You are on page 1of 21

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

net/publication/262259177

A New Algorithm of Radar to ADS-B Registration

Article  in  Air Traffic Control Quarterly · July 2011


DOI: 10.2514/atcq.19.3.191

CITATIONS READS
2 1,093

1 author:

Jifeng Ru
University of New Orleans
18 PUBLICATIONS   399 CITATIONS   

SEE PROFILE

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

Sensor Registration View project

All content following this page was uploaded by Jifeng Ru on 13 May 2014.

The user has requested enhancement of the downloaded file.


A New Algorithm of Radar to ADS-B Registration
Robert Sittler, Jifeng Ru, Siva Sivananthan, David Vader
ARCON Corporation, 260 Bear Hill Road, Waltham, MA, USA

Abstract
In the Air Traffic Control (ATC) systems, multisensor based fusion is essential for accurate
target tracking and display. With the development of Automatic Dependent Surveillance –
Broadcast (ADS-B) techniques in the ATC, it is necessary to detect and correct the position and
time biases between a radar system and ADS-B. In this paper, a new approach for the radar to
ADS-B registration is developed based on alternating optimization technique. The algorithm uses
ADS-B reports as a trusted reference of target position to compute not only the range and
azimuth biases but also the time bias in radar measurements. It provides a robust and efficient
solution for calibrating radar biases. With three scenarios, the simulation results show that the
proposed algorithm can accurately estimate the radar biases with fast convergence.
Key Words: radar registration, ADS-B, alternating optimization, ATC

1. Introduction
The current ATC systems employ multiple sensors to improve the tracking performance. The
benefits of this multisensor fusion process, however, can be lost if there are some systematic
biases in the sensor measurements. Measurements with registration errors could degrade tracking
accuracy, and in extreme cases, generate false tracks during fusion due to correlation failures.
Therefore, sensor registration, a process to remove bias errors from multisensor measurements,
becomes a prerequisite for sensor fusion to ensure the accuracy of target tracking and display
[Dana, 1990].
The common radar registration errors contained in radar measurements are range, azimuth and
time biases. The range bias is a constant deviation in range measurement while the azimuth bias
is a misalignment of the radar north mark with respect to the true north mark. Both biases are
assumed to be uniform over the coverage area of each sensor. The radar time bias could be
caused by clock offset, and/or processing, storage, and transmission delays. Furthermore, the
time bias could introduce more errors into the range and azimuth biases through target velocities.
As a result, it is important for a registration method to isolate time bias so that range and azimuth
biases can converge to true values.
The current ATC systems in the United States use the NAS (National Airspace System)
registration algorithm [FAA, 2009] to calculate range and azimuth errors simultaneously. This
method considers a pair of sensors at a time and requires collecting four reports (two from each
sensor) from two targets that are on the opposite sides of the line connecting the sensor pair.
These reports, projected to a common stereographic plane, form four simultaneous equations. In
order to mitigate the effect of random errors in the measurements, a number of such four
equation sets are formed using the reports from multiple targets and accumulated over a time
period. The resulting sums of the equations are solved to find the four unknown biases: the range
and azimuth bias corrections for two radars. The geographical constraints, imposed by the NAS
registration to ensure the convergence of the algorithm, may result in longer time for some
sensor pairs to collect enough samples from the targets of opportunity.
As a core element for the Next Generation Air Transportation System (NGATS), ADS-B
provides a new technology solution to improve the safety, capacity and efficiency of aviation
while accommodating future air traffic growth. However, the deployment of ADS-B into the
NAS would not immediately replace radars and for some period of time, non-ADS-B equipped
aircraft will share the airspace with equipped ones. Therefore, the fusion of ADS-B data with
radar data is necessary. ADS-B targets provide more accurate geo-referenced aircraft position
with precise time of applicability (TOA) measurement. If radars are properly registered to the
ground and have no time bias, it is expected that radar data should align with ADS-B data. The
NAS registration, based on radar-to-radar, does not calculate the time bias and hence cannot be
used to ensure the time alignment between radar reports and ADS-B reports. Thus, an algorithm
that simultaneously corrects the range, azimuth and time biases in the radar is a prerequisite if
ADS-B data is to be integrated with radar data for multi-sensor tracking.
Some registration methods have been proposed in the literature, with optimization techniques
including maximum likelihood [Zhou, 1997], least square (LS) [Leung, 1994], and extended
Kalman filter (EKF) or unscented Kalman filter (UKF) [Li, 2004, Zhou, 2004, Nabaa, 1999].
However, none of these algorithms take time bias into account and the computational complexity
based on EKF/UKF makes these methods impractical for real time ATC applications. To reduce
the computation, some decoupling strategy is applied for practical development. In [Jaramillo
2002], two separate LS/KF filters are designed to estimate radar bias (range and azimuth only)
and aircraft transponder bias separately. The time bias has been considered in the recently
developed ARTAS system [Hogendoorn, 1999]. Instead of optimally estimating all target states
and sensor biases using one large KF filter, the ARTAS tracker decouples it into a number of
separate filters: one KF for each target, and one KF for the estimation of all sensor related
systematic errors [Doom, 1993]. The Eurocontrol’s SASS-C system also considers the time bias
into radar bias estimation using the Muratrec interpolation system [Pchelintsev, 2009]. An off-
line bias correction system is proposed in [Besada, 2009], which estimates radar range, azimuth
and time biases using KF filters.
In this paper, we propose an algorithm based on alternating optimization technique to
simultaneously estimate radar range, azimuth and time biases. The algorithm calculates these
biases using quality ADS-B positions and velocities as approximate true quantities. To
differentiate a time bias from a range/azimuth bias, the algorithm selects track samples in
different velocity quadrants with respect to the radar under consideration. Once two of the
opposite velocity quadrants attain the specified sample size, the algorithm iteratively computes
all three bias values and smoothes the calculation into the system correction for that radar. Note
that, although ADS-B, as currently designed, is subject to biases and should be quality-controlled
[Heidger, 2008, Krozel, 2005, Besada, 2009], we assume here that ADS-B data are accurate and
represent the truth with negligible errors. In practice, properly selecting high accuracy ADS-B
data is necessary for radar alignment.
The paper is organized as follows. Section 2 formulates the radar to ADS-B registration problem
in the stereographic plane. The proposed radar to ADS-B algorithm is described in Section 3. A
set of simulations are presented in Section 4 and the conclusions are provided in Section 5.
2. Problem Formulation
All bias correction computations take place in stereoplane coordinates so the reverse
transformation of ADS-B data to radar coordinates is not required. The stereoplane is an
imaginary plane where each Cartesian coordinate on the plane represents a point on the Earth’s
surface. Radar target reports are converted from polar coordinates to the stereoplane while ADS-
B target reports’ position and velocity in geodetic coordinates are also converted to the
stereoplane. Each radar report is linked with an appropriate ADS-B report. This linked pair
represents one target and, together with its velocity, provides the raw data for the registration
calculations.
Consider the geometry of synchronous ADS-B report and biased radar report, represented by the
blue dots in Figure 1. Without loss of generality, we assume that the radar is located at the origin
of the stereoplane coordinate system. Let ( x radar , y radar ) and ( x adsb , y adsb ) denote the system x
and y positions of the targets reported by the radar and ADS-B, respectively. Let (  ,  ) be the
true range and azimuth in the polar coordinates, then we have   xadsb
2
 y adsb
2
,
  tan 1  y adsb x  . Denote ( x adsb , y adsb ) as the converted ADS-B velocity in the system
 adsb 

plane. Let b = ( b , b , bt ) denote the biases of the radar in the range, azimuth and time,
respectively. The purpose of this study is to find the bias b in the radar coordinate system based
on the accurate ADS-B reports.

( xadsb , yadsb ) ( xradar , yradar )


b  bt

 b  bt

b

 x
radar 

Figure 1. The geometry of position errors between radar and ADS-B


In the radar coordinate system, considering that the errors in range and azimuth are statistically
independent, we can easily model the target dynamics to the first order in two separate
components,
 (t )   0  t    (1)

t
 (t )   0    (2)

where  0 and  0 are initial range and azimuth,  and  are the radial and tangential velocities
in nmi/sec, respectively, and   and  are two independent noises added to the measurements.
Radial and tangential velocities are available from the reports as

  ( x adsb x adsb  y adsb y adsb ) 


 

  ( y adsb x adsb  x adsb y adsb ) 


 

From the equations above and Figure 1, we see that the radar biases originate from two different
sources: first is the range or azimuth bias itself; second is the bias proportional to the
tangential/radial velocity and the time bias. Thus the registration bias can be modeled as
D   b   bt    (3)

D  b  bt   (4)

where D  and D are the radial and tangential position discrepancies between ADS-B reports
and radar reports, b is the arc to approximate the tangential position error, and   and  are
two independent random errors that account for the measurement and modeling errors. Radial
and tangential position discrepancies are obtained from ADS-B reports and radar reports as
follows

D 
 y adsb ( y radar  y adsb )  x adsb ( xradar  xadsb ) 

D 
 y adsb ( x radar  x adsb )  x adsb ( y radar  y adsb ) 

The aim here is to estimate the registration bias b = ( b , b , bt ) by minimizing the random
error components of (3) and (4), defined below
J   2  2 (5)

3. Alternating Optimization based Radar to ADS-B Registration


In this section, we first discuss the necessary target selection rules and then we derive the
iterative optimization algorithm to estimate the radar biases.
3.1 Target selection rules
In order to prevent confounding a time bias with a position bias, some target report selection
criteria is required. As an example of this problem, suppose that only targets moving in the radial
direction outward from the radar are available. Then a time correction has the same general
effect as a range bias correction. Similarly, confounding is a problem if targets proceed around
the radar in tangential paths. Another problem arises if track velocities are zero or too small, then
infinite or large time corrections are calculated, which are mostly the magnification of
measurement errors. Thus, the imposition of a minimum acceptable velocity magnitude is also
suggested. In addition, the target is discarded if its ADS-B range is smaller than the reasonable
minimum target range.
The necessary selection rules are illustrated in the Figure 2. Each valid report is classified as
lying in one of the quadrants, A1, A2, B1 or B2. To be eligible for inclusion in the report data set
for bias calculation, the report must be accompanied by another report in the opposite quadrant
(A1 with A2, or B1 with B2). A set of such pairs will tend to balance out the confounding effects
described above.
Radial velocity

B2 A1

Tangential
velocity
A2 B1

Figure 2. Target selection quadrants in velocity space


 

3.2 Radar to ADS-B registration Algorithm


In this section, we propose an iterative optimization algorithm featuring an alternating projection
approach, which finds the best approximation to any given point in the intersection of a finite
number of subspaces, using a sequence of alternating projections [Deutsch, 1992]. This method
can be conducted in various ways. The simplest iteration procedure would consist of adjusting
range and azimuth on one computational pass alternating with a time adjustment on another.
Each pass follows hill-climbing. Therefore, we decompose the optimization of the objective
function J in equation (5) into two minimization problems: 1) find the time bias bt to minimize
J, assuming b  and b are predetermined; 2) find the range bias b  and azimuth bias b for the
given time bias, provided bt is fixed. Note that the order of these two subproblems can be
swapped. Next we discuss these two subproblems in detail.
In equation (4), using the tangential velocity with the time bias for targets at different radii and
accumulating them will skew the azimuth bias calculation. Hence, all targets need to be scaled
by a fixed radius before accumulation. As such, the weighted tangential discrepancy appears as if
referred to a standard range. The equation (4) is rewritten as
c 
D   c b  c bt   (6)
 
where  c is a predefined range for scaling.

Considering a sufficient number of N targets are collected, the objective function in equation (5)
is rewritten as
  c c   
2

E   (    )   D (i)  b   (i )bt   


N N
D (i)   c b   (i)bt  
2 2 2
(7)
i 1 i 1 
   (i)  (i)  

where E is the sum of squares of position discrepancies between the radar and ADS-B over all N
selected targets, and i is the index of the target.
Step 1) Estimate radar time bias. Here, we just assume that the range and azimuth biases are zero
or preset and then find the time bias bt that will minimize E. This is achieved by taking the
partial derivative of equation (7) with respect to bt and equating it to zero. It yields

N   c  
2 
   (i ) D  i    
  (i )   i D  i 
i 1    

bˆt1  (8)
N 
 
2
 
  i 2    (ci) i  
i 1    

where bˆt1 denotes the time bias estimate in the first iteration.

Step 2) Estimate the range and azimuth biases. First the range and azimuth discrepancies should
be updated by subtracting the time bias calculated in (8)

D  1 (i )  D  (i )  bˆt1  (i ) (9)

D 1 (i )  D (i )  bˆt1(i ) (10)

Assuming that there is still some residual time bias bt 2 left, then applying the bias models (3) and
(6) again to D  1 and D 1 , we have

D 1 (i )  b   (i )bt 2    (11)

c 
D 1 (i )   c b  c (i )bt 2   (12)
 (i )  (i )
Then, similarly as earlier, assume that the time bias is zero or preset and then find the range and
azimuth biases that will minimize E. This is achieved by taking the partial derivative of
equation (7) with respect to b  and b , respectively, equating to zero, where D  1 and D 1
replace D  and D . This will result in the following estimates of the range and azimuth biases:
N
1
bˆ 1 
N
 D  1 (i ) (13)
i 1

1 N
 D 1 (i ) 
bˆ 1 
N
 
i 1   (i ) 
 (14)
Correcting discrepancies for the results of bˆ 1 and bˆ 1 yields

D  2 (i )  D  1 (i )  bˆ 1 (15)

D 2 (i )  D 1 (i )  bˆ 1 (i ) (16)

The next step is to repeat the above adjustments on the same samples with D 2 and D 2 taking
the previous roles of D  and D . The resulting second step bias corrections are bˆt 2 , bˆ 2 and bˆ 2 .
This iteration process can be continued until the convergence criterion is satisfied.
The bias corrections obtained from this iterative correction process can be deemed as a
geometrical decreasing series whose rate can be calculated from the available data. When
accelerated convergence is desired, we can have the sequential reduction factor K to allow the
computation of a series of bias corrections in a few short steps. The K factor is given by,
2 2
 N   N  
   (i )     c i 
 i 1   i 1  (i ) 
K (17)
N 
 c   
2

N   i     i  
2

i 1 
   (i )  

then the final bias corrections are given by


bˆtcor  bˆt1  t 2
(1  K )

bˆcor  bˆ 1   2 (18)
(1  K )

bˆcor  bˆ 1   2
(1  K )

Typical values of K range between 0 and 0.1, with values close to 0 indicating that corrections
beyond the second iterations don’t contribute significantly to the final bias. In some cases, where
target velocities are unevenly distributed, this factor can be close to 1. The proof of K is given in
Appendix A.
The recommended method described here is applied separately for each radar from which beacon
reports of targets that also have ADS-B are available. It must be applied iteratively since certain
approximations as well as random target measurement errors will need to be reduced by multiple
correction cycles and smoothing. Such iterations must also be continued to follow any
subsequent changes in the range, azimuth and time biases.

3.3 Smoothing of Bias Corrections


To reduce the noise effects of the bias estimates, a smoothing scheme is applied
bˆs ( k )  bˆs ( k  1)   (bˆcor  bˆs ( k  1)) (19)

where α is a weighting factor, b̂ cor is the bias estimate (range, azimuth or time) from the kth data
batch, bˆ s ( k  1) and bˆ s ( k ) are the smoothed bias estimates from the (k-1)th and kth data batch,
respectively. bˆ s ( k ) is applied as the bias correction in the radar target reports.

4. Performance Evaluation
In this section, we present some simulation results to illustrate the performance of the proposed
registration algorithm.
4.1 Simulation Scenarios
Three scenarios are designed to generate simulated data using radar information of the
Philadelphia ATC site (PHL), which has three short range radars and one long range radar. Their
characteristics [FAA, 2010] are shown in Table 1. The scan rate of ADS-B data is 1 sec and the
standard deviation in x and y coordinates is 0.005 nmi, i.e. the Navigational Accuracy Category
for Position (NACp) is 10.

Table 1. Simulated radar characteristics

radar num radar name range std (nmi) azimuth std (deg) scan rate (s)
0 PHL 0.04 0.23 4.61
1 QIE 0.08 0.23 12.0
2 NKX 0.04 0.23 4.95
3 WRI 0.04 0.23 4.80
       (PHL=Philadelphia, QIE=Gibbsboro, NJ, NKX=Willow Grove, PA, WRI=Wrightstown, NJ)

Scenario 1: 150 targets flying in random straight trajectories for 1000 sec. Initial position within
70 nmi of PHL radar is randomly generated. Heading is random as well. Speed is randomly
chosen between 120 and 400 knots. Target trajectories are shown in Figure 3, where black dots
represent the radar locations.
Scenario 2: 80 targets all on an east-bound trajectory, as shown in Figure 4. Fast targets are
located in the upper-right and upper-left quadrants with respect to the PHL radar flying at 450
knots. Slow targets are in the lower quadrants flying at 55 knots. This scenario is specifically
designed to test the robustness of registration, when the targets and their velocities are unevenly
distributed, i.e., the sum of radial and tangential velocities is non-negligible.
Scenario 3: same as Scenario 2, but with several azimuth bias jumps of 0.087 deg for the PHL
radar.
The imposed bias values, designed to cover various possible values in our scenarios are given in
Table 2. For all scenarios, a minimal sample size of n = 30 in two opposing quadrants is required
before the registration calculation is performed. An alpha filter with α = 0.1 is used to smooth out
bias corrections. When the difference between the previously smoothed bias and the newly
calculated bias is too large, the smoothing process is reset, giving full weight to the new
corrections.

Table 2. Simulated radar biases

radar name range bias (nmi) azimuth bias (deg) time bias (s)
PHL 0.1 -0.131 1
QIE 0.00 0.0 1
NKX -0.15 0.174 1
WRI -0.05 0.226 1

Figure 3. Target trajectories for scenario 1 (random velocity distributions)

Figure 4. Target trajectories for scenario 2 (uneven velocity distributions)


4.2 Simulation Results
The root mean square error (RMSE) is used to evaluate the bias estimation accuracy, i.e,

 bˆ 
M
1 2
RMSE (bˆs ( k ))  s
j
(k )  btrue
j
(k ) (20)
M j 1

where M denotes the total number of Monte Carlo Runs and here is set to 5, bˆsj ( k ) is the
smoothed bias estimate (range, azimuth, or time) from kth data batch for the jth independent run,
j
and btrue (k ) is the corresponding truth.

Scenario 1 is designed to check the performance for all four radars. The RMSEs of the
range/azimuth/time bias estimation for scenario 1 for all four radars are shown in Figure 5 to
Figure 7. For the range bias, there are larger variations for the long range radar (QIE) as
compared to the short range radars, which is expected due to the bigger range noise
characteristics of the long range radar. For azimuth and time biases, estimate variations are
comparable for long and short range. Overall, all bias estimates are accurate with fast
convergence.

Figure 5. Range bias estimation errors for scenario 1


Figure 6. Azimuth bias estimation errors for scenario 1

Figure 7. Time bias estimation errors for scenario 1


Scenario 2 is designed to test the robustness of the proposed registration algorithm with unevenly
distributed targets and target velocities. These targets are generated around the PHL sensor. The
RMSEs of the range/azimuth/time bias estimation for PHL radar are shown in Figure 8 to Figure
10. Compared to the Scenario 1 with randomly distributed targets, the algorithm needs a little bit
more time to get stabilized for the range bias. Uneven target distribution seems to have no impact
on azimuth and time bias estimates.
Figure 8. Range bias estimation errors for scenario 2 (PHL radar)

Figure 9. Azimuth bias estimation errors for scenario 2 (PHL radar)


Figure 10. Time bias estimation errors for scenario 2 (PHL radar)

Scenario 3 is designed to test the performance of the algorithm with azimuth bias jumps, which
could happen when radars change the mode of operation. The azimuth bias jumps are inserted at
the time instant 300 second with -0.087 deg, 600 second with 0.087 deg and 900 second with
0.087 deg. The RMSEs of the azimuth bias estimation for PHL radar are shown in Figure 11. As
shown, bias jumps are handled quite well by the registration algorithm.

Figure 11. Azimuth bias estimation errors for scenario 3 (PHL radar)
4.3 Discussions
The simulations show that the algorithm is robust and converges rapidly under various situations,
even when targets have significant uneven velocity distributions or in the presence of bias jumps.
Based on the standards specified in [FAA, 2010], which require the range residual bias to be
within 0.032 nmi for short-range radars and 0.125 nmi for long-range radars, and the azimuth
residual bias to be under 0.087 deg 95% of the time, the estimation accuracy provided by the
proposed algorithm under different scenarios is well within the prescribed guidelines. Although
the algorithm is fundamentally an iterative method, we have also observed that one iteration is
sufficient in most scenarios.
As the proposed algorithm requires a certain number of target reports collected for each quadrant
before the registration calculation can proceed, we conducted extensive studies through
simulations on how the sample size affects the estimation performance. As may have been
expected, we observed that smaller samples cause more fluctuation in the estimates, while larger
samples tend to improve accuracy but slow the convergence. Based on our analysis, a quadrant
sample size of 30 gives optimal registration results, which is the value we chose for our present
work. Note that the optimality criteria may be dependent on the prescribed guidelines, as well as
the specific adaptation, such as radar number and locations, used for the simulations.

5. Conclusions
The biases in radar target reports, if uncorrected, would degrade the tracking accuracy.
Therefore, in multisensor surveillance systems, such as ATC, accurate registration of the radars
is a prerequisite for sensor fusion. We have proposed a new radar to ADS-B registration
algorithm based on alternating optimization, where the range, azimuth and time biases in radar
targets are calculated with ADS-B reports as a true reference. The algorithm introduces a
sequential reduction factor to accelerate the series of bias corrections in one final step with little
additional computational load. The various simulation results demonstrate that the proposed
scheme provides a simple and robust solution for radar registration in terms of accuracy and
convergence.
Appendix A: the proof of sequential reduction factor K
From Section 3.2, based on equations (15) and (16), repeating the same process we have

N   c  
2 
   (i) D  2 i     (i)   i D 2 i 
 
i 1    
bˆt 2  (21)
N 
 
2
 
  i 2    (ci) i  
i 1    

then the range and azimuth discrepancies are updated by subtracting the time bias bˆt 2

D  3 (i )  D  (i )  bˆ 1   (i )bˆt1   (i )bˆt 2 (22)

D 3 (i )  D (i )  bˆ 1  (i )  (i )bˆt1  (i )bˆt 2 (23)

Assuming that there is still some residual time bias bt 3 left, then we apply the following bias
models,
D 3 (i )  b 2   (i )bt 3    (24)

c 
D 3 (i )   c b 2  c (i )bt 3   (25)
 (i )  (i )
Estimating range and azimuth bias as before,
N
1
bˆ 2 
N
 D
i 1
3 (i ) (26)

1 N
 D 3 (i ) 
bˆ 2 
N
 
i 1   (i ) 
 (27)

Rewriting equation (15) as

D  2 (i )  D  1 (i )  bˆ 1  D  (i )   (i )bˆt1  bˆ 1 (28)

then we have
N N N N
 D  2 (i)  (i)   D  (i)  (i)  bˆt1  (  (i)) 2  bˆ1   (i) (29)
i 1 i 1 i 1 i 1

Substituting bˆt1 from (8) into (29), it yields


N N

 D 2 (i)  (i)  bˆ1   (i)


i 1 i 1
(30)
Similarly,
2
N
c N
   c 

i 1  (i )

 (i )D2 (i )   ˆ
b 1   (i ) (i )
  (i )  (31)
i 1  
Hence, substituting (30) and (31) into (21), we have
2
N N
  
 bˆ 1   (i )  bˆ 1   (i )i  c 
bˆt 2 
i 1 i 1   (i ) 
(32)
N 
 c   
2

  i     i  
2

i 1 
   (i )  

In the same manner, it can be shown that


2
N N
  
 bˆ 2   (i )  bˆ 2   (i )i  c 
bˆt 3 
i 1 i 1   (i ) 
(33)
N 
 c   
2

  i     i  
2

i 1 
   (i )  

where

1 N
1 (bˆt1  bˆt 2 ) N
N
bˆ 2 
N
 D
i 1
3 (i ) 
i 1 N
 D (i)  bˆ
N
1 
i 1
 (i )
(34)
1 N 1 N bˆ N (bˆ  bˆt 2 ) N bˆt 2 N
  D  (i )   D  (i )  t1   (i )  t1  
 (i )     (i)
N i 1 N i 1 N i 1 N i 1 N i 1

1 N D  3 (i ) D (i ) ˆ
1 N
(bˆ  bˆt 2 ) N (i )
bˆ 2 
N

i 1  (i )

i 1  (i )
N
 b 1  t1
N
i 1  (i )
(35)
1 N D (i ) 1 N D (i ) bˆt1 N (i ) (bˆt1  bˆt 2 ) N (i ) bˆt 2 N
(i )
           
N i 1  (i ) N i 1  (i ) N i 1  (i ) N i 1  (i ) N i 1  (i )

Therefore, substituting (34) and (35) into (33), finally we have


2
N N
  
 (  (i ))    c i 
2

i 1   (i )  bˆ  Kbˆ
bˆt 3 
i 1
t2 t2 (36)
N 
 c   
2

N   i     i  
2

i 1 
   (i )  

So the definition of K in equation (17) is proved as


2 2
 N   N  
   (i )     c i 
 i 1   i 1  (i ) 
K
N 
 c   
2

N   i     i  
2

i 1 
   (i )  

Continuing in this manner, we would get bˆt 4  Kbˆt 3 , bˆt 5  Kbˆt 4 , …, thus the total time bias
estimate is


bˆtcor  bˆt1  bˆt 2  Kbˆt 2  K 2 bˆt 2  ...  bˆt1  t 2
(1  K )

Similarly,


bˆcor  bˆ 1   2
(1  K )

bˆcor  bˆ 1   2
(1  K )

References
[1] Dana, M. P. (1990), “Registration: a Prerequisite for Multiple Sensor Tracking,” in Y. Bar-
Shalom (Ed.), Multitarget-Multisensor Tracking: Advanced Applications, Artech House.
[2] Deutsch, F. (1992), “The Method of Alternation Orthogonal Projections,” in Approximation
Theory, Spline Functions and Applications, S. P. Singth, Ed. Kluwer Academic Publishers.
[3] Leung, H., Blanchette, M., and Harrison, C. (1994), “A Least Squares Fusion of Multiple
Radar Data,” in Proceedings of RADAR, Paris.
[4] Li, W., Leung, H., and Zhou, Y. (2004). “Space-time Registration of Radar and ESM using
Unscented Kalman Filter,” IEEE Transactions on Aerospace and Electronic Systems, 40(3).
[5] Nabaa, N. and Bishop, R. (1999), “Solution to a Multisensor Tracking Problem with Sensor
Registration Errors,” IEEE Transactions on Aerospace and Electronic Systems, 35(1).
[6] Zhou, Y. (2004), “A Kalman Filter Based Registration Approach for Asynchronous Sensors
in Multiple Sensor Fusion Applications,” in Proceedings of the IEEE International Conference
on Acoustics, Speech, and Signal Processing, Quebec, Canada.
[7] Zhou, Y., Leung, H., and Yip, P. (1997), “An Exact Maximum Likelihood Registration for
Data Fusion,” IEEE Transactions on Signal Processing, 45(6).
[8] FAA (2009), “Surveillance Source Input Processing”, NAS-MD-636, A6.05/A2.09.
[9] FAA (2010), "Surveillance and Broadcast Services (SBS) Integration into ATC Automation
Processing Requirements Document", Version 1.17.
[10] Doom, B.A. and Blom, H.A.P. (1993), "Systematic Error Estimation in Multisensor Fusion
Systems," in Proceedings of SPIE Conference on Signal and Data Processing of Small Targets,
Orlando, USA.
[11] Hogendoorn, R.A, Rekkas, H.C., and Neven, W.H.L. (1999), "ARTAS: An IMM-based
Multisensor Tracker," in Proceedings of International Conference on Information Fusion,
California, USA.
[12] Jaramillo, A.S., Portas, J.A., Corredera, J.R. and Garcia J.I. (2002), “On-line Bias
Estimation of Secondary Radar Networks for ATC,” in Proceedings of the 7th International
Conference on Information Fusion, Stockholm, Sweden.
[13] Heidger, R. and Mathias, A. (2008), "Multiradar Tracking in PHOENIX and its Extension to
Fusion with ADS-B and Multilateration", in Proceedings of the 5th European Radar Conference,
Amsterdam, The Netherlands.
[14] Krozel, J. and Andrisani, D. (2005), “Independent ADS-B Verification and Validation”, in
Proceedings of the AIAA 5th Aviation, Technology, Integration, and Operations Conference
(ATIO), Arlington, USA.
[15] Pchelintsev, A. (2009), "Radar Alignment and Accuracy Tool: RASS-R Radar Comparator
Dual", http://www.intersoft-electronics.com/Publications/IE-SUP-00042-001Radar
AlignmentandAccuracyTool.pdf.
[16] Besada, J.B., Miguel, G., Tarrio, P., Bernardos, A., and Garcia, J. (2009), "Bias Estimation
for Evaluation of ATC Surveillance Systems," in Proceedings of the 12th International
Conference on Information Fusion, Seattle, USA.

Biographies
Dr. Robert Sittler, Ph.D. (MIT), Vice President of ARCON Corporation, is a founder in the art of
multi-sensor tracking with 50 years of experience beginning with design and analysis of
surveillance algorithms for the SAGE air defense system. Later work has concentrated on
surveillance algorithms for civilian air traffic control. He designed the tracking algorithm
currently used by the FAA in terminal areas and participated in the early conceptual design of the
en-route area surveillance system. His latest work has been the analysis, simulation, and test of
various aspects of tracking surveillance for the FAA's Advanced Automation System.
Dr. Jifeng Ru is a Senior Staff Engineer at ARCON. She received her Ph.D. degree in Electrical
Engineering from University of New Orleans, in 2005, where she worked on many projects
supported by various U.S. government agencies in the field of signal estimation and detection as
well as target tracking. Since 2005, she has been working on ATC Systems in support of the
FAA’s Common ARTS, STARS and RWSL programs. Dr. Ru has been responsible for the
development, testing and evaluation of tracking and detection algorithms for terminal automation
systems. She has also been involved in the design and implementation of new algorithms to
support the integration of the ADS-B data into the terminal surveillance systems in order to
enhance system performance.
Dr. Sivananthan is the Manager of the ATC systems division at ARCON. He received his Ph.D.
degree in Electrical Engineering from the University of Massachusetts, Lowell in 2011. Dr.
Sivananthan has extensive experience working with terminal and en-route ATC systems. He has
also worked with GPS receiver systems, ADS-B and Multilateration systems. He has been
involved in developing enhancements to ultra-tight coupled GPS receivers to improve their
frequency lock and code lock performance. His research interests are also in the area of statistical
signal processing with applications to radar signal processing.

Dr. Vader is a Staff Engineer at ARCON. He received his Ph.D degree in Engineering and
Applied Sciences from Harvard University, Cambridge, MA in 2010. From 2004 to 2009, his
graduate research at Harvard has focused on three-dimensional imaging and mechanical
characterization of biopolymer matrices, such as collagen and fibrin. In 2004 and 2005, Dr.
Vader was also teaching assistant in applied statistics for engineering students at Harvard. From
2005 to 2007, he helped design and carry out microscopic Magnetic Resonance Imaging (micro-
MRI) experiments at Massachusetts General Hospital to study microscopic brain tumor growth
in a collagen matrix. Now he is involved in the testing, characterization and improvement of
various algorithms used in ATC Systems.

View publication stats

You might also like