Effective Use of Magnetometer Feedback for

Smart Projectile Applications

JONATHAN ROGERS and MARK COSTELLO

Georgia Institute of Technology, Atlanta, GA, 30332

THOMAS HARKINS
US Army Research Laboratory, Aberdeen, MD, 21001

MOSHE HAMAOUI
Data Matrix Solutions, Inc., Aberdeen, MD, 21001

Received September 2010; Revised March 2011

ABSTRACT: The use of magnetometers for orientation estimation on rapidly-spinning flight bodies is analyzed.
Specifically, the effect of spin-induced magnetic field distortion is discussed, with particular attention to its impli-
cations for magnetometer-based orientation estimation. First, the nature of spin-induced field distortion is
described and it is shown that, if not properly accounted for, distortion can lead to significant estimation errors
in artillery projectiles. Then, an orientation estimator is constructed driven by magnetometer, gyroscope, and
GPS feedback. A novel feature of this algorithm is its compensation for spin-induced distortion of the Earths
field. The algorithm also incorporates in-flight magnetometer calibration performed simultaneously with projec-
tile orientation estimation. The comprehensive algorithm is built as a coupled set of Extended Kalman filters.
Observability of the estimation problem is discussed and unobservable modes are identified. Finally, example
results and Monte Carlo simulations compare estimation performance to algorithms which neglect spin-induced
distortion effects or do not perform in-flight calibration. These results demonstrate that magnetometer-based sys-
tems on-board spinning projectiles should incorporate corrections for field distortion, and that overall accuracy is
greatly enhanced by performing in-flight calibration.

INTRODUCTION The use of magnetometers to obtain orientation

information is not a new idea and has been used for
The availability of low-cost, lightweight sensors
many types of flight vehicles, from satellites to
and digital microprocessors has enabled weapons
UAVs. Typically, an algorithm is employed to deter-
designers to equip artillery projectiles with full
mine a solution for the direction cosines matrix
guidance and control capability. However, in com-
(DCM), Euler angles, or quaternions of the vehicle
parison to electronics packages on-board missiles,
based on a set of magnetometer measurements.
projectile sensor suites and guidance units must be
Wahba [1, 2] initially proposed a batch least squares
able to withstand large acceleration loads at launch
technique for determining the DCM of a satellite
and high spin rates while maintaining low cost.
based on vector measurements. Other researchers
One challenge routinely faced by designers has
proposed purely deterministic methods [3] and re-
been the development of attitude estimators for
cursive algorithms [4] for computing the DCM. Bar-
these vehicles. While accurate attitude information
Itzhak and Oshman [5] extended this work by devel-
is critical to control system performance, it is
oping a method for recursively estimating quatern-
often impossible to obtain by integrating outputs
ions from a set of vector measurements. Similar
from low-cost rate sensors. One common solution
algorithms [6] were developed for Euler angle deter-
is to use magnetometer measurements to provide
mination. Most recently, Mortari [7] has shown that
periodic updates to integrated rate sensor data.
when using more than two vector measurements,
Magnetometers are attractive due to their low
the optimal Euler axis and the principle Euler angle
power requirements, rugged construction, and low
can be obtained separately without iterative proce-
cost.
dures. Psiaki [8] developed a magnetometer-only
attitude and angular rate estimator for low-cost
NAVIGATION: Journal of The Institute of Navigation
Vol. 58, No. 3, Fall 2011
spacecraft and showed reasonable performance
Printed in the U.S.A. using flight data.

203
 While much of the above research has been those outlined in [18] are typically not an option
directed toward spacecraft applications, several in- during projectile flight. Standard pre-flight sensor
vestigators in the projectile community have re- calibration procedures using reference sensors are
cently developed algorithms specifically tailored to too expensive given the low-cost nature of gun-
smart projectiles. Ohlmeyer, Fraysse, and Pepitone launched munitions, and cannot account for cali-
[9] incorporated magnetometers in a low-cost INS bration changes after launch due to large shocks.
along with accelerometers and GPS. Magnetometer Although it may be possible to develop sensors that
bias was estimated in-flight to improve perform- maintain calibration through launch and do not
ance. Wilson [10] also proposed the use of magneto- require extensive calibration, cost considerations
meters as the primary attitude sensor on-board for projectile applications make this solution less
smart projectiles. He showed that complete attitude attractive. Thus, new techniques are required to
solutions could be obtained only by combining mag- perform in-flight estimation of as many nuisance
netometers with additional sensors such as acceler- parameters as possible.
ometers or solar sensors. Most recently, Lee et al. The contributions of this article are twofold.
[11] developed a roll attitude estimator for smart First, an in-depth analysis of the problem of mag-
munitions using magnetometers based on an netic field distortion inside projectile bodies is pre-
unscented Kalman filter. sented. An analytical model for this field distortion
Two recurring problems have consistently hin- is built and compared to experimental results.
dered efforts to implement magnetometer-based A new magnetometer model is built which incorpo-
estimators on-board guided artillery shells. The first rates this distortion, and is used to generate an
is that distortion and attenuation of the Earths mag- Extended Kalman filter that estimates field distor-
netic field can be significant inside the body of a spin- tion parameters. The second main contribution is
ning projectile due to the formation of eddy currents the development of a coupled set of Extended Kal-
within the conducting metallic body. Harkins [12] man filters that estimate projectile orientation, nui-
has explored this phenomenon experimentally, con- sance parameters, and distortion effects simultane-
cluding that these effects can lead to significant ously. Using this filter, a direct analysis of the bene-
error in magnetometer-based estimators if proper fits of in-flight nuisance parameter and field
compensation is not used. The second, and perhaps distortion estimation is performed. This is accom-
less application-specific, is that bias, scale factor, plished through example and Monte Carlo simula-
and misalignment errors can often have significant tions in which performance of estimators with and
detrimental impact on overall attitude estimation without the capability to perform in-flight nuisance
performance. Calibration values, especially bias, can and distortion parameter identification is directly
change during launch or in flight. Several techni- compared. Results show a significant benefit to in-
ques have previously been developed to mitigate this corporating distortion effects and performing nui-
problem by performing on-line estimation of these sance parameter estimation in flight. The article
nuisance parameters (autocalibration). Lerner and begins by describing magnetic field distortion within
Shuster [13] first developed a method to estimate rapidly-spinning projectile bodies. Then, algorithms
magnetometer nuisance parameters on-board space- are developed to estimate distortion effects, magne-
craft given a priori knowledge of attitude. Alonso tometer nuisance parameters, and projectile orien-
and Shuster [1416] have proposed so-called tation simultaneously. An observability analysis of
attitude independent autocalibration methods for the entire estimation problem is then performed and
spacecraft that rely on changes in the Earths mag- unobservable parameters are identified. Finally,
netic field magnitude over one orbit cycle. Crassidis simulation results show that attitude estimation
et al. [17] expanded on this work, developing three performance is almost always improved when dis-
algorithms to perform real-time magnetometer cali- tortion effects are incorporated and nuisance param-
bration based on observed differences in field magni- eter estimation is performed.
tude. Most recently, Gebre-Egziabher [18] developed
an autocalibration algorithm for UAVs by fitting an
ellipsoid to measured magnetic field data. Magne-
SPIN-INDUCED MAGNETIC FIELD DISTORTION
tometer data used to define this ellipsoid is gener-
ated by rotating the vehicle through prescribed When a conductive body is subjected to a chang-
turns during the calibration process. ing magnetic field, eddy current effects occur inside
Previously-developed autocalibration algorithms the body. Equivalently, a spinning cylinder im7-
are for the most part unsuitable for projectile appli- mersed in a transverse uniform magnetic field is
cations for several reasons. First, the Earths mag- essentially subject to two radial oscillating fields
netic field does not change enough throughout which are 90 degrees apart in space and time. By
flight to employ attitude-independent solutions. Faradays law, these oscillating magnetic fields will
Second, prescribed calibration maneuvers such as induce electric fields according to

204 Navigation Fall 2011

 ~
~ 2 @B
r3E (1)
@t
If the cylinder is conductive, eddy currents will
be induced to flow along the length of the cylinder.
These currents in turn give rise to a secondary
magnetic field which, in superposition with the
external field, creates a distorted magnetic field in
and around the cylinder [19]. Because on-board
electronics packages used for smart weapon guid-
ance and control are located within a rapidly spin-
ning projectile body, magnetometer-based orienta-
tion estimators must compensate for this field dis-
tortion to avoid significant error.

Modeling Spin-Induced Distortion:

Infinite Cylinder
Maxwells equations lead to a diffusion equation
for the magnetic vector potential describing the
Fig. 1Magnetic field lines for infinite cylinder rotating counter-
time evolution of the magnetic field, given by clockwise in an external magnetic field

@~
A
r2 ~
A lr (2) distorted field for an infinite cylinder, and results
@t
were compared to those generated with the closed-
where ~ A is the magnetic vector potential, l is the form solution. The magnetic fields generated by
magnetic permeability, and r is the electrical con- these two models matched to within 1%. Then, the
ductivity. Note that in the above equation the sim- FE model considered a nonmagnetic hollow cylinder
plifying assumption that the displacement current composed of 6061 Al, this time with inner and outer
in Maxwells equations is negligible is made, since radii of approximately 50 mm and 57 mm respec-
projectile spin rates correspond to wavelengths that tively and length of 175 mm. This finite-length cyl-
are orders of magnitude larger than typical projec- inder was immersed in a uniform transverse mag-
tile body dimensions. Recently, Ziolkowski and Gra- netic field of 1 Gs and spun at selected frequencies.
bowski [20] imposed appropriate boundary condi- Figure 2 shows example magnetic field results gen-
tions on Equation (2) to obtain an analytic solution erated with this FE model for this finite cylinder
for the case of an infinite, hollow cylinder spinning spinning at a rate of 80 Hz.
in a uniform, transverse magnetic field. Figure 1
shows magnetic field lines for an infinite, non-
magnetic, conducting cylinder of inner radius 50
mm and outer radius 75 mm rotating counterclock-
wise at a rate of 20 rad/s placed in a constant trans-
verse external field (computed using the expressions
from reference [20]). Note that, for an infinitely-long
conducting cylinder, the field inside the body is uni-
form (i.e., constant distortion angle throughout) and
attenuated with respect to the external transverse
field. Any axial component of the external field is
unaffected for cylinders of infinite length.

Modeling Spin-Induced Distortion: Finite Cylinder

For cylinders of finite length, no closed-form solu-
tion for the magnetic field is possible. Thus, finite
element methods must be employed to model mag-
netic field distortion inside rotating bodies of finite
length. To analyze distortion effects for finite
bodies, a finite element (FE) model was constructed
using ANSYS software. First, for validation pur- Fig. 2Magnetic field vectors near a cylinder spinning at 80 Hz,
poses, the FE model was used to generate the spin- generated by FEA. The cylinder spins about ~IP

Vol. 58, No. 3 Rogers et al.: Effective Use of Magnetometer Feedback for Smart Projectile Applications 205
 Figure 2 highlights some key differences between
distortion effects from infinite- and finite-length
cylinders. First, in contrast to the infinite-length
case, the field inside finite-length cylinders is not
uniform. Second, in the finite-length case, the dis-
torted field inside the cylinder has a noticeable
axial component. This axial component is zero
along the spin axis and in the transverse plane of
the centroid, but grows considerably near the edges
of the cylinder.
Experiments were conducted to validate results
from the FE model. A Helmholtz coil was used to
generate a near-uniform rotating magnetic field
near a cylinder composed of 6061 Al with dimen-
sions approximately equal to those described above.
The experiment was performed at selected frequen-
cies between 0 and 250 Hz, and the magnetic field Fig. 3Attenuation factor vs spin rate
was measured approximately at the centroid of the
cylinder.    
m~ zD
1 1 m
~z
In order to quantitatively compare results be- cD tan  tan (5)
myD
~ m
~y
tween the analytical model, FE model, and experi-
ment, three variables are defined that describe ~ xD  m
m ~x
spin-induced distortion: the attenuation factor (AF), fD q (6)
the transverse distortion angle, cD, and the induced m~ 2y m ~ 2z
axial component, fD. Attenuation factor (AF) is a
value representing the attenuation of the compo- Figures 3 and 4 show attenuation factor and dis-
nents of the Earths field transverse to the projec- tortion angle as a function of spin rate calculated
tile spin. Distortion angle (cD) represents the by the analytic model and the FE model as well as
change in direction of the transverse components of values obtained from experiment. These figures
the field (as shown in Figure 1). The induced axial show results for distortion parameters both at the
component (fD) is a value between 0 and 1 that centroid (both models and experiment) and at a
introduces distortion in the axial direction as a per- selected example point away from the centroid (FE
centage of the overall transverse field strength. Let model only). The selected example point is located
the components of the Earths magnetic field vector 34 mm in the radial direction and 25 mm in the lon-
inside the projectile body expressed in the body ref- gitudinal direction from the centroid of the cylinder.
erence frame be given by m ~ xD ; m
~ yD ; m
~ zD. Therefore, Furthermore, Figure 5 shows the induced axial
component (fD) as a function of spin rate at the
8 9 2 38 9 selected example point generated by the FE model
<m~ xD = 1 0 0 < m~x =
~ yD 4 0 ccD scD 5 m
m ~ y 1  AF (experimental data at this example location is not
: ; : ; currently available).
m
~ zD 0 scD ccD m~ z 1  AF
8 q 9
>
< fD m~ 2y m ~ 2z > =
0 (3)
>
: >
;
0

where m ~ x; m
~ y; m
~ z are components of the Earths
magnetic field expressed in the body reference
frame. Note that in Equation (3), and in the re-
mainder of this article, ca denotes cos(a), sa denotes
sin(a), and ta denotes tan(a). Given both the nomi-
nal and distorted sets of magnetic field components,
it is also possible to solve Equation (3) for AF, cD,
and fD such that
q q
m~ 2y m ~ 2z  m ~ 2yD m2zD
AF q (4)
m ~ 2y m~ 2z
Fig. 4Transverse distortion angle vs spin rate

206 Navigation Fall 2011

 tile is rotated at various spin rates in the presence
of a known external field, and the magnetic field
within the projectile carefully measured at the
magnetometer location.
With valid approximations for distortion parame-
ters at the location of the magnetometer sensor, it
is possible to compensate for and even estimate
these distortion parameters in flight. The remain-
der of this article focuses on how magnetometers
can be used effectively for real-time orientation
estimation in spinning projectiles, and compares
estimation performance to algorithms that neglect
spin-induced distortion.

Fig. 5Induced axial component vs spin rate at selected example ESTIMATOR DESIGN
point (FE model)
The projectile orientation estimator proposed
here consists of three coupled Extended Kalman fil-
The results shown in Figures 35 demonstrate ters: one to estimate Euler angles; one to determine
that all distortion parameters vary smoothly as a magnetometer scale factor, bias, and misalignment;
function of spin rate. Furthermore, even for projec- and, one to determine magnetic field distortion pa-
tiles exhibiting relatively low spin rates, these rameters. These three filters are coupled as shown
effects can be considerable. For spin rates between in Figure 6, creating a so-called integrated filter
40 rad/s and 120 rad/s (the range of roll rates for design. The three filters operate somewhat inde-
the example projectiles considered here), polyno- pendently, with the nuisance parameter and distor-
mial curve fits were performed to fit FEA data at tion fit parameter estimation updated at specified
the selected example point, and were found to pro- intervals.
duce reasonably accurate approximations. Note The choice to implement the three filters shown
that a cubic fit proved adequate to describe AF as a in Figure 6 as a coupled set of filters rather than as
function of roll rate while a quadratic fit proved one large filter stems from several factors. First,
adequate to describe both cD and fD as a function of note that because the derivatives of nuisance and
roll rate. Therefore, distortion fit parameters are assumed to be zero,
the integration portions of the filters naturally
AF  a3 p3 a2 p2 a1 p a0 (7)
decouple. Second, the integrated filter provides a
cD  w2 p2 w1 p w0 (8) simple mechanism for turning off nuisance and dis-
tortion fit parameter estimation in poor estimation
fD z2 p2 z1 p z0 (9) geometries, such as when flight occurs along the
magnetic field vector. Third, the update rates
The coefficients in Equations (7)(9) can be easily required for Euler angle estimation are at least one
determined for a specific projectile configuration order of magnitude higher than for nuisance and
through a calibration process in which the projec- distortion fit parameter estimation, and thus signif-

Fig. 6Integrated filter schematic

Vol. 58, No. 3 Rogers et al.: Effective Use of Magnetometer Feedback for Smart Projectile Applications 207
icant computation time can be saved by avoiding
high-frequency updates to these parameters.
This section describes the design of each of the
three coupled Extended Kalman filters (EKFs). Kal-
man filters generally use a process model to propa-
gate state estimates and a measurement model to
update state estimates whenever measurements
are available. Note that the sensor suite considered
here consists of a three-axis magnetometer, a three-
axis rate gyroscope, and GPS-derived inertial veloc-
ity feedback. Signals from the rate gyroscope are
used directly to drive Euler angle dynamics, and
thus scale factor errors, biases, and misalignments
of the gyroscopes are not estimated.
The first section describes sensor and projectile
geometry, while subsequent sections discuss the Fig. 7Magnetometer sensitive axis definition
distortion fit, magnetometer nuisance parameter,
and orientation state estimators, respectively. In Note that for projectile applications, the require-
the final section, an observability analysis of the ment for low-cost sensors coupled with high spin
entire problem is performed and unobservable rates prohibit use of pure angular rate integration
states are removed. for attitude estimation.
Estimation of all three Euler angles using only
magnetometer and rate gyro feedback is not an
Sensor and Projectile Geometry
observable problem if we are to avoid using angular
Feedback from three sources are assumed to be rate integration to obtain one of the angles. Numer-
available, namely a three-axis magnetometer ous techniques have been developed to complete the
(located near the projectile centerline), a three-axis magnetometer-based orientation estimation prob-
rate gyro, and an inertial velocity estimator. In lem, including use of derivatives of magnetometer
order to maintain as general a formulation as possi- signals, feedback from other less accurate roll sen-
ble, the three-axis magnetometer is treated as three sors, or approximation of pitch and roll angle using
single-axis magnetometers. It can be shown that inertial velocity feedback [10]. For projectile appli-
the number of misalignment angles to be estimated cations, the last technique proves to be an attrac-
for three single-axis sensors is the same as for one tive option for two reasons. First, inertial velocity
three-axis sensor, and thus no estimation penalty is feedback is typically readily available anyway from
incurred by this treatment. The magnetometers GPS, an IMU, or a combination. Second, projectiles
sensitive axis, denoted by ~ s, is described by two typically experience small angles of attack, even
rotation angles and can be obtained by first rotating during maneuver, allowing pitch and roll to be
the ~IP  ~
JP plane by /S about ~ IP , and then rotating determined from velocity with reasonable accuracy.
s off the projectile centerline by wS as shown in Fig-
~ Thus, given mass center velocity components in the
ure 7. The output of a single axis magnetometer inertial frame, vx ; vy ; vz , and assuming zero angle of
can be written as attack, direct measurements for pitch and yaw
  angle are possible, given by
m S m ~ x cwS m
~ y c/S swS m~ z s/S swS b n (10) 0 1
 
where m ~ x; m
~ y; m
~ z are the components of the exter- B vz C vy
hm sin1 @qA; wm tan1 (12)
nal magnetic field resolved into the body reference v2x v2y v2z vz
frame, S is scale factor, b represents bias, and n
represents zero-mean Gaussian white noise. Note that the above expressions are only valid for
Angular rate feedback is provided by a three-axis small angles of attack. However, in the presence of
rate gyroscope assumed to be aligned with the body angles of attack beyond a few degrees, hm and wm
axes. The output from the gyro is given by can be viewed as noisy measurements and subse-
8 9 2 38 9 8 9
< xx = SGx cpq cpr < p = < bp = quent error can be accounted for by proper tuning
x 4 cqp SGy cqr 5 q bq of filter gains.
: y ; : ; : ;
xz crp crq SGy r br
8 9
< np = Distortion Fit Parameter Estimator
nq (11) With the knowledge that AF is approximately a
: ;
nr quadratic function of roll rate, and cD and fD are

208 Navigation Fall 2011

   1 
approximately cubic functions of roll rate, an
Extended Kalman filter (EKF) is designed to esti- K D PD I  CTD RD CD PD CTD CD PD CTD R1
D
mate polynomial fit parameters in real time. First,
(18)
note that the magnetometer equation now changes
from the expression shown in Equation (10) to Note that in Equation (18), CD is the linearized
  output matrix given by
q
m S m~ x c wS f D c wS m~ 2y m ~ 2z b n h  i
m
~ @m~  @m
~  @m
~  @m
~  @m
~  @m
~  @m
~  @m
~  @m
~
CD @@a0 @a1 @a2 @a3 @w0 @w1 @w2 @z0 @z1 @z2
  (13)
S1  AF ccD c/S swS m ~ y s /S s wS m~z (19)
 
S1  AF scD s/S swS m ~ y  c /S s wS m~z
and RD is a user-specified gain matrix correspond-
Equation (13) serves as the measurement model ing to the estimated measurement covariance.
for this filter and those described in subsequent sec- Finally, state covariance updates are computed
tions. The state vector estimated by this filter is according to
given by  1
T 1
PDNew P1
D CD RD CD (20)
8 9
>
> a0 >>
>
>
>
> a1 >>
>
>
> >
> 2>
>
> a >
>
> > Magnetometer Nuisance Parameter Estimator
>
>
> a3 >>
>
< =
w0 A separate Extended Kalman filter is used to
xD
~ (14)
>
> w1 >> estimate nuisance parameters for each single-axis
> >
>
>
> w2 >>
> magnetometer. The state vector estimated by this
>
> >
>
>
> z 0 >
> filter is given by
>
> >
>
>
> z 1 >
>
: ; 8 9
z2 S >
>
> >
< =
b
It is assumed that all fit parameters do not vary xM
~ (21)
> /S >
> >
with time, and thus the state and covariance propa- : ;
wS
gation equations are given by
It is assumed that all nuisance parameters are
x_D 0
~ (15) constant, and thus the state and covariance propa-
gation equations are given by
_D Q
P (16)
D

where PD is the state covariance matrix and QD is x_M 0

~ (22)
a gain matrix corresponding to the estimated pro-
cess model error covariance. The measurements _ M QM
P (23)
used by the filter are the three magnetometer out-
puts corresponding to Equation (13), and thus the where PM is the state covariance matrix and QM is
state update is given by a gain matrix corresponding to the estimated pro-
cess model error covariance. The sole measurement
8 9 8 9 used by this filter is the magnetometer output m*,
>
> a0new >> >
> a0 >
>
>a > > a1 > and thus the state update is given by
>
>
> 1new >>
>
>
>
>
>
>
>
>
> a >
> >
> a >
>
>
> 2new >
> >
> 2 >
> 8 9 8 9
>
> >
> >
> >
> > Snew > > S >
>
> a >
> > a >
> > > > >
< 3new
= > < 3
=   
<
bnew
= <
b
=  
w0new w0 K M m  mpred (24)
KD m ~ pred
~ m (17) > / > > / >
> w1new >
> > >
> > w1 >
> : Snew >
> ; > : S> ;
>
>
> >
> > w2 >
> >
> wSnew wS
>
>
w 2new > >
> >
> >
>
>
> z0new >> >
> z0 >
>
>
> >
> >
> >
>
>
>
> z1new >>
>
>
>
> z1 >
>
> In Equation (24), KM is the Kalman gain matrix
: ; : ; given by
z2new z2
  1 
~  and m
where m ~ pred are the actual and predicted T T
K M PM I  CM RM CM PM CM CM PM CTM R1
M
magnetometer outputs respectively, and KD is the
Kalman gain matrix given by (25)

Vol. 58, No. 3 Rogers et al.: Effective Use of Magnetometer Feedback for Smart Projectile Applications 209
where where CE is the linearized output matrix of the sys-


tem computed from Equations (11) and (12) and RE

@m @m @m @m is a user-specified gain matrix. Then, the state vec-
CM (26)
@S @b @/S @wS tor is updated according to
8 9
and RM is a user-specified gain matrix correspond-
8  9 8 9
>
>
> m1  mpred
1
>
>
>
> pred >
ing to the estimated measurement covariance. /
< new = < = / < m2  m2 >
>  =
Finally, state covariance updates are computed hnew h K E m  mpred (32)
according to :  ; : ; >
> 3 3 >
>
wnew w >
>
> h h >
>
>
 1 : m ;
T 1 wm  w
PMNew P1M C R
M M CM (27)
where the superscript (pred) indicates predicted
Note that three separate instances of this filter measurements. The state covariance matrix is
(one for each single-axis sensor) actually comprise updated according to
the nuisance parameter estimator.  1
T 1
PENew P1E CE R E CE (33)
Orientation State Estimator
The orientation estimator uses an EKF frame- Observability Analysis
work to combine feedback signals from rate gyros,
magnetometers, and the velocity estimator. The The distinguishability of each parameter in the
process model used by this filter is given by the estimation problem can be analyzed by considering
nonlinear strapdown equations, a state vector consisting of /, h, w, S, b, /S, wS, AF,
cD, and fD. Three of these states can be identified
2 3 2 38 9 8 9 immediately as indistinguishable due to the identi-
_
/^ 1 s/^ th^ c/^ th^ < xx = < xx =
6 _ 7 40 c/^ s/^ 5 xy D xy cal way in which they affect sensor outputs: /, /S,
4 h^ 5
: ; : ; and cD. As the projectile rolls, the magnetometer
_ 0 s/^ =ch^ c/^ =ch^ xz xz
w^ sensor cannot distinguish between a change in roll
(28) angle, a change in magnetometer roll angle align-
^ h;
^ w^ are the ment, and a change in magnetic field distortion
Note that in the above equations, /; angle. This is because the magnetic field is the sole
  
estimated Euler angles and xx ; xy ; xz are gyro truth source of the estimator, and other sources of
measurements. As mentioned before, signals from roll angle feedback would be required to distinguish
the rate gyroscope are used directly to drive Euler between these factors. Therefore, since a primary
angle dynamics and thus are not used for measure- goal of the system is to provide an estimate of roll
ment updates. The Jacobian of [D], omitted here for angle, /S and cD are eliminated from the estimation
brevity, is used to propagate the state covariance problem and magnetometer roll alignment angles
matrix PE according to and the model for distortion angle must be held
constant throughout flight.
_ E AE P E PE A T Q
P (29)
E E The distinguishable nature of each variable in
the estimation problem can be verified mathemati-
where AE is the Jacobian of [D] and QE a user- cally by examining the magnetometer measure-
specified gain matrix. The measurement vector, ment equation shown in Equation (13). Consider
given by the observability map of the nonlinear system [21]
8 9 which can be used to determine distinguishability
>
> m1 >>
> m >
> > of one parameter with respect to another. For
< 2=
instance, to compute whether / is distinguishable
ME m3 (30)
>
> >
> with respect to /S, the observability map would be
>
> h m >
>
: ; computed according to
wm
2 3
@m @m
is used to update the estimated state vector at a
6 @/ @/S 7
rate of 100 Hz. State and covariance updates are H6
4 @m_
7
_ 5 (34)
@m
performed by first computing the Kalman gain ma-
trix KE according to @/ @/S
  1  If H is always singular, then / is indistinguish-
K E PE I  CTE RE CE PE CTE CE PE CTE R1
E able from /S everywhere in the state space. By
computing H for each pair, indistinguishable combi-
(31) nations of the state vector can be identified.

210 Navigation Fall 2011

 observer developed here and one that does not com-
pensate for magnetic field distortion. The second
case demonstrates performance in a real-world sce-
nario including all uncertainty parameters. All
cases in this section are based on the same nominal
trajectory in which the projectile is launched with a
quadrant elevation of 28.7 deg, muzzle velocity of
860 m/s, and roll rate of 5 rad/s. For the second
example, to examine the effects of control inputs on
the projectile, 67 N control pulses of 0.1 sec dura-
tion are exerted after apogee in alternating opposite
directions in a no-roll frame every second, resulting
in angle of attack perturbations but minimal trajec-
tory alteration. These control pulses are added so
as to create realistic control-induced angle of attack
perturbations that reduce the accuracy of inertial
Fig. 8Unobservable geometries velocity-based orientation estimates. Without the
addition of these pulses, the projectile would fly a
very smooth ballistic flight path and inertial veloc-
The observability map was computed for each
ities could be used by themselves to determine
pair of variables and it was verified that /, /S, and
pitch and yaw with extreme accuracy. Such a trajec-
cD are indistinguishable from one another. All other
tory is not representative of smart weapons flight
combinations proved to be distinguishable. For non-
profiles.
linear systems, the statement that two variables
Figures 912 show trajectory time histories for
are distinguishable in general does not guarantee
both nominal cases (in which control pulses are and
local observability with respect to one another. In
are not incorporated). In Figure 10, deflection
other words, while two parameters may be observ-
occurs without control inputs due to gravitational
able in general, it may be impossible to distinguish
interaction with projectile spin. In Figure 11, con-
between them in certain vehicle orientations. For
trol inputs result in pitch and yaw angle perturba-
example, while scale factor, bias, w, and wS are in
tions on the order of a few degrees. The no-control
general observable with respect to one another,
case is overlaid with the controlled example, except
there are numerous geometries in which combina-
without the pitch and yaw angle perturbations.
tions of these quantities are indistinguishable. Fig-
Throughout this section, it is assumed that
ure 8 demonstrates two such geometries (dotted
the projectile body has similar dimensions to the
vector represents ~ s). In the leftmost diagram, two
cylinder described in the second section of the pa-
different combinations of w and wS will produce the
per (neglecting the nose cone and fins) and thus
same magnetometer output in the given field. In
similar distortion characteristics. Furthermore, it is
the rightmost diagram, two different combinations
assumed that the magnetometer is placed in the
of bias and scale factor will produce the same mag-
selected example location as also described in the
netometer output in the given field. There are a
large number of similar unobservable geometries
experienced by the projectile during a typical tra-
jectory. However, even though in some cases unob-
servable geometries can be momentarily encoun-
tered, the integrated filter continues estimation
through these points and can recover accuracy
when the geometry improves.

RESULTS
Example estimation results are shown for a 105
mm-diameter fin-stabilized projectile with mass,
axial inertia, and transverse inertia given respec-
tively by 17.606 kg, 0.0377 kg-m2, and 0.8530 kg-
m2. Two example cases are presented to demon-
strate estimator performance. The first case exem-
plifies the need to incorporate distortion effects
by comparing estimation performance between the Fig. 9Altitude vs range for both nominal trajectories

Vol. 58, No. 3 Rogers et al.: Effective Use of Magnetometer Feedback for Smart Projectile Applications 211
 Fig. 10Deflection vs range for nominal trajectories Fig. 12Pitch and yaw angle vs time for example trajectories

case and standard deviations used for the Monte

Carlo case in the next section. In addition, mag-
netic field distortion fit parameters are listed in
Table 2. Note that for the Monte Carlo simulation,
only a0, w0, and z0 were varied indicating that
actual distortion curves are likely to have the same
general shape as the nominal model obtained from
calibration, with uncertainty taking the form of a
reasonable additive offset.
Realistic values for the Earths magnetic field are
generated for all cases using the World Magnetic
Model, an industry-standard code produced jointly
by the US National Geospatial-Intelligence Agency
and the UKs Defence Geographic Centre [22]. For
the Monte Carlo cases, magnetic field vectors were
generated at random from locations between 2458
and 458 latitude. Uncertainty in the magnetic
Fig. 11Projectile roll rate vs time for nominal trajectories
field is also included in the second example simula-
tion and is described in Table 1. Note also that nor-
malized units are used for all magnetic field calcu-
second section. Performance of the integrated atti- lations, obtained by normalizing all components by
tude-nuisance parameter estimator is examined by the magnitude of the Earths field for that particu-
considering both example and Monte Carlo simula- lar case.
tion results. The following section outlines two sets Two example cases are shown in order to charac-
of example results, while the subsequent section terize the effects of nuisance parameters and spin-
describes Monte Carlo simulation results. induced field distortion as well as overall perform-
ance of the integrated estimator. The first case
examines the effect of field distortion by consider-
Example Results
ing two orientation estimators: one using a magne-
Three primary feedback sources are assumed to tometer model that neglects spin-induced distortion
provide measurement data to the integrated esti- (based on Equation (10)), and the other using the
mator: a three-axis magnetometer aligned with the estimator developed here (based on Equation (13)).
projectile body axes (modeled as three single-axis In order to isolate the effect of field distortion, both
magnetometers), a three-axis gyroscope aligned estimators are assumed to have perfect knowledge
with the body axes, and an inertial velocity estima- of magnetometer nuisance parameters while the
tor. Feedback errors associated with each measure- latter estimator is assumed to also have perfect
ment source are listed in Table 1, as are their ran- knowledge of field distortion as a function of spin.
domly-generated values for the second example Furthermore, in both cases, gyro and inertial veloc-

212 Navigation Fall 2011

 Table 1Feedback Error Parameters
Std. Deviation for
Feedback Source Description Parameter Example Case 2 Value MC Simulations

Magnetometer Scale Factor S1, S2, S3 (nd) 0.998, 1.033, 0.990 0.03
Bias b1, b2, b3 (norm. 0.236, 20.017, 20.212 0.7 Gauss
units)
/S Alignment Angle /S1, /S2, /S3 (deg) 2.48, 0.430, 91.4 1.0
wS Alignment Angle wS1, wS2, wS3 (deg) 0.905, 89.28, 89.88 1.0
Noise Std. Dev. n (norm. units) 0.01 N/A

Gyros Scale Factor SGx, SGy, SGz, (nd) 1.052, 0.991, 0.993 0.03
Bias bp, bq, br (rad/s) 1.529, 20.787, 1.007 5.0, 1.0, 1.0
xx Cross Axis cpq, cpr (nd) 20.0038, 0.0072 0.01
xy Cross Axis cqp, cqr (nd) 0.0016, 0.0087 0.01
xz Cross Axis crp, crq (nd) 0.0190, 0.0056 0.01
Noise Std. Dev. np, nr, nq (rad/s) 0.05 N/A

Velocity Estimator Bias (m/s) bvx, bvy, bvz, (m/s) 0.95, 20.60, 3.40 1.52, 1.52, 3.40
Noise Std. Dev. nv (m/s) 3.05 N/A

Earths Magnetic Field N, E, D mN, mE, mD (norm. 0.362, 20.054, 20.930 N/A
Field components units)
Errors in Knowledge Dmmag (norm. 20.045 0.03
Magnitude of Field units)
Errors in Knowledge Dmaz, Dmel (deg) 20.246, 20.120 0.2
of Az. and El. of
Field

Table 2Magnetic Field Distortion Curve Fit Parameters

Parameter Value Parameter Value Parameter Value
23 22
a0 23.1110 w0 21.4310 z0 21.581023
a1 8.1231025 w1 8.0431023 z1 25.1331024
a2 2.4631025 w2 21.4231025 z2 3.4631026
a3 27.9931028
Initial a0 21.0231022 Initial w0 (Example 2) 23.8731022 Initial z0 (Example 2) 23.6731023
(Example 2)
Init. a0 Std. Dev. 9.3431023 Init. w0 Std. Dev. 4.2931022 Init. z0 Std. Dev. 4.7331023
(Monte Carlo) (Monte Carlo) (Monte Carlo)

ity feedback is assumed to be perfect and the pitch angle estimation error on the order of a few
Earths magnetic field is assumed to be known degrees occurs when distortion is not accounted for.
exactly. The purpose of this example case is to dem- The integrated estimator developed here is able to
onstrate expected estimation error for magnetome- properly account for all distortion effects and thus
ter-based observers that neglect spin-induced dis- demonstrates estimation error near zero for all
tortion. For all example cases in this section, the angles. Note that this example case represents an
projectile is launched at an azimuth angle of ideal scenario in which all nuisance parameters
approximately 43 deg West of North and estimation were assumed to be zero and other sensor feedback
commences 3 sec into the trajectory at a rate of was assumed to be perfect. In a realistic environ-
100 Hz. ment, roll angle estimation error for an observer
Figures 13 and 14 show roll, pitch, and yaw esti- that does not account for distortion would likely be
mation error exhibited by both estimators for this even worse than demonstrated here. Furthermore,
example case. Figure 13 demonstrates that the esti- note that typical magnetometer nuisance parame-
mator that does not account for magnetic field dis- ters (scale factor, bias, and misalignment) cannot
tortion suffers roll estimation errors consistently properly compensate for magnetic field distortion
between 20 and 40 deg, while Figure 14 shows that effects since these appear largely as a transverse

Vol. 58, No. 3 Rogers et al.: Effective Use of Magnetometer Feedback for Smart Projectile Applications 213
 Fig. 13Roll estimation error vs time for example case 1 Fig. 15/ Estimation error vs time for second example case

Fig. 14Pitch and yaw estimation errors vs time for example

Fig. 16h Estimation error vs time for second example case
case 1

distortion angle (i.e., cD is the dominant effect). The updates to their respective states at a rate of 100
only parameter that could account for such an Hz. In Figures 1517, this case is referred to as
effect is the magnetometer roll alignment angle, /S, Using NP Estimation.
which was shown to be unobservable and thus This estimation case is compared to a case in
removed from the estimation problem. The example which the nuisance parameter and distortion pa-
results shown here demonstrate that incorporation rameter filters are not activated. In this case, orien-
of spin-induced distortion effects is critical to ensur- tation estimation begins at 3 sec and nuisance pa-
ing estimation accuracy for magnetometer-based fil- rameters and distortion fit parameters are held
ters in rapidly-spinning flight bodies. constant at their initial values (referred to in Fig-
The second example demonstrates estimator per- ures 1522 as Not Using NP Estimation). Figures
formance in the presence of reasonable sensor 1522 show comparisons of filter performance for
errors and uncertainty in the Earths magnetic these two trajectories. Figures 15, 16, and 17 show
field. Error and uncertainty parameters for this time histories of / estimation error, h estimation
example case are listed in Tables 1 and 2. As in the error, and w estimation error, respectively. For the
first example, orientation estimation begins at 3 sec case in which nuisance parameter estimation was
into the trajectory using nominal guesses for mag- used, Figure 18 shows a time history of the differ-
netometer nuisance parameters. Then, at 6 sec ence between measured and predicted magnetome-
both the nuisance parameter estimator and the dis- ter outputs for each sensor, while Figures 19, 20,
tortion parameter estimator begin to provide and 21 show time histories of scale factor estima-

214 Navigation Fall 2011

 Fig. 17w Estimation error vs time for second example case

Fig. 19Scale factor estimation error vs time for second example

case

Fig. 18Difference between measured and predicted magnetome-

ter output for second example case

tion error, bias estimation error, and wS alignment

angle estimation error for each magnetometer, Fig. 20Bias estimation error vs time for second example case
respectively. In Figures 1921, the dotted lines rep-
resent 1r error covariance estimates output from
the Kalman filter. Finally, Figure 22 shows a time
history of percent error in AF and fD estimates. to less than 7 deg, maximum error in h is reduced
Figures 1522 demonstrate very interesting, and from about 8 deg to less than 4 deg, and maximum
to a certain extent, unexpected phenomena when error in w is reduced from about 5 deg to less than
nuisance and distortion parameter estimation is 3 deg (before control activation). This error reduc-
performed. First, Figures 1517 demonstrate that tion, especially in roll angle estimation, likely
attitude estimation performance is greatly would have considerable positive impact on control
enhanced after nuisance and distortion parameter performance.
estimation is activated (after 6 sec). Specifically, Figure 18 demonstrates that, as expected, after
maximum error in / is reduced from about 20 deg nuisance and distortion parameter estimation

Vol. 58, No. 3 Rogers et al.: Effective Use of Magnetometer Feedback for Smart Projectile Applications 215
 each estimated nuisance parameter is mathemati-
cally observable, in many flight scenarios (such as
the example shown here) many combinations of
nuisance parameters will lead to similar signal out-
puts. Thus, although nuisance parameter estimates
might converge to incorrect values, magnetometer
signals may more closely reflect predicted measure-
ments. As shown in Figures 1921, by the end of
the trajectory 1r covariance estimates are small for
both scale factor and wS because the observed and
predicted magnetometer outputs match reasonably
well even though some nuisance parameters con-
verged to incorrect values.
Figure 22 shows that error in AF and fD actually
grew after the distortion parameter estimator was
activated. Again this is the result of poor observ-
ability between AF, fD, scale factor, and errors in
knowledge of the magnetic field (which, in this
case, represent errors in a truth source and cannot
be compensated for). Thus, errors in knowledge of
the magnetic field and magnetometer nuisance pa-
Fig. 21wS Estimation error vs time for second example case
rameters can contaminate attenuation factor esti-
mates and again cause convergence to the wrong
value.
The most important result demonstrated by this
example case is that, even if nuisance parameters
and distortion fit parameters converge to incorrect
values, attitude estimation performance almost uni-
versally improves. This is because, if nuisance and
distortion parameters are fixed, the orientation es-
timator is forced to alter /, h, and w when discrep-
ancies are observed between predicted and actual
magnetometer outputs. When nuisance and distor-
tion parameter estimation is incorporated, the inte-
grated filter can tune these parameters such that
predicted data is closer to measured data, thus
removing the burden on the orientation filter. Since
the primary goal of the integrated estimator is to
obtain estimates for /, h, and w, nuisance and dis-
tortion parameter estimation pays off in spite of the
tendency for some of these parameters to display
Fig. 22Percent error in distortion parameters vs time for second poor observability and converge to incorrect values.
example case

Monte Carlo Simulation

begins, the error between measured and predicted To demonstrate that orientation estimation per-
magnetometer outputs decreases significantly. This formance invariably improves when nuisance and
is the result of the nuisance parameter and distor- distortion parameter estimation is incorporated, a
tion parameter estimators tuning their respective Monte Carlo simulation is performed. The con-
values so that the measured data matches pre- trolled example trajectory was run 1,000 times, and
dicted data. However, while Figure 20 demon- in each case magnetometer, gyro, and velocity feed-
strates that the filter accurately estimated all mag- back error parameters were perturbed according to
netometer biases, scale factor and wS converged to the standard deviations shown in Tables 1 and 2.
incorrect values for each sensor. This type of behav- For each case, the Earths magnetic field vector was
ior was observed throughout many example cases, selected at random from a location between 2458
and is the result of poor local observability condi- and 458 latitude using the World Magnetic Model,
tions between these two parameters. As described and launch azimuth angle with respect to North
in the Observability Analysis Section, although was varied uniformly between 0 and 360 deg. The

216 Navigation Fall 2011

Fig. 23Histogram of RMS errors in / estimation (1,000 total Fig. 25Histogram of RMS errors in w estimation (1,000 total
cases) cases)

tortion parameter estimation performs better than

initial guess for pitch angle was 0.46 rad in each
the filter without it.
case, while the initial guess for roll angle was var-
Tables 3 and 4 provide a statistical summary of
ied uniformly between 0 and 2p rad. Two cases
the Monte Carlo results (No NP denotes that no
were run for each set of initial conditions: one case
nuisance and distortion parameter estimation was
in which nuisance and distortion parameter estima-
used; NP denotes that nuisance and distortion pa-
tion was performed, and one case in which nuisance
rameter estimation was used). Note that, for esti-
and distortion parameter estimation was not. For
mation of each Euler angle, all statistical measures
each trajectory, RMS errors were calculated for /,
show significant improvement when nuisance and
h, and w estimation.
distortion parameter estimation is incorporated. In
Results of the Monte Carlo simulation are shown
Table 3, mean RMS errors for all angles show
in Figures 2325. These figures show histograms of
greater than 50% reduction and error variances for
RMS estimation error for /, h, and w, respectively,
all angles show greater than 75% reduction when
for the cases in which nuisance and distortion pa-
nuisance and distortion parameter estimation is
rameter estimation is and is not included in the
performed. The confidence intervals shown in Table
estimation process. All figures clearly demonstrate
4 also clearly demonstrate that estimation of all
that the filter that incorporates nuisance and dis-
angles improves considerably when parameter esti-
mation is incorporated. However, note that in most
cases nuisance parameter convergence to incorrect
values occurred to some extent due to poor observ-
ability scenarios. Thus the primary result demon-
strated by this Monte Carlo example is that, in gen-
eral, online magnetometer calibration results in
improved estimation even though convergence of
parameters to their true values is not guaranteed,
since the filter can tune alternative parameters in
the measurement equation without having to dis-
rupt Euler angle estimates.

Table 3Mean and Variance of Monte Carlo Results

Mean (deg) Variance (deg2)
No NP NP % Reduction No NP NP % Reduction

/ 20.26 9.89 51.2 220.8 50.7 77.0

h 3.97 1.54 61.2 5.80 0.21 96.4
Fig. 24Histogram of RMS errors in h estimation (1,000 total w 3.58 1.70 52.5 5.06 0.19 96.3
cases)

Vol. 58, No. 3 Rogers et al.: Effective Use of Magnetometer Feedback for Smart Projectile Applications 217
 Table 4Confidence Intervals of Monte Carlo Results (Results in Degrees)
50% Confidence Interval 75% Confidence Interval 90% Confidence Interval
No NP NP % Red. No NP NP % Red. No NP NP % Red.

/ 16.06 7.81 51.4 23.51 13.11 44.2 38.92 19.00 51.2

h 3.34 1.48 55.7 5.17 1.79 65.4 7.04 2.13 69.7
w 2.88 1.66 42.4 4.52 1.96 46.4 6.70 2.25 66.4

CONCLUSION 6. Bar-Itzhak, I. Y. and Idan, M., Recursive Attitude

Determination from Vector Observations: Euler Angle
The use of magnetometers for orientation estima- Estimation, Journal of Guidance, Control, and
tion on-board projectiles is considered. In particu- Dynamics, Vol. 10, No. 2, MarchApril 1987, pp. 152
lar, the effect of spin-induced magnetic field distor- 157.
tion is discussed. Analytical, finite element, and ex- 7. Mortari, D., Euler-q Algorithm for Attitude Determi-
perimental techniques are used to characterize nation from Vector Observations, Journal of Guid-
magnetic field distortion effects for typical projectile ance, Control, and Dynamics, Vol. 21, No. 2, March
configurations. Then, an estimation algorithm is April 1998, pp. 328334.
8. Psiaki, M. L., Global Magnetometer-Based Space-
developed that demonstrates how compensation for
craft Attitude and Rate Estimation, Journal of Guid-
field distortion effects can be performed. The esti- ance, Control, and Dynamics, Vol. 27, No. 2, March
mation algorithm is responsible for orientation esti- April 2004, pp. 240250.
mation as well as distortion and magnetometer nui- 9. Ohlmeyer, E., Fraysse, J., and Pepitone, T. R.,
sance parameter estimation, and consists of three Guidance, Navigation, and Control Without Gyros:
coupled Extended Kalman filters. An analytical A Gun-Launched Munition Concept, AIAA Guid-
observability analysis is then performed to identify ance, Navigation, and Control Exhibit, Monterey, CA,
globally indistinguishable states, which are then re- August 58, 2002, Paper AIAA-20025025.
moved from the estimation problem. An example sim- 10. Wilson, M. J., Onboard Attitude Determination for
ulation demonstrates that standard magnetometer- Gun-Launched Projectiles, 43rd AIAA Aerospace Sci-
based observers that do not compensate for spin- ences Meeting and Exhibit, Reno, NV, January 10
13, 2005, Paper AIAA-2005-1217.
induced distortion are likely to suffer significant
11. Lee, H., Kim, K., and Park, H., Roll Estimation of a
estimation error. Finally, a Monte Carlo simulation Smart Munition Using a Magnetometer Based on an
directly compares estimators with and without Unscented Kalman Filter, AIAA Guidance, Naviga-
nuisance and distortion parameter estimation capa- tion, and Control Exhibit, Honolulu, HI, August 18
bility, and it is shown that parameter estimation 21, 2008, Paper AIAA-2008-7460.
enables significant attitude estimator performance 12. Harkins, T., On the Viability of Magnetometer-Based
increases. Projectile Orientation Measurements, US Army
Research Laboratory Technical Report ARL-TR-4310,
November 2007.
13. Lerner, G. M. and Shuster, M. D., In-Flight Magne-
REFERENCES
tometer Calibration and Attitude Determination for
1. Wahba, G., Problem 651: A Least Squares Estimate Near-Earth Spacecraft, Journal of Guidance and
of Satellite Attitude, SIAM Review, Vol. 7, No. 3, Control, Vol. 4, No. 5, SeptemberOctober 1981, pp.
July 1965, p. 409. 518522.
2. Wahba, G., et al., Problem 65-1 (Solution), SIAM 14. Alonso, R. and Shuster, M. D., Complete Linear
Review, Vol. 8, August 1966, pp. 384386. Attitude-Independent Magnetometer Calibration,
3. Shuster, M. D. and Oh, S. D., Three-Axis Attitude Journal of the Astronautical Sciences, Vol. 50, No. 4,
Determination from Vector Observations, Journal of OctoberDecember 2002, pp. 477490.
Guidance and Control, Vol. 4, No. 1, Jan.Feb. 1981, 15. Alonso, R. and Shuster, M. D., Centering and Observ-
pp. 7077. ability in Attitude-Independent Magnetometer-Bias
4. Bar-Itzhak, I. Y. and Reiner, J., Recursive Attitude Determination, Journal of the Astronautical Sciences,
Determination from Vector Observations: Direction Vol. 51, No. 2, AprilJune 2003, pp. 133141.
Cosine Matrix Identification, Journal of Guidance, 16. Alonso, R. and Shuster, M. D., Attitude-Independent
Control, and Dynamics, Vol. 7, No. 1, Jan.Feb. 1984, Magnetometer-Bias Determination A Survey,
pp. 5156. Journal of the Astronautical Sciences, Vol. 50, No. 4,
5. Bar-Itzhak, I. Y. and Oshman, Y., Attitude Determi- OctoberDecember 2002, pp. 453475.
nation from Vector Observations: Quaternion Estima- 17. Crassidis, J. L., Lai, K., and Harman, R. R., Real-Time
tion, IEEE Transactions on Aerospace and Electronic Attitude-Independent Three-Axis Magnetometer Cali-
Systems, Vol. AES-21, No. 1, January 1985, pp. 128 bration, Journal of Guidance, Control, and Dynamics,
136. Vol. 28, No. 1, JanuaryFebruary 2005, pp. 115120.

218 Navigation Fall 2011

18. Gebre-Egziabher, D., Magnetometer Autocalibration netic Conducting Cylindrical Shells, IEEE Transac-
Leveraging Measurement Locus Constraints, Journal of tions on Electromagnetic Compatability, Vol. 51, No.
Aircraft, Vol. 44, No. 4, JulyAugust 2007, pp. 13611368. 3, August 2009, pp. 720724.
19. Perry, M. P. and Jones, T. B., Eddy Current Induc- 21. Nijmeijer, H. and van der Schaft, A. J., Nonlinear
tion in a Solid Conducting Cylinder with a Trans- Dynamical Control Systems, Springer-Verlag, New
verse Magnetic Field, IEEE Transactions on Mag- York, 1990, pp. 9596.
netics, Vol. MAG-14, No. 4, July 1978, pp. 227232. 22. US National Geospatial-Intelligence Agency, Techni-
20. Ziolkowski, M. and Gratkowski, S. R., Shielding cal Document for the World Magnetic Model (WMM),
from External Magnetic Fields by Rotating Nonmag- Released February 3, 2010.

Vol. 58, No. 3 Rogers et al.: Effective Use of Magnetometer Feedback for Smart Projectile Applications 219