation Results
Figure 5 - True and Estimated 8 versus Time
An interceptor aircraft, a bomber target and the six-state fil-
ter have been simulated in ACSLz2. The interceptor flies initially
North (parallel to the inertial X axis) at 242.6 d s e c (Mach 0.8) at
an altitude of 10.668 km (35 kft), while the target flies South with
a velocity of 242.6 d s e c at an altitude of 9.144 km (30 kft), without
any maneuver.
At an initial X separation of 92.659 km (50 nm), the inter-
ceptor detects the jamming target and begins a weaving maneuver
with a turn acceleration of 2 g horizontally to 45 deg West of due
North, followed by a short straight leg, a 90 deg turn to the right, a
30 sec straight leg, a 90 deg left turn, etc. See the trajectory in Figure
4.
\
u
INT
TGT
i
100 200 ZOO 150 100 -50 0
-2500-
I
I
A-1 Plant Eauations for Six-State Filter case cos$, cos9 sin$, -sin9
The velocity w is given by:
cosg (A-1 3)
1
s i n e cos$ s i n 9 sin$ c oo s 0
ya = - 2 +B, - r w 2 t a n e - YO
(A-1 4)
Atz ~mdz
- d+
dt
= - W
case
=
In order to find the derivatives of e and w, it is useful to ex- which is the derivative of e and is therefore one row of the vector
press the relative coordinates X, Y and Z in terms of r, )I and 13 and plant equation.
then to find their first and second derivatives. First:
Using the second row of Equation (A-13) and Equations
A-10 through A-12, it is found that:
X = r cos0 cos$ (A-4)
Z = - r sin0 (A-6)
The relative Cartesian velocities are: Differentiating Equation (A-1) yields the auxiliary equation:
and the covariance of the error after the i-th measurement is:
2
@ 4 , 1 ( t l,O) = w 6 sec 6 t '
(8-23)
Where the elements of iP apply to the measurement interval ti-l to
(B-17) ti. From experience with other applications of the Singer model, it
The integral of this matrix is: appears unnecessary to compute and utilize the other elements of
Qi. R. J. Fitzgerald of Raytheon has pointed out that this is mathe-
matically equivalent to assuming an impulsive change in the second,
fourth and sixth diagonal elements of Mili.1 just prior to the mea-
surement at t+
Assuming that the target is an incoming bomber, the pro-
cess-noise parameters are chosen as:
Let the maneuver time constant 7, be far less than the measurement
time T, (which is 2.5 sec in subsection 3.1.2), let there be a fixed
value of the integral in Equation (B-1 S), and let the correlation func-
- tion by approximated by an impulse matrix:
where the rms measurement noise for this jamming target is as-
sumed, admittedly optimistically, to be 1 mrad. The two angular-
rate elements are: