Professional Documents
Culture Documents
15 Art Error Radar With Ads
15 Art Error Radar With Ads
org
Published in IET Radar, Sonar and Navigation
Received on 9th July 2012
Revised on 11th November 2012
Accepted on 27th December 2012
doi: 10.1049/iet-rsn.2012.0199
ISSN 1751-8784
Abstract: Systematic error of radar cannot be accurately determined solely with the use of an automatic dependent surveillance
broadcasting (ADS-B) device. This is largely, because of the device’s inability to obtain the exact transmitting time of ADS-B
data packets. In order to study and establish an efficient and accurate error registration algorithm for radars based on the ADS-B
system, a joint systematic error estimation model of the radar and ADS-B was built. The authors showed that when receiving time
is known, the time difference between radar measuring time and ADS-B data transmitting time is consistently biased, a feature
that can be used to calculate the systematic error of the ADS-B receiving system. Then, they established an error estimation model
based on the joint systematic error of the radar and ADS-B. The generalised least squares estimation method was adopted to solve
the model. Finally, data samples verified the accuracy and effectiveness of the error registration algorithm.
IET Radar Sonar Navig., 2013, Vol. 7, Iss. 4, pp. 361–370 361
doi: 10.1049/iet-rsn.2012.0199 & The Institution of Engineering and Technology 2013
Authorized licensed use limited to: Universidad de los Andes. Downloaded on April 02,2020 at 04:45:29 UTC from IEEE Xplore. Restrictions apply.
www.ietdl.org
practice. The worst case for time delay within a transmitted
position message occurs when a position update arrives
with maximum delay, that is, just under the 100 ms limit
outlined in the requirements. ADS-B airborne position
messages may implement estimation techniques such as
alpha–beta trackers or Kalman filters to satisfy the intent of
the 100 ms position requirement [18]. So we can ignore this
time delay for the sake of simplicity.
It can be proved that the time lag Δt is constant with
respect to radar measurement period and ADS-B message
transmitting period. In practice, the ADS-B messages are
not periodically transmitted, but this does not affect its use
in correct periodic radar measurement. The ADS-B message
can be easily reconstructed to transmit with any wanted
period through the smoothing and interpolation method
[19]. Owing to the high accuracy of GPS information and
the assumption of uniform rectilinear motion of airliner,
Fig. 1 ADS-B receiving and processing system errors introduced by smoothing and interpolation method
can be ignored. So we have the following analyses.
Let the transmitting period of a ADS-B data packet be TA,
We can adjust the time at which ADS-B receives the the measurement period of radar be TR, the transmitting time
message alignment to the time at which the airliner is of the ADS-B data packet be t1A , t2A , t3A , . . . , the measurement
detected by radar with the device shown in Fig. 1. Fig. 2 time of the radar be t1R , t2R , t3R , . . . , the time lag be
shows that when the radar and the ADS-B receiving system Dt1 = t1R − t1A and Dt2 = t2R − t2A . Their relationship is
are located in the same place, after the ADS-B transmitting shown in Fig. 3.
subsystem transmits a position message, the airliner will Then we have
still fly some time before detected by radar. This means that
there is a time lag Δt between the transmitting time of
ADS-B data packet tA and the measuring time of radar tR. TA − Dt1 + Dt2 = TR (1)
In fact, time delay occurs between the time at which the
target transmits the ADS-B information and the time at When TA = TR, we have
which the avionics calculate the position. According to
industry standard, the ADS-B transmitting subsystem Dt1 = Dt2 (2)
attempts to provide the position data accurate to within a
specified time tolerance of the time of transmission in the Equation (2) indicates that when the transmitting period of an
non-GPS/global navigation satellite system (GNSS) Time ADS-B data packet is equal to the radar measurement period,
Mark Coupled Case. Moreover, in the GPS/GNSS Time time lag between the transmitting time of ADS-B data packet
Mark coupled case, the delay can be well compensated for and the radar measurement time is a constant that, in turn, can
by advancing the position in the direction of travel. It has be seen as the time bias of the ADS-B receiving system. So
been found that the ADS-B transmitting subsystem provides the time lag Δt can be seen as the systematic error of the
position in the non-GPS/GNSS Time Mark coupled case in ADS-B receiving system. This was combined with range
362 IET Radar Sonar Navig., 2013, Vol. 7, Iss. 4, pp. 361–370
& The Institution of Engineering and Technology 2013 doi: 10.1049/iet-rsn.2012.0199
Authorized licensed use limited to: Universidad de los Andes. Downloaded on April 02,2020 at 04:45:29 UTC from IEEE Xplore. Restrictions apply.
www.ietdl.org
where TR is the rotation matrix
⎡ ⎤
−sin ls −sin Ls cos ls cos Ls cos ls
T R = ⎣ cos ls −sin Ls sin ls cos Ls sin ls ⎦ (7)
0 cos Ls sin Ls
Fig. 3 Relationship between radar measuring time and ADS-B 3.2 Model building
data receiving time
Fig. 4 shows the geometrical relationship among the ADS-B
data packet transmitting location, the radar measurement and
bias, azimuth bias and elevation bias of radar as a joint the true position of the airliner at time k. (xEA, yEA, zEA) is
systematic error to build an error registration model in this the location of the airliner when the ADS-B data packet
paper. transmitted at time k − ΔtA, which is also the airliner
location when the ADS-B receiving system received the
3 Error estimation model and its solution data packet at time k. (xETr, yETr, zETr) is the true position of
airliner at time k. (xERm, yERm, zERm) is the radar
This section will establish a joint systematic error estimation measurement at time k. ΔtA is the time bias of ADS-B
model. GPS information in the ADS-B data packet is based receiving system. The airliner is assumed to be uniform
off of geographic coordinates, whereas the radar rectilinear motion in this paper.
measurement system is based off of a radar-centred local In Fig. 4, after transmitting the ADS-B data packet, the
coordinate system. Thus, coordinate transformations are airliner will fly the duration ΔtA before it is measured by
needed first in order to produce a unified coordinate system. the radar at position (xETr, yETr, zETr). Moreover, the radar
In this paper, an Earth-centred Earth-fixed (ECEF) measurement is (xERm, yERm, zERm), which deviates from
coordinate system is used as the unified coordinate system. the true position because of systematic error and the
measurement noise of the radar. So the ECEF coordinate of
3.1 Coordinate transformation the airliner obtained by the ADS-B receiving system at
time k is
Let the accurate position of an airliner obtained by GPS in an
ADS-B system be (LA, λA, HA), where LA is the latitude, λA is ⎡ ⎤ ⎡ ⎤
xETr (k) xEA (k )
the longitude and HA is the height above sea level. Its ECEF ⎢ ⎥ ⎢ ⎥
coordinate (xEA, yEA, zEA) is ⎣ yETr (k) ⎦ = ⎣ yEA (k ) ⎦ + T A (k )
⎧ zETr (k) zEA (k )
⎨ xEA = (C + HA ) cos LA cos lA ⎡ ⎤
vA (k )DtA cos 1A (k) sin uA (k )
y = (C + HA ) cos LA sin lA (3) ⎢ ⎥
⎩ EA × ⎣ vA (k )DtA cos 1A (k) cos uA (k ) ⎦
zEA = [C(1 − e2 ) + HA ] sin LA
vA (k )DtA sin 1A (k)
where C is defined as ⎡ ⎤ ⎡ ⎤
xEA (k) DxA (k)
⎢ ⎥ ⎢ ⎥
Eq = ⎣ yEA (k) ⎦ + T A (k ) × ⎣ DyA (k) ⎦ (8)
C= 1/2 (4)
1 − e2 sin LA 2 zEA (k) DzA (k)
IET Radar Sonar Navig., 2013, Vol. 7, Iss. 4, pp. 361–370 363
doi: 10.1049/iet-rsn.2012.0199 & The Institution of Engineering and Technology 2013
Authorized licensed use limited to: Universidad de los Andes. Downloaded on April 02,2020 at 04:45:29 UTC from IEEE Xplore. Restrictions apply.
www.ietdl.org
where (see (9)) According to (6), its ECEF coordinate is
⎡ ⎤
−sin lA (k) −sin LA (k) cos lA (k) cos LA (k) cos lA (k)
T A (k) = ⎣ cos lA (k) −sin LA (k) sin lA (k) cos LA (k) sin lA (k) ⎦ (9)
0 cos LA (k) sin LA (k)
⎡ ⎤
∂(Dx(k)) ∂(Dx(k)) ∂(Dx(k))
⎢ ∂r′′ (k) ∂u′′R (k) ∂h′′R (k) ⎥
⎢ R ⎥
′ ′ ⎢⎢ ∂(Dy(k)) ∂(Dy(k)) ∂(Dy(k)) ⎥ ⎥
∇C′′ f Ck , b = ⎢ ′′ ⎥ = Fk (15)
k ⎢ ∂rR (k) ∂u′′R (k) ∂h′′R (k) ⎥
⎢ ⎥
⎣ ∂(Dz(k)) ∂(Dz(k)) ∂(Dz(k)) ⎦
∂rR′′ (k) ∂u′′R (k) ∂h′′R (k) C′′ =C′ ,b=b′
k k
⎡ ⎤
∂(Dx(k)) ∂(Dx(k)) ∂(Dx(k)) ∂(Dx(k))
⎢ ∂Dr ∂DuR ∂DhR ∂DtA ⎥
⎢ R ⎥
′ ′ ⎢⎢ ∂(Dy(k)) ∂(Dy(k)) ∂(Dy(k)) ∂(Dy(k)) ⎥
⎥
∇b f Ck , b = ⎢ ⎥ = Gk (16)
⎢ ∂DrR ∂DuR ∂DhR ∂DtA ⎥
⎢ ⎥
⎣ ∂(Dz(k)) ∂(Dz(k)) ∂(Dz(k)) ∂(Dz(k)) ⎦
∂DrR ∂DuR ∂DhR ∂DtA C′′ =C′ ,b=b′
k k
364 IET Radar Sonar Navig., 2013, Vol. 7, Iss. 4, pp. 361–370
& The Institution of Engineering and Technology 2013 doi: 10.1049/iet-rsn.2012.0199
Authorized licensed use limited to: Universidad de los Andes. Downloaded on April 02,2020 at 04:45:29 UTC from IEEE Xplore. Restrictions apply.
www.ietdl.org
where ∂Ck = (C′′k − C′k ) denotes the measurement noise According to the GLS algorithm, the solution of (19) is
−1
∂Ck = Rr (k), ur (k), hr (k) (18) −1 −1
b̂ = X S X T
XT S Y (26)
j j
Fk∂Ψk can be seen as the error caused by measurement noise,
Gkβ as error caused by systematic error, and Gkβ′ − f(Ψ′k, β′)
as pseudo-measurement. With these assumptions, the joint The estimation covariance is
systematic error estimation model can be established as −1 −1
Xb + j = Y (19) cov(b̂) = X T S X (27)
j
⎡ ⎡ ⎤ ⎤
∂2 Dx ∂2 Dx
⎢
⎢
⎢ ′′
′ T
T ⎢ ∂C′′k 2 ∂C′′k ∂b ⎥ ′′
⎥ Ck − C
′
k ⎥
⎥
⎢ Ck − Ck b − b′ ⎢ 2 ⎥ ′ ⎥
⎢ ⎣ ∂ Dx ∂2 Dx ⎦ b − b ⎥
⎢ ⎥
⎢ ∂b∂C′′ ∂b2 ⎥
⎢ ⎡ 2 k ⎤ ⎥
⎢ ⎥
⎢ ∂ Dy ∂ Dy
2
⎥
⎢ ⎢ ∂C′′ 2 ′′ ⎥ ′′ ⎥
1⎢ ′′ T T ⎢ ∂Ck ∂b ⎥ Ck − C ′ ⎥
O= ⎢
⎢ Ck − C′k b − b′ ⎢ 2 k ⎥ ′
k ⎥
⎥ + On (30)
2⎢ ⎣ ∂ Dy ∂ Dy ⎦ b − b
2
⎥
⎢ ⎥
⎢ ∂b∂C′′ ∂b 2 ⎥
⎢ ⎡ 2 k ⎤ ⎥
⎢ ∂ Dz ∂ Dz
2 ⎥
⎢ ⎥
⎢ ⎢ ⎥
⎢
T ⎢ ∂C′′k 2 ∂C′′k ∂b ⎥ ′′
⎥ Ck − C
′
⎥
⎢ C′′k − C′k T b − b′ ⎢ 2 ⎥ ′
k
⎥
⎣ ⎣ ∂ Dz ∂2 Dz ⎦ b − b ⎦
∂b∂C′′k ∂b2 C′′ =C′k ,b=b′
IET Radar Sonar Navig., 2013, Vol. 7, Iss. 4, pp. 361–370 365
doi: 10.1049/iet-rsn.2012.0199 & The Institution of Engineering and Technology 2013
Authorized licensed use limited to: Universidad de los Andes. Downloaded on April 02,2020 at 04:45:29 UTC from IEEE Xplore. Restrictions apply.
www.ietdl.org
As shown in Fig. 5 the second-order term is very small Using the N × N matrix D, ξL can be whitened as a
compared with the first-order term. The linearisation normalised, independent and identically distributed noise
approximation error has little influence of Taylor expansion Dξ. So the generalised linear model can be transformed as
of (14), so the remainder term O of (14) can be omitted.
Based on the assumption above, estimation error (the Y ′ = DY L = DX L bL + DjL = X ′ bL + j′ (35)
second kind error) is taken into consideration. According to
Bar-Shalom et al. [20], the linear model is always defined as where ξ′ ∼ N(0, I). According to the least squares estimation,
the minimum variance and unbiased resolution of (35) is
Y L = X L bL + jL (32) −1 −1
b̂L = X ′ X ′ X ′ Y ′ = XQ−1 X XQ−1 Y
T T
(36)
where ξL is the random measurement noise, which is assumed
to be independent and identically distributed zero-mean
Gaussian noise with a variance s2n I. The least squares and its estimation covariance is
estimation is the minimum variance and unbiased −1
estimation for this linear model. cov b̂L = XQ−1 X (37)
If the covariance of ξL is not in scale with s2n I (I is an
identity matrix), such as that of (21), where ξL ∼ N(0, Q), So the GLS estimation (26) and (27), which have the same
this linear model is defined as a generalised linear model. form of (36) and (37), respectively, can be understood to be
In this case, the least square estimation is no longer the the minimum variance and unbiased estimation. In addition,
minimum variance and unbiased estimation [20]. In order to the covariance of this algorithm achieves the Cramér–Rao
obtain the minimum variance and unbiased estimation, the lower bound.
measurement noise should be independent and identically
distributed zero-mean Gaussian noise. Assuming Q is a 4 Real data test
positive definite matrix, applying Cholesky factorisation, it
can be written as In this section, real data of a 3D-phased array radar and
ADS-B will be used to test the performance of the model
Q = DT D (33) and algorithm. Fig. 6 shows the real data before error
registration. Data were collected during a 10 min duration
Thus of time. It can be seen that radar measurements apparently
deviate from GPS data. In practice, not all the ADS-B
−1 messages can be received at the same time. In the real data
E (Dj)(Dj)T = DQDT = DD−1 DT DT = I (34) test, a smoothing and interpolation method was used to
make ADS-B data periodically distributed in data preparation.
Using the model and algorithm in this paper, the estimation
Table 1 Parameters of radar results are shown in Table 2.
Then estimation results can be used to register the radar
Parameter Value measurements. Fig. 7 shows the registration results. Fig. 8
shows error between radar measurement and GPS after error
measurement noise of range 50 m
measurement noise of azimuth 0.0087 rad registration. Fig. 9 shows the standard deviation of the
measurement noise of elevation 0.0087 rad systematic error of the radar and ADS-B receiving system.
systematic error of range 1800 m According to Figs. 7 and 8, radar measurements after
systematic error of azimuth 0.0087 rad registration are generally very similar to the GPS data, a
systematic error of elevation 0.0175 rad
systematic error of time 0.5 s
fact that can be used to verify the performance and
application of the model. Fig. 9 shows that the algorithm is
well convergent. Owing to the measurement noise, there are
still some differences between radar measurement and GPS
data, which does not obviously affect the performance of
the algorithm and can be improved during the next data
processing stage, such as the application of the Kalman filter.
In order to add more support for the algorithm, another
batch of observations using the same radar were made,
shown in Figs. 10–12.
From Figs. 10–12, we can see that the registration
algorithm is useful in making the radar measurements
align with the GPS data. However, there are some
obvious differences between them at the low-altitude
366 IET Radar Sonar Navig., 2013, Vol. 7, Iss. 4, pp. 361–370
& The Institution of Engineering and Technology 2013 doi: 10.1049/iet-rsn.2012.0199
Authorized licensed use limited to: Universidad de los Andes. Downloaded on April 02,2020 at 04:45:29 UTC from IEEE Xplore. Restrictions apply.
www.ietdl.org
IET Radar Sonar Navig., 2013, Vol. 7, Iss. 4, pp. 361–370 367
doi: 10.1049/iet-rsn.2012.0199 & The Institution of Engineering and Technology 2013
Authorized licensed use limited to: Universidad de los Andes. Downloaded on April 02,2020 at 04:45:29 UTC from IEEE Xplore. Restrictions apply.
www.ietdl.org
flying stage, which may be the reason for the poor focused on systematic error estimation in two dimensions,
performance of the radar at low altitudes. In addition, in order to make a comparison between the algorithms in
measurement noise of the radar has also affected the [15–17] and our algorithm, the algorithm in [3] is used as a
registration during low-altitude flying of airliner, in which substitute where ‘Radar 1’ in [3] is substituted with ADS-B
the algorithm does not work well. (they are both assumed to be unbiased), and ‘Radar 2’ is
Now a performance comparison between our algorithm and substituted with the real radar in this paper. Only range
other algorithms which did not consider time bias of ADS-B bias, azimuth bias and elevation bias of the radar are
into consideration will be made. Besada et al. [15–17] did not considered. The compared algorithm is denoted as
take the time bias of ADS-B into account, which makes the algorithm without time bias (AWTB) here. Results using
estimation less accurate. As the algorithms in [15–17] the first batch of data are shown in Figs. 13–15.
368 IET Radar Sonar Navig., 2013, Vol. 7, Iss. 4, pp. 361–370
& The Institution of Engineering and Technology 2013 doi: 10.1049/iet-rsn.2012.0199
Authorized licensed use limited to: Universidad de los Andes. Downloaded on April 02,2020 at 04:45:29 UTC from IEEE Xplore. Restrictions apply.
www.ietdl.org
From Fig. 13, it can be seen that there is an obvious shift
between ADS-B data and radar measurement. Fig. 15
shows that estimation error of radar systematic error is
much higher than the algorithm presented in this paper. So
the comparison shows that the time bias of ADS-B is very
important to the registration of systematic errors of radar,
and the estimation model and algorithm in this paper is
very accurate and applicable.
In general, the algorithm presented in this paper has good
performance and practical applicability.
5 Conclusions
This paper studied the radar systematic error registration
problem using civilian airliners with the ADS-B system
Fig. 12 Error between radar measurement and GPS location after installed when the transmitting time of ADS-B data packets
error registration was not available. A joint systematic error estimation model
Fig. 13 ADS-B data and radar measurement after error registration with AWTB
IET Radar Sonar Navig., 2013, Vol. 7, Iss. 4, pp. 361–370 369
doi: 10.1049/iet-rsn.2012.0199 & The Institution of Engineering and Technology 2013
Authorized licensed use limited to: Universidad de los Andes. Downloaded on April 02,2020 at 04:45:29 UTC from IEEE Xplore. Restrictions apply.
www.ietdl.org
was built and its GLS solution was given. There are two major 7 Dong, Y.L., He, Y., Wang, G.H.: ‘A generalized least squares
positive results to be noted regarding our model. registration algorithm with Earth-centered Earth-fixed (ECEF)
coordinate system’. Third Int. Conf. Computational Electromagnetics
and Its Applications, 2004, pp. 79–84
1. The joint model presented in this paper can be accurately 8 Zhou, Y.F., Leung, H.: ‘An exact maximum likelihood registration
calibrated using real radar data, which verifies the accuracy algorithm for data fusion’, IEEE Trans. Signal Process., 1997, 45, (6),
and effectiveness of the estimation model. With the help of pp. 1560–1572
the ADS-B system, it is a very convenient method to 9 Dong, Y.L., He, Y., Wang, G.H.: ‘A modified exact maximum
likelihood registration algorithm’. Third Int. Conf. Computational
monitor systematic errors of radar in real time. Electromagnetics and Its Applications, 2004, pp. 85–88
2. In this paper, only two random experiments are provided. 10 Nabaa, N., Bishop, R.H.: ‘Solution to a multisensor tracking problem
Future work could research the possibility of fusing a large with sensor registration errors’, IEEE Trans. Aerosp. Electron. Syst.,
number data and the error field analysis based on a 1999, 35, (1), pp. 354–363
11 Besada, J.A., García, J., de Miguel, G.: ‘New approach to online optimal
large number of real data. An efficient fusion method for a estimation of multisensor biases’, IEE Proc., Radar Sonar Navig., 2004,
large number of data sets could be developed to make 151, (1), pp. 31–40
results more accurate and applicable. 12 Besada, J.A., de Miguel, G., Bernardos, A.M., Casar, J.R.: ‘Generic
multisensor multitarget bias estimation architecture’, IET Radar Sonar
Navig., 2012, 6, (5), pp. 365–378
6 Acknowledgment 13 Haim, K.H.T.S.: ‘Sensor registration using neural networks’, IEEE
Trans. Aerosp. Electron. Syst., 2000, 36, (1), pp. 85–101
This work was supported in part by the National Natural 14 Lian, F., Han, C., Liu, W.: ‘Joint spatial registration and multi-target
Science Foundation of China under Grants 61032001 and tracking using an extended probability hypothesis density filter’, IET
61102116. Radar Sonar Navig., 2011, 5, (4), pp. 441–448
15 Besada, J.A., Garcia, J., De Miguel, G., Jimenez, F.J., Gavin, G., Casar,
J.R.: ‘Data fusion algorithms based on radar and ADS measurements for
7 References ATC application’. Record IEEE 2000 Int. Radar Conf., 2000,
pp. 98–103
1 Burke, J.J.: ‘The SAGE Real time quality control function and its 16 Besada, J.A., Garcia, J., De Miguel, G., Casar, J.R., Gavin, G.: ‘ADS
interface with BUIC II/BUIC III’. MITRE Corporation Technical bias cancellation based on data fusion with radar measurements’. Proc.
Report, No. 308, 1996 Third Int. Conf. Information Fusion, 2000, vol. 2, pp. WEC5/
2 Leung, H., Blanchette, M.: ‘A least squares fusion of multiple radar 23–WEC5/30
data’. Proc. Radar 1994, Paris, 1994, pp. 364–369 17 Besada, J.A., Garcia, J., De Miguel, G.: ‘Radar bias correction based on
3 Stefano, F.A.F., Fulvio, G., Antonio, G., et al: ‘Least squares estimation GPS measurements for ATC applications’, IEE Proc., Radar Sonar
and Cramér–Rao type lower bounds for relative sensor registration Navig., 2002, 149, (3), pp. 137–144
process’, IEEE Trans. Signal Process., 2011, 59, (3), pp. 1075–1087 18 RTCA, Inc.: ‘Minimum Operational Performance Standards for 1090
4 Zhou, Y.F., Leung, H., Blachette, M.: ‘Sensor alignment with MHz Extended Squitter Automatic Dependent Surveillance –
earth-centered earth-fixed (ECEF) coordinate system’, IEEE Trans. Broadcast (ADS-B) and Traffic Information Services – Broadcast
Aerosp. Electron. Syst., 1999, 35, pp. 410–418 (TIS-B)’, 2009
5 Zheng, Z.W., Zhu, Y.S.: ‘New least squares registration algorithm for 19 Atkinson, K.: ‘An introduction to numerical analysis’ (Wiley,
data fusion’, IEEE Trans. Aerosp. Electron. Syst., 2004, 40, (4), New York, 1989, 2nd edn.)
pp. 1410–1416 20 Bar-Shalom, Y., Li, X.R., Kirubarajan, T.: ‘Estimation with applications
6 Dana, M.P.: ‘Registration: a prerequisite for multiple sensor tracking’ to tracking and navigation: theory algorithms and software’ (John Wiley
(Artech House Press, 1990) & Sons, Inc., 2001)
370 IET Radar Sonar Navig., 2013, Vol. 7, Iss. 4, pp. 361–370
& The Institution of Engineering and Technology 2013 doi: 10.1049/iet-rsn.2012.0199
Authorized licensed use limited to: Universidad de los Andes. Downloaded on April 02,2020 at 04:45:29 UTC from IEEE Xplore. Restrictions apply.