Unbiased Converted Measurements For Tracking (1998)
Unbiased Converted Measurements For Tracking (1998)
Unbiased Converted Measurements For Tracking (1998)
CORRESPONDENCE 1023
Authorized licensed use limited to: University of Pretoria. Downloaded on February 29,2024 at 13:09:27 UTC from IEEE Xplore. Restrictions apply.
several approaches to approximate the true covariance. B. 3D Case
The one presented in [2]replaces Rt by the average
covariance R, = E[R, I r,,P,]. The average covariance The unbiased conversion for the 3D case can be
R, will be, in general, larger than the true covariance obtained similarly. The observations in the 3D case
Rt conditioned on the true range and bearing [2]. are
Another procedure (which is simpler, but also more
accurate, as will be seen later) is given here. Namely, r, = r + ur; 0, = p + up; E, =E + uE (8)
the squared error is averaged directly conditioned on
where Y, 0, and E are the true range, bearing, and
the observations. In this new approach, the elements
elevation angles of the target, v,., vp, and vEare
of the covariance, denoted as R,, are given by
independent zero-mean noise with the standard
deviation n,., ap, and a,, respectively. The unbiased
conversion for this case is given, similarly to the 2D
case, as
xk = Xp'X;'r,cosp,cosE, (94
= E{Xp2r:,cos2P,,, + (r,,,- uT)' cos2(P,, - v0) yk = Xp'X;'r,sinP,cosE, (9b)
- 2Xp1rT,,(r,,t +
- v ~ ) c o s ~ , , [ c o s ~ , , c o s vsin@,,&
p sinvml}
zk = X;'Y, sin&,. (9c)
= Xp2r: cos2P,,, + (r: + a
;
)( 1 /2)E[1+ cos 2(Pm- vp)l
The elements of the covariance R, for the unbiased
- 2Xj ' r i cos2PmE[cosup] converted measurements (9) are given by
1024 IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 34, NO. 3 JULY 1998
Authorized licensed use limited to: University of Pretoria. Downloaded on February 29,2024 at 13:09:27 UTC from IEEE Xplore. Restrictions apply.
The averaging procedure of the covariance, measurements as
previously presented in [2], approximates the true
covariance Rt by the averaged R, = E[R, I rm,Pm,&,J.
The consistency of R, for the unbiased conversion (9)
is checked later via simulations. After some algebraic
manipulations, the elements of R , are given by The converted measurements (14) are slightly biased,
but acceptable for practical noise levels. Taking the Assumption
RL1 = $Xp2AF2(ri+ 2a,2)(1 + (Ab)2cos2P,) approximation that 2 M 1 + x, the coefficients in (14)
are approximately equal to those in (13),
x (1 + (Ay cos 2 4
(1 + -e-";) M 1 + a i / 2 M eu:l2. (15)
- $ (r; + a;)( 1 + Ab cos 2&)( 1 + A: cos 2&,)
= J~cos2u;?adu
1
It has been shown above that the exact debiasing In this section the statistical consistency of the
procedure depends on the compensation factors unbiased conversion given above is tested. The details
A, and Ab (as well as A, and A: for the 3D case). of the statistical consistency evaluation procedure
These factors can be determined from the distribution can be found in [2 or 31. It should be noted that this
of the bearing (and elevation) noise. The bearing statistical consistency check is based on a theorem
measurement error is usually assumed as zero-mean in statistics that the sum of independent zero-mean
Gaussian noise, that is, up N(O,ao). In this case the
N unity-variance Gaussian variables squared has a
above factors are given by [2] chi-square distribution. So the evaluation here is
approximate since the converted measurement errors Limitation
are not Gaussian in general. It is interesting also
to test the statistical consistency of the debiased
conversion [2] under non-Gaussian bearing noise, as
well as the two different averaging approaches for
Thus the unbiased converted measurements in the
the covariance. The number of Monte Carlo runs in
Gaussian noise case are
each simulation is 3000 and the plot indicates the
xk = ecr02 12r,cosp,,,; yk = e0i12r,sinp,. (13) chi-square 0.99 probability bounds.
The difference between the above unbiased conversion A, Consistency Check for 2D Case
in the Gaussian noise case and the debiased
conversion [2] is small only if up is not very large. The simulation here is similar to that in [2]. The
The debiased conversion [2] gives the converted true target position is at 10,000 m with azimuth
CORRESPONDENCE 1025
Authorized licensed use limited to: University of Pretoria. Downloaded on February 29,2024 at 13:09:27 UTC from IEEE Xplore. Restrictions apply.
2.2
0 8 ,
-unbiased 2.2
I . ,
2. I 2.1
v)
w Y
2 2 =ai 2
d
z 1.9
2
I .9
1 ' : : : : : I 1.8
1.81 : : : : : 1
0 5 10 20 25 30
15 0 2 4 6 8 1 0
Uniform Distribution Bound (deg.)
Standard Deviation (deg.)
Fig. l(c). Average NES for unbiased conversion under uniformly
Fig. l(a). Average NES for unbiased conversion under Gaussian
distributed noise.
noise.
0 5 10 15 20 25 30 0 2 4 6 8 1 0
Standard Deviation (deg.) Uniform Distribution Bound (deg.)
Fig. l(b). Average NES for debiased conversion under Gaussian Fig. l(d). Average NES for debiased conversion under uniformly
noise. distributed noise.
45 deg and the standard deviation of range error is the approach of [2] has shown good consistency
What the
experiment 100 m. The bearing error varied in standard deviation in the 2D case for practical noise levels. In this
entailed and distribution, First the bearing error is assumed subsection the consistency of the two covariances
to be zero-mean Gaussian noise. The results of the for the unbiased conversion (9) is checked. The true
consistency check in this case for the unbiased and the target position in a radar-centered coordinate system
debiased conversion are shown in Figs. l(a) and l(b), is (50 km, 50 km, 1 km). The standard deviation
respectively. The debiased conversion performs well of range error is 50 m and the measurement errors
for practical noise levels and becomes inconsistent in bearing and elevation angles are assumed to be
Limitation zero-mean Gaussian noise with the same standard
only for very large noise levels owing to a small bias
in (14). To show the dependence of the debiasing deviation a.The value of a is varied to check the Variable used to
procedure on the noise distribution, the bearing noise consistency of the unbiased conversion for the 3D measure success
is assumed to be the uniformly distributed within case. As can be seen from Figs. 2(a) and 2(b), the of model
[-a,a], with the standard deviation ap = a / d . The covariance Rp shows good consistency with the
How performance
conversion errors, while R, does not. baseline is derived
compensation factors for the unbiased conversion are
from experimental
given by (16a) and (16b). The bound a is varied to results
check the statistical consistency of the unbiased and IV. CONCLUSIONS
the debiased conversions. The debiased conversion
is shown in Fig. l(c) to be inconsistent when the It has been shown that the exact compensation
uniformly distributed bearing error bound exceeds procedure for the bias in the classical polar or
spherical to Cartesian conversion is multiplicative
2.0 deg, while the unbiased shows in Fig. l(d) good
in its nature and depends on the statistics of the
consistency, even for very large noise levels.
cosine of the angle measurement noise, which can be
determined from the noise distribution. An unbiased
B. Consistency Check for 3D Case Under Gaussian conversion and a new approach to obtain of the
Noise covariance of the conversion have been presented in
this paper. The previously offered debiased procedure
There are two averaging approaches for the does not completely eliminate the bias and shows a
Variables used covariance of the unbiased conversion for the sensitivity to some non-Gaussian noises. The new Conclusion
to measure 3D case, which give the expressions R, and Rp, unbiased conversion and the new covariance show made
success of respectively. The average covariance R, given by
model
good consistency and robustness.
1026 IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 34, NO. 3 JULY 1998
Authorized licensed use limited to: University of Pretoria. Downloaded on February 29,2024 at 13:09:27 UTC from IEEE Xplore. Restrictions apply.
3.5 I
Comments on "Properties and Performance of
Extended Target Motion Analysis"
CORRESPONDENCE 1027
Authorized licensed use limited to: University of Pretoria. Downloaded on February 29,2024 at 13:09:27 UTC from IEEE Xplore. Restrictions apply.