A Transfer Alignment Algorithm Study

Based on Actual Flight Test Data

From a Tactical Air-to-GroundWeapon Launch
Christopher C. Ross Timothy F. Elbert
WL/MNAG Sverdrup Technology, Inc.
Eglin AFB, FL Eglin AFB,FL

--
Abstract This paper provides results from a joint study attitude navigation states. The following is a discussion
of the T/A process and the hardware requiredto perform
of various Transfer/Alignment (T/A) techniques
including velocity matching, integrated velocity a transfer/alignment.
matching, and doubly integrated Velocity matching
approaches. The goal is to determine how the various Transfer Alignment
algorithms perform when driven with actual flight test
data. The study addresses key T/A Kalman filter issues A T/A method is the process of transferring the high
such as the effect of wing flexure vibration on accuracy navigation information from an aircraft INS to
instrument error estimation and rapid alignment a weapon INS which is typically located on a pylon
concepts. A description of how a T/A Kalman filter is station of the aircraft‘swing or on a launcher located in
implemented and used in inertially guided weapons is a remote weapons bay. A direct transfer of aircraft INS
included. The data used in this study was acquired from navigation states (position, velocity, and attitude) would
the Air Force Operational Concept Demonstration not account for the fact that the weapon is not pointed in
(OCD) Program. The OCD effort was performed by the the exact same direction as the aircraft. In other words
Air-To-Surface Weapons System Program Mice (SPO) the aircmft sensors and the weapon sensors are not at the
at Eglin AFB in late 1992 and early 1993. The Eglin same tilts (pitch, roll, and heading) with respect to the
SPO initiated the flight test program to demonstrate the local vertical coordinate frame. Not compensating for
effectiveness of GPS/INS integration in a tactical these tilt differences would lead to erroneous velocity
weapon. The test was conducted using a Honeywell and position estimates, resulting in undesired weapon
Integrated Flight Management Unit (IFMU) and an navigation performance. This limitation in weapon
Interstate Electronics 5-channel P-code GPS receiver in accuracy can be greatly reduced if the weapon has its
a GBU-15 airframe on board a block 40 F-16 aircraft. own INS and access to aircraft INS data. A T/A Kalman
Telemetry data from this OCD weapon was transmitted filter is typically nm on the weapon CPU and receives
continuously from the beginning of weapon alignment aircraft navigation information via a 1553 interface bus.
to final impact on the ground. The OCD telemetry The T/A Kalman filter utilizes the measurements from
stream contained INS and GPS data from both the the aircraft INS and a weapon INS to estimate the
aircraft and weapon during captive carriage T/A misalignment between the two systems (both with
maneuvers. After launch, the weapon data alone was respect to the local vertical), and, as a residual effect, to
transmitted during the free flight to the target. These calibrate the weapon INS sensors.
messages provide the data set used in the T/A study.
Objectives
INTRODUCI~ON
There are many issues concerning the T/A process.
Today’s users of aircraft weapons are requiring that After a minimal discussion of the Kalman filter as it
munitions be “smart.“ One genre of smart weapons applies to the alignment process, this paper will focus on
includes those capable of navigating to a predetermined the more signifkant issues concerning T/A algorithm
target coordinate autonomously. The navigation design including how long a weapon should align, the
performance of such a weapon is dependent on the effect of vibration on sensor calibration, the effect of
proper initialization of the vehicle navigation states aircraft to weapon lever arm miscalculations, and the
which include position, velocity, and attitude. The T/A choice of T/A measmment methods.
approach to initializing a weapon is the most accurate
technique presently used to initialize the weapon

431
0-7803-1435-2/94 $3.00 0 1994 IEEE

Authorized licensed use limited to: J.R.D. Tata Memorial Library Indian Institute of Science Bengaluru. Downloaded on February 20,2025 at 22:54:38 UTC from IEEE Xplore. Restrictions apply.
 KALMANF ~ T ECONCEP~S
R APPLIED estimate the majority of the alignment error from the
TO TRANSFERALIGNMENT input velocity differences. To illustrate this point Figure
1 shows a filter's ability to estimate the misalignment
For the purposes of this study, it is assumed that the angle between the aircraft INS and weapon INS using
reader is familiar with the standard Kalman filter only a 6-state velocity matching filter. These estimates
equations [l]. For stability reasons, a Joseph's form are compared to the OCD 15-state integrated velocity
Kalman filter is implemented and a state correction matching filter that was initialized at t=O seconds and
vector is applied to close the loop [2]. The challenge is has stabilized. It is worth noting that the pitch and roll
to apply the Joseph's form Kalman filter equations to the Euler angle misalignments are observable within 5-10
T/A problem. When a Kalman filter is applied to the seconds due to the influence of the gravity vector. The
T/A process, there are two major benefits which result; azimuth Euler angle misalignment became observable at
alignment error estimation and sensor calibration. t=440 seconds when the aircraft executed a 60 degree
heading change. All subsequent misalignment angle
Alignment Error Estimation plots will compare the T/A study results with the OCD
stabilized estimates.
The most important elements of the T/A Kalman filter
are found in the 6 by 6 element sub-matrix of the system
dynamics matrix composed of three tilt <Yx,Y, ,Yz ) O M
Euler Anpk Uhallpnmnnl Estlmatn : OCO n O-!jtolr VU flw

error equations and three velocity (6V,, 6V, ,SV,) error Note 0'3 flrta Itorled at 1-0 and ha8 rmchmd rlmdy rtaIn.
0.03.
equations. The sub matrix of the system dynamics modinp
matrix is given below.

0 -T3 T2 I 0 0 0.01

[T3 0 -T,i 0 0 0
-------------,--------------
-T2T,
A ( r ) X ( r ) = 0 -a,-5 Z2
O i O
ay I 0
0

a,
-ag
o
a,
-a,
o
I Z,
1 -z, Z,
O -"I
o
-0ML
420 440 460 480 Hx) 520 540
I
5M)

where: nc

Figure I.
A (t ) = KF State Dynamics Matrix
X ( t ) = KF Error State Vector Sensor Calibration
Y = Misalignment Error Estimation Vector
6 V = Veiociv Error Estimation Vector Sensor calibration can be achieved by adding sensor
a , , a y, a , = Local Level Acceleration Vector error states to the Kalman filter. It would be ideal to
model all of the known sensor errors in the filter;
T,=-(p*-R,) however, such an endeavor reduces computational
T2 = P y + Q y capability and requires more memory and throughput in
T3 = - a , the weapon CPU. The alternative is to model only the
z, = -p, - 2 Q x most significant sensor error states and compensate for
2, = p y +2Ry the unmodeled states by increasing the appropriate
parameters in the Kalman filter process noise matrix.
z, = 2R, The sensor error states modeled in the Kalman filter
p, ,p, = x and y Local Level Craft Rate in radfsec used for this study include accelerometer bias and gyro
R, , a yR,
, = Local Level Earth Rates in radfsec bias.

Kalman Filter Measurements

The (%,%,uJ vector elements (local level frame
A Kalman filter requires a measurement as input to
accelerations) in the state dynamics matrix provide the allow the estimation of tilt and sensor errors. The typical
necessary mathematical coupling in the Kalman filter to T/A Kalman filter processes the difference between the

432

Authorized licensed use limited to: J.R.D. Tata Memorial Library Indian Institute of Science Bengaluru. Downloaded on February 20,2025 at 22:54:38 UTC from IEEE Xplore. Restrictions apply.
 velocity estimates of two navigation systems. The filters make up the plant. The model presented below
implemented for this study incorporate the difference represents the doubly integrated velocity matching plant
between aimaft INS velocity measurements and with the associated m r state vector. This plant was
weapon INS velocity measurements. modified to support analysis of the velocity and
m et
m velocity matching techniques. Detailed
There are three methods [l] of processing the velocity infonnation on these techniques are presented in
measurements recently evaluated through simulation for reference [l].
the Naval Weapon Center (NWC). They are the
velocity matching 0, integrated velocity matching
0, and doubly integrated velocity matching
(DIVM) techniques.

Although simulations were accomplished in the NWC

study, it was recommended that further work was
needed to study the effectiveness of the three techniques
in the vibration environment encountered under the
wing of a fighter aircraft. Recent flight testing
performed during the Opention Concept Demonstration
(OCD)program at Eglin AFB,FL provided an ideal data
set for testing T/A algorithms in a realistic wing flexure
and v i b " environment 161.

TESTINOTRANSFTWALIGNMENT

OCD Flight Test Data

The mission of the Operational Concept Demonstration
(OCD)Program was to demonstrate the effectiveness of
INS/GPS integration in a tactical air-to-surfaceweapon.
To provide a testbed for such a demonstration, designers
modified a GBU-15 TV guided glide bomb. The
standard video seeker in the GBU-15 was removed and
replaced with a tightly coupled INS/GPS system.

The OCD flight test program involved six missions with

modified weapons actually dropped on targets. There
were also numerous simulated- drops in which the
weapon was not ejected and the aircraft flew a nominal
trajectory to the target area. A T/A was accomplished
in the weapon INS using aircraft INS data provided via
1553 bus communications. Both the aircraft navigation
data and a large set of weapon data was telemetered to a
ground station and recorded. The raw weapon data
included Inertial Measurement Unit 0 AV's(change
E]6L

in velocities) and AVs (angle increments) which provide

the input to the navigation algorithms and alignment
filters analyzed in this effort.
Measurement Vector
The Kalmun Filter Plant Model
Today's Kalman filter designers must choose the
The Kalman filter plant or system dynamics matrix is measutements inputted to the filter. In the past,
the engine of the Kalman filter. All statistical differences between the velocity measurements from the
interactions are govemed by the state equations which aircraft and the weapon were accepted as adequate.

433

Authorized licensed use limited to: J.R.D. Tata Memorial Library Indian Institute of Science Bengaluru. Downloaded on February 20,2025 at 22:54:38 UTC from IEEE Xplore. Restrictions apply.
 Modem researchers have become interested in alternate good thing, then to integrate twice should be even
measurements such as angle rate matching, integrated better. The goal is to further reduce the random noise
velocity matching, and doubly integrated velocity on the measurements, that should allow more accurate
matching. In this study, the angle rate matching estimation of tilt mrs and sensor biases as presented in
approach is not investigated due to the complexity of [l]. The doubly integrated velocity measurement is
derivation and implementation. Research is being given below.
performed under an Air Force contract with TASC (The
Analytical Sciences Corporation) to investigate the
effectiveness of angle rate matching [5]. dDj,=j
T
2-0
I
T
t=o
Vm*dt - IT IT
f=O z=o
Va*dt
Velocity matching is the standard approach of
differencing the velocity measurements of the aircraft mi,,= Doubly Integrated Velocity
and the weapon. The input to the Kalman filter is given
below. Matching Measurement

AV, = V,,,- V, Figures 2.3 and 4 illustrate the performance of the three
V,,,= Missile Velocity alignment methods (VM. IVM, DIVM) employing a 1
second filter cycle time implemented for this study.
V, = Aircraji Velocity Hgure 2 compares the pitch error estimates for the three
AV,,, = Velocity Matching Measurement methods and includes the OCD real-time estimate as a
baseline. The three methods agree roughly within 1-2
The inputs are processed by the Kalman filter and milliradians. The velocity matching method shows an
estimates of the tilt errors and accelerometer and gyro initial overshoot which is quickly corrected. This
bias are estimated. The argument over which phenomena was not seen on most of the data runs and is
measurement type to use centers around the ability to attributed to initialization errors. A steady state
estimate accelerometer and gyro bias. estimate of the pitch misalignment for all three
approaches is achieved in approximately five seconds.
Integrated velocity matching is similar to velocity There is no discernible difference between the three
matching with the exception that the measurement is the methods that would suggest one is better than the others
difference between the integral of the aircraft and the for pitch estimation.
weapon velocity estimates. In effect, the measurement
becomes the difference of position changes estimated
over the measurement cycle time. The advantage of
differencing the integral of the velocity measurement is
that the random noise which corrupts the velocity
measurements will be attenuated due to the integration
process. A reduced measurement noise should improve
filter performance in the estimation of instnunent errors
such as accelerometer and gyro bias. The integrated
velocity measurement equation is given below.

Ai,, =IT Vm*dt-coVa*dt

t=O
Ai,, = integrated velocity measurement
T = Measurement Cycle Time

The doubly integrated velocity measurement approach

simply applies, as a measurement, the difference Figure 3 compares the roll estimates for the three
between the aircraft velocity estimate integrated twice methods. The velocity matching and the integrated
and the weapon velocity estimate integrated twice. The velocity matching approaches converge in
assumption is that if integrating the measmment is a approximately seven seconds and the doubly integrated

434

Authorized licensed use limited to: J.R.D. Tata Memorial Library Indian Institute of Science Bengaluru. Downloaded on February 20,2025 at 22:54:38 UTC from IEEE Xplore. Restrictions apply.
 velocity matching converges approximately three to five Measurement Update Rates
seconds later. The three methods converge on the OCD
estimate of misalignment angle without any visible The choice of the measurement technique is influenced
differences. by the T/A filter update rate. The velocity malchmg
technique degrades at the longer filter cycle times (3-5
seconds) [l], therefore a common filter update rate of
one second was chosen to evaluate the three
measurement techniques. The OCD program which is
used as a baseline, implemented a five second update
rate with integrated velocity as the measurement.

k v e r Arm Corrections

A lever arm vector is the relative distance between the

aircraft INS and the weapon INS. The lever arm
compensation is necessary to account for centripetal
4Jnl I effects when the aircraft p e r f m a turn during
120 440 1110 1110 sa0 5% 540 !!a
alignment. When the aircraft turns,the weapon INS and
10
the aircraft INS will experience different velocity effects
Figure 3. due to the moment arm between them. The different
effects need to be accounted for in the measurements
Figure 4 compares the heading misalignment angle before they are input into the Kalman filter. The
estimates for the three methods. Heading is the accuracy to which lever arm should be calculated is
parameter of greatest interest when estimating the tilt debatable. Table 1presents T/A results using OCD data
errors. The reason an aircraft maneuvers during as the input and varying the lever arm errors in the body
alignment is to provide the observability through y and z axis by f5 feet. Compared with the nominal
horizontal accelerations which allows the estimation of values for tilt error (no lever arm error), the lever arm
the heading misalignment. The relatively large errors contributed no more than 0.21 milliradians to the
difference between the three filter methods and the OCD RMS tilt error estimates as Summarized in Table 1. For
baseline (approximately 3 milliradians) is believed to be the maneuvers experienced in the OCD program, the
real. The T/A study used a relatively short alignment effect is shown to be negligible.
time (150 seconds) compared to the OCD flight test
(560 seconds) and a different measurement ensemble to Table 1. Lever Arm mor Effect
obtain its steady state estimate of the heading
misalignment angle. Lever Ann Error (feet) Euler Angle Est. Error (mad)
X Y 2 Pitch Roll Hdg

0.0 0.0 0.0 0.00 0.00 0.00

0.0 -5.0 0.0 0.10 -0.10 0.01
0.0 5.0 0.0 4.10 0.z0 0.0z
0.0 0.0 5.0 4.0z 0.07 0.20
0.0 0.0 -5.0 0.01 0.07 -0.20

VibrationEffects

Modern smart munitions face a hostile environment

un&r the wing of today’s fighter aircraft. Wmg flexure
and vibration of the aircraft due to wind buffeting and G
loaded turns provide vibration inputs to the weapon
Figure 4. IMU that are not measured by the aircraft INS. The

435

,.. .. , . . .. .. .- .,. , . I . ._I-_(” .... .-

Authorized licensed use limited to: J.R.D. Tata Memorial Library Indian Institute of Science Bengaluru. Downloaded on February 20,2025 at 22:54:38 UTC from IEEE Xplore. Restrictions apply.
 introduction of wing flexure vibration tends to impede
the transfer alignment process. The estimation of sensor
errors such as accelerometer and gyro bias is hindered
by modeled aircraft wing dynamics such as flexure
and vibration. The promised benefits of the integrated
measurement techniques are that the input to the T/A
KaLman filter has reduced noise properties. This
reduction in noise should result in more accurate
estimation of sensor biases. Figures 5 and 6 illustrate
the vibration experienced by a weapon system before
and after launch as measured by the weapon
accelerometers. Figure 7 shows the acceleration
characteristics through launch, including the peak -1 L A
588 588 570 572 574 576 578 580
acceleration experienced when the aircraft ejector
n
cartridges are fired.
Figure 7.

A secondary problem with vibration is that the weapon

will experience a different magnitude and frequency
02\ Std Drv - 0.0791 G I ii spectnuri due to vibration after launch. Modern
accelerometers, such as the RBA 500 used on the OCD
program, may have bias characteristics which change
significantly with aircraft/wing vibration induced
acceleration magnitude and frequency. A significant
change in accelerometer bias due to vibration (e.g. 5
mg) can nullify the effort of estimating the bias during
the alignment process. If the accelerometer exhibits a
sensitivity due to the vibration experienced during
captive carriage, the T/A filter will produce a bias
I , I .II I estimate that is not necessarily correct for free flight.
552 554 556 558 560 582 564
This problem was experienced during the OCD program
"2
and resulted in discarding the accelerometer bias
Figure 5. estimated during alignment. Figures 5 and 6 illustrate
the change in the vibration levels between captive and
025
Y Bcdy Acceleration : Frw Fllght free flight. Figures 8 and 9 provide a look at the
corresponding change in the power spectral density
02 -
0.11 - std Dw - 0.0205
between captive and free flight.

0.1 -

0.05 -

4.15t
i 4t I il -I
-02
-025 I
t I
592 594 596 598 600 602 604

I C

Figure 6.
0 5 1 0 1 5 2 0 2 5 3 0 3 5 4 0 z J 5 0

Hz

Figure 8.

436

Authorized licensed use limited to: J.R.D. Tata Memorial Library Indian Institute of Science Bengaluru. Downloaded on February 20,2025 at 22:54:38 UTC from IEEE Xplore. Restrictions apply.
 4-

3.5-

3-

2.5-

2-

1.5

0
1-

0
-

5 1 0 1 5 2

Rapid TransferiAlignment Techniques

0 2

Figure 9.
5
Hx

Typical T/A approaches in the past required that the

aircraft perform a five to seven minute T/A which
3
I4 50
00 33 55 11 oo ** 5 0

This study has described many of the issues concerning

included a low dynamic S-turn maneuver to excite the design and implementation of a transfer/alignment
heading errors. The length of time was chosen because Kalman filter. Three measurement techniques were
gyro and accelerometer errors can be estimated only discussed including the velocity matching, integmted
after 2-5 minutes. When facing a hostile target, pilots velocity matching, and the double integrated velocity
would much prefer not to be performing five minute matchmg approaches. Figure 11 includes the estimates
maneuvers before launching a weapon, therefore, rapid of pitch, roll, and heading error for all three approaches
approaches to alignment make sense. Modem discussed. The integrated approaches exhibited greater
approaches [5] concentrate on rapidly estimathg the tilt sensitivity to maneuvers at initialization, however, this
errors and not on the estimation of the sensor errors. fact did not appear to improve the alignment process at
Alignment time has been demonstrated through the one second filter update rate.
simulation in the 10-30 second time range through
performing a rapid aircraft maneuver and then leveling
off. firther studies are planned for using the OCD data
to investigate rapid alignment techniques.
0.03 -
0.02 -
Sensitivity To Initial Conditions

The convergence of the alignment filter methods I! :nO

investigated all showed a sensitivity to vehicle dynamics -OD1 -

present during initiakation. Figure 4 presented the T/A
-0.02 -
filter response for non-dynamic initial conditions (i.e.
straight and level flight). Hgure 10 shows the response -0.03 -
of the various alignment techniques when initialized -0.04 1 I
4 2 0 4 4 0 4 8 0 4 M 5 0 0 5 2 0 5 u ) s g )
while the aircraft was experiencing minor horizontal
accelerations (0.1 to 0.2 g). Recall that the OCD .IC

reference in Figure 10 has already achieved steady state. Figure 11.

The doubly integrated velocity approach exhibited the
greatest sensitivity to the initial conditions yielding an The estimates of accelerometer bias were hindered by
overshoot 55 milliradians over the other approaches. the wing flexure vibration of the aircraft before launch.
The large overshoot lead to a significant steady state Lever arm measurements were found to be acceptable
error in the heading misalignment estimate. with 5 feet of error for the y and z body axis. Finally,

437

_.. - _^.I. .

Authorized licensed use limited to: J.R.D. Tata Memorial Library Indian Institute of Science Bengaluru. Downloaded on February 20,2025 at 22:54:38 UTC from IEEE Xplore. Restrictions apply.
 the filters were all found to be sensitive to horizontal [6] Timothy F. Elbert, Paul F. Torrey, David M. Gaskill
accelerations at the initialization point. The integrated "INS/GPS OCD Final Report", Contract Number:
approaches exhibited greater sensitivity initial FO8635-91-C-OOO2, Sverdrup Technology, Inc., P.O.
conditions leading to larger overshoots of the steady Box 1935 Eglin AFB, FL 32542-5000, October 1993.
state values which led to a bias in the final steady state
estimate.

Further studies need to be done to provide a complete

analysis of the integrated velocity approaches in a T/A
Kalman filter. This report is by no means exhaustive in
the endeavor to decide which method is superior and it
leads to more questions than answers.

The potential benefits of the three approaches presented

should be quantified through implementation and
comparison at different update rates to determine the
optimal mix of measurement types and filter cycle
times.

The OCD program provided an extensive data set that

includes both captive carriage and free flight inertial
instrument data that is acceptable for evaluating rapid
alignment concepts. The OCD data should continue to
be exploited for transfer alignment research.

[l] Strapdown Associates et. al., "Transfer Alignment

Applications for Air Launched Missiles," document
(NWC-007-003) prepared for Naval Weapons Center,
China Lake, CA, 93555, October 1990.

[2] Arthur Gelb, Applied Optimal Estimation.

Cambridge Mass., and London England M.I.T. Press,
1988.

[3] Paul G. Savage, "Strapdown Inertial Navigation

Lecture Notes," Strapdown Associates, Inc., 13805 F i t
Ave North, Plymouth, MN 55441, February 1990.

[4] Timothy F. Elbert, "Weapon Attitude Measurement

System ( W A M S ) Development Final Report," Contract
Number: FO8635-91-C-0002, Sverdrup Technology,
Inc., P.O. BOX1935 Eglin AFB, FL 32542-5000, April
1992.

[5] Jim Kain, "Advanced Alignment Concepts,"

AFATL-TR-89-50,The Analytic Sciences Corporation,
55 Walkers Brook Drive, Reading, MA 01867, October
1989.

438

Authorized licensed use limited to: J.R.D. Tata Memorial Library Indian Institute of Science Bengaluru. Downloaded on February 20,2025 at 22:54:38 UTC from IEEE Xplore. Restrictions apply.