Professional Documents
Culture Documents
Systems Analysis
INERTIAL NAVIGATION
SYSTEMS ANALYSIS
K E N N E T H R. B R I T T I N G , Sc. D.
Lecturer in Aeronautics and Astronautics
Measurement S y s t e m Laboratory
Massachusetts Institute of techno loo?^
Copyright
To the memory of
KATHERINE ANNE
Foreword
viii
FOREWORD
WALTERWRIGLEY,SC.D.
Professor of Instrumentation and Astronautics
Educational Director, Charles Stark Draper Laboratory
Massachusetts Institute of Technology
Preface
Part of this book evolved from a set of lecture notes prepared in the Department of Aeronautics and Astronautics a t the Massachusetts Institute of
Technology for a one-semester graduate course in inertial navigation systems.
In addition, a portion of the book was adapted from my doctoral thesis.
Since the lecture notes were prepared for students who had already completed a comprehensive introductory course in classical mechanics, kinematics, inertial instrument theory, and inertial platform mechanization, a
fairly advanced level of preparation is assumed. Nevertheless, i t has been
found through experience that the book is reasonably self-contained, allowing the student to follow the development with a modicum of referral t o
the technical literature.
The material is intended principally for the avionics system engineer who
wishes to compare the performance of the various types of system mechanizations. While i t is applicable t o spacecraft and undersea navigation, the
thrust of the book is aimed at terrestrial applications on, or slightly above,
the surface of the earth. Because of the current interest in navigation
systems for aircraft, the relevant navigation equations are developed for this
application.
Perturbation techniques are extensively used t o develop linearized system
equations whose solutions closely approximate those obtained by a solution
of the nonlinear differential equations. Since linear systems theory is applicable to linearized system equations, these equations are quite amenable
to physical interpretation, providing an insight into system behavior not
readily obtainable from computer solution of the nonlinear equations. The
developed linear system models are, of course, essential to the application
of optimal filtering techniques which are currently being applied to aided
inertial systems.
Chapter 1 emphasizes concepts common to all inertial navigation system
PREFACE
PREFACE
xi
Contents
I Introduction
The Concept of Inertial Navigation, 1
Types of Inertial Navigation Systems, 3
A Critique of Previous Analysis Techniques, 4
A Unified Approach to the Error Analysis, 7
2
Notational Conventions, 12
The Time Derivative of the Direction Cosine Matrix, 16
Column Matrix Time Derivatives, 17
Analogies to Vector Analysis, 18
Perturbation Techniques, 20
Symbology, 24
3
Reference Frames
Inertial Frame, 30
Geographic Frame, 33
Earth Frame, 34
Geocentric Frame, 34
Body Frame, 34
Tangent Frame, 35
Reference Frame Relationships, 35
Platform, Accelerometer, and Gyro Frames, 38
4
xiv
CONTENTS
5
5.1
5.2
5.3
5.4
Principle of Operation, 65
Dynamic Model for SDE" Gyro, 69
Uncertainty Torque Compensation, 74
Instrument and System Redundancy and Reliability, 75
6
Self-Alignment Techniques
System Description, 2 17
Derivation of System Differential Equations, 219
Solution of System Differential Equations, 220
Approximations to the Solutions, 224
Development of an Error Model, 227
CONTENTS
Appendix B
B.l
33.2
B.3
B.4
xv
229
235
24 1
Index
245
Inertial Navigation
Systems Analysis
CHAPTER
7
Introduction
"If a man will begin with certainties, he shall end in doubts;
but if he will be content to begin with doubts he shall end
in certainties. "
FRANC1S BACON
1.1
T H E CONCEPT O F INERTIAL N A V I G A T I O N
INTRODUCTION
1.2
TYPES OF INERTIAL N A V I G A T I O N S Y S T E M S
INTRODUCTION
inertially nonrotating coordinate system is commonly referred to as a spacestabilized inertial navigation system (SSINS). Chapter 6 is devoted to a
detailed discussion and error analysis of the SSINS.
The other class of instrumented coordinate frames which are in general use
in current navigation systems comprises the local level coordinate frames.
I n these frame mechanizations the sensitive axes of two accelerometers and
two gyros are constrained to lie in the local horizontal plane. Because the
accelerometers' sensitive axes lie in the local horizontal plane, explicit calculation of the gravitational field vector is avoided. Semianalytic systems that
physically instrument local-level coordinate frames are called local-level
inertial navigation systems (LLINS). These systems are discussed in Chapter 7.
Analytic inertial navigation systems do not physically instrument a
reference frame but rather use the gyro outputs to calculate analytically the
relative orientation between the system's initial and present state. This type
of system mechanization is commonly called a strapdown inertial navigation
system (SDINS). Strapdown system design is currently a subject of intense
technical interest since computer limitations have until quite recently limited
their application. The lack of gimbal structure in strapdown systems affords
a reduction in system size, weight, power consumption, and, potentially,
cost. Because the strapdown system imposes a very large dynamic range on
the inertial instruments, i t does not yet challenge the two types of semianalytic systems from an accuracy standpoint. Recent advances in component development may soon overcome this disadvantage. Strapdown
systems are discussed in Chapter 8.
1.3
A CRITIQUE
The principles of inertial navigation, which have been well understood for
many years, possess a simplicity and elegance that border on the sublime.
Consider that a properly initialized inertial navigation system can completely
specify the position, velocity, and attitude of an arbitrarily moving vehicle
via a simple measurement of force, providing that the force measurement is
made in a known reference frame and gravitational field. It is somewhat
enigmatic, therefore, that it is necessary to use rather sophisticated mathematical techniques and notation in order to describe the dynamic behavior
of these systems. On the other hand, it should be recognized that the complete specification of the dynamic behavior of an inertial navigator requires
a ninth-order differential equation since there are three degrees of rotational
freedom and the three translational equations are second order. While this
book advances no panacea in regard to the problems of notation and algebraic
INTRODUCTION
1.4
U N I F I E D A P P R O A C H TO T H E E R R O R A N A L Y S I S
INTRODUCTION
A U N I F I E D A P P R O A C H TO T H E ERROR ANALYSIS
where (eihT,FE,
~ =
)
(eN, e,,
d L = latitude error
61 = terrestrial longitude error
ah, = inertially computed alt,itude error
The attitude error will be defined as the orthogonal transformation error between platform and geographic axes, coordinatized in geographic axes. Note
t.hat since this definition corresponds to the transformation error involved in
resolving a physical measurement made on the platform into geographic axes,
the attitude errors do not, in general, correspond t o the platform error angles.
n =2,3;
all
10
INTRODUCTION
where f
?a =
For the purposes of the latitude, longitude, and altitude calculations, the
estimator was of the same form as above:
? = (?a)a(fi)l-a,
all a
CHAPTER
2
M a t h e m a t i c a l Notation and
Techniques
"A good notation has a subtlety and suggestiveness'which a t
times make i t seem almost like a live teacher."
BERTRAND RUSSELL
The notation used in this book has evolved from the vector notation of
Wrigley and Hollister,'l from the work of Broxmeyer12 a t the Draper Laboratory a t M.I.T., and from the lively discussions the author has had with his
colleagues and with students enrolled in the study of inertial systems a t
M.I.T.
Since the purpose of this book is to describe the operation of systems as
opposed t o components, a matrix method of notation seems most suitable.
Matrix notation lends itself to very concise formulations which, for the threedimensional navigational problem, are almost essential if the necessary
algebraic manipulations are to be carried out. I n addition i t is recognized
that in current inertial system design, a digital computer which deals with
scalars and not vectors is relied on t o perform the necessary calculations. For
example the computed inertially referenced geocentric position vector exists
as three numbers in the computer: P,, FY, and P,, which represent the estimated components of the actual vector in the inertial coordinate frame.
These three components can be grouped into a vector array, denoted as fi,
such t h a t matrix manipulations can be performed. The grouping of these
three scalars into vector, or more precisely, column matrix form is merely
a matter of mathematical convenience and carries with i t no philosophical
import.
Although the computer is programmed to carry out arithmetic operations
based on an assumed reference coordinate frame, the computer knows
nothing of reference frames and merely carries out a series of scalar operations. A "computed" reference frame55is sometimes introduced which "is a n
12
2.1
NOTATIONAL CONVENTIONS
Vectors
2.1.2
Column Matrices
Note that when the components are explicitly stated in an array, as above,
the frame denoting superscript is omitted since the frame reference is denoted
by the subscript of the components. I n the example x, y, z refers to the
components in an inertial frame (see Section 3.1). The heavy braces { )
NOTATIONAL CONVENTIONS
13
and rour form are introduced t o save space in the written text and should be
distinguished' from a row matrix.
Besides vectors, any appropriate group of quantities can be arranged in
column matrix form such that matrix algebra techniques can be used.
Example
6n
Coordinate Transformations
Note the cancellation feature of this notation takes into account t h a t the
order of transformation is important, t h a t is, c;c,"rb # CrC;rb.
As indicated by its name, the DCM is an array of direction cosines :
where, for this particular case, cj, = direction cosine between the j t h axis in
the i frame and the kth axis in the b frame. If each of the two frames is
orthogonal then the inverse of the DCM is equal t o its transpose where the
transpose of a matrix quantity is indicated by a superscript T.
Example
c; = (CqlT
14
Similarity Transformations
Angular Velocity
wpb= ( y R 0
my)
where w:b = angular velocity of the b frame relative t o the i frame coordinatized in the b frame.
Because they are vector quantities, angular velocities follow the usual rules
of vector addition. If rotations are occurring between a number of coordinate
frames, the subscript notation facilitates the statement of the mathematical
relationship.
Example
NOTATIONAL CONVENTIONS
15
~ t is. seen that in this case the notation allows the inner subscript indices t o
"cancel." Changing the direction of the rotation vector merely reverses the
order of the subscript indices.
Example
-aib
= Wbi
a:,= c;npbcp
2.1.6
As mentioned in the introduction to this chapter it is necessary to distinguish between physical vectors and arrays of scalar quantities in the
computer. It is also convenient to denote quantities measured by the
instruments in a special manner.
A quantity t h a t is measured by the instruments is denoted by a "tilde,"
(- 1Example
&pb
(G,, o?,,
o?,)
where &ib, is the array of three rate gyro outputs in a strapdown inertial
navigation system.
A quantity which is computed on the basis of the instrument measurements
and on other geometric considerations is denoted by a "hat," ( A ) .
Example
A
w z =
A
(jcos L, -L -A
sin
(see Eq. 3-8) where oi",is the array of the three geographic components of the
angular velocity of the n frame with respect to the i frame.
16
2.2
T H E T I M E DERIVATIVE OF T H E DIRECTION
C O S I N E MATRIX
C: = lim
+ At) - C:(t)
AC:
c:(t
- = lim
heb=
hey
[-AOp
-AOzO
--he,
AO.;
Ae,
sin
he,,
Ic
R,P , Y
where heR, AOp, AOy are the "small" rotation angles through which the b
frame has rotated in time At about its positive roll, pitch, and yaw axes.
Note t h a t because t h e rotation angles are small in the limit as At --t 0, smallangle approximations are valid and the order of rotation is immaterial.
Substituting Eq. 2-2 into Eq. 2-1 shows that
..
c; = C:(t)
neb
lim at+o
At
But in the limit as At -+0, Aeb/At is the skew-symmetric form of the vector
angular velocity of the b frame relative to the i frame during time At. Note
that because of the limiting process the angular velocity can also be referenced t o the i frame. Thus
Aeb
lim - =
A t - 0 At
17
+ At)
R(t 4- At 1
F i g u r e 2.1
cbi- c@p,
where from Section 2.1.5
2.3
COLUMN M A T R I X T l M E D E R I V A T I V E S
18
where the product rule of differential calculus has been invoked. But from
thus Eq. 2-5 becomes
Eq. 2-4, c', = CLP:,;
Equation 2-6 is the matrix formulation of the familiar Coriolis law of classical
mechanics.
A second time differentiation of Eq. 2-6 yields an expression for the inertially referenced acceleration as a function of geographically referenced
quantities :
i;"
CCl,(pn
2QIn,in
hpnrn P Z P L P )
(2-7)
Note the presence of the familiar Coriolis, tangential, and centripetal acceleration terms in Eq. 2-7.
Observe that vectors must be written in column matrix form before
derivatives can be taken since in the notation of this book the symbol i.
has no meaning. This requirement would be a serious constraint if general
vector relationships, independent of coordinate frames, were to be developed.
For the purposes of inertial system design and analysis, however, the coordinatization requirement is not in the least constraining since it is impossible
to make physical measurements without referring these measurements to a
coordinate frame.
2.4
2.4.1
Dot Product
The dot or inner product of two column matrices is constructed by transposing one of the column matrices and performing an ordinary matrix
multiplication. The order of multiplication is unimportant. Thus
a-b=aTb=bTa
Example
19
Note that the product (Pi) (Pi)T, the dyadic product, also has definition in
with the rules of matrix multiplication.
I n general, the individual matrices in any matrix product must be conformable, that is, the number of columns in the first matrix must equal the
,umber of rows in the second.
2.4.2
Cross Product
SZ&rb =
where I
2.4.3
identity matrix.
(2-12)
(2-13)
Note that parentheses are not needed on the left-hand side of the equation
since A(Bd) = (AB)d. As a consequence the second triple product of vector
20
analysis is written as
(a x b) x c = (Ab)*d
(2-14)
(2-15)
2.5
PERTURBATION TECHNIQUES
v=v+6v
A
where v
computed velocity
true velocity
dv
When substitutions of the type above are made for dependent variables
in the nonlinear differential equations and products of error quantities are
neglected, linear differential equations involving only the error quantities
emerge. These error equations, which may be timevarying, are simpler in
form than the original differential equations and are analytically more tractable. It is within the framework of this philosophy that products of the error
variables and other "small" quantities such as the earth's ellipticity and
higher order terms in the gravitational field equations will be negligibly
small and consequently will not appear in the final error equations. Computer
solution of the nonlinear system equations and direct analytical solution of
the navigator equations under certain restrictive conditions (see Appendix
A) have confirmed the validity of the linearized approach. It is t h e author's
experience t h a t perturbation analysis is to be preferred over direct computer
studies because of the insight gained into the system behavior by examining
only the linear error response. However, i t must be kept in mind t h a t the
linearized solutions are only valid for "small" perturbations around the true
solutions such that extrapolations are not made and conclusions not drawn
outside of the region of validity.
2.5.1
21
PERTURBATION TECHNIQUES
e,".
v,
62 cos L ,
v,
-dL,
and
vD
==
-62 sin L
The parameters VN, vE, vD are interpreted as the error angles about the
positive north, east, and down directions, respectively, which account for the
transformation error in 63.
22
a:b
e: = c; + sc;
(2-19)
e:
If
is not explicitly orthogonalized, then it follows that 6 ~ will
: in general be
nonorthogonal and of the general form :
I n order to maintain the very desirable manipulative and interpretive propcan be orthogonalized using the formula
erties of orthogonal matrices,
et
(e;),
The bracketed term on the right-hand side of Eq. 2-21 is seen t o be skew
symmetric since i t involves the difference between a matrix and its transpose.
Unfortunately, there are no general rules which can be applied in determining the square root of a matrix. Equation 2-21 is a possible algorithm if i t is
used frequently enough such that 6 ~ satisfies
;
a "small" quantity criterion.
Equation 2-21 can be written:
where
23
PERTURBATION TECHNIQUES
~t is seen from t h e expressions above for pX,P,, and Bz that since, in general,
the elements of C; (the cij; i = 1, 2, 3 ; j = 1, 2, 3) are time varying it is
inconvenient to associate the /3's with the elements of 6~:. Fortunately, for
the purposes of this development it is sufficient t o work directly with the
error angles, P,, P,, and P,.
2.5.1.3
Nonorthogonal Transformations. I n this case a transformation is computed between a nonorthogonal and orthogonal set of axes
such as the accelerometer-platform transformation of Eq. 3-34 :
I-
=I -
(AC:)~
+ 8(ACa)T
P
(2-24)
where
0
s(Ac:)T
-30x2
w,
&,
As expected six independent error angles are necessary t o specify the transformation error for a nonorthogonal transformation. Since in Eq. 2-24 the
~ desired transformation matrix, it is again rnathequantity I - ( A C ~ is) the
matically convenient to substitute Eq. 2-24 into Eq. 2-23 and postfactor the
desired transformation matrix, yielding
24
2.6
SYMBOLOGY
2.6.1
Coordinate Frames
The following is a list of the coordinate frames defined and used in this
document.
Frame
Sub/Superscript
Components
Location
Inertial (absolute)
Inertial (operational)
Geographic
Earth
Geocentric
Body
Tangent
Platform
Accelerometer
Gyro
Gyro housing (case)
Gyro float
Mechanized
Computation
Note. I n the writing of the final gyro error equations in Chapter 8, the
error components are designated with x, y, z subscripts.
I n the following, the symbol "( )" is used to designate a general quantity
to which the symbology in question is applied.
25
SYMBOLOGY
-
Description
Symbol
Location
2.6.3
It is frequently necessary to describe the relationship between the coordinate frames which are almost coincident such as described in Section
2.5.1.1. This relationship can be equivalently described either as a skewsymmetric matrix or as a "small-" angle rotation vector. The general notational procedure will be t o designate the skew-symmetric matrix by a n
uppercase symbol and the rotation vector by a lowercase symbol. The rotations are defined to occur about the positive axes of the coordinate frame
in which the rotation takes place.
Transformation
Rotation Vector
Location
26
2.6.4
Symbol L i s t
Symbol
Description
Accelerometer scale factor error matrix
Scale factor error of accelerometer k
Gyro float output axis rotation
Magnetic field strength
Accelerometer bias error
Six element matrix relating nonorthogonal
instrument axes to platform axes
Gyro viscous damping coefficient
Coordinate transformation relating the
subscript axes, j, t o the superscript
axes, k
Nonorthogonal transformation
Mass density of earth
Deviation of the normal
Deviation of the normal a t the point on the
earth directly under the system's location
Earth's ellipticity
Specific force exerted on instrument set
Accelerometer measurement uncertainty
Gravitational acceleration due to the earth
Gravitational acceleration due t o the kth
body in universe
Gravitational acceleration a t earth's center
of mass due t o the kth body in universe
Radial component of G
Colatitude component of G
Gravitational acceleration associated with
reference ellipsoid
Deviation of gravitational field from t h a t
associated with reference ellipsoid
Gravity acceleration due t o earth
Gravity magnitude associated with reference
ellipsoid
Gravity anomoly
Gyro float angular momentum
Altitude above the reference ellipsoid
Altitude based on inertial computations
Error in h
Location
SYMBOLOGY
Symbol
Description
Identity matrix (1's on diagonal, 0's
elsewhere)
Principal moment of inertia of gyro float
Experimentally determined gravitational
field coefficients
Gyro compliance along the j t h axis due to a
force acting along the kth axis.
Gyro torquing sensitivity
Uncertainty in knowlege of K t ,
Equatorial gravitational field constant
Polar gravitational field constant
Eccentricity of reference ellipsoid
Sensitivity vector
Sensitivity vector
Geographic latitude
Initial geographic latitude
Geocentric latitude
Geocentric latitude a t the point on the earth
directly below the system
Terrestrial longitude
Initial terrestrial longitude
Change in terrestrial longitude
Torque applied to the gyro float
Gyro output axis uncertainty torque
Torque applied by gyro torque generator
Gyro temperature torque coefficient
Gyro float nondeterministic, random torque
Magnetic torque coefficient
Sensitivity matrix
Sensitivity matrix
Sensitivity matrix
Mass of the earth
Universal gravitational constant
Power spectral density of white noise
Navigation error vector
Unbiased white noise
kth Legendre polynomial
Differential operator, dldt
Distance from origin of I frame t o origin
of i frame
27
Location
28
Symbol
Description
Attitude forcing function for platform system
Attitude forcing function for strapdown
system
Group of vector terms
Group of vector terms
Forcing function for canonical error equation
for the jth mechanized frame
Forcing function for two accelerometer
local level system
Distance from hypothetical origin of I
frame to system's location
Distance from origin of earth-centered
frame to system's location
Fixed gyro torque
Geocentric position vector to the point on
the earth directly below the system
Equatorial earth radius
Polar earth radius
Position vector magnitude based on noninertial information
Position vector magnitude based on inertial
computations
Gyro torquer scale factor uncertainty matrix
Gyro temperature deviation from calibration
temperature
Time
Mass unbalance along gyro's positive input
axis
Mass unbalance along gyro's positive spin
axis
Earth's gravitational potential
Earth referenced velocity
Skew-symmetric form of v
Accelerometer random uncertainty
System weighting function
Equatorial projection of the earth's radius
vector
Polar projection of the earth's radius vector
Error state for general terrestrial navigator
Initial value of x
Location
29
SYMBOLOGY
Symbol
Description
Location
(7-44)
5.2.2
(6-15)
(4-34)
(4-34)
(4-23)
5.2.2
(3-30)
3.8.4.1
2.2
(3-36)
C.l
3.8.4.2
(8-126)
(6-8)
(8-46)
2.1.5
2.1.5
(A-11)
6.3.2
(3-7)
(7-15)
5.2
(8-110)
(7-44)
(8-95)
Fig. 4.2
Fig. 4.2
C.3.1
CHAPTER
3
Reference Frames
Tous les problemes de geometrie se peuvent facilement reduire
a tels termes, qu'il n'est besoin par apr6s que de connaitre la
longeur de quelques lignes droites, pour les construire."
3.1
INERTIAL FRAME
31
this principle7 i t is impossible t o distinguish instantaneously between gravitational and inertial forces since inertial forces that are measured in a noninertial frame of reference are, in fact, gravitational forces exerted by the
stars.
The question arises as t o how one can refer measured forces and motions
to an inertial frame t h a t has physical significance t o the problem of navigation in the vicinity of t h e earth. As will be
a frame that has its origin
a t the mass center of the earth and that is nonrotating relative t o the stars
can be considered to be an inertial frame for measurements made in the
vicinity of the earth.
Consider the vector output of an ideal set of accelerometers whose sensitive
axes are mutually orthogonal. I n its simplest form an accelerometer can be
conceived of as a suitably damped, spring-restrained reference mass whose
displacement is measured relative to a coordinate frame fixed in the instrument case. From Newton's second law the output from such an instrument
package is proportional to the difference between the inertially referenced
acceleration and the net gravitational accelerations a t the instrument's
location. This difference between inertial and gravitational acceleration is
identically equal t o the contact force exerted on the accelerometer set by its
support structure.71 Thus the accelerometer output is symbolically written
as :
where
C;
R
'
= inertially
referenced acceleration
32
REFERENCE FRAMES
and
c; = cqc;
where R = vector from hypothetical inertial frame origin to the instrument
location.
vector from earth-centered frame origin to the instrument
location.
vector from hypothetical inertial frame origin to the earth's
center of mass.
coordinate transformation matrix relating the inertial axes,
I , to the earth-centered inertially nonrotating axes, i.
Equation 3-1 is thus written:
It is next observed that since the mass center of the earth is in free fall, then
the acceleration of the earth is equal in magnitude and opposite in direction
to the net gravitational accelerations a t the earth's center, that is,
"
But because the earth centered, i frame, is nonrotating relative to the inertial,
I frame, then C ~ ? I= Fi. Finally, the effect of the earth's gravitational field
is taken out from under the summat,ion in Eq. 3-4, yielding
fa = Cq$-
Ga
[G,. - G,4]
k'
where k' = summation over all the bodies of universe except the earth.
G = gravitational acceleration a t the instrument location due to the
earth.
Equation 3-5 demonstrates that the gravitational effect of all of the other
bodies in the universe on the accelerometer output appears as a term which
is the difference between the gravitational acceleration a t the center of the
earth and that a t the instrument location. Fortunately, these difference
terms are on the order of lo-' JG1 for the bodies in the universe causing the
GEOGRAPHIC FRAME
Local meridian
(x, y, 2)-inertial;
largest effect, namely the moon and the sun. Thus for instruments whose
resolution does not extend down t o lo-' earth G7s, which is t h e case for
practical navigational instruments, the output of the accelerometer triad
can be approximated as:
fa
C;fi - Ga
(3-6)
Equation 3-6 points out the important fact t h a t the vector output of a n
accelerometer triad will be proportional t o the nonfield specific force, coordinatized in the particular frame t h a t happens t o be mechanized by t h e
accelerometers.
Thus the inertial frame of importance for practical terrestrial navigation
systems has been shown t o consist of a frame t h a t is nonrotating relative t o
th.e stars and t h a t has its origin a t the earth's center. The inertial frame is
illustrated in Figure 3.1. The axis directions have been chosen such t h a t t h e
x and y inertial axes lie in the equatorial plane and the z axis is coincident
with the earth's angular velocity vector.
3.2
The geographic frame is a local navigational frame which has its origin a t
the system's location and its axes aligned with t h e north, east, and down
34
REFERENCE FRAMES
The earth frame is a coordinate frame which has its origin a t the earth's
center of mass and has axes which are fixed in the earth. The axes are
arranged such that the earth and inertial frames are coincident a t a time,
t = 0, the navigation starting time. Specifically, from Figure 3.1 i t is seen
that a t t = 0, the inertially fixed reference meridian, earth frame meridian,
and local meridian are coincident, such that the following relationship holds :
I = I,
+ il - miet
(3-7)
il = celestial longitude
I, = initial terrestrial longitude
m i , = earth's inertial angular velocity
t
3.4
time
The local geocentric frame has its origin a t the system's location, coincident
with the origin of the geographic frame. The x , axis lies in the direction
opposite to the geocentric position vector, r. The y, axis lies along east, and
the x, axis, which lies in the local meridian plane, completes the righthanded orthogonal axis set. This frame is also illustrated in Figure 3.1.
3.5
The body frame constitutes the familiar vehicle axes of roll, pitch, and
yaw which has its origin a t the vehicle center of mass. Note that the origin
of the body frame does not, in general, coincide with the location of the
navigation system. As illustrated by Figure 3.2, the roll axis points forward,
the pitch axis points out the right-hand side, and the yaw axis points down,
all with respect to the vehicle.
35
Figure 3.2
3.6
Body frame.
3.7
REFERENCE F R A M E R E L A T I O N S H I P S
w:,
-sinLcosil
-sin
L sin 1
cos L
where L = geographic latitude.
--sinil
cos 1
-cosLcosA
-COS
L sin il
-sin L
36
3.7.2
REFERENCE FRAMES
l nertial-Eart h
i
W i e = Wie,
(0, 0,w i e )
wiet -sin q e t
wiet
cos wiet
1-
0
3.7.3
Inertial-Geocentric
wfc =
wj, =
(1 cos L C ,- L C , - A
(3-13)
(3-14)
sin L C )
-sin A
sin LCsin A
cos A
cos LC
- cos LCsin
-sin LC
3.7.4
-cos LCcos A
(3-15)
Geographic-Geocentric
w;, = w:,
CO;
c;
(0,
A, 0 )
L-sin D
( L - L,)
D = deviation of the normal
where
3.7.5
Earth-Geographic
sin D-
cos D-
wz = { t cos L, -L, -t
sin L)
i)
-sin A1
-cos L cos A1
cos A1
- cos L sin A1
-sin L
(3-21)
(3-22)
= change in terrestrial longitude from start of navigation ( t = 0)
37
REFERENCE F R A M E R E L A T I O N S H I P S
3.7.6
Inertial-Body
b
Wib
(3-23)
my)
here m ~ cop,
, c o y are the vehicle's inertially referenced roll, pitch, and yaw
angular velocities, respectively. The coordinate transformation relating the
body frame to the inertial frame is, of course, a function of the complete time
history of the vehicle's angular motion and normally cannot be specified a
3.7.7
= {wR? u
Inertial-Tangent
t
wit
= (mi, cos Lo, 0,-mi,
sin Lo)
- sin Lo cos m,,t
- sin Losin m,,t
cos Lo
Cit =
-sin
wiet
cos miet
-sin Lo-
Tangent-Geograp
hic
o:n=
{i
cos L,
-L,
-,?sin L)
cos (1 - lo)
The transformation matrix above can be approximated through series expansion to apply t o situations where the origin of the tangent and geographic frames are separated by only a short distance. The second-order
approximation t o Eq. 3-27 is given by :
AL2
A1
1 - - - sin2Lo2
2
A12
4
A1 sin Lo
-AL
A1
- -sin 2Lo
4
A1 cos Lo
-Al(cos Lo- AL sinL,)
AL2
AZ2
1 - -- -cos2Lo
2
2
38
REFERENCE FRAMES
where AL = L - Lo
A1 = 1 - I,
-AL
3.8
-A1 sin Lo
A1 cos Lo
-AlcosLo
AL
1
(3-29)
P L A T F O R M , ACCELEROMETER, A N D G Y R O FRAMES
Platform Frame
(p frame;
x,, y,
z, axes)
z, a x e s )
The gyro frame, like the accelerometer frame, is defined by the input or
sensitive axes of each gyro. All of the comments made about the accelerometer frame apply to the gyro frame.
3.8.4
39
where
Fi gu re 3.3
Platform-instrument geometry.
REFERENCE FRAMES
Figu r e 3.4
and I
identity matrix
Oij = component of 8, resulting from a rotation about the j t h platform
axis.
=
The six independent angles, Bij, are estimated via the instrument alignment
procedures mentioned previously.
It is insufficient to merely multiply the specific force measurements by the
Cg transformation of Eq. 3-30 to obtain the platform referenced specific force.
This situation arises because a physical vector measured by a nonorthogonal
set of instruments and coordinatized in orthogonal axes will not be the same
vector as that measured by an orthogonal set of instruments.
To illustrate this concept, consider Figure 3.4, which shows the twodimensional situation resulting from instrument misalignment about the
x axis only. I n the figure, recall that an accelerometer will measure the
orthogonal projection of specific force along its sensitive axis. Then the two
accelerometers will sense the quantities :
Using Eq. 3-30, these components transform into the platform frame as
follows :
cos
+fez, sin y
sin y
+ fe,,
cos y
where it was noted that Ow,, as shown in Figure 3.4, is negative in Eq. 3-30.
Carrying out the matrix multiplication yields
41
~ u the
t force as measured by an orthogonal set of instruments would be
given by :
F" = f (cos y, sin y )
It is clear that the force, as measured in the nonorthogonal frame and transformed into the orthogonal platform frame, c;fa,is not equal to t h e force as
would be measured in platform axes, p p .
For the three-dimensional case, this transformation error can be evaluated
analytically by expressing f p in terms of cZfa using Eq. 3-30:
where it was noted that the specific force transforms from the orthogonal
platform frame t o the nonorthogonal accelerometer frame via the transpose
of Eq. 3-30:
fa = (C,)P T-P
f
(3-32)
Note that the transformation properties between nonorthogonal and orthogonal frames differ from those between two orthogonal frames since
C:(C;)T
=I
nc: (AC,) # I
Nevertheless for the two-dimensional case illustrated by Figure 3.4, i t is
evident that Eq. 3-32 holds. The amount that the matrix product, C,P(Cz)T
differs from the identity matrix is the transformation error. Thus
Cz(CE)T
%Z
0 Z Y - %Y
OYZ
BZz - OY,
BY,
%,
(3-33)
0
The transformation used in resolving the force measurements into platform
%Y
- ex,
- 6ZX
f
'
Y
X
axes, a,nd which recovers the original force vector, Cz, is given by subtracting
the bracketed symmetric matrix term of Eq. 3-33 from 3-30 :
*P
or
c,
C,P - [Cap(C,)
I]
Note that Eq. 3-34 reduces to t h e usual small-angle transformation for the
orthogonal case of Ox, = 8,,, 8,, = O,,, and OZX = O,,.
Thus t o transform specific force measurements made in a nonorthogonal
axis set t o an orthogonal platform axis set, the following relationship is used :
42
REFERENCE FRAMES
where Aeg is given by the computed version of Eq. 3-34, f is the specific
force as measured in the nonorthogonal instrument axes, and f p is the
measured specific force so resolved in the orthogonal platform frame that the
physical specific force vector is recovered.
A
*
&" = Caw
p-CJ
[I - ( A e : ) T ] ~ g
(3-36)
where +ij = component of +iresulting from a rotation about the jth platform axis; i = x, y, z, j = x, y, z , and $=, +,, and 4, denote the angles between x, and x,, y, and y,, and z, and z,, respectively; but in this case,
G g is the inertially referenced angular velocity vector as measured in the
nonorthogonal gyro axes and
is the measured angular velocity resolved
in the orthogonal platform frame such that the physical angular velocity
vector is recovered.
The situation is slightly different if it is desired to so command the platform a t a specified angular velocity that a particular reference frame is
instrumented by the platform axes. I n this case appropriate rate commands
must be supplied to each gyro via torquing signals, recognizing the fact that
the gyro input axes are not orthogonal. Analogous to Eq. 3-32, the platform
inertially referenced angular velocity in platform coordinates transforms
into gyro coordinates via
a*
Thus to achieve a given platform angular velocity the desired angular velocity is premultiplied by
:
(ez)T
where
W:
&:,
43
3.8.4.3
Uncompensated Instrument-Platform Transformation.
In situations where no attempt is made t o measure the six misalignment
angles of Eq. 3-34, the instrument outputs are accepted as being coordinatized in platform axes. I n this case the specific force measurements are given
by Eq. 3-32 as:
P" = fa = (C,)P Tf P= [I (AC,)D TIfP
(3-39)
where, as before,
0: =
G:P
But since the commanded platform angular velocity actually takes place
in the nonorthogonal gyro axes, the inertially referenced platform angular
velocity is given by :
coyP = C6)WZ = (I
+ Ac:)GrP
CHAPTER
Geometry of t h e E a r t h
"Gravity must be caused by an agent acting constantly
according t o certain laws; but whether this agent be material
or immaterial I have left t o the consideration of my readers."
ISAAC NEWTON
4.1
T H E GEOCENTRIC P O S I T I O N VECTOR
Perhaps the most natural and convenient way t o specify the inertially
determined position relative t o the earth is in terms of the system geocentric
position vector, r. The use of the geocentric position vector is particularly
appropriate because the gravitational field compensated accelerometer
outputs are proportional t o the second time derivative of the inertially
referenced geocentric position vector. (See Eq. 3-6.) From Figure 3.1, the
inertially referenced geocentric position vector is given by :
ri
celestial longitude
A, r sin LC)
(4- 1)
45
Local
meridian
plane
Figure 4.1
From this same figure it is seen that the resolution of the geocentric
position vector in the geographic frame is
r n = ( - r sin
D ,0, - r cos D)
where D = L - LC
D = deviation of the normal
The geocentric position vector can be written in terms of the geocentric
earth radius and the altitude above the reference ellipsoid, as illustrated
by Figure 4.1.
Since
'
r$ =
{-yo
and
hn = (0'0, - h ) ,
(4-4)
(4-5)
rn =
{-yo
(4-6)
An exact expression for the square of the geocentric radius vector is given
by the sum of the squares of the components of Eq. 4-6:
r2 = r$
+ 2roh cos Do + h2
(4-7)
46
GEOMETRY O F T H E EARTH
+ h)2
[(rO
+ h) term yields
2hr0(l - cos Do)
r = (r, + h)
0-0
+ h)2
The terms beyond first order in the expansion above will be a maximum a t
latitude 45" where Do
&-rad. Even for an altitude of 200,000 ft, an
evaluation of the expression above yields
Thus the error involved in evaluating the geocentric radius magitude with
the expression
r=ro+h
is less than 1 f t for aircraft altitudes.
4.2
T H E DEVIATION O F T H E N O R M A L
The deviation of the normal is defined as the angle between the geocentric
and geographic verticals, that is,
D =A L - L C
(4-3)
47
T H E E A R T H R A D I U S MAGNITUDE
r,
h is on the order of 1 ft. Furthermore, the equatorial projection
of the earth radius vector is equal to
x,
r, cos LC,
(4-10)
where
e = re
rp
ellipticity
(4-12)
re
Substituting Eqs. 4-8, 4-10, and 4-11 into 4-9 yields
sin D = e
ro
Do = e sin 2L
where e = - (e2/2)sin 2L
+E
(4-14)
E =
-e sin 2L(e/2
100,000 f t
e sin 2L
(4- 15)
4.3
The earth radius vector, for the purpose of inertial navigation computations, is defined as the vector extending from the center of the earth t o the
48
GEOMETRY O F T H E EARTH
but
x: = r t cos2LC,
and
(1
I
(4-20)
where e is the ellipticity as defined by Eq. 4-12. Equation 4-20 can be written
in a more convenient form by noting that the relationship between the
polar and equatorial radii is given by
rP = re(l - e)
Thus Eq. 4-20 becomes.
e
(1 - cos 2L)
+ &e2(1
- cos 4L) -
r,
as
is on the order of less than 150 ft, assuming that re can be precisely specified.
The expression for r, which includes the e2 terms, Eq. 4-21, is accurate t o
49
4.4.1
+, Al) = N
SSL
"I
50
G E O M E T R Y O F T H E EARTH
Figur e 4.2
+,
FIELD
51
4, Al) = 2
"O
k=O
Ak Pk(cos 4)
r
+ periodic functions of 8
and
For the reference ellipsoid model, symmetry exists about t'he polar axis, 2,.
Thus D ( p ; p, 8 ) = D(p, /3) and the periodic functions in 8 drop out of the
potential function, which can now be written :
JflP
cos p dm
and
52
GEOMETRY O F T H E EARTH
and
where
The coefficients, J k , are determined experimentally such as by the observation of satellite orbital deviations from orbits which would theoretically
result if the earth were s p h e r i ~ a l . ~ O ~ ~ ~
A consistent set of values for use in the expression above for the earth's
potential is based on References 40 and 14 and is as follows:
Note that in calculating these values from the numbers given in the
references, the conversion used in converting from kilometers to feet was
The constants above are consistent with a value for the earth's ellipticity
given by :
1
e=
298.30
Note that for k = 2 , 3, and 4, the Legendre polynomials are given by:
The even harmonics are seen to be symmetric about the pole giving rise to
the oblate terms, while the odd harmonics are antisymmetric giving rise to
the so-called pear-shaped term. Writing out the expression for the potential
through the J4 term:
53
In summary, Eq. 4-26 is the desired analytical expression for the earth's
gravitational potential derived under the following three assumptions :
..
.
4.4.2
where i, and i4 a t the unit vectors in the directions of increasing radius and
colatitude, respectively. Note that the expression above for V excludes the
derivative with respect to the terrestrial longitude because symmetry has
4 in degrees
30
60
90
120
I
150
1
180
q5 in degrees
Figure 4.3
54
GEOMETRY O F T H E EARTH
C$
0.00021
0
Figure 4.4
I
30
in degrees
90
120
6 in degrees
60
150
180
been assumed about the earth's polar axis. The result is of the forw:
G = C,i, + C,i,
where the radial component is given by:
G7-
=--
1-
r2
$J~(:F~
eos2
+ - 1) - 25,
(4-27)
(r:
THE
55
4.4.3
(G,
+ a, t a n L,)
'
r
rz
(G, - G, cot LC)-
?- -
Gx = - - l + B J 2(?)Z[lr2
QY = - { I
r2
z
= -~
r2
+J
$J~(?T
( 1
0(:1])2 +
r
error
((7.)
+ error ( G ~ )
- 5 (:)]}2 + error (GJ
r
- 5(:)1]]5
(4-32a)
(4-32b)
(4-32~)
GEOMETRY O F T H E EARTH
South
1.0
0.5
-1.0
-1.5
-2.0
-1.0
Figure 4.5
North
pole
Equator
pole
-0.5
0.5
1.O
The error incurred in neglecting the effects of J , and J , in Eqs. 4-32a and
4-323, the equatorial components, is plotted in Figure 4.5 for r,/r =
Because of polar symmetry the maximum G, and Gv errors are identical.
As is indicated the G, and G, errors are scaled by cos A and sin A, respectively.
The error for the polar component is also shown in this figure. The maximum
error in the equatorial components is seen to be 1.2 x
G a t LC= 64O
G a t the north pole.
and the maximum polar component error is 2.0 x
Thus i t has been shown that for accelerometers with resolution in the
10 m i c r o 4 range, Eq. 4-32 is a reasonable approximation to the gravitational
field vector.
4.5
The gravity field is defined as the acceleration field arising from the
combined effects of the earth's gravitational field and the inward centripetal
acceleration due to the earth's rotation :
Polar axis
Figure 4.6
aplane l
a,,
a t mean sea
58
GEOMETRY O F T H E EARTH
East coast
West coast
900
1200
1500
1800
2100
Figure 4.7 Prime and meridian deflections of the vertical along a west-to-east path
across United States.
59
-8
200
400
600
800
1000
1200
deflection curve. One easily discerns the effects of the continental shelf, the
Appalachian range, the gentle rise of the Great Plains ending in the Rockies,
the Sierra's, and finally the depths of the Pacific. These effects are not seen
in the meridian deflection curves because the major topographical features
have a north-south bias across the continental United States.
Autocorrelation functions for the prime and meridian deflections, based
on the information in Figure 4.7, are plotted in Figure 4.8. These curves
show that the root mean squared (rms) values for the prime and meridian
deflections are about 3.9 arc-sec and 2.2 arc-sec, respectively. Note that the
mean square value is just the value of the autocorrelation function at zero
distance.
The gravity vector can be expressed as a function of the gravity and
gravitational fields associated with the reference ellipsoid. The substitution
where Ag
9 = s,
=
gravity anomaly
ge = gravity magnit;de
+ A9
(4-35)
is made in Eq. 4-34 and the result is substituted into Eq. 4-33.Carrying
out the matrix multiplication in geographic coordinates yields
Lge+ A
~ J
LG,
(4-36)
60
GEOMETRY O F T H E EARTH
1
100
200
300
400
500
600
700
- 1
800
States.
But the gravitational field vector associated with the reference ellipsoid is
given by :
+ r w t cos L cos L C )
(4-37)
4.5.1
61
g,
Go - r ~ : cos
, L cos L C
The vertical component of the gravitational field vector is found by transforming the radial and colatitude components given by Eq. 4-30 into the
pographic frame using Eq. 3-17. This operation results in
GD = G+ sin D - G, cos D,
here G!, and G4 are given by Eqs. 4-28 and 4-29, respectively. If the
trigonometric functions above are expanded in series, then GD can be
approximated as follows without incurring an error of more t h a n about
1 0 - ~g :
GD
-G,
G4e sin 2L
(4-41)
where Eq. 4-15 was used t o approximate the deviation of the normal.
Equations 4-28 and 4-29 are now substituted into the expressions above
yielding
3 cos 2L)]
+8
4.6
A N A L Y T I C E X P R E S S I O N S F O R T H E SPECIFIC F O R C E
VECTOR
62
GEOMETRY O F T H E EARTH
Inertial Frame
The analytical form for the specific force vector in inertial coordinates is
found from Eq. 3-6 by a simple coordinate transformation into the inertial
frame via c:. ~ h u s
f i = pi - Gi
(4-44)
where
f i = (fx, f,, f z l
2i = {Fx7rV7
rz]
Gi
An alternate form for Eq. 4-44 is found by substituting into the above the
and
analytical expression for ri in terms of the spherical coordinates r , LC,
A, given by Eq. 4-1. This procedure, however, results in an unnecessarily
L
messy" expression which is difficult to work with.
6
4.6.2
Geographic Frame
63
substituting Eq. 4-33 for Gn and the equation above into Eq. 4-45 yields
an expression for the specific force of the form :
fn =
where
f
wk + 2-g
+" +
+ 252G)vn - gn
(4-48)
4 fN , f D )
= {(I + 2wie)cos L, -L, -(i
f
+ 2 w i e )sin L )
fN
f,
fD
irN
= 6,
=
6,
+ v,(t+
2wie) cos L
(4-48a)
(4-48b)
+ LvN - g
(4-48c)
vn =
(fN -
r D i , - ( r D cos L
(4-49a)
where i t was observed t h a t the geocentric position vector is given by Eq. 4-6
as
rn =F { - r 0 sin Do, 0 , -ro cos Do - h )
(4-6)
Differentiating Eq. 4-6 and substituting into Eq. 4-4 gives
v,
v,
v,
=
=
+
+
(r, cos Do
h ) i - f 0 sin Do - roB0cos D,,
(rOcos Lc0 h cos L ) t
-h - iocos Do r , ~ sin
, Do - roL sin Do
(4-50a)
(4-50b)
(4-5Oc)
vn = { ( r L
(4-51)
1 - 2e cos 2 L
e2
+( 1 - 4 cos 2 L + 7 cos 4L) - . .
4
64
GEOMETRY O F T H E EARTH
fN = ( r ,
f,
= (T,
fivr
rX + 2hL
f,
fo
-X
(4-53a)
(4-53b)
(4-536)
CHAPTER
Gyroscope Performance
Inertial navigation systems utilize gyroscopes t o provide a frame of reference
to which the force measurements can be referred. While a variety of gyroscope
configurations have evolved over the years, such as the two-degree-of-freedom
gyro,58the electrostatic gyro,15 and the laser gyroY4*
the single degree of
freedom (SDF), floated integrating gyro continues t o be the most ,widely
used instrument in present-day applications of inertial technology. This
instrument was originally developed by C. S. Draper and his associates a t
the Instrumentation Laboratory a t M.I.T.18 and design refinements made
over the years a t M.I.T. and by industrial organizations have resulted in
the evolution of the SDF gyro into one of the most precise devices in the
field of instrumentation.
The following is intended t o be a brief development of the S D F gyro's
salient characteristics as they pertain t o the study of inertial systems. See
Reference 7 1 for a more complete discussion of the material which follows
and for detailed information concerning the sciencelart of current gyro
design.
5.1
PRINCIPLE OF O P E R A T I O N
The operation of the SDF gyro is visualized with the aid of Figure 5.1,
which is a simplified cutaway view of a typical instrument. The inputoutput-spin (I,0 , S ) axis system constitutes a n orthogonal axis set which
is fixed in the gyro housing or case. The gyro rotor and its gimbal are housed
in the float which is immersed in a high density, high viscosity fluid which
serves t o provide support through floatation. I n an effort to minimize output
axis bearing friction, a magnetic suspension is provided to supplement the
floatation system; however, other means such as jewelled pivots, flexure
supports, and vibratory schemes are sometimes used t o provide the float
66
SINGLE-DEGREE-OF-FREEDOM
GYROSCOPE PERFORMANCE
Magnetic
suspension
Case
(gyro housing)
momentum
Figure 5.1
Single-degree-of-freedomgyroscope.
= torque
PRINCIPLE O F O P E R A T I O N
67
In addition to the inertial frame defined in Chapter 3 and the gyro case set
defined, i t will be convenient to define a float frame which has
a common origin with the case frame, namely at the float's center of mass,
but which is fixed in the float. The float and case frames differ by the small
~ u t p u axis
t
rotation, A,, as shown in Figure 5.1. The output angle is maintained a t a small value via servo techniques which are discussed later. The
transformation between float and case axes is given by:
Now, the objective of this derivation is to relate the applied torques to the
output angle, A,, and the inertial angular velocity of the case in the steady
state. To this end, the transient rate of change of angular momentum term
in Eq. 5-3 is neglected. I n addition i t shall be assumed that the gyro dynamics
can be completely characterized by the gyro spin angular momentum. Thus
Next, the angular velocity of the float with respect to inertial space, Qif,
is expressed as the sum of the angular velocity of the float with respect to
the case plus the angular velocity of the case with respect to inertial space:
c;q,c; + a;,
68
SINGLE-DEGREE-OF-FREEDOM
GYROS.COPE P E R F O R M A N C E
Substituting Eqs. 5-2 and 5-4-5-8 into Eq. 5-3 and transforming the resulting
equation into the case frame yields
Mtg - C A ,
+- (u)M
As with the input axis torque, the torque applied about the spin axis, Ms,
is applied through the float's hydromechanical support. Thus Eq. 5-10
becomes
(5-1l a )
(5-11b )
(5-11c)
The steady-state gyro operating characteristics are obtained directly
from Eq. 5-11 . Equations 5-1l a and 5-1l c are the equations for the torques
which are applied t o the float by its support to sustain the angular velocities
on the right-hand side of these equations. Equation 5 - l l b is the equation
of importance for this development, since if os = 0, it becomes
where t o summarize
A,
DYNAMIC
69
T ~ U Sit
5.2
While the development leading t o Eq. 5-12 serves t o illustrate the SDF
gryro7s operating characteristics, the gyro dynamic behavior was, for the
sake of simplicity, ignored. The dynamic characteristics can be investigated
using a mathematical development similar t o that previously used t o determine the gyro's steady-state behavior. Although the following derivation
exposes the gyros principal dynamic characteristics i t should be understood
that certain simplifying assumptions will be involved :
The float can only rotate relative to the case about the output axis,
t h a t is, the float gimbal is rigid.
The gyro rotor gimbal is rigid.
All motions take place along principal float axes such that products
of inertia can be ignored.
The gyro rotor is maintained a t a constant speed relative to the float.
Derivations which take into account the effects of rotor and gimbal deflections are to be found in References 26 and 64.
The angular momentum will no longer consist solely of the rotor angular
momentum but is given by:
Note t h a t J, includes the moment of inertia of the rotor since perfect rotor
speed control has been assumed. Since the case and float frames share a
common axis along the case output axis, then in E q . 5-13, J o = J,,.
If Eqs. 5-6 and 5-13 are substituted into Eq. 5-3, and the resulting equation
is transformed into the case frame, the output axis equation is given by
70
SINGLE-DEGREE-OF-FREEDOM G Y R O S C O P E P E R F O R M A N C E
@)Multiplier
Figure 5.2
I n the above, all of the terms on the right-hand side except the first two are
unwanted and must be compensated or otherwise accounted for. The Hco,A,
term is referred to as "cross coupling" torque, the A , ( w ~- o g ) ( ~, ~Jz,)
term is referred as "anisoinertia coupling" torque, the ( J Z f- J Z f ) ~ , w s
is referred to as "anisoinertia" torque, and the Johois referred to as "output
axis acceleration" torque. Equation 5-14 can be arranged in block diagram
form as shown in Figure 5 . 2 . I n Figure 5.2, note that the gyro time constant,
r g , has been written for the quantity (J,/C). It as been demonstrated
experimentally that for certain gyros the gyro time constant is not accurately
specified by J,/C, and that it is necessary to consider the gimbal flexibility
along the input axis. Taking into account this effect, i t is seen that the time
constant is increased and is given by:
where KZfis the elastic restraint coefficient along the x, float axis.
71
5.2.1
i b
Wb,
1/C
P ( T ~ P 1)
Signal
generator
H
I
Vehicle
Platform
torquer
Figure 5.3
72
that is, the inertial angular velocity of the platform about the gyro input
axis is proportional to the negative of the commanded angular velocity,
Mt,/H, and the uncertainty angular velocity, ( u ) M / H .
5.2.2
The SDP gyro can be mechanized as a rate gyro if the gyro float is torqued
with a signal which is proportional to the output angle, A,. The torque
generator output is then given by:
Mtg = -K
Ag
tL7
where Kt, is the torquing sensitivity. For inertial applications, pulse torquing
techniques are used because of the difficulties inherent in analog torquing.
Neglecting the unwanted error torques in Eq. 5-14, the gyro operation in
the rate gyro mode can be illustrated as shown in Figure 5.4.
Pulse
torquer
Figure 5.4
control
Pulse output
Gyro operation in digital rate mode.
73
From Figure 5.4 it is seen that in steady state the applied torque, M,,,
is just sufficient t o balance the gyro torque caused by the input axis angular
velocity, wI, and the spurious output axis torques, (u)M. That is,
w, = A
-K,,A,,
A,
where 6Ktg = b,,- Kt,. It is seen from Eq. 5-18 t h a t if the estimate of the
scale factor is too high (positive 6Ktg),then too little torque is being applied
to the gyro float. Thus it is conventional t o state the scale factor error in
terms of the torque being applied t o the instrument by letting
T =
6%
-= instrument scale factor error
Kt,
Thus a positive scale factor error (scale factor too high) gives rise t o an
underestimate in the measured angular rate. I n addition, it is seen that
uncertainty torques applied about the positive output axis gives rise t o
an overestimate in the measured angular rate.
For the more usual case of digital torquing the gyro output pulses, suitably
scaled, represent changes in the input angular velocity; t h a t is,
where Ay = gyro output pulse. The relationship above holds because of the
presence of the integration in- the forward loop of Figure 5.4, but since
w,
dO,/dt, then it is seen that each output pulse represents an increment
of rotation about the gyro's input axis :
74
SINGLE-DEGREE-OF-FREEDOM G Y R O S C O P E PERFORMANCE
5.3
The single degree of freedom gyroscope has been studied for many years
and reliable error models have been experimentally confirmed. Spurious
torques about the gyro output axis, represented by ( u ) M in Eq. 5-14, lead
t o errors jn the indicated angular velocity for the case of rate gyros, and to
a physical platform rotation for the case of a gyro stabilized platform. The
error torques consist of fixed torques, mass unbalance torques, anisoelastic
torques, and torques due to magnetic and temperature effects. The output
axis error torques are given by :28*41
= I,
d: S
K j k = compliance along the jth axis due to a force acting along the
kth axis
MT = temperature torque coefficient
6T
MB
B
6M
= magnetic
field strength
= nondeterministic
random torque
The numerical values for the various torque coefficients in Eq. 5-20
depend, of course, on the particular gyro design and are evaluated via a series
of specialized gyro test procedure^.^^ Having established the values for the
B
I N S T R U M E N T A N D SYSTEM R E D U N D A N C Y A N D RELIABILITY
75
5.4
I N S T R U M E N T A N D SYSTEM R E D U N D A N C Y AND
RELIABILITY
= time
Thus t o achieve a reliability of 0.95 for 1 year requires a gyro mean time to
failure of 59 years. I n a commercial application some consideration should
76
chexad triad -
sin a
-sin a
cos a
sin a
cos a
-sin a
cos a
cos a
cos a
(5-23)
0
sin a
-sin
where a = one half the great circle angle between gyro input axes =
3148'2.8". Both the tetrad and hexad arrays are capable of effecting a
solution if any three gyros are operating. Both systems have selficontained
failure detection and isolation capability, an advantage over systems
consisting of two redundant triads.
Having established the symmetric arrays as optimal, the task remains of
computing the configuration reliabilities. If we take the tetrad as an example
I N S T R U M E N T A N D SYSTEM R E D U N D A N C Y A N D RELIABILITY
77
P ( 4 operate) = R4 = e-4At
(5-24)
while the probability that any combination of three will operate is given by
eFAt)
Reliability plots-perfect
failure isolation.
78
Then the configuration reliability is given by the sum of Eqs. 5-24 and 5-25
(union of mutually exclusive events) :
-
Rtetrad -
4e-3At - 3e-4At
(5-26)
Similar reasoning can be used to show that the reliability for the hexad array
is given by
Rhexad
= e-3At(20- 10e-3At+ 36e-2At- 45evAt)
(5-27)
Figure 5.5 shows plots of Eqs. 5-21, 5-26, and 5-27. I n addition, reliability
curves are shown for systems consisting of:
Two redundant triads
Three redundant triads
Six orthogonal gyros
Nine orthogonal gyros
The plots are made under the assumption that any failure can be detected
and isolated. Note that the reliability of the nonorthogonal arrays is quite
superior to that of the redundant orthogonal arrays.
CHAPTER
6
T h e Space-Stabilized Terrestrial
Navigator
6.1
D E S C R I P T I O N O F SYSTEM
= Cqfi -
Ga
(3-5)
where f = specific force (nonfield contact force per unit mass) exerted on
instrument triad
Hi = inertially referenced acceleration
80
ri
= r(cos
(4-1)
It is seen from Eq. 4-1 that the system's latitude and longitude can then be
calculated. The terrestrial longitude, I , is calculated from the celestial
longitude, it, using the following relationship from Chapter 3 :
where ve = earth-referenced velocity. The velocity is conveniently coordinatized in geographic axes to obtain the north, east, and down components :
vn = {vN,vE, vD) = C:ie
(6-2)
Equation 6-2 can be written in terms of ri by noting that since
- C:(ji
- Qieri)
CP( j i - B t r i )
(6-3)
Figure 6.1 shows a functional block diagram for an ideal space-stabilized
inertial navigator. Notice that the navigation computer processes only force
data t o obtain the desired navigational information. However, the system
must be supplied with suitable initial conditions and a time reference which
is implicitly contained in the computer be it digital or analog.
Figure 6.1 indicates that the system's altitude, h, is calculated. As is well
kn0wn,~8if the altitude computation is based only on inertially derived
information, the computation loop is mathematically divergent. This
problem is investigated in Reference 7 and two methods of bounding the
altitude computation are discussed, one involving the use of an altimeter and
{vN,vE, vD)
MECHANIZATION EQUATIONS
Initial conditions
f p
3-
Accelerometer
triad
fa
Navigation
computer
'
: L
-'I
= h
lnertially
stabilized
platform
Figure 6.1
6.2
MECHANIZATION EQUATIONS
82
6.2.1
Platform Commands
6.2.2
Prom Eq. 3-5, it is seen that if the specific force measurements are transformed to the inertial frame and if gravitational field compensation is provided, then the inertially referenced acceleration can be calculated. Thus
.Pi. = et p p
..
where f
hi
&
The computed accelerometer-platform matrix, CE, is in general a nonorthogonal transformation and requires special treatment. (See Section 3.8.4). The
determination of the computed platform-inertial transformation, q,via
appropriate system alignment techniques is a subject unto itself 7s4g and
is discussed a t length in Chapter 9. I n general self-alignment techniques are
preferred since the error in the system alignment matrix is then correlated
with the instrument errors, resulting in a lower final system error.55
MECHANIZATION EQUATIONS
6.2.3
83
In this equation the question arises as t o how the geocentric position vector
magnitude, 8, is calculated. This question is just another statement of the
altitude problem discussed in Section 6 . 1 , since if P is calculated using f i
alone, the computation of the navigational quantities is divergent. The
~ r o b l e mcan be formulated in a general manner by introducing a weighting
factor, K, into the calculation of P3 in the denominator of Eq. 6-7 via the
nonlinear estimator :
p3 = (+,)"(+i)3-"
(6-8)
where ri., = position vector magnitude based on external information such as
an altimeter
Pi
weighting factor
pi
:;(
+ 9: + p)"
(6-9)
The calculation of the position vector magnitude based on external altitude information is complicated by the fact that the supplied altitude is
usually referenced t o the surface of the earth. It is shown in Chapter 4, Eq.
4-8, t h a t the geocentric position vector magnitude can be calculated via the
84
T H E SPACE-STABILIZED TERRESTRIAL N A V I G A T O R
expression :
Pa
= Po
+i
See Figure 4.1 for a geometric interpretation of Eq. 6-10. The earth radius
magnitude is calculated based on Eq. 4-21 written as a function of the
geocentric latitude, LC:
where i t was recognized that sin LC= r,/r. Additional computational accuracy can be achieved if the nontruncated version of Eq. 6-11, which includes
the e2 effects, is used. Considering the uncertainty in re, the gravity anomalies,
and the instrument uncertainties, a more complex expression for Po than
Eq. 6-11 does not seem warranted.
The +=,/Pa term on the right-hand side of Eq. 6-11 is treated in like manner
t o the P,/P and re/? terms which occur in Eq. 6-7. Because these terms are
second order, being multiplied by either J , or e, an evaluation of P using the
previously introduced weighting factor, K , does not appear justified. The
calculation of P will therefore be taken to be the same as that for ri., of Eq.
6-10.
To summarize, the gravitational field vector is calculated using the
following version of Eq. 6-7 :
where Pi is calculated using Eq. 6-9, Pa is calculated using Eq. 6-10, and the
weighting factor, K , remains to. be specified.
85
MECHANIZATION EQUATIONS
6.2.4
Gn = Q(jf- f i : e f i )
(6-13)
here Pi and Pi result from integration of the equation for the inertially
referenced acceleration, Eq. 6-6. The skew-symmetric form of the earth
rate vector is given by Eq. 3-11 as
47;
-sin
-cos
6.2.5
2 cos 2
cos
-cos
2 sin 2
-sin
I t is shown in Chapter 4, Eq. 4-15 that the deviation of the normal is related
to the geographic latitude via the expression, D
e sin 2L. Thus the equation above for L becomes
2 = sinw18,
e sin 2 2
(6-14)
r
In Eq. 6-14 the choice again exists as to how the geocentric position vector,
?, is calculated, since either Pi of Eq. 6-9, Fa of Eq. 6-10, or some combination
of both can be used. The calculation of i can be formulated in a general
manner by introducing another weighting factor, a, such t h a t :
The ramifications of the choice of a will be left to the error analysis.
86
T H E SPACE-STABILIZED TERRESTRIAL N A V I G A T O R
i = f, + sin-'
h
4
(P:
+P
- mi'$
where the estimated position vector magnitude is again given by Eq. 6-15.
6.2.6
Mechanization Diagram
6.3
ERROR ANALYSIS
88
sources :
Gyro drift rate errors
Accelerometer errors
Accelerometer alignment error
System alignment error
Altimeter error
Gravity anomalies and deflections of the vertical
Gyro-torquing uncertainty is not a major error source in the space-stabilized mechanization since the gyros are free of earth, vehicle, and body rate
torquing commands which are necessary in noninertial frame mechanizations.
Low-level gyro-compensation torques are, of course, provided to account for
the effects of gyro mass unbalance, anisoelasticity, and so on. (See Section
5.3.) Errors due to torquing uncertainty in applying these compensation
torques will therefore constitute second-order quantities and will be neglected
in the error analysis. Similarly gyro alignment errors, which account for the
inability to specify the relationship between the nonorthogonal gyro frame
and the platform frame (see Section 3.8.4.2) are second order for the spacestabilized mechanization since only low level compensation commands are
applied to the gyros and since the only motion of the platform is due t o gyro
uncertainty or drift.
6.3.1
O i D=
Because of the gyro errors, the platform will possess a finite angular velocity,
o:,. This angular velocity is related to the individual instrument uncertainties through Eq. 5-15 which shows that the platform's inertial angular
velocity about the gyro's input axis, ol, is .gi'ven by
ERROR ANALYSIS
89
Thus the platform rotation is related to the individua.1 gyro torque uncertainties as follows :
Note that the sub-subscript notation for the gyro frame; x,, y,, 2,; has for
simplicity been replaced by denoting the gyros as the x, y, z gyros. Hopefully
no confusion will result from this simplification since i t is obvious that the
instrument uncertainties, (u)M, must be referred to the individual instrument sensitive axes. The reader may reasonably inquire as to why, if gyro
compensation torques are being provided, the inertial angular velocity of the
~ l a t f o r mis proportional to ( u ) M ,the uncompensated gyro output axis error
torque specified by Eq. 5-20. The major reason for this treatment of the gyro
errors is to maintain generality in the error analysis such that the results of
the analysis can be used to establish sensitivities to the various classes of
gyro errors such as fixed, mass unbalance, anisoelastic, and so on. Moreover,
even if the gyro error coefficientsare determined and compensation torques
provided, there will invariably be shifts in the values of these coefficients.
If in Eq. 6-18 the substitution, (u)cuk= -(u)Mk/Hk, k = x, y, z is made,
then the inertial angular velocity of the platform is written :
If the platform has the angular velocity given by Eq. 6-19, then the direction
cosine matrix relating the platform and inertial frames is given by the
general relationship of Eq. 2-4:
with the initial conditions :
90
T H E SPACE-STABILIZED TERRESTRIAL N A V I G A T O R
6.3.1.2
Specific Force Measurement Errors. The specification of a
detailed accelerometer error model depends, of course, on the particular
instrument design. As was argued for the case of gyro error modeling, the
analysis is as general as possible such that the results can be used to establish
error sensitivities. To this end the output of the three accelerometers can be
arranged in vector form as follows :
A" =
0'
0
a,
where a,, k = x, y, z, is the scale factor uncertainty associated with the kth
accelerometer, expressed as the ratio of two numbers. Note that as with the
gyros, the accelerometer sub-subscript notation has been replaced in favor
of simply denoting the individual instruments as the x, y, and z accelerometers.
ERROR ANALYSIS
91
e:,
where
error :
c,
92
written :
bi = G: + 6Gi
(6-26)
To determine an expression for 6Gi, the two position vector magnitude equations for ri and r,, Eqs. 6-9 and 6-10, respectively, are first perturbed by letting
in Eq. 6-9,
(6-27)
Fk = rk ark
k = x, y,z
and in Eq. 6-10,
(6-28)
F, = r,
6r,
and
il= h
6ha
where 6h, = error in the measured altitude of the system above the reference
ellipsoid.
+
+
It is found t h a t
and
where products involving higher order error terms and products of an error
with the earth's ellipticity, e, have not been included because of their insignificantly small values.
Substitution of Eqs. 6-27, 6-29, and 6-30 into Eq. 6-12 yields the desired
expression for the error in the computed gravitational field :
(ri) 6riri - 6ri]
r2
(6-31)
where
6ri
6.3.1.6
ca,
ERROR ANALYSIS
93
(6-35)
Gi = G% C ~ A G "
where t h e column matrix AGn contains the deflection of the vertical and
gravity anomaly terms, namely,
AGn
( E g , -q9,
As,
Thus when Eq. 6-35 is substituted for Gi and Eq. 6-31 for dGi, Eq. 6-34
becomes
I n arranging Eq. 6-36 in the form shown, the identity from vector analysis
given by Eq. 2-13 was employed:
94
?k = r,
+ 6r,
= x,y , z
sin1-
6r
dh,
+ cos L ( l + a tan2L ) 2
- a -t a n L
r
r
(6-37)
(6-39)
Note that if all inertial information is used for the calculation of the geocentric position vector magnitude, then a = 0 and ? = ti. The navigation
errors are seen to be related t o the geocentric position error vector by the
following expression :
95
E R R O R ANALYSIS
obviously, other choices of the weighting factor, a , will yield different relationships between the variables. The important point is that the method of
the latitude, longitude and altitude aflects the error response.
This is a point which is invariably overlooked in the standard references on
inertial navigation systems. This fact can be used to advantage under certain
circumstances; for example, by choosing a = 1 the latitude error can be
completely uncoupled from the two equatorial channels.
To explore this point in more detail recall that the latitude, longitude, and
&itude are found directly from the computed geocentric position vector,
since
ri = (rz, ry, rz) = { r cos LCcos A, r cos LCsin
A, r sin LC)
(4-1)
It is evident from the equation above t h a t there are many possible relationships between the inertially referenced position components and the latitude,
longitude, and position vector magnitudes. For example, the latitude could
be extracted either with the computation scheme of Eq. 6-14 or with the
following trigonometric relationship :
Similarly the longitude could have been extracted using, instead of Eq.
6-16, the following :
i = i, + sin-'
f,
(P2 - fZ)%
- w,,t
6 L = -(sin L
+ a cos2L csc L )
+ sin A -
6% +
+ (1 - a ) cos L r
GC
6ha cot L
r
(6-37a)
and
sec L
61 = -[-(1
- a ) sin
A 6r,
(6-38a)
96
Thus it is seen that unless all inertially derived information is used in extracting the latitude and longitude from the geocentric position vector components (a = 0 ) , then the error response depends on both the choice of a
and the chosen computation scheme.
The general expression relating the latitude, longitude, and altitude errors
to the geocentric position errors and the altimeter error is given by combining
Eqs. 6-37, 6-38, and 6-39 directly:
where
-a tan L'
-a
It is seen from the form of the M, matrix in the equation above that i t will
not be possible to solve for the vector 6ri as an explicit function of 6L,
61, and 6h because the inverse of M, contains a singularity a t a = 1. Thus
the relationship given by Eq. 6-42 cannot be inverted such that dr, and 6r,
can be independently specified. This difficulty is avoided if the variable dh
is written as a function of the error in the inertial computation of altitude,
6hi, and dh,. The inertially derived altitude is calculated using the relationship specified by Eq. 4-8 :
hi = Fi - i.o
(6-43)
where Fi is given by Eq. 6-9 and F0 is given by Eq. 6-11. But from Eq. 6-29,
(rilTri/r and if second-order quantities are neglected, f0 = ro.
i.i = r
Thus
r
where ah, = hi - h = error in the inertial computation of altitude above the
reference ellipsdid.
Substituting Eq. 6-44 into 6-39 shows that
Equation 6-45 is now substituted for 6h in Eq. 6-42, yielding
A
6ni =
I r 61
cos L
= M2C: 6ri
+ k2dh,
(6-46)
ERROR ANALYSIS
where
: :]
M 2 = b
-cctanL
-atan
and
k 2 [
6.3.1.8 Earth-Referenced V e l o c i t y Errors. The error in t h e computation of velocity is found directly from Eq. 6-13 by substituting
and from Eq
Thus the velocity error is given by
Equation 6-47 is nothing more than a relationship between the velocity error
vector and the navigation errors as given by 6ii, 6ri, and Nn.
For the case of a = 0 in the computation of the navigation errors, Eq.
6-40 shows t h a t
8ri = Ch Sn = CC(r 6L, r Sl cos L, -6h)
In addition from Eq. 4-51,
vn = ( ( r L
-6L,
-61 sin L)
where i t was noted t h a t if the initial conditions are ignored, 61 = 62. Substituting 'the expressions above for 6ri, vn, and vn into Eq. 6-47 yields after
some algebra and neglecting of second-order terms:
where the differential operator, p = dldt, has been employed. For most
applications, Eq. 6-48 can, with sufficient accuracy, be written as the three
98
uncoupled equations :
-- r 6 i
6UE -- r 6i cos L
dun-
duo
-ah
Note that for the case of a = 1, i t is impossible t o express the velocity errors
as an explicit function of 6L, 61, and 6h. This difficulty is seen from the form
of the MI matrix of Eq. 6-42 as previously mentioned.
The alternate expression given by Eq. 6-46 can be used t o relate the velocity errors t o 6L, 61, and ah,. Substitution of Eq. 6-46 into 6-47 yields
'P
-ri sin L
r cos Lp
.
sec2 L
-Z
(1 - a sec2L)L - a t a n Lp(1
- ai tan L
+ t a n Lp
sin2L sec L
L tan L
6ha
99
E R R O R ANALYSIS
Ci CpCEfa
ez,
substituting Eq. 2-16 for e;, 6-25 for ei, 3-30 for
and 6-21 for C ~ into
O the
above yields
f * = [I - N n - CZ0DpC:o - C;ZiCi,]Czfa
(6-51)
A
. ..
(6-53)
.
,$"
100
cn - v n = C,dn
+ C?ri
where
zn = {+,
"=)
6.3.2
As was previously mentioned, solution of Eq. 6-36 will by itself yield the
position errors in inertial coordinates. Equation 6-42 or 6-46 could then be
used t o compute the corresponding latitude, longitude, and altitude errors;
Eq. 6-47 could be used to compute the velocity errors, and Eq. 6-52 could be
used t o compute the level and azimuth errors. This was basically the
approach taken in Reference 7. The approach taken herein is to combine these
equations in a canonical form having an error state vector composed of the
attitude and position errors ; namely,
The velocity errors can then be extracted from the computed state vector
via Eq. 6-48. The formulation above of the state vector appears to be a reasonable compromise between the concise formulation represented by solution of
Eq. 6-36, for
and a more complex formulation which includes the velocity errors, particularly since the velocity and position errors are so simply related by Eq. 6-48.
Please note that the vector x is not a "state vector" in the usual sense,
since i t will be seen that nine initial conditions will be needed to solve for x.
The three additional initial conditions are required because the differential
equations describing the behavior of dL, 61, and 6 h will be shown to be second
order.
The system equations are manipulated into the canonical form by making
use of the relationships between the inertially referenced position errors and
the navigation errors as given by Eq. 6-40. Note that as previously discussed
in Section 6.3.1.8, the error equations cannot be written as a function solely
of the dependent variables 6L, 61, and 6 h for the case of a = 1.
101
ERROR ANALYSIS
- 2)1
(K
- 3)
(6-57)
K-2
102
giving
I n the equation above, 6n is given by Eq. 6-40, and win, is given by Eq. 3-8.
These expressions are substituted into Eq. 6-58 and the necessary time differentiation and matrix algebra is performed. I n addition the analytic expressions for Nn given by Eq. 2-17 and for f n given by Eq. 4-53 are utilized
in the algebraic manipulations, resulting in Eq. 6-59 shown on Pg. 103.
I n obtaining Eq. 6-59 the definition of the Schuler frequency was employed,
namely,
w, = J g / r
Jp/r3
Note that the terms appearing in Eq. 6-59 are first-order error terms as
described in Section 2.5. Terms involving the earth's ellipticity, e, or equivalently, those involving the centripetal acceleration due to the rotation of the
earth are second order when multiplied by the error variables and are therefore neglected. Generally speaking the errors due to approximations in the
final error equations are commensurate with the uncertainty in the sensed
specific force for navigation quality accelerometers, about 2 x 10-5g.
Since the coefficients of the error equation are functions of the system's
motion over the earth, then the 2 x 10-5 g criteria is found to apply only
if the vehicle motion is that which one would expect to encounter in an
application such as the supersonic transport.' For an application such as a
hypersonic glider the equations would have to be modified to include secondorder effects in order to maintain the specified precision. On the other hand,
for a reentry vehicle application, the time of flight is so short that the errors
resulting from a violation of the 2 x 10-5 g criteria will not be appreciable.
I n any application care must be exercised to ensure that the equations are
not being applied outside the region of their validity.
Equation 6-59 is three equations in six unknowns. The other three necessary equations are those which describe the behavior of the attitude errors.
The differential equation for the attitude errors is found from Eq. 6-55 by
premultiplying by Cz and performing a time differentiation :
104
T H E SPACE-STABILIZED TERRESTRIAL N A V I G A T O R
But since :
C is approximately constant, C;
equation above is written :
i n - V"
,
:
o
C F C ~= c:c:P:,,
- yn) =
(I cos L, -L, -I
and the
Cpd
n .TJ
sin L )
then
[i
sin L
-<"
-'OS
- Qinnun =
LplT d ~ 1
A cos L
sin Lp
The differential equation for the attitude errors is then given by
cos L
C;(u)wTJ
(6-61)
sin ~p
Equations 6-61 and 6-59 form the desired error differential equation in
canonical form for the space-stabilized inertial navigation system. The
equation set is valid for the mechanization scheme shown in Figure 6.2.
Equations 6-61 and 6-59 are written in the form:
where the forcing function for the space-stabilized system is given by
The characteristic matrix A is shown in Figure 6.3 and the error state vector
is given by:
x = ( E EE
~ ED 6L 61 6h)
Since Eq. 6-62 is valid only for the case of a = 0, the variable 6h in the
above expression can be replaced by 6h,
6.3.2
Solutions to Eq. 6-62 are available for the case of constant gyro drift end
for several constant terrestrial longitude rates. The navigation errors for this
106
T H E SPACE-STABILIZED TERRESTRIAL N A V I G A T O R
case, 6r" expressed in inertial coordinates, are given by Eq. A-18 in Appendix
A. The corresponding latitude and longitude errors are found by applying
Eq. 6-40:
6 n = :C 6ri
(6-40)
The operation above is carried out in Reference 6, and the results are shown
in Figures 6.4 and 6.5 for the case where the gyro drifts along all three axes
STATIONARY
SCHULER CELESTIAL
LONG. RATE
_I
12
18
24
TIME (hours)
Figure 6.4
Latitude errors.
ZERO CELESTIAL
LONG. R A T E
SCHULER CELESTIAL
LONG. RATE
STATIONARY
TIME (hours)
Figure 6.5
Longitude errors.
108
T H E SPACE-STABILIZED TERRESTRIAL N A V I G A T O R
= coy = mZ = w =
constant
=0
i = -oi,
= w, - mi,
i >> W ,
Note that for the large celestial longitude case, only the error envelope is
shown since the longitude rate is not explicitly specified. Note t h a t constant
gyro drift results in linear growth of the latitude and longitude errors with
an upper bound given by the product of the drift rate and elapsed time.
CHAPTER
7
T h e Local- Level T e r r e s t r i a l
Navigator
7.1
D E S C R I P T I O N O F SYSTEM
C:[fi
- Gi]
where
(4-48)
(4-51)
where rL and rl are the meridional and prime radii of curvature, as defined in
109
110
r, "1
%
+ qg
(4-51)
(1 cos L, -L, - A
where
sin L)
(3-8)
+ w,,
Figure 7-1 shows a functional block diagram for the local-level terrestrial
navigator. As indicated by Figure 7.1, the system computes the altitude.
The stability of this calculation is investigated in the error analysis.
The mechanization described in this chapter is just one of a class of inertial
navigation systems which have the common characteristic of maintaining
two of its accelerometers in the reference horizontal plane and a third along
the normal to the reference ellipsoid. The local-level system described herein,
which instruments the geographic coordinate frame, is undoubtedly the
most successful of all the inertial navigation system configurations. The class
of local-level systems today constitute the majority of operational inertial
navigation systems.
Initial conditions
S-
~ccelerometer
triad
fn
Navigation
computer
r~
Z
Geographically
stabilized
platform
command
Figure 7.1
MECHANIZATION EQUATIONS
111
MECHANIZATION EQUATIONS
7.2.1
Platform Commands
112
T H E LOCAL-LEVEL TERRESTRIAL N A V I G A T O R
7.2.2
For the local-level mechanization, it will be assumed that both the gyro
and accelerometer frames are nominally aligned with the platform frame
which, in turn, is nominally aligned with the local geographic frame. The
specific force measurements are not assumed to be compensated for the
effects of the accelerometer's input axis nonorthogonality. Since the assumption is made that the platform is aligned with the geographic frame, an
explicit alignment matrix is not necessary. The geographically referenced
specific force is then given by
A
fn
= fa = p p f r a
e: ez
ID
(74
where
=
= I.
The initial alignment of the platform frame with the geographic frame is
accomplished using either self-contained techniques such as leveling and
gyrocompassing or external techniques such as optical alignment.
7.2.3
Earth-Referenced Velocity
The estimates of the earth-referenced velocity are found from the integration of the appropriately compensated specific force data. Using the relationship between the specific force and velocity given by Eq. 4-48, the time
rate of change of the velocity is given by
113
E R R O R ANALYSIS
7.2.4
Gra.vity Calculation
As is indicated by Eq. 7-3, the vertical channel requires an explicit computation of the magnitude of the earth's gravity field, g. An analytic ex~ressionfor the gravity field vector magnitude associated with t h e reference
ellipsoid is derived in Section 4.5.1 :
g e = lu
- [l - %J2(1
- 3 cos ZL)] - rcu:
r2
cos L cos LC
(4-43)
It is seen from the equation above that the choice again exists as t o how the
geocentric position vector magnitude, r, is calculated. The approach will again
be taken t o introduce the weighting factor, K , to allow a general treatment of
the mixing of external altitude information and inertially derived altitude
information. Thus
-
~2
(e,)"(~~)~-"
(7-4)
Fa
and
Po
= Po
+h
(6-10)
(6-11)
ei = eo + hi
(7-6)
7.2.5
114
T H E L O C A L - L E V E L TERRESTRIAL N A V I G A T O R
I n the computation of the latitude and longitude, the altitude associated with
the radii of curvature is calculated by introducing the second weighting factor,
a,to mix the altimeter-derived and inertially derived altitudes:
GE
[jZ
hi = -6,
(%)"(Fri)l-']00s
f.
(7-9)
(7-10)
The latitude, longitude, and altitude are obtained by integrating the above
equations with the suitable initial conditions :
7.2.6
Mechanization Diagram
7.3
E R R O R ANALYSIS
116
T H E LOCAL-LEVEL TERRESTRIAL N A V I G A T O R
Thus the error sources listed in Section 6.3 are supplemented by the following two sources for the local-level navigation system :
Gyro torquing error
Gyro alignment error
7.3.1
where
117
ERROR ANALYSIS
where
Note that the time derivative operator, p = dldt, was introduced into the
w,,, p61 = p61.
quation above and i t was recognized that, since 1 = t
The platform error angular velocity is found by substituting Eq. 7-16
into 7-15 and noting that the desired platform angular velocity is given by
W i n - Thus
(7-17)
d ~ =
: (T
~ D ACz)w& h i " ,
(u)w"
where 6wTp = platform error angular velocity.
7.3.1.2
Specific Force Measurement Errors. The accelerometer
error mode1 for the local-level system is taken as identical to t h a t for the
space-stabilized system ; namely,
where
and the various terms are defined in Section 6.3.1.2.
7.3.1.3
Accelerometer Alignment Errors. If the accelerometer
outputs are accepted as coordinatized in an orthogonal axis set, then an
explicit calculation of the transformation from accelerometer axes to platform axes is not needed. Thus
=1
(7-18)
The relationship between the accelerometer frame referenced and platform
frame referenced specific force is shown by Eq. 3-32 to be given by the
following :
fa = [I
(AC:)T]f"
(7-19)
7.3.1.4
A t t i t u d e Errors. The attitude errors are defined in exactly
the same fashion as for the space-stabilized system, that is, the orthogonal
transformation error between the platform and geographic axes coordinatized
in geographic axes. For the local-level mechanization, the platform is assumed
to be aligned with the local geographic frame. Thus the attitude errors are
equal to the platform errors as defined in Section 7.3.1. l .
The differential equation for the attitude errors is found by applying the
general relationship between the time rate of change of & direction cosine
matrix relating two frames and the frames' relative angular velocity. Thus
if Eq. 2-4 is applied to this case,
e:
where
Ci
= Cp:,
Cg
=I
+ En
(7-20)
(7-21)
Thus
WE,
+ Eno,",
(7-22)
119
ERROR ANALYSIS
Sv"
Vn(SO,",
+ 2 6 0 2 ) = bgn + 6fn
(7-26)
where
Vn
an
e n + 2QG
+ 2 q e )sin L
0
-L
-
(i + 20,,) cos L
(i + 2 ~ , , )cos L
and
i n =
V T
f D + (AC,,
f
(7-28)
The platform-referenced specific force is expressed in terms of the geographically referenced specific force via the attitude matrix of Eq. 7-21, CE = I En.Thus the geographically referenced specific force perturbation is found
by substituting Eqs. 7-21 and 7-28 into 7-27, giving, after neglecting secondorder terms :
dfn = -Enfn
(ACz)Tfn (u)fa
(7-29)
07 gel
(7-30)
where the gravity field magnitude associated with the reference ellipsoid,
ge7is calculated using Eq. 7-5. The expression for cje is perturbed by substituting into Eq. 7-5, the quantities:
=
Po
h,
Pi = 9,
A,, i,= LC
*
64, 2 = L
6L, h = h
aha, and gi = h
Sh,. If second-order quantities are, as usual, neglected as being negligibly small in comparison t o the
remaining terms,
g
+
+
120
Equation 7-7 yields the same relationship as that obtained for the spacestabilized system, Eq. 6-45
6h
(1 - a ) 6hi
+ a ah,
+ (1 - a
6vN = rp 6L
) 6h,
~
+ a t Sh,
(7-32)
r cos Lp 61 - rt sin L SL
(7-33)
(7-34)
As usual second-order terms such as 6rL and 6r, were neglected in obtaining
the expressions above. It was also noted in the expressions above t h a t
rP
-ri
sin L
0
(1 - a ) L
r cos Lp
(1 - a),!cos
-P
,
(7-35)
Note that the vertical channel could be modified utilizing Eq. 6-45 to eliminate hi from Eq. 7-35. If this is done, Eq. 7-35 takes the alternate form :
121
E R R O R ANALYSIS
gotice that for the formulation of Eq. 7-36, a singularity is introduced into
tlie relationship when 6h approaches ah, as a approaches a value of one.
Therefore Eq. 7-35 will be used since i t is valid for all values of a.
Comparing Eq. 7-35 with its counterpart for the space-stabilized system,
~ q6-50,
.
i t is seen that it is only for a value of a = 0 that the two expressions
are identical.
7.3.2
The error equations of Section 7.3.1 are now combined in the canonical
form having an error state vector composed of the attitude and position
errors :
x = {eN,e ~E, ~6L,
, 61, ahi)
The velocity errors are extracted from the state vector using Eq. 7-35
and, if desired, the altitude error based on the weighted combination of
inertially computed altitude, hi, and altimeter computed altitude, ha, can
be found from Eq. 6-45.
The canonical error equation is derived in two steps. The attitude error
equation, Eq. 7-23, is first used to obtain an expression for three equations
in the five state variables: sn, bL, and 61. The velocity error equation, Eq.
7-26, is then used t o obtain a second set of three equations in the six state
variables: sn, bL, 61, and ah,.
Equation 7-23 is written in the desired form by substituting on the righthand side Eq. 7-17 for dwTDand Eq. 7-16 for 60;. This operation yields
2 sin L
- A sin L
i5
A cos L
--A
-A
cos L
P
sin L
23
1 cos L
sin L p
The other three necessary equations are found from the velocity error
equation, Eq. 7-26, with bgn given by Eq. 7-31 and 6fn by Eq. 7-29. Equation
7-35 is next substituted for 6vn and a simplified version of Eq. 4-51 for the
velocity, consistent with the first-order error analysis,
vn = ( T L , r,! cos L, -h}
is used in the subsequent, lengthy algebraic manipulations which result in
Eq. 7-38.
123
Equations 7-37 and 7-38 comprise the error equations for the local-level
mechanization. For the case where all inertial information is used in the
calculation of the latitude and longitude rates, or = 0, these two equations
can be written in the canonical form as follows:
where
It is t o be emphasized that Eq. 7-39 is only valid for the case of cc = 0 and
that a vertical accelerometer is present in the system. Naturally, all of the
remarks made regarding A in Chapter 6, also apply t o this case.
Notice in t h e forcing function, Q,, that the major error sources are not
frequency modulated as they were for the space-stabilized system since the
instrument outputs are nominally aligned in the frame in which t h e errors
are expressed, the geographic frame. This fact accounts for the major
differences in the error propagation between the two sytems, but further
comparisons will be made in Chapter 8.
7.4
T H E TWO-ACCELEROMETER
LOCAL-LEVEL .SYSTEM
124
T H E LOCAL-LEVEL TERRESTRIAL N A V I G A T O R
7.4.1
gD
= -%.
Mechanization Equations
There is, of course, no explicit gravity calculation, and the latitude and
longitude rates are given by Eqs. 7-8 and 7-9, respectively, by letting a = 1.
The mechanization diagram for the two-accelerometer local-level terrestrial
navigator is shown in Figure 7.3.
7.4.2
Error Equations
126
velocity error and the latitude, longitude, and altitude errors is found from
Eq. 7-35 by setting oc = 1 and letting 6hi = aha, yielding
rP
-Ti sin L
r cos L p
[z]1
+
o
c:
L] aha
(7-42)
-2,
The error equations resulting from the processing of the specific force data,
the velocity error equakions, are considerably modified with the elimination
of the vertical accelerometer. The two-channel version of Eq. 7-26 for the
velocity error equations is written via a simple deletion of the vertical equation. Equation 7-42 is next substituted for 6vn and the simplified analytic
expression for vn given by Eq. 4-51 is substituted. Thus Eq. 7-26 becomes
fo
-f~
rA sin 2Lp
rrp2+ 2 h
rp
++ f i ( i +2wie)sin2L
2 ~ p +
=(u)fa -AGn+(ACz)Tfn-
cos L(2Ap
+1 -2
2 tan
~ L)
6ha
(7-43)
(7-44)
= Qnl
where
2Lp
and
A, is given
by Figure 7.4.
6haJ
128
Thus the quartic fa.ctors into two pairs of imaginary roots with frequencies :
o,
o,,sin L. The proximity of the imaginary pairs gives rise t o the
characteristic Schuler frequency with period of 84 min and the Foucault
frequency caused by a beating phenomena which has a period of
-+
( 2 n / o i esin L ) m 34 hr
7.4.3
at
45O.
The solution of the differential equations represented by Eq. 7-44 give the
error response for the two-accelerometer local-level navigator for arbitrary
vehicle motion within the constraints implied by a "first-order" analysis.
An analytic solution of Eq. 7-44 would be quite tedious since the coefficients
of the differential equations are time varying except for the case of constant
terrestrial longitude rate ; t = constant ; constant latitude, L = constant ;
and constant altitude rate, h = constant.
Considerable simplification occurs if the stationary case is examined where :
giving
-
mi, sin L
mi, sin L
-cos Lp
-mi, sin L
-mi, cos L
oieCOS L
mi, cos L
sin L p
-g
rp2
r o , , sin 2 L p
- 2 r o , , sin Lp
r cos Lp2
129
Note that a nominal initial alignment of the instruments with the geographic
frame has been assumed such t h a t the subscripts x, y, and z have been
replaced by N, E, and D , respectively. The initial condition errors, 6L(O),
~ L ( o )61(0),
,
&0), and the initial misalignment errors, ~ ~ ~E ~( ( O0 ) ,~
) ~~ ( 0 ) .
are accounted for by taking the Laplace transformation of Eq. 7-46 :
-
wie sin L
wie sin L
- s cos L
-mie sin L
--aiecos L
wie cos L
mie cos L
s sin L
-9
rs
-2rw,, sin Ls
s2r cos L -
- cos L 61(0)
+ ~ ~ (-0sin) L 61(O)
gg + r[s 6L(O) + 6L(0)] + rwie sin 2L 62(0)
(UWD(u)fN -(u)&
csin
~
L)
where s is the Laplace operator, TN, TE, T~ are constant, and t h e overbar
denotes a Laplace transformed variable.
Note that the forcing functions due to the gyro nonorthogonality and the
altimeter rate error have been ignored in going from Eq. 7-46 t o 7-47. The
signal flow diagram corresponding t o the equation above is shown in Figure
!7.5,and the characteristic determinant for Eq. 7-47 is given by Eq. 7-45.
i
An alternate approach t o the solution of the error equations using state
transition matrix methods is developed in Appendix B.
131
7.4.3.1
Navigation and Level Errors for Constant Gyro Drift.
Considering the stationary case and letting constant gyro drift be the sole
error source, we have from Ey. 7-47 that
s
oiesin L
oiesin L
-mie sin L
- m i , cos L
mi, cos L
w i , cos L
--g
ys2
-2rwie sin L s
-s cos L
s sin L
Toie
sin 2 L s
s2r cos
0 L
k] I);
EN
ED
(U)WN/S
(u)mD/a
ax
(7-48)
where ( u ) c o(u)cuE,
~,
and ( U ) C Oare
~
the constant gyro drift rates associated
with the north, east, and azimuth gyros, respectively. Because of the
Foucault modulation, Eq. 7-48 is best solved via use of an analog or digital
computer. The results of such a solution a t latitude = 45" are shown in
Figures 7.6, 7.7, and 7.8. I n Figure 7.8 the level errors were f d n d t o be so
12
18
Time
Figure 7.6
'L
24
Hours
30
36
Time
Figure 7.7
'L
Hours
133
12
Time
Figure 7.8
18
24
30
Hours
Time
2,
Hours
Figure 7.9 Navigation and level errors for constant north gyro drift at
terrestrial velocity of 1900 knots.
2 for
east
The system modes are seen to be the space rate mode and the Foucault
modulated Schuler frequency. For this case of A = 4wie the space rate period
is 6 hours while the Foucault modulation now occurs with a period of about
8.5 hr instead of the 34-hr period for the stationary case. These 6-hr and
8.5-hr modes are easily identified in the figures.
Perhaps the most important feature revealed by the comparison is that the
latitude and azimuth error sensitivities are reduced from the stationary case
by the factor w,/i, or in this situation for 2 = 40,,, by a factor of 4. For
the cases that exhibit a longitude error which grows with time, namely the
responses to (u)o, and (u)w,, the vehicle motion aanears t o have little
135
effect on the error growth. On the other hand, the sensitivity Gl/(u)oE,
which is bounded for the stationary case, is reduced by the factor oi,/A.
The level error sensitivities in response t o level gyro drift are seen t o remain
Unchanged while the level error response t o azimuth gyro drift is seen t o
emerge from the computer noise. A digital computer solution has revealed
~
E ~ / ( u ) c o ~have
,
in fact increased
that these error sensitivities, E ~ / ( Z L ) Wand
by the factor j/w,,. Examination of the signal flow diagram (Figure 7.5)
reveals t h a t the coupling sensitivity between the azimuth and east level loop
is increased by the required factor of j/oi,.
An interesting limiting case arises when the vehicle is flying west with
i = -mi,. This case is readily analyzed by setting mi, to 0 in Figure 7.5,
thereby eliminating the Foucault and space rate coupling. The level error
sensitivities remain unchanged sans the Foucault modulation, but the
latitude, longitude, and azimuth errors grow in proportion to the product of
Time
Hours
Figure 7.10 Navigation and level errors for constant east gyro drift a t east terrestrial
velocity of 1900. knots.
136
T H E LOCAL-LEVEL TERRESTRIAL N A V I G A T O R
12
18
Time
24
30
Hours
Figure 7.11 Navigation and level errors for constant azimuth gyro drift at east
terrestrial velocity of 1900 knots.
%
the drift rate and time. Specifically, for times greater than a Schuler period,
61 w -(u)mNt sec L
E=
m (u)wNttan L
+ (u)wDt
(7-51)
(7-52)
137
leading t o Eq. 7 - 4 3 t h a t if the accelerometer compensation is performed without error, the appropriate equations corresponding t o Eq. 7 - 4 3 for the
stationary case is given by
'6L - geE
-Ey
+ ga,
r,~g
r cos L
at'
(7-53)
(u)fiV
+ (u)f E
(7-54)
mi, sin L
mi, sin L
- s cos L-
-mi, sin L
- w i a cos L
mi, cos L
mi, cos L
--g
rs2
s2r cos L -
+~
~ (-0sin
) L 81(0)
s sin
I-
TDmiesin L
WGD-
(7-55)
Solution of Eq. 7-55 for the case of constant gyro drift yields
138
EE
6E
mie sin L s .
( s 2 m:,(s2
m:mie sin L
s(s2
+ m:)(s2 +
( U ) ~ L ) N
( u ) ~ N
(s2
(,2+%
+ mWi2,)
m",
(COS
1
miet - cos ~ , t ) ( ~ ) m ,- sin mst(u)mE
ms
tan L
cD ~5-sin wiet(u)co,
coie
sec L
139
(1 - cos coiet)(u)coE
mie
1
+sin wiet(u)coD
mi e
6L
rsi
1
+sin wiet(u)co,
mie
coiet cos L
L
+ sin2
cos L
L
+ sin
--- (coiet - sin coiet)(u)coD
Q'i e
>
6LRSs = -- JZ
sec L
"RSS
(U)W
*ie
O
W
(1 - COS Wiet)l/i
mi e
[ ~ : ~ t 2(1
~ - cos w i e t ) p
a t L = 45"
(7-69)
T H E LOCAL-LEVEL TERRESTRIAL N A V I G A T O R
Time, hr
Time, hr
&
4-
25
" 20 d
(
*
[wie2t2
Oie
+ 2 (1 - COS uiet)]%
10
12
14
16
18
20
22
24
Time, hr
Figure 7.12 Local vertical INS navigation errors (perfect coriolis compensation);
root sum squared.
Note that the level, azimuth, and latitude errors are bounded, but that the
longitude error increases without bound with a rate which is approximately
given by the gyro drift rate uncertainty. These plots also describe the situation where the gyro drift rates can be described as members of an ensemble
of constant functions with an RMS value of 1 meru. See Appendix C for a
description of the statistics of the ensemble of constant functions.
141
7.4.3.2
If
bias is the sole error source, we have from Eq. 7-47 that
w i e sin L
- a i e sin
w i e sin
0
-mi,
cos L
-8
mi, cos L
-9
rs2
- 2 r w i e sin L s
mi, cos
cos L
s sin
r w i , sin 2 L s
s2r cos
L -
here (u)fNand (u)fEare the constant north and east accelerometer biases,
respectively. Figures 7.13 and 7.14 show the results of a computer solution of
12
18
Time
F i g u r e 7.13
'L
24
Hours
30
12
Figure 7.14
24
18
Time
n
,.
30
Hours
Eq. 7-70. Note that the Schuler mode predominates since the accelerometer
bias directly excites the relatively "high gain" level loops. The Schuler
oscillations are modulated a t the Foucault mode frequency of 1 cycle134 hr.
The maximum sensitivity of latitude error to accelerometer bias is seen t o
be in the range of 7 arc-min/mg bias. Similarly, the longitude sensitivity has
a maximum value of about 9 arc-min/mg bias.
If the effect of the accelerometer compensation is neglected, as was done
in obtaining analytic solutions for gyro drift, Eq. 7-55, the following solutions are obtained :
EN
4 f
= ( 1 - COS cost)(
-
&E =
4 f
- ( 1 - COS cost) (
5'
= -tan
4 f
L ( l - cos cost) (
n
36
Time
Hours
Figure 7.15 Navigation and level errors for accelerometer bias at east terrestrial
velocity of 1900 knots.
144
N~E
61 = sec L ( l - cos cost) (-
(7-75)
g
Note that these solutions neglect the effects of the Foucault modulations,
first-order effects. I n addition, the cross coupling effects shown in Figures
7.13 and 7.14 are completely neglected. If, however, the analytic solutions
above are compared with the computer generated solutions, i t is concluded
t h a t neglecting the accelerometer compensation yields results which are quite
accurate for periods of time up to about one Schuler period (84 min). Thus if
one is interested in modeling a local vertical inertial navigation system for
short periods of time, which would be the case for an aided inertial system,
the simplified model obtained by neglecting the accelerometer compensation
would be perfectly adequate.
E'igure 7-15 shows the effect of a 1900 knots east terrestrial velocity on the
error response t o accelerometer bias. The Foucault modulating frequency is
increased by a factor of j/co,, = 4 and the error sensitivities are seen to remain unchanged. I n the limiting case mentioned previously when 1 = -mi,,
the Foucault modulation disappears completely leaving a pure Schuler
oscillation. In addition, the cross coupling is eliminated and the response is
accurately described by Eqs. 7-71-7-75.
7.4.3.3
Latitude and Longitude Rate Errors. Figure 7.16 shows
computer derived plots of latitude and longitude rate errors for the case of
constant gyro drift and accelerometer bias. These errors are easily related t o
the north and east velocity errors, since for the stationary case, Eq. 7-42
shows that
BUN = r d j
(7-76)
6v,
=r
Bt cos L
(7-77)
where 6vN is the north velocity error and 6vE is the east velocity error.
It is seen, therefore, that the north and east peak velocity sensitivity to
1.7 ft/sec), while
level gyro drift is about 1.35 nm*/hr/meru drift (1nm/hr
the sensitivity t o azimuth gyro drift is about 0.75 nmlhrlmeru drift. Peak
velocity errors due to accelerometer bias are seen to be about 1.25 nm/hr/10e4
g bias. Note the particularly interesting effect of the three system modes of
oscillation in response to level gyro drift.
Latitude and longitude rate error plots are shown in Figure 7.17 for the
case of a constant east terrestrial velocity of 1900 knots. Comparison of
Figures 7.17 and 7.16 reveals that the rate error magnitudes are unaffected
nm = nautical mile.
Time
Figure 7.16
Hours
by the vehicle motion, a result which is not too surprising since the level
error magnitudes were previously shown to be virtually unaffected by vehicle
motion. Note that for the case of this rather high terrestrial longitude rate,
Eq. 7-77 does not yield the total east velocity error. I n particular, for
A # o,,,
the appropriate expression for the east velocity error is given by
Figure 7.17
12
Time
18
Hours
Latitude and longitude rate errors a t east terrestrial velocity of 1900 knots.
T i mesHours
Figure 7.18
12
18
Time
Figure 7.19
24
Hours
149
12
18
Time
Figure 7.20
'L
24
30
3'6
Hours
150
T H E LOCAL-LEVEL TERRESTRIAL N A V I G A T O R
12
Time
Figure 7.21
18
24
30
Hours
for the case of Figure 7.7 and the same numerical sensitivity with the appropriate units for the case of Figure 7.16.
For the purposes of design, i t is convenient to obtain analytic expressions
36
151
12
18
Time
F i g u r e 7.22
'L
24
30
Hours
for the response t o initial condition errors. As before, this solution is most
conveniently effected by solving the matrix Equation 7-47 with the Foucault
modes omitted. The results of such a solution for arbitrary finite constant
celestial longitude rate, A = constant, is given by the following equation:
where x2 = {
E ~ E, E , E
6 ,L , 61, 6 ~61),
36
CHAPTER
8
Development of a Unified Error
Analysis
I n Chapters 6 and 7, two inertial system mechanizations were described in
detail and their mechanization equations were developed. A linear, perturbation type error analysis has shown t h a t these two systems, while mechanizing
and computing in two different reference frames, have identical characteristic equations, providing t h a t position estimates are obtained using
inertially derived altitude information. An external source of altitude information, mixed with the inertially derived altitude information, was used in
the gravitational field calculation. The major error sources, t h e gyro and
accelerometer errors, have been shown t o force the system dynamical equation
a t frequencies which depend on the relative motion between platform and
geographic coordinates. These results indicate that i t ought t o be possible
to develop a generally applicable error analysis for a wide class of inertial
navigation systems, providing that the general system model is properly
formulated.
8.1
A G E N E R A L TERRESTRIAL N A V I G A T O R M O D E L
153
154
0
Thus the forcing function, Qj, must be evaluated for each of t h e system
configurations. Note t h a t under the assumptions above, both the mechanized
frame (the frame tracked by the platform) and the computation frame (the
frame in which the measurements are resolved to extract the position and
velocity) are arbitrary.
A functional block diagram, incorporating the assumptions above, for the
general terrestrial inertial navigation system is shown in Figure 8.1. An
accelerometer triad is mounted on a platform which can be either stabilized
or considered as part of the vehicle, in which case rate gyroscopes are
available t o measure the vehicle's angular velocity. Mixed platform-strapdown systems which are not stabilized along all three axes are included in
this model. The measured specific force, Fa, is first transformed into the
h
platform frame obtaining ", which is used to provide gyro compensation.
(See Chapter 5).Next a transformation is made into the mechanized reference
frame, the j frame, via an initial system alignment procedure. Note that the
transformation can represent either a computed transformation matrix
which was the case for the space-stabilized system treated in Chapter 6
or a physical transformation where the platform and mechanization frames
ci
triad
torquers
Gyro stabilized
platrorm
rrraa
6:
lj
ijk
compensation
A
rk
ik
===+
Velocity
computation
Position and
angular rate
computation
Platform command
Gyro
nonorthogonality
Initial system
alignment
Accelerometer
nonorthogonality]
Gravitational
field
calculation
Altitude
measurement
(2i iil
Physical coupling
longitude
altitude
conversion
156
are assumed t o be aligned, which was the case for the local-level system
treated in Chapter 7. The mechanized frame-referenced specific force is then
transformed into the computation frame, the 7c frame, via a coordinate
transformation matrix, e:, which is computed on the basis of the system's
computed position and/or on the rate gyro measurements if the system is
of the strapdown variety. Note that in general a transformation matrix is
computed relating the gyro axes to platform axes,
as discussed in Section
3.8.4.2. I n the error analysis, however, it will be assumed t h a t both the
and
matrices are equal t o the identity matrix such that the effects of
gyroscope and accelerometer nonorthogonality can be evaluated.
The computation frame referenced specific force, f k , is compensated for
the gravitational field effects t o obtain the computational frame referenced,
inertial acceleration, e:gi. The acceleration is then integrated twice with
appropriate initial conditions in velocity and position t o obtain a n estimate
I
of the earth's geocentric position vector, coordinatized in the k reference
frame. The position vector, 1", is used to compute the system's latitude,
longitude, and altitude and, with the aid of external altitude information,
the gravitational field compensation. The system's earth-referenced velocity
is calculated based on computation frame-referenced velocity, and position
information and angular velocity information is utilized t o provide the
platform commands and t o update the
matrix.
I n deference t o John Harrison,ll a n accurate clock is needed for a complete
solution of the navigation problem. The clock is, of course, implicitly present
in the guise of the computer.
e:,
e;
e,9
8.2
GENERALIZED M E C H A N I Z A T I O N A N D ERROR
EQUATIONS
The mechanization and error equations for the general terrestrial inertial
navigation system model will now be developed. Note that both the computation frame, k, and the ideal platform mechanization frame which shall be
referred to as the j frame are completely arbitrary. Typically the j and k
frames will correspond t o one of the reference frames dischssed in Chapter 3,
but other reference frames can be used as well.
8.2.1
As illustrated in Figure 8.1, the specific force measurements are transformed from accelerometer axes, a , t o computation axes, k, via
157
cz, is mechanized
e"
eEO
e;
e;e&
e:
e:
ef
If the system is of the strapdown type, then rate gyros are available t o
measure the vehicle's inertial angular velocity. I n this case the vehicle or
body frame corresponds to the platform frame and the platform and j
frames are coincident. The angular velocity,
can then be computed
based on the rate gyro measurements and the platform-computation frame
transformation can be computed via
(8-4)
158
D E V E L O P M E N T O F A U N I F I E D E R R O R ANALYSIS
mi",= [I
+ (AC,) 10%
v T
Note that the angular velocity, c&, is ideally equal to 0 but because of the
system's errors, has a small but finite magnitude. The computed estimate of
the desired platform inertial angular velocity can be written as the sum of
the platform's ideal angular velocity and an error angular velocity :
Substituting Eqs. 8-7 and 8-8 into 8-6 and neglecting products of error
quantities shows that the platform error angular velocity is given by
c; = c;s;v
Ci(0) = CL0
159
CgO= [I
+ D"]
(8-12)
The elements of DP are found from Eq. 8-9 by substituting Eq. 8-11 and
recognizing t h a t
(8-13)
cox = (d,, d,, d,] = d"
Thus Eq. 8-9 yields a vector differential equation for the platform error
angles of the form :
eg,
ez
c,
eg,
Because of the system's errors, the relative angular velocity between these
two frames cannot be precisely calculated, and the transformation will be
in error. The computed transformation in general is given by
where, as usual, the elements of the skew-symmetric matrix, I?, are the
error angles which represent t h e transformation errors between the
mechanized and computation frames. The computed angular velocity can
be written as
= C:(Q:,
m:,)Cj
(8-17)
160
f k = (I -
rk- c;zjc;)cD0r
k
(8-20)
Thus if Eq. 6-22 and the analytic expression for fa given by Eq. 7-19 are
substituted into Eq. 8-20, there results
161
Finally it is noted in the equation above that CEofD = CioC7fj. Thus the
analytic expression for C; given by Eq. 8-11 can be used' to introduce the
of the platform error angles into the expression above, resulting in
a$
ez
and letting
and
6fi:k
162
If Eqs. 8-23, 8-24, and 8-25 are substituted into Eq. 8-4, the resulting expression can be written in vector form as follo.cvs:
The computed inertial velocity of the platform, &&, is found directly from
t h e rate gyro outputs. From Section 5.3.2, i t is seen t h a t t h e output of the
rate gyro triad can be expressed as
As is seen from Eq. 8-27, positive uncertainty torques [+(u)M] give rise to
a n overestimate of the platform's inertial angular velocity. Thus t o conform
t o the sign convention used for the platform system Section 6.3.1. l , it is
necessary t o let
Substituting of Eqs. 8-5 and 8-28 into Eq. 8-27 and recognizing t h a t G&, =
o i p shows t h a t
(8-29)
6w$, = (AC:)Tw:P - TPw;! - (u)wP
The angular velocity of the reference computation frame relative t o the
inertial frame is based on the system's computed position unless, of course,
t h e inertial frame or some other "nonlocal" frame is chosen as the computation frame. (See Section 8.2.7) Thus a n analytic expression for t h e error in
this computation, 6wFk, cannot be expressed in terms of the navigation
errors without first specifying the computation frame. It is seen, however,
t h a t this angular velocity, O:k, must be coordinatized in platform axes for
use in Eq. 8-4. Thus
*P
Wik
ekwik
PAk
e:
163
P k
(8-30)
6wTk = C z 6wtk
C$ wfk
substitution of Eqs. 8-29 and 8-30 into 8-26 shows that the vector differential
equation for the platform-computational frame error angles is given by
k - P
(8-31)
[jk S2kpk= Sw?,
C,LT - (AC:)T]w&
C:(u)wp
The initial condition for the differential equation above, pIc(O), is the initial
system misalignment error arising from the inability t o specify t h e precise
initial relationship between the platform and computation frames.
The expression for the computed computational frame-referenced specific
force, f k , is expressed in terms of the error sources by substituting Eq. 8-23
into 8-1. Note t h a t from Eq. 8-22, ei = I, and that
= I since accelerometer
nonorthogonality is not compensated. Thus
A
e:
f '" = [I - Pk
+ C:(AC:)TC:]C,Lfp + C,k(u)fa
(8-32)
It is seen from the equation above that the transformation error is due t o
the error arising from the solution of the transformation matrix relating
platform t o computational coordinates, Eq. 8-4, and the error caused by
accelerometer nonorthogonality. The transformation matrix, Pp, has been
shown to satisfy Eq. 8-31 and is therefore a function of the error in the
computed inertial angular velocity of the computational frame, the
gyros' torquing uncertainty, input axis nonorthogonality, and drift
uncertainty.
8.2.2
The attitude error for the general terrestrial navigator is defined as the
orthogonal transformation error between platform and geographic axes,
coordinatized in geographic axes. This definition corresponds the transformation error involved in expressing a physical measurement made on the
platform in geographic axes. As was pointed out in Section 6.3.1.9, the
attitude errors do not, in general, correspond t o the platform errors, although
this correspondence holds for the local-level system treated in Chapter 7.
The attitude errors are derived separately for platform systems and strapdown systems in what follows; however, i t is shown that the resulting error
equations are independent of the system mechanization.
164
For either type of system, the evaluation of the attitude errors proceeds
from an evaluation of the transformation from platform coordinates to
geographic coordinates,
It is necessary to define a transformation
between computational and geographic coordinates, eg, which is calculated
based on the system's navigational information. It is assumed t h a t 6; is,
or is constrained t o be, orthogonal such that the following differential
equation holds
eg.
C;
C;PLk,,
(8-33)
ea
= SL;,
and
where Yn is the skew-symmetrical form of the error angle vector, rLn, which
describes the transformation error. These error angles are taken t o be positive
as measured about positive geographic axes. If the expression for the computed angular velocity and Eq. 8-34 are substituted into Eq. 8-33, i t is seen
that the error angle differential equation is given by
ei;
e:;
e; = ( I - En)C:
where the attitude error, En, is given by
The differential equation for the attitude error is found by writing Eq. 8-38
in vector form, premultiplying by Cg and performing a time differentiation.
are then used t o arrive at
Equations 8-18 for f k , 8-19 for dp, and 8-35 for
the intermediate result :
ILn
165
Because the mechanized reference frame, j,is rotating relative t o the platform
frame, p , with a "small-" error angular velocity, the terms involving
are second-order and o F j can be replaced by o&
in t h e differential equation
for the attitude errors. These modifications result in
Thus the differential equation for the platform system's attitude errors is
given by
i n+
- CE do:,
+ AC:)wrP + Cg(u)wv
- S2rn$n = Cg(TD
(8-39)
I n carrying out the matrix algebra involved in obtaining Eq. 8-39, products
of error quantities were neglected.
It is again noted that the p frame corresponds t o the body or vehicle frame
for the strapdown system, and that the p and j frames are identical. Thus if
Eq. 8-23 is substituted for
and Eq. 8-34 is substituted for
Eq. 8-40
.can be written as
e:
q = (I
e;,
En)C;
(8-41)
En = Y n+ C g Pk Cnk
(8-42)
166
second-order terms is
kn
+ a:,& - c;
as,.,
+n
+ c ~ ( u m V(8-43)
23 T
- (AC,)
= c;[T'
Comparison of Eqs. 8-39 and 8-43 reveals that the attitude error expression
for platform systems is identical in form to its counterpart for strapdown
systems, differing only in the gyro nonorthogonality forcing term.
8.2.3
fk = [I - CE(En - Wn)C:
+ C;(AC:)*C$]fk + CpL(w)fa
(8-44)
where the differential equation for En is given by either Eq. 8-39 or Eq.
8-43 and that for Yn is given by Eq. 8-35.
8.2.4
1, 2, 3 ;
all
(8-45)
It was shown that the resulting system equations could be written in the
following canonical form :
Ax = Qj
where for the space-stabilized system, j = i, and Q, is given by Eq. 6-63;
and for the local-level system, j = n, and Q, is given by Eq. 7-40. The
canonical matrix, A, as given by Figure 6.3 was shown to apply to both
167
systems for an arbitrary value of K , but only for the case of a = 0. I n other
words, for the canonical form t o be applicable it is necessary to use all inertial
information in the calculation of the system's latitude, longitude, and
&itude, but in the gravitational field calculation, any ratio of inertial t o
external altitude data may be used.
These seemingly inconsistent results are obtained because the navigational
calculations are, in general, nonlinear in nature. I n addition, i t should be
recognized that inertial systems which use external information t o stabilize
the vertical channel or t o calculate the system's position are properly
classified as nonlinear hybrid systems. Thus it is not too surprising, for
instance, t h a t in the calculation of latitude, longitude, and altitude from t h e
system's geocentric position vector magnitude (Section 6.3.1.7), it is found
that the linearized error relationships depend on both the choice of a and the
chosen computation scheme. It is then reasonable t o expect t h a t the error
response will depend on the method of combining the inertial and external
altitude information, that is, on the analytic form of the ri- estimator.
Obviously, other forms for the ri- estimator could have been chosen.
Instead of the nonlinear estimators which were used in the analyses of
Chapters 6 and 7, consider the linear estimator given by
where the weighting factor, v, can have the range of values 0 < v 5 1,
with v = 0 corresponding t o the all inertial case and v = 1 corresponding
to the all external information case. If the equation above is perturbed by
substituting from Chapter 6,
.i., = r
+ 6ha
It is seen that
Thus it is concluded that both the nonlinear and linear estimators yield
identical linearized error equations for all values of the weighting factors in
the range 0 < ( a = v) _( 1.
168
8ri
2v aha]
r
and
Thus it is seen that error equations obtained using Eqs. 8-49 and 8-50 will
differ from the corresponding equations obtained using Eqs. 8-51 and 8-52
even though there is a one t o one relationship between K and v, since for the
calculation of i2, K = 2v, and for the calculation of i3, K = 3v. Note, however,
that u is limited t o values between 0 and 1 while K can take on any value,
although as a practical matter K 2 2 t o ensure a stable altitude calculation,
a fact t h a t was previously gleaned from Figure 6.3.
The system error equations for the case where P is calculated using the
linear estimator of Eq. 8-46 are found by letting K = 3v in Eq. 6-62 for the
space-stabilized system and K = 2v in Eq. 7-39 for the local-level system.
This substitution, however, results in an expression for A which is slightly
different for t h e two systems, the A,, term containing the term,
for the local-level system. I n order t o keep the final error equations as general
as possible, i t is therefore recommended that the nonlinear estimator of the
form given by Eq. 8-45 be used in the gravitational field calculation:
p = (+a)K(+i)n-K
n = l,2,3,
Pa
+, + h
...
all
(8-45)
169
(8-53)
The calculation of the gravitational field vector for the general terrestrial
inertial navigator proceeds from the analytic expression given by Eq. 4-32,
which can be written as follows:
where the equatorial and polar constants, K , and K,, are given by
The analytic expression for the gravitational field in the arbitrary k computational frame is found by a simple coordinate transformation of Eq. 8-54
into the 7c frame. Thus
K,
0
0
(8-55)
r3
where
A
= G" -
G"
170
The gravitational field associated with the reference ellipsoid is found from
Eq. 4-38 to be given by
(8-59)
G: = Gk - CtAGn
where
(4-39)
AG" = { E g , -qg, 4 7 )
Substitution of Eq. 8-59 into 8-58 shows that
A
Gk = Gk
8.2.6
+ dGk - C S G n
The acceleration computation proceeds from the fact that the computation
frame-referenced specific force is ideally given by
But the inertially referenced acceleration is related to computational framereferenced acceleration via the relationship given by Eq. 2-7 :
F
c;(rk + 2ZL%tk+
+ PtkP:gk)
(8-61)
$k
=f
+ GL - 2fi:kik
A
h:kp -fifkfgkP
(8-62)
Fk
= rk
+ ark
a:k+
and substituting Eq. 8-60 for Gk, Eq. 8-44 for ?k, and Eq. 8-57 for 6Gk,
giving
171
where
Rk = skew-symmetric form of rk
and
Ws
8.2.7
JG
JP/+
Schuler frequency
This functional relationship might well be nonlinear as was the case for the
space-stabilized system as shown by Eqs. 6-14, 6-16, and 6-17. The important
point is t h a t this relationship can be specified in the computational frame.
172
(8-66)
Note t h a t 6h of Eq. 6-40 has been replaced by 6hi in the above t o emphasize
the fact that the altitude is calculated using the components of the computed
position vector. Equation 8-66 has the simple physical interpretation that
6L, r61 cos L, and ahi are, respectively, the north, east, and down projections
of the computation frame-referenced error vector. The navigation error
vector 6n should not be looked upon as being coordinatized in the geographic
reference frame, since t h e elements of 6n are in fact generated by three scalar
calculations involving inertially referenced position components. The errors
resulting from these scalar calculations have been arranged in t h e form
shown by Eq. 8-66 strictly for mathematical convenience.
The array of error quantities denoted as 6n is not equal t o the error vector
6rn, defined in the usual sense. To illustrate the distinction, consider the
case where the computed geocentric position vector is transformed from
inertial t o geographic axes. Since a computed coordinate transformation
matrix must be used t o effect the transformation,
The error vector 6ri can be found in terms of 6rn from the equation above
by letting Pn = rn
6rn, Pi = ri
6ri, and substituting :
C = (I - Nn)Cf
from Eq. 2-16, yielding
It is observed t h a t
6n # 6rn
Thus Eq. 8-66 is properly interpreted as a relationship between error vectors
rather than a relationship between the components of the computed position
vectors, 3" and Pi.
If the computation frame is a local frame as previously discussed, it is
necessary to transform the computed position vector, Pk, into an auxiliary
reference frame in order t o extract the latitude, longitude, and altitude.
It would then appear t h a t a n error relationship of the form given by Eq8-66 might not apply because of the possible introduction of transformation
errors. To investigate the possibility, consider the case where the computed
position vector is transformed into inertial axes and the methods of Section
6.2.5 are used t o compute the latitude, longitude, and altitude. The inertially
173
The error vector, 6ri, can be found in terms of 6rk by perturbing the expresmust first be specified.
sion above in the usual manner, but the form of
Recall that the computation, k, frame, while completely arbitrary, is
related t o the inertial reference frame in some definite manner. This translates
into the practical requirement that the computer must be able t o calculate
8: based on the system's computed position and/or on predetermined
relationships between the two reference frames. There are three possibilities
the first being that the k frame is related to the i frame
for the matrix
through some fixed relationship, that is, ei is a matrix of constants. Since
this transformation does not depend on computed navigational quantities,
can be calculated without error, that is, ci = c:. The second possibility
is that the k frame is rotating with respect to the i frame with some arbitrary
but a priori specified angular velocity, an example being the case where the
computation frame is taken t o be the earth frame, k = e . It is again argued
that since the transformation does not depend on the navigational quantities,
= c;. See Eq. 3-12 for the example of k = e. The third possibility
then
is t h a t the orientation of the k frame relative t o the i frame depends on the
system's computed position; that is, the computation frame is a local frame.
If this is the case then the
transformation can be written as a matrix
product such t h a t previously derived relationships can be used; t h a t is,
c:
e:,
e:
e:
e;
ci
c,"
jrn
\in
yn)
k
- C k n 6wik
(S-71)
where
h
k
8wik
= wik - wik
174
If on the other hand the computation frame is a local frame, then 60Fk # 0.
The computed angular velocity can in this case be written as
because, if the frame is local, then Gnk can be computed without error. It
follows, therefore, that
= 6-ik, = Ck, 80in, and
$"= 0
k local
+"
+ QLvn = dw,,
k local
(8-74)
Sri = C: 6rk
+ CL( N n - Yn)rn
(4-46)'
fkrk)
(8-77)
Gn
= vn
+ 6vn,
e; = (I - Yn)C;.
175
d;rn = C t 6n
k nonlocal
(8-80)
6vn = 6 i
+ a:nd n - N n v n
(8-82)
6rk = C t 6n - C t N n r n
k local
(8-83)
k" i
"k
"k
Because the computation frame is local, conk
= wk,, and
&i
i
An
n
" k
k
6wTny and w e , = W e k 6 ~ : Thus
~ .
i e = O i e yW i n = Oin
en = C,.
k
C;Nhwje
Also
(8-84)
Substituting Eqs. 8-83 and 8-84 into Eq. 8-79 and observing t h a t
rn = vn - Sznrn
from Eq. 4-49 yields
where it is recalled from Chapter 2 t h a t the notation ( ) * denotes the skewsymmetric form of the quantity within the brackets. The bracketed quantity
premultiplying rn in Eq. 8-85 is now examined and found t o be equal t o 0.
To obtain this result, the order of this quantity is reversed and it is observed
from Eq. 2-10 t h a t -S2,",Nn
NnS2:n = -Qznvn. Thus
-+
176
D E V E L O P M E N T O F A U N I F I E D ERROR ANALYSIS
for 6x1,Eq. 4-51 for vn, and Eq. 2-17 for the elements of Nn into Eq. 8-82
shows that
8.3
C A N O N I C A L FORM O F T H E ERROR E Q U A T I O N S
i t will be necessary to group the attitude error equations, Eq. 8-39 for
platform systems and Eq. 8-43 for strapdown systems, with the position
error equation given by Eq, 8-65. To accomplish this grouping, the error
,
and
equations must be written as a function of the attitude errors, E ~ cE,
cD; and the navigation errors, 6L, 61, and ahi.
Examination of the attitude error equations reveals that a relationship is
needed between the computation-geographic frame transformation error,
$", and the error in the computed geographic-inertial angular velocity,
6mfn. The required relationship is found from the basic relationship given by
Eq. 2-4 when applied to the geographic-inertial transformation :
a;
c?:PL:nc:
e;
Gn
+ SFnvn=
C:(I
+s
+ C i dm,,
where in obtaining Eq. 8-88 it was noted from Eq. 2-10 that
and
-+ Nn) from
P yields
~
(8-88)
177
Equation 8-88 is a very useful relationship between vn and $" since i t can
be used to eliminate *n from the 'attitude error differential equations.
Equation 8-88 can also be used t o effect a closed form solution for the
geographic-inertial error angles, since
"k
+ v n ) ( ~ ; , + SW:,)
enwin = C:(I
k A n
+ 6wFn, then
= w:,
SOL= C2(6w:,
+ 'YnwYn)
(8-89)
Substituting Eq. 8-89 into 8-88 shows that the differential equation for vn
is given by
Si",vn = 6 ~ : ~
(8-90)
+" +
min =
{jcos L,
-i,-;i sin E )
Thus
Sw:,
(-A
sin L 6 L
-6L,
-61 sin L )
(8-91)
Note that the initial longitude error has been ignored such t h a t 61 = 6A.
The solution for vn is, of course, the same expression as that obtained from
direct perturbation of the C r matrix of Eq. 3-10. (See Section 2.5.1.1.)
Equations 8-88 and 8-91 are substituted into Eqs. 8-39 and 8-43 showing
that the attitude error equation can be written as
-1
sin
-L
A sin L
-2
A cos L
A sin L
cos L
-c;
Lp] r n 6L
A cos L
61
sin Lp
q j = q,
+ C:TDwrD + CEACzwiP,
= C%(u)wD
qj = Pa
= C;(u)wb
Note that the body frame, b , designation has replaced the platform frame
designation in the strapdown system forcing function given by Eq. 8-94.
The forcing functions for the two classes of systems are seen t o differ only
178
where Yn = V n- Nn, and it is recalled from Section 8.2.7 that if the computation frame is local, then Yn = -Nn, otherwise Yn = 0.
Equation 8-95 is first substituted into Eq. 8-65 and the resulting expression
is premultiplied by C i . Similar to the development of Eq. 6-57 i t is noted
that
The bracketed terms premultiplying 6x1 and Yn in Eq. 8-97 are simplified
by observing that
and that
179
+ LRS,:,:
+ w:Mg)
6n
(8-100)
)Yn =
(a&,
-I- S2:nQL + w:M3)Yn
(8-101)
+2
P 3
+ d?,rn + SZrnI,Pnrn - Gn
(8-103)
- Gn)
(8-104)
6ii
6n
+ (En - Nn)fn + 4, = q1
(8-105)
where
q2 = 2(Jin
+ nPnyn- YnS2Yn+ C:
6Qkzkck)Gn
n
4- [ y n
zn
2n
6QfkCz
(8-106)
yn + aYnyn
+ Bipn - yna;n
- ynfi:,
- -C:[(Qfk
nfn)
- Q:,)]C;
(S-107)
Thus substitution of Eqs. 8-106 and 8-107 into the expression for q, shows
that
q, = c . o : ~ , Y ~ r ~ YnGn
-wsr2
Thus q, is written as
q2 = w:(M3Yn - Yn)rn
180
Substituting the expression for M, from Eq. 8-96 shows that, independent
of the values assigned to the error angles of Yn,
This equation is recognized as being identical to that obtained for the spacestabilized system as a comparison with Eq. 6-58 shows. Thus Eq. 8-109 is
written in component form as Eq. 6-59. Combining Eq. 8-92 for the attitude
errors and Eq. 6-59 for the position errors results in the error equation for
general terrestrial navigator :
Ax = Qj
where the characteristic matrix, A, is given by Figure 6.3, the state vector
x is given by
(8-111)
x = ( E ~ eE,
,
E ~6L,
, 61, ahi)
and the forcing function, Qj, has the general form:
Note in the expression above for the forcing function that it was necessary
to distinguish between the strapdown and platform cases when evaluating
the effects of gyro nonorthogonality. I n particular, it is seen from Eqs. 8-93
and 8-94 that for the platform case the nonorthoganality term appears with
the plus sign, while for the strapdown case this term appears with the minus
sign and the transpose.
For the case of K = 0,that is, the case where all inertial information is
used t o specify the magnitude of the gravitational field vector, the characteristic matrix can be partitioned into four 3 x 3 matrices:
where Fnis the skew-symmetric form of f n given by Eq. 4-53 ando:, is given
by Eq. 3-8. Note that A is a function of only two quantities, o
:
,
, and fn,
181
which points out the dependence of the system's motion over the earth on
the unforced error equation.
It has been shown, therefore that the error equations for the general
terrestrial inertial navigation system can be written in the form shown by
Eq. 8-110. Note that the dynamical behavior of all systems is identical, as
evidenced by the fact that the characteristic matrix, A, does not depend on
either the system's mechanization or its computation frame. Equation
8-110 is a complete specification of the system's dynamical behavior in
response to error sources. Since the equations for 6L, 61, and ah, in Eq. 8-110
are second order, a total of nine initial conditions are needed t o effect a
solution :
The velocity errors are then related to the navigation errors through Eq.
8-86.
The forcing function, Qj, is characterized by the fact that the major sources
of system error, due to gyro and accelerometer uncertainty, C;(u)wP and
C;(u)fa, are frequency modulated a t a rate which depends on the angular
motion between the instrument coordinates and the geographic coordinates.
This fact accounts for the major differences in performance between the
various mechanizations. Errors due to gyro torquing uncertainty, C;T%&,,
are similarly modulated by the rotation of the platform with respect to the
geographic frame, but in addition are proportional to the inertial angular
velocity of the platform. Error effects due to gyro input axis nonorthogonality, -&C:(AC:)',~~%, depend on the commanded inertial angular velocity
of the platform for the case of platform systems and on the inertial angular
velocity of the vehicle for the case of strapdown systems. Errors due t o
accelerometer input axis nonorthogonality , Ci(ACz)TCzfn, are identical for
all systems depending on both specific force and platform-geographic frame
rotations. The effects due to altimeter uncertainty depend on the value of
the weighting factor, K , and are seen t o be independent of the motion of the
system. Similarly, effects due to the inability to specify an exact analytic
expression for the earth's gravitational field, AGn, do not depend on the
system's motion.
The dynamic behavior of the inertial navigation system can be investigated by examining the system's characteristic determinant which, for the
stationary case, is given by the following ninth-order polynomial.
182
,
The dynamical behavior of the system depends on both the nature of the
forcing function, Qj, and the roots of the characteristic determinant. It is
immediately seen that the terrestrial navigator has an imaginary pair of
roots a t p = &oje, the so-called space rate frequency. I n addition to the
root a t the origin, p = 0, the other six roots depend on the gravitational
field weighting factor, K . See Eq. 8-56.
For a value of K = 0, the system uses all inertial information t o calculate
the earth's gravitational field. The characteristic determinant is given by
1 A1
= --r2
cos Lp3(p2
K=2
The quadratic formula can be used to factor the quartic polynomial in the
expression above, yielding, for the case where the system is not near the
equator,
The factorization given by Eq. 8-116 is the same as that obtained for the
two-accelerometer local-level system. (See Section 7.4.2.) This system, like
the two-accelerometer local-level system, will therefore exhibit both the
Schuler and Foucault frequencies.
For a value of K = 3, the gravitational field magnitude is calculated using
a stable mix of external and inertial altitude information. For this case the
characteristic-determinant is given by
183
8.4
SPECIALIZATION O F T H E
GENERALIZED THEORY
The error equations for the general terrestrial navigator having been
obtained, i t is now a relatively simple matter t o write down the error equations for a variety of inertial navigation system configurations. The specification of the error equations has been reduced t o the specification of the
forcing function, Qj, given by Eq. 8-112. Note that because the error equations are completely independent of the chosen computation frame, it is
only necessary t o evaluate various platform configurations. For example, a
space-stabilized system which computes in inertial coordinates will behave
in identical fashion to a space-stabilized system which computes in geographic
coordinates.
I n writing down the final error equations for a particular system i t is
instructive t o formulate the forcing function in terms of the various types
of gyro errors, such as mass unbalance and anisoelastic. (See Chapter 5.)
Since certain of these errors depend on the orientation of the gyro relative
to t h e platform, i t is necessary t o specify a gyro orientation. It is t o be
emphasized t h a t the gyro orientation chosen here is completely arbitrary
and, in a practical situation, is chosen based on the expected dynamic
environment. For example, if i t were desired t o minimize errors due t o gyro
mass unbalance in a local-level system, the level gyros would be mounted
with their output axes along the vertical, thereby aligning the level gyros'
spin-input planes orthogonal t o the gravity vector. Designating the gyro by
the direction of its sensitive axis, the orientation of the output and spin axes
184
where
Please note that both the choice of instrument orientation and the choice
of the "major" terms from Eqs. 5-14 and 5-20 are rather arbitrary. If gyros
of known design are to be analyzed then all of the terms in Eqs. 5-14 and
5-20 must be evaluated to determine the "major" terms. This procedure is
in general quite complex since the platform's dynamics, the gyros' temperature control systems and magnetic environment, and the vehicle's dynamic
behavior must all be evaluated.
If the terms given in Eq. 8-119 are indeed the major terms, then compensation is usually provided which theoretically reduces the gyro's deterministic
uncertainty to a second-order quantity. The gyro uncertainty is then modeled
185
where
For the purposes of modeling the accelerometer errors, the model given by
Eq. 6-23 is used; t h a t is, the accelerometer uncertainty is modeled as the
sum of a bias, scale factor, and random uncertainty.
8.4.1
The error equations for the space-stabilized inertial navigator have been
independently derived in Chapter 6, Eq. 6-63 being the analytic expression
for Qi. The general expression given by Eq. 8-112 is now applied to the
space-stabilized mechanization, and the detailed instrument uncertainties
given by Eqs. 8-119 and 6-23 are applied. It is assumed for simplicity t h a t
the platform is nominally aligned with the inertial frame; thus p is replaced
t
by i in Eq. 8-112 and Qiis written as
la:(u)fa - AGn
(8-120)
It is seen, as previously noted, t h a t the gyro torquing and gyro nonorthogonality terms are negligible for this mechanization.
I n evaluating Eq. 8-119 for this case, it is noted that the terms which are
proportional t o angular velocity and angular acceleration drop out, since
mi,= oZi
= 0. It should be recognized, however, that in practice i t is not
186
Local-Level Mechanization
I n Chapter 7 the local-level system was described in detail and its error
equations were specified. Equation 8-112 is now applied and the detailed
instrument uncertainties specified. It is again assumed for simplicity that
the platform and instrument frames are nominally aligned with the mechanized frame. Replacing p with n in Eq. 8-12 shows that
and an analytic expression for fn is given by Eq. 4-53. Also the inertially
referenced platform angular velocity is given by Eq. 3-8 as
w
,:
= mi", =
(CON, W E , COD) =
(1. COs L, - L ,
-j, sin L )
187
It is seen from Eq. 7-40 t h a t the gyro error terms are not frequency modulated and thus do not excite the system a t one of its natural frequencies.
Note, however, t h a t this system mechanization is sensitive to gyro torquing
scale factor uncertainty. Note also that with the chosen gyro orientation
that the north and east gyros are insensitive t o gravity-induced mass
unbalance errors.
8.4.3
Free azimuth systems are essentially local-level systems with a spacestabilized vertical or azimuth channel; that is, the platform is commanded
t o remain in the local horizontal plane but is uncommanded in azimuth.
The relationship between the platform coordinate frame and the geographic
frame is specified by defining an azimuth rotation angle measured clockwise
from indicated north to a fiducial line on the platform. Thus the coordinate
t,ransformation relating the platform frame to the geographic frame is given
ios
[JOtA
sin L dt]
sin[Jot~sinLdt]
- sin
[c
sin L dt]
0-
c o ~ [ ~ ~ ~ ~ s i n l ;0d t ]
1-
The angular velocity of the platform frame with respect t o the inertial frame
is found by transforming w,",, given by Eq. 3-8 sans the vertical component,
into the platform frame using the equation above, that is,
w;D = c;{A cos L, -L, 0)
(8-124)
where C," is given by Eq. 8-122 and f" is given by Eq. 4-53.
Because the platform of the free azimuth system is inertially nonrotating,
the system has the obvious advantage of being insensitive t o azimuth gyro
torquing uncertainty.
8.4.4
188
Since, as ha.s been previously pointed out, inertial navigation systems behave
a s low pass filters in response t o error sources, the frequency modulated
instrument errors will be attenuated in the ratio of the Schuler frequency
t o the modulation f r e q ~ e n c y . ~The
'
accuracy and alignment requirements
for the level instruments is thereby reduced. The system is not without its
problems, however, since utmost care must be exercised to assure that the
level gyro's sensitive axes are truely orthogonal t o the azimuth rotation
angular velocity vector.
While other mechanizations utilizing azimuth rotation are possible,l it
is assumed t h a t the inertial angular velocity of the platform is given by
where
Thus the constant azimuth rotation rate, 4, is defined as relating the platform
and geographic coordinates through the relationship :
r cos $t
G: = -sin 4t
sin $t
cos 4t
01
o
Note that the azimuth torquing uncertainty error will be relatively large
for this mechanization since the uncertainty is multiplied by the large
angular rate
- A sin L. This error can be reduced if t h e system is
mechanized t o provide a constant level of azimuth gyro torquing.*
8.4.5
Strapdown Mechanization
SPECIALIZATION O F T H E GENERALIZED T H E O R Y
189
190
D E V E L O P M E N T O F A U N I F I E D E R R O R ANALYSIS
where o:Pis given by Eq. 8-127 and the platform-referenced specific force
is given by
f P= C:fn
(8-129)
The Cz matrix depends on the vehicle's motion and cannot be specified in
advance. For the purposes of error analysis a trajectory must be assumed
and an attitude specified.
I n addition to the error sources treated for gimbal systems, the strapdown
mechanization gives rise to several other error sources which are usually
m ~ d e l e d . ~These
. ~ ' errors can be analyzed by expressing the error contribution as an equivalent angular velocity and including in (u)oP.A partial list
of additional sources would include the following :
Gyro torquing asymmetry
Noncommutivity errors arising from the use of angular data
Truncation error
Gyro and accelerometer quantization error
Computer round-off error
It is first noted from Figures 5.2 and 5.4 that in the rate gyro mode the
signal generator output, which is a voltage proportional to A,, is affected by
ol, A,ws, and tho. The gyro output signal must therefore be compensated
for these dynamic effects if high accuracy is to be achieved. This compensation is readily accomplished since coo and ws are obtainable from the other
gyros. Note, however, that the addition of this cross coupling between the
gyros introduces the possibility of a computational i n ~ t a b i l i t y . ~ ~
8.4.5.1
Gyro Torquing Asymmetry. Errors due t o torquing asymmetry arise from the fact that the gyro torquer may have a scale factor
error which depends on the polarity of the torque being applied. Symmetric
scale factor error effects are discussed in Sections 5.2.2 and 8.2.1.2. The
effects of torquing asymmetry are best seen if an angular vibration environment is considered. For the symmetric case, i t is seen from Eq. 5-19 that
a sinusoidal angular oscillation about the gyro's input axis results in a zero
mean angula; velocity error equal to
(u)w = -w,
sin wt
191
( u )=
~ -T+wIAJ
mt dt
(8-130)
Thus for each vibration cycle an angular error results which is proportional
t,o scale factor asymmetry. Let us evaluate an example to see what the magnitude of this buildup might be. Let us say that the vehicle is vibrating a t
cu = 10 Hz(cps) with an amplitude of 1 arc-min. Thus if we assume that
the per cycle angular error is
(7- - T+) =
arc-min
( u p = 2 x 10-~
cycle
For a 2-hr flight, the accumulated error is equal to
[(U)O],,
,=2
.Y
arc-min
10-~
cycle
10Hz x 2 h r x
3600 sec
hr
1.44 arc-min
This effect would appear to be quite significant and probably requires that
the designer make a detailed evaluation of the angular vibration environment. I n a particularly severe case, shock mounting would probably have
to be employed. For a constant sinusoidal vibration along each gyro input
axis, Eq. 8-130 in vector form can be expressed as an angular velocity
uncertainty :
= vibration frequency
Obviously, for Eq. 8-131 to be used effectively, the angular vibration spectrum must be known. This type of data are rather scarce for aircraft applications and, in addition, would tend to be strongly influenced by the aircraft
type, mission, inertial measurement unit location, and so on.
192
DEVELOPMENT O F A UNIFIED
E R R O R ANALYSIS
8.4.5.2
N o n c o m m u t i v i t y Errors. Noncommutivity effects result
from the fact that the attitude matrix computer processes finite sized angular
outputs from the rate gyros. See Section 5.2.2 for a description of the digital
torquing process. To investigate the form of the error introduced into the
calculation of the attitude matrix as a result of noncommutivity, consider
the case of three successive rotations about the body's positive x, y, and z
axes. The coordinate transformation relating the rotated coordinates to the
original body coordinates is given by
cos 6 ,
where b' denotes the rotated frame.
If the rotation angles are equal to the A0 pulse sizes, we can expand the
expression above, keeping up to second-order terms. We get an expression
of the form:
C,bf = (1
where
+ Ae1)(I + Ae2)
-Ae,
Ae,
A
0
AO,]
A8,
and
The off-diagonalterm ABj A8, occurs if a rotation about the j t h axis precedes
a rotation about the kth axis. The second-order term represents the noncommutivity error. Thus a direct error, as given by Eq. 8-134, occurs in
the computation of the attitude matrix.
Since the commutivity error is on the order of Aei, we choose angle increments as small as possible, consistent with computer speed and roundoff
error considerations. Unfortunately, the prediction of the commutivit~
error with time requires a complete time history of the input angular velocity.
Farre1122has evaluated the error buildup in response to angular oscillations
and finds the commutation error to be quite significant if the A8 pulse
sizes are not kept below about 20 arc-sec. Systems are currently being built
w i t h ~ u l s esizes in the 1 to 2 arc-sec range.
193
8.4.5.3 Truncation Error. Truncation error results from approximations in the computational algorithm which is used t o update the attitude
matrix as given by Eq. 8-4. Although we shall be concerned with the direct
updating of the direction cosine matrix, it should be pointed out t h a t other
schemes can be used t o effect the coordinate transformation such as fourparameter techniques and Euler angles. W e i n e F has studied the available
choices and concludes that for SDF, digitally torqued instruments utilizing
a DDA computer, the direction cosine approach requires minimum computation, although other i n v e ~ t i g a t o r s ~have
l . ~ ~ indicated a preference for the
Euler four-parameter method.
The direction cosine transformation is found quite easily for systems
which utilize electrostatic gyroscopes,15since ingenious pickoff schemes allow
the direction cosines t o be read directly from the instrument. The output of
each electrostatic gyro (ESG) pickoff is the direction cosine between the
spin axis and an appropriate fiducial line on the instrument case. Although
there are three of these pickoffs per ESG, in general, only two of the three
pickoffs provide useful information a t any given time. Thus if two ESG's
are used in a system, only four direction cosines will be available for computation a t any instant. The remaining five are found through application of
the orthogonality relationship for coordinate transformation matrices :
The solution of Eq. 8-4 for the direction cosines can, as mentioned previously, be performed in a variety of ways. If digitally torqued S D F instruments are used, i t is shown in Section 5.2.2 t h a t each output pulse is proportional t o the integral of the input axis angular velocity. Thus the output
of the instrument represents an incremental angular rotation about the
input axis equal to A6. This property can be exploited in the solution for
the direction cosine matrix if a Taylor series expansion for C,k in At is considered :
1
c(t At) = c(t) ~ ( t At
)
@ ( t ) At"
-c
+ . . (8-135)
3!
where C z = C for notational simplicity.
If the basic relationship given by Eq. 2-4 is substituted into the equation
above, there results
at3
C(t
S2 At
A9
194
D E V E L O P M E N T O F A U N I F I E D E R R O R ANALYSIS
8.4.5.4
Quantization Error. Quantization error, which is t o be distinguished from commutation error, is defined to be the error which results
from the digital measurement or conversion of continuous physical
quantities such as the angular position of the gyro float assembly. I n the case
of the gyros, quantization effects can result in a t most one bit of information
being lost during a mission. This low sensitivity to quantization effects
occurs because the gyro is itself an analog device with a physical "memory."
Treated statistically, the resulting error appears in the form of a random
phase shift. Thus by appropriate choice of quantization levels, the requiting
navigation errors can be reduced to negligible proportions.
Quantization effects become very important during alignment, however.
It is readily seen that for fixed base alignment the pulse rate is likely to be
very low. Thus long filtering times are necessary to smooth the data. In
addition, complications can be introduced by instruments which limit cycle
because they are being pulse torqued.@
8.4.5.5 Corn puter Round- Off Error. Round-off error is associated
with the fact t h a t the compute'r7sword length is finite. Each time a computation is performed the computer must approximate the final digit. This effect
is readily analyzed using statistical methods in order to determine the word
length required to yield a specified RMS error after a specified number of
computer iterations.67
195
8.5
EFFECT O F A L T I M E T E R U N C E R T A I N T Y
a&,,
Latitude = 45"
Navigation time, hr
Figure 8.2
~ = 5 4
K
= 2.5
Latitude = 45'
Navigation time, hr
Figure 8.3
6000-
K=
13
Latitude = 45'
Navigation time, hr
Figure 8.4
197
EFFECT O F A L T I M E T E R U N C E R T A I N T Y
the R M S
latitude and longitude errors are quite sensitive to the altimeter uncertainty.
On the other hand, it is seen from Figure 8.4 that the altitude error decreases
for large values of K. A value of K = 3 appears to offer a reasonable compromise between these two opposing trends.
K,
CHAPTER
Self-Alignment
Techniques
9.1
ANALYTIC COARSE A L I G N M E N T M E T H O D
199
The gravity and angular velocity vectors transform according to the following expressions :
n
gp = Cng
oi",= c;w;
If v is defined as v = g x o i e , we also have
up = c;vn
Thus the alignment matrix is uniquely defined provided that the inverse
indicated above exists. This inverse exists if no one row of the matrix is a
linear combination of the remaining rows. This condition is always satisfied
if the two vectors g and oieare not collinear. These vectors coincide only
at the earth's poles, where the analytic alignment procedure is useless. For
the general case, the inverse in Eq. 9-1 is given by
oie cos L
--aiesin L
gwie cos L
-1
-tan L
- sec L
mie
1
gmie
sec L
200
SELF-ALIGNMENT TECHNIQUES
Kasper38 shows that for fixed base alignment, the analytic scheme compares
favorably with the existing optical alignment methods.
Note that this method is not restricted to the case where the gravity and
earth rate vectors are the measured quantities, but can be applied to situations where sequential measurements are made of only one vector whose
components are changing with time. For example, in a platform system
such as the space-stabilized system (Chapter 6) the earth rate vector is not
a directly measurable quantity. The components of gravity are, however,
time varying in the platform frame. Thus the analytic alignment method
could be applied to determine the transformation between the platform
frame and a reference inertial frame. Note also that the method requires
that a vector quantity be measured and is, therefore, generally restricted to
three-accelerometer systems.
9.1.2
Error Analysis
An error analysis for this alignment scheme, which takes into account
the effect of instrument uncertainties and base motion is not readily amenable
to analytic methods. The analysis which follows is intended to indicate an
approach which will result in equations which are best solved on a digital
computer. The equation for C,"can be written in the form:
E:
MQ
where
The elements of M are constant a t any latitude and are given by Eq. 9-2,
but Q contains measurement uncertainties. The equation above can therefore
be written as
= M(Q
6Q)
where
A
e::
--
Thus
h
&ye,
201
When Eqs. 9-2 and 9-4 are substituted into Eq. 9-5 the resulting M 6QCg
matrix is, of course, not in the desired skew-symmetric form, since
Ma&=
where
and
Moreover, it is difficult t o directly apply the constraints so t h a t the lengths
of the measured vectors and the angles between them are constant. This
problem is resolved in a practical manner by requiring that the computed
transformation, C;, be orthogonal. This is accomplished, as shown in Section
2.5.1.2, by forming the matrix product :
where
(e3,T
e:
[(e;),
[(e;), e;]
where products of error quantities have been neglected. If the square root
is expanded in series, there results
202
SELF-ALIGNMENT T E C H N I Q U E S
and (E*),, k = N, E , D are the misalignment angles which result after the
orthogonalizing procedure and are given by
sec L )
mie
(ED),
=-
fy tan L +
9
sec L
mie
Thus i t is seen that one can expect to see a north level error of about
3.4 arc-minlmilli g east specific force uncertainty. The east level error due
to north specific force uncertainty is -1.7 arc-minlmilli g ; to down specific
force uncertainty,
1.7 arc-min tan Llmilli g ; and to azimuth angular
velocity uncertainty -1.7 arc-min sec Llmeru *. The aximuth error angle
due to the east specific force uncertainty is given by -3.4 arc-min tan Ll
milli g, and due to east angular velocity uncertainty is given by +3.4 arc-min
sec Llmeru. It should be pointed out that the uncertainties 6f, and 6wk,
k = x, y, z , represent the total specific force and angular velocity measurement uncertainties, respectively. These errors result from both the instrument
uncertainties and instrument nonorthogonality effects. See Section 3.8.4
for a discussion of these nonorthogonality effects.
If these results are compared with those which would-be obtained with
an acceleration coupled platform gyrocompass (Section 9.2), i t is seen that
the north level error, (EN),, is identical to that which would be obtained
using a platform gyrocompass. The east level error, (E,),, contains additional
terms due to the orthogonalizing procedure which are functions of the vertical
accelerometer and azimuth gyro uncertainties. Finally, the azimuth error,
(ED),, is seen to be identical for the two gyrocompassing systems.
0.015 deglhr.
203
It is also interesting to compare the results of Eq. 9-9 with those which
would be obtained if the instrument package could be rotated such t h a t two
accelerometers were nulled and the angular velocity about indicated east
were zero. I n this situation the conditions for equilibrium are found to be
dfv - g e =
~ 0
6f,
+g
&=
~0
dm, - mi,cos L E -~mi, sin L e N = 0
These results are seen to be the same as those obtained for the acceleration
coupled gyrocompass of Section 9.2. Note the apparent discrepancy in sign
in the azimuth sensitivity to east angular velocity uncertainty when the
results of this section are compared to those of Section 9.2. This sign difference comes about because for the analytic method the angular velocity is
directly measured, while for the physical gyrocompass method the gyros are
operating in the platform mode. See Section 5.2.1 and 5.2.2 for further
discussion on this point.
9.2
'
&DmieCOS L
where GE is the angular velocity about the east axis and cD is the
azimuth misalignment.
4. Since the signal from the north accelerometer is proportional to the
azimuth misalignment above, then that signal can theoretically be used
to drive the azimuth error to zero. Conceptually then, one could
mechanize a gyrocompass by (I) supplying earth rate commands to
the platform ; ( 2 ) providing tight level control ; (3) providing azimuth
nulling via the north accelerometer.
204
SELF-ALIGNMENT TECHNIQUES
Gravity coupling
East
accelerometer
North
gYro
KO
North
accelerometer
BE -
KE
- Instrumented
/
north
Azimuth
gYro
- Instrumented
azimuth
East
gYro
-- Instrumented
east
Gravity coupling
Figure 9.1
9.2.1
P l a t f o r m Gyrocompass E r r o r Analysis
where E
I
p
n = geographic axes
angles
205
Now the angular velocity of the platform axes with respect to inertial space
coie cos L
-E D W i e cos L
EEOie
EEWie
sin L
+ iN
- E N C L ) ~sin
~
L
+ ZE
+ iD
I::[
mi, cos L - K d e ,
+ 6mN + K N 6 f B
+ dmE K E 6fN
sin L - KDgeE + 6mD - K D d f N
-K..E~
(9-19)
-mi,
GD
where 6 0 " = ( d ~ r )6~w, E , dmD) = angular velocity uncertainty
6 f n = ( 6 f N ,d f E )
Note that i t is being assumed that the earth rate terms are being supplied
exactly. Note also that the signs of K N , K E , and K D have been chosen to
drive the 6 terms to 0. Equating Eqs. 9-18 and 9-19 yields
I:'[
+ BCVN + K N d f E
= [-KEgcE
+ e N m sinL + e D m eCOSL+ amE - K E 6fN
-KDgeE 3- eEwie cos L + dmD - K D 6fN
-KNgeN - eEmie sin L
(9-20)
g~
Taking the Laplace transformation of Eq. 9-20 and arranging in matrix form :
SELF-ALIGNMENT T E C H N I Q U E S
n
Figure 9.2
where s denotes the Laplace variable and the superbar denotes a Laplace
transformed variable. The signal flow diagram for this system is shown in
Pigure 9.2. Note the similarity between Figure 9.2 and Figure 7.5, the signal
flow diagram for the local vertical navigation system.
Since the determinant associated with the characteristic matrix above is
given by
Eq. 9-21 is readily solved using Cramer's rule. The equations are first solved
for the steady-state errors assuming constant angular velocity and specific
force uncertainties by invoking the final value theorem :
s-bo
207
Solution yields
&N,,
-[ B U N
9Kiv
KN
- - tan L
-
Caw,
K ~ s
'DSS
GfE]-
KN
mi, sin L
K,dK,g
afE] - mi, 1
COS
[dm, - K E GfNl
(9-26)
K N ~K' E ~KDS
,
>> mie
where the symbol ">>" implies a several orders of magnitude inequality.
Routh's criteria shows that for the inequality constraints above there is no
stability problem for this system providing that the K ' s are positive. Note,
tan L
--
K N ~
-
mi, COS L
'dm,
-mi, sin L
K i v g ( K ~ g mi, cos L
tan L
--
9
KDmiesin L
KNg(KDg mi, cos L
Figure 9.3
208
SELF-ALIGNMENT TECHNIQUES
Applying Eq. 9-26 to the steady-state coefficients of Figure 9.3, the coefficients take a simplified form as shown in Figure 9.4. Figure 9.4 points out
the important fact that the sensitivity E ~ / ( u ) Eis~ independent of system
gain. I t s magnitude of
- 3.4 arc-min meru drift
cos L
COie COS
- wiesin L
K,K,g2
1
9
wie sin L
KNg2
Figure 9.4
209
9.3
A L I G N M E N T O F STRAPDOWN SYSTEMS
(9-31)
mib = w i e Od
where the subscript, d, indicates disturbance quantities. Secondly, gb and
b
oie
become functions of time to some extent. This can be seen from the fact
that since &Te = 0,
b
b
hie
= -QLbmie
where the elements of the skew-symmetric matrix nibare given by the
components of w,. It is, therefore, necessary to introduce some filtering in
order to reduce the effects of these vibrations. A simple low-pass filter could
210
SELF-ALIGNMENT TECHNIQUES
be used t o obtain the average values of the measured quantities. This would
tend t o give t h e average alignment matrix. It is clear, however, t h a t the
instantaneous position of the body frame can vary considerably from its
average position, depending upon t h e motion of the aircraft. As a result, a
large initial misalignment could exist when the system is switched t o the
navigation mode of operation if only a n average alignment were achieved.
If the statistics of the aircraft vibrations were available, a more elaborate
optimal filtering scheme could be constructed. However, i t could prove
difficult t o separate the perturbations of w:e and gb from the disturbances
o,and f, by linear filtering, since it is very likely t h a t these components
contain the same frequencies. I n addition some time lag would be introduced
by t h e filter. The analytic alignment method is therefore mainly useful as
an average alignment, which is a rapid way of obtaining an initial estimate
of the transformation matrix.
The second or "corrective" alignment stage refines the initial estimate of
the transformation matrix by using estimates of the error angles between
the known reference frame and the corresponding computed frame.
9.3.1
+ fd
&g
J.
Body mounted
accelerometers
Zb
>fn
Ggb
J.
p)}
eb"
J.
ek
2 -
Filter
6ub
W e
+a d
mounted
g Y ros
Figure 9.5
;;b
6;
Gn
211
9.3.2
ii;
cr is updated using t h e
e;n;,
(9-32)
ate,
I,na
but
w:~,
W:
+ awb +
w:e
(e;)-lo;
(9-35)
where E is the skew-symmetric error angle matrix, defined in the usual way.
Thus Eq. 9-34 becomes
b
&:a = w,,,
+ w, + Sw
b
- Ebwie
(9-36)
(9-37)
212
SELF-ALIGNMENT TECHNIQUES
and
c,"= c;a;
zn = -olE",,
- 60" - a 2 c n
where it was noted that
-Eno;
n n
= Glie&
n
*cmd
=K&n
Note that Eq. 9-43 represents three scalar differential equations which are
coupled through the term nin,sn, which represents earth rate coupling.
The elements of 2" in Eq. 9-43 remain to be specified. A direct indication
of the three components can be obtained from the computed horizontal
components of g and the computed east component ofo,,. Specifically since
and
then
A
=
g
f~
+6 f
+ faa + a
E f~f d N
-BEN
(9-44)
f ~
(9-45)
where fdN and fdz are the north and east components of the disturbance
specific force vector, and 6fN and 6fE are the north and east components of
the accelerometer uncertainties, respectively. The remaining element, ED,
21 3
(9-47)
where mdl and 60, are the east components of the disturbance angular
velocity and gyro uncertainty, respectively.
The system is designed t o process the fN, fE, and G E measurements assuming t,hat there are no error sources.
sec~9
mie
The error in the estimation, &, is found by substituting
(9-48)
E = E + ~ E
f 6.f~)
(9-49)
(fa,
L
+ 6 f ~-) sec
mie
+ 8 w ~- )
(mdz
KE
0
-COie
COS
cos L
&I
O
(9-50)
214
SELF-ALIGNMENT TECHNIQUES
(P
+ K D ~ =D
tan L
Wie
215
di
2 - I
-+K & ( s ) =
e - K ( t - T ) ~ ( Td~
)
S:: : {
~ ~ (=t e -) K ~ t eK.'
-sec L[wdE(7) d w E ( ~ ) ]
Since the base motion is not specified, it is best to treat Eqs. 9-55 statistically. We find the mean-squared value by squaring Eqs. 9-55 and taking
the mathematical expectation of the result. If the statistics of the independent
variables are uncorrelated, that is, if the various random processes are
independent and if no more than one is biased, then the cross coupling
terms will drop out when the mathematical expectation is taken. This
laborious task is best left for computer solution.
We investigate the system dynamics for the simple case of zero base
motion, constant accelerometer uncertainty, and constant gyro uncertainty :
fdE(t) = f d N ( t ) = wdEW = 0
d f k ( t ) = dfk
constant
6 w k ( t ) = 6wk = constant
k:=N,E
=N,
E, D
f
dmE - tan L L
9
"D)
( 1 - .-KDt )
--
KD
E~(O)~-"D~
(9-56c)
216
SELF-ALIGNMENT T E C H N I Q U E S
0
0
l/g
0
F i g u r e 9.6
-'lKE
o
-119
0
l / m i e COS L
-'IK,
-tan L / g
0
These equations are summarized in Figure 9.6. Comparison with Figure 9.4,
which shows comparable information for an acceleration coupled physical
gyrocompass, reveals striking similarities between the two systems. Note
t h a t the primary error sources and sensitivities are the same for both systems.
That is the level errors are caused primarily by the accelerometer uncertainties, and the azimuth error is caused primarily by the east gyro drift. If
should be emphasized, however, that the effect of base motion is likely to
be very significant in the alignment of a practical system, whether one uses
a physical or analykic gyrocompass scheme.
APPENDIX
A
Development of a System Error
Model
I n this appendix the differential equations describing the dynamic behavior of
a space stabilized inertial navigation system in response to gyro uncertainty
are derived for arbitrary motion over and above the earth. Since the object
of this development is to establish a standard against which perturbation
techniques can be evaluated, certain simplifications are made. The equations
are derived using the approximation that the earth is a spherical homogeneous body. The errors that result from this assumption are on the order
of the products of the earth's ellipticity and the navigation errors. In addition
it is assumed that for the purposes of calculating the magnitude of the
earth's gravitational field vector, the altitude of the system above the earth's
surface is precisely known. I n a practical situation, this latter assumption
necessitates the use of an errorless altimeter.
The derived equations are then solved for the case of constant gyro
uncertainty and constant east west velocity over the earth, and the position
errors are computed. After these exact solutions are evaluated for the case
of "small-" angle platform rotation with respect to the inertial frame, tlhe
principle of superposition is used to obtain an expression for the position
error as a function of all three gyro uncertainties. A linearizing scheme is
then developed which duplicates the results obtained from the exact solutions.
The development of the exact solutions is adapted from Reference 6.
A.l
SYSTEM DESCRIPTION
The inertial navigation system being considered consists of an uncommanded gyro stabilized inertial platform upon which are mounted threesingle-degree-of-freedom accelerometers with mutually orthogonal input
axes. A computation capability is assumed such that the specific force data
can be processed.
218
The ideal accelerometer output for this system consists of three signals
which are proportional to the nonfield specific force exerted on each
accelerometer along its sensitive axis. From Newton's second law the specific
force is equal to the difference between the inertially referenced acceleration
and the net gravitational acceleration a t the instrument's location. It is
sufficient to consider only the gravitational effect of the earth as shown in
Chapter 3. Thus the ideal output of the ideal accelerometer set can be
written :
f * = cpi;i- Gp
(A-1)
where
fp
CF
Zi
= inertially
Gi
= gravitational
It is a consequence of the fact that the specific force measurements are made
in platform coordinates that the inertially referenced acceleration is also
coordinatized in this coordinate system. Since the center of the earth is
taken as the origin of the inertial coordinate frame, the vector r is interpreted
as the position vector extending from the center of the earth to the navigation
system's location.
The computation scheme for extracting position from the specific force
data is readily seen from Eq. A-1, since the position vector, ri, can be
obtained from two integrations of the gravitational field-compensated
specific force signals. I f the effects of gyro uncertainty are considered, then
the transformation CF is seen to be time varying since the platform is now
rotating relative to the inertial frame. The computer, on the other hand, is
programmed to process the data based on the assumption that the platform
frame is inertially nonrotating. Errors are thereby introduced into the
computation ; letting
219
Yp
Accelerometer
triad
ej
ei
lnertially
stabilized
platform
II
Figure A . l
Gravitation
computer
A.2
D E R I V A T I O N O F SYSTEM DIFFERENTIAL E Q U A T I O N S
The system differential equations are found directly from Eq. A-2. The
gravitational field is computed from this expression :
where, as has been previously mentioned, the magnitude of the radius vector,
Irl, is available from an external source such as an altimeter. Clearly there
is a choice as t o how this external altimeter information is combined with
the computed information as is discussed in Section 8.3.4. For the purpose
of developing the perturbation model, however, Eq. A-3 is used.
An expression for the measured specific force in platform axes is given by
Eq. A-1, since i t is assumed that the measurement is errorless.
220
Noticing t h a t the constant p/r3 is the square of the Schuler frequency, and
that GP = C:Gi = -(p/r3)C?ri, then Eq. A-4 becomes
where co,2 = ,u/r3, square of Schuler frequency for spherical earth model.
Notice t h a t in the derivation of Eq. A-5, which represents three uncoupled,
linear, differential equations, no assumptions have been made as t o the form
of the matrix product e i C ? , no linearization techniques have been used
(although the equations are linear), and no restrictions have been placed
on the motion of the system.
The subsequent algebraic manipulations can be considerably simplified
if the platform frame is assumed t o be initially aligned with t h e inertial
= I, the identity matrix. This initial alignment
frame, in which case
assumption is by no means restrictive in nature and is introduced only for
clarity in this development. Notice that although the initial alignment has
been precisely determined, the navigation system is in error because the
platform is rotating relative t o the inertial frame. The transformation
between inertial and platform coordinates can therefore be written as a
function of the gyro uncertainties alone, Eq. A-5 being written as
e;
A.3
S O - L U T I O N O F SYSTEM DIFFERENTIAL E Q U A T I O N S
To solve the system differential equation, Eq. A-6, the forcing functions
must be evaluated in terms of the time history of t h e gyro uncertainties, the
inertially referenced acceleration, and the system position. An analytic
expression for the inertially referenced acceleration is found by time
differentiating the expression for the geocentric position vector, ri, expressed
in spherical coordinates :
ri = ( r cos L cos A, r cos L sin
A, r sin L )
(A-7)
3,
celestial longitude
The solution t o Eq. A-6 becomes unwieldy if general system motion over
and above the earth is allowed. For this reason the assumption is made that
the system is moving in an arbitrary east-west path a t constant velocity
221
where
f
i
constant
constant
(I
If Eq. A-7 is substituted into Eq. A-6, the system differential equation
becomes
[(w? - P) cos L cos i t
P + co:i?=
Note t h a t the bracketed term on the right-hand side of Eq. A-8 is just the
specific force fi. The initial conditions associated with the equations above
consist of the computed initial position and velocity of the system :
P(0) = ( r cos L, 0, r sin L)
(A-9)
(A-10)
where, for the assumed initial alignment the x, y, x platform axes coincide
with the inertial axes. Note that the angular velocity of Eq. A-11 results
from constant negative gyro uncertainties along the x, y, and x platform axes.
A.3.1
0
cos co,t
0
-sin
sin o,t
wx2i cosOw,t
222
D E V E L O P M E N T O F A SYSTEM E R R O R M O D E L
a, + w;+,
..
P, + w:P,
- i 2 ) cos L cos At
= r(w:
r
2
= - (w,
- i2)
cos L[sin ( I + w,)t
(A-l2b)
COS
(A
- C0,)t-J
(A-12c)
Solving Eqs. A-12 with t h e initial conditions given by Eqs. A-9 and A-10
yields solutions :
(A-13a)
FZ = r cos L cas i t
i, = r{;
cos L - -
2ws[w: - ( 1
- (~ 0 2 , - i ~2 ) ( i - w , ) c o s ~
2ws[o: - (I-
+ w,)~]
x
L(w: - 12)
sin ( 1 + w,)t
+- 2rrcos
4-(I +
r ~ ,sin
2 L
r cos L(w: - 1 2 )
sin w,t
sin ( I - w,)t +
+ 2ro:
-(I(0:
0:)
4-
r(m: - i2)
cos L
cos ( I - w,)t
210: - ( I rco: sin L
2 COS
cc):
- 0,
r(w: - 1 2 )cos L
2 r d - (1
w,t
c: =
sin w,t
-sin w,t
cos w,t
( A + w,)t
+ ~ , ) cos
~ l
223
..
i.,
..
2v
(A-14a)
(A-14b)
- sin
( A - w,)t]
(A-14c)
Solving Eqs. A-14 with the initial conditions of Eqs. A-9 and A-10 yields
W,2
P, = r cos L
21o: -
rcoywssin L
2
0,
2
COY
- A2
(It w,)']
sin cost
W,z
2[w,2 -
(2
- A2
( A - Coy)2]
cos cost
+ w,)t
cos (1
r cos L ( w : - A2)
roz sin L
cos (A - w,)t sin co,t
+ 2[w:
- (1- w , ) ~ ]
- w;
2
(A-15a)
0,
?v = r cos L sin At
9,
(A-15b)
+
l
2wSb: - (1- ~ , ) ~ 2WS[w:
- (A +
r cos L(w: - A')
sin ( A + w,)t
+ 2[w,2
- (1+
=r
cos L
(w: -
i2)(i
- cow)
ro,2 sin L
+
0:
A.3.3
2
Cc),
( A - wv)t
2 COS
w: - w,
w,t
sin w,t
0-
cos w,t
1-
I sin wst
224
D E V E L O P M E N T O F A SYSTEM E R R O R M O D E L
..
( A - 16a)
Pz
(A-l6b)
(A-16c)
Solving Eqs. A-16 with initial conditions given by Eqs. A-9 and A-10 yields
P,
=r
(m: - X 2 )
0: - (1- w,12
cos L
=rcosL{L0.
cos w,t
+r
(i
- wz)(wz- A2) sin cost
ws[w:- ( A - w , ) ~ ]
w: - ( A - w z )
+r
P, = r sin L
A.4
cos ( A - w,)t
(A-17c)
A P P R O X I M A T I O N S TO T H E S O L U T I O N S
are treated herein. The following approximations can be made in Eqs. A-13,
A-15, and A-17.
225
The error incurred in making the approximations above are on the order of
( A / W , ) ~ ,which for the generally applicable terrestrial navigation case is
approximately equal to the square of the earth's ellipticity or l/105. I n
addition, the small-angle assumption can be applied to any trigonometric
term having w, as its argument.
Thus the solutions become as follows. For the oxcase:
P,
=r
Py
cos L cos At
r cos L sin At
(A-13d)
+ rw,t
sin L - r w
-x sin L sin wst
Ws
P,
=r
(A-13f)
os
?, = r cos L sin At
Pz = r sin L
lo,sin At
+ r o y t cos L cos At - r cos L - sin o s t +- 2r co1s L -
os
For the o, case :
?, = r cos L cos At
Py = r c o s L s i n h - - r ~ z t c o s ~ c o s ~ t + r c o s*,
~ - s i n w s t 2- r c o sAco,
~ ~ s i n l t
*s
W s
(A-17e)
P, = r sin L
Notice that the leading terms in the nine equations above are the exact
solutions which would be obtained by an errorless navigation system.
Following the exact solutions are error terms which grow linearly with time
plus error terms which exhibit Schuler and celestial longitude rate modulation.
If the exact solutions are subtracted from Eqs. A-13d, e, f, A-15d, e, f and
A-17d, e, f, the navigation errors can be written in the following compact
cos L t sin i t
+2
where the navigation error has been defined as the difference between the
computed solutions and the actual system position:
d = - rk
k = x,y, 2
The errors can be understood by inspection of Eq. A-8, the system vector
differential equation, which as was previously recognized, has the forcing
Time, (hr)
Figure A.2
DEVELOPMENT O F AN ERROR M O D E L
227
function C:fi = C;(f,, f,,f,). The most significant terms (those which are
linearly increasing) arise due to quadrature coupling of fi into the drifting
platform frame. Thus the major x axis drift errors arise due to coupling of
f, and f, into the x and y accelerometers, respectively, and similarly for the
y and z axis drift errors.
Equation A-18 is plotted for the stationary case ( A = mi.) in Figure A.2.
Note that a drift of 1 meru was assumed for all plots. Errors resulting from
other drift rates can be read directly from the curves since amplitudes of
both the fundamental (low frequency) and Schuler modes are a direct
function of drift magnitude. One interesting feature is that the Schuler
mode is not superimposed on all the error curves. Analysis of the differential
equations reveals that the Schuler modes appear due to the initial velocity
conditions. That is, if initial velocities were used which took drift rate into
account, the Schuler mode amplitudes would become negligible. The appropriate initial velocity would be Ci r- 2 , (0) instead of i (0). Thus the z channel
(dr,) for the x axis drift case and the x channel (dr,) for the z axis drift case
do not contain the Schuler mode since the initial velocity conditions were
correct for these cases.
'
A.5
D E V E L O P M E N T O F A N ERROR M O D E L
An error model is now developed which yields results which are identical
to those obtained via an exact solution of the system differential equations.
Starting with the system differential equation (Eq. A-S), the position error
is defined as before by the quantities:
Before Eq. A-19 can be substituted into Eq. A-8 to obtain the error differential equation, the error vector must be differentiated. Thus the question
arises as to the coordinate frame associated with the computed position
vector Pi. The point of view taken herein is that the symbol Pi represents an
array of three scalar quantities which exist as computer outputs and which
represent the computed coordinates of the position vector which we treat
as being coordinatized in the inertial frame. No attempt is made to associate
a "computed" reference frame with this array. With this point of view in
mind, i t follows that 6Pi = gi - f and Eq. A-8 becomes
228
where
is the skew-symmetric form of the angular velocity of the platform
frame relative to the inertial frame :
+ wzrb = fi,
Solution of the error differential equation shows that the error vector is
given by Eq. A-18, the solution obtained directly from the system e&ations.
Thus it has been shown that the perturbation techniques leading t o the
system error equation (Eq. A-22) are valid. 1n particular, the concept of not
associating a computer coordinate frame with the computer position vector
has been validated.
APPENDIX
B
State Transition M a t r i x for
Inertial
Navigation Systems
I n this appendix, state transition matrix methods are developed which
are applicable to the general class of two-dimensional local-level navigators
discussed in Chapter 7, Section 7.4. I n order to minimize the algebraic
complexity the development is confined t o the case in which the Foucault
modulation of the Schuler frequencies are neglected. As discussed in Section
7.4.3.1 care must be exercised such that this simplified model is not applied
to situations where the Foucault modulation is a first-order effect.
The error equation for the class of system under consideration can be
written as follows :
4 x 1 = Q1
23
-A
A sin L
sin L
0
P
cos L
0
-i
cos L
-9
9
0
The error state vector is given by
X1
= ( & N ,EE,
A sin L
P
- p cos L
0
p sin L
cos L
rp2
r cos Lp2
ED,
dL, d l )
Since this equation applies to the most widely used class of inertial
navigation systems, it is advantageous to use state space methods to obtain
229
230
- A sin L
A cos L
- A cos L
0
0
0
g-
- g sec L
sin L
A =
--Asin~ 0
0
-1
cos L
-sin L
0
0
0
0
0
0
0
0
1
0
-AcosL
and
B.2
STATE T R A N S I T I O N M A T R I X
The solution to Eq. B-2 is given in terms of the state transition matrix,
a(t)= eAt as
t
= @ ( t - to)x(to) + J ; ( t
- o).(o)u(o) do
*(O) = I
where I is the identity matrix, and the composition law:
* ( t ) = *(t - to)*(to)
from which it follows that
a-l(t)
=*(-t)
The transition matrix is found from the relationship :
a(t)= 9-*(Is - A)-I
where s = Laplace operator
9 - I = inverse Laplace transformation
( )-I = matrix inversion operator
Applying Eq. B-7, the state transition matrix is found to be given by
(B-6)
232
8.3 S T A T E T R A N S I T I O N M A T R I X F O R S H O R T
SAMPLING TIMES
The state space approach is used when optimal filtering techniques are
applied to the inertial navigation system. I n this situation, the state transition matrix is used to model the system's behavior over the sampling time,
T. Thus small-angle assumptions can be made in the expression above:
e
< 10%
for
< 6 min
< 10%
for
< 16 min
e<10%
for
T<12min
< 10%
for
< 23 min
cos At N 1
< 10%
for
sin At NXT
< 10%
for
cos w,t N 1
-1--
wZT2
2
- wsT - --N
= w,,
where e is the maximum error associated with the approximation. Thus for
update times of less than 6 min ( T < 6 min), the following state transition
matrix should give adequate results :
-At
At sin L
8.4
sin L
1
0
At cos L
-hsinL
0
tcosL
-t
cos L
m:t
--w2t sec L
-at
- ~ C O S L
-tsinL
EXAMPLES
The following examples are included t o illustrate the use of the state
transition matrix in the error analysis of the two-accelerometer systems.
EXAMPLES
B.4.1
233
The solution for the initial condition errors is made by inspection of Eqs.
B-6 and B-8. Thus
where * ( t ) is given by Eq. B-8 and x 2 ( t ) is given by Eq. B-3. This method is
used in Section 7.4.3.4t o determine the initial condition errors for the twoaccelerometer local-level system.
B.4.2
where ( u )fN and ( u )fE are the north and east accelerometer biases, respectively. Thus in state space notation
1
r
sec 1
r
Thus from Eq. B-6 the error response to accelerometer bias starting a t
to = 0 is given by
234
Thus it is seen that this method yields results very quickly and efficiently
compared with solving Eq. B-1 via Cramer's rule. The result above is seen
to agree with the analytical results of Section 7.4.3.2.
APPENDIX
C
Statistical Error Analysis M e t h o d s
C.l
R E S P O N S E O F A L I N E A R S Y S T E M TO R A N D O M I N R U T S
(C-2)
236
The mean squared value a t any time, t, is found by replacing the upper limits
by t . Thus
The RMS error is, of course, found by taking the square root of the expression
above. To summarize, Eq. C-7 yields the mean squared error as a function
of time for a linear, time-invariant system when the input statistics are
time-stationary. Analog computation techniques are available to solve the
equation ab0ve,~6but it is instructive t o examine the solution for several
simple cases.
C.2
R E S P O N S E TO T H E E N S E M B L E O F
CONSTANT FUNCTION'S
+xx(r2
- 7 1 ) = xo
(C-8)
When Eq. C-8 is substituted into Eq. C-7 it is seen that the input autocorrelation function can be taken out from under the integral signs and the
But since
It is recognized from Eq. C-2 that the integral expression in Eq. C-10 is
just the output of the system to a unit valued constant input, the unit step
response. Thus t o find the system's response to this class of input functions,
one merely computes the system's unit step response and scales the result
by the appropriate root-mean-squared value of the input forcing function.
This interpretation can be directly applied to the results of Section 7.4. For
example if one assumes the individual gyro drifts to be uncorrelated with an
RMS value of 1 meru, then Figure 7.12 can be interpreted to be the RMS
attitude and navigation errors resulting from gyro drifts of this statistical
class. It is noted in passing that in analogous fashion to the above development, the system's response to an input function which is a member of the
ensemble of constantly increasing functions is found from the system's unit
ramp response.
C.3
R E S P O N S E TO W H I T E N O I S E
71) =
Nd(~2
- 71)
To evaluate the effect of this class of input, Eq. C-11 is substituted into Eq.
C-7. Because of the presence of the delta function in the integrand, the mean
squared error is given by
238
C.3.1
Example
(C-14)
the numerical value for N is easily specified from empirical data. Note from
Eq. C-13 that random walk is given by the time integral of white noise.
Since the analytical expression for the mean squared error as given by Eq.
C-12 requires that the input to the system be white noise, the system weighting function must be appropriately modified such that the input given by
Eq. C-13 can be accommodated. This modification involves the insertion of
a so-called shaping filter, in this case a mathematical integration. Equation
C-12 is therefore modified to read
sin At
where
=t
+ mi,
= terrestrial longitude
5,
Figure C.l
Applying the derived expression for the mean-squared error to this case,
we have
which integrates to
-
N 6L2 = 27
2 3 ( I t - f sin z i t )
(C-IS)
18 -
16 14 E
I=
r12-
2
L
w
g3 10 .+-,
m
w
u,
8 -
z 64-
Easterly flight
10
Navigation time, hr
Figure C.2
240
The square root of the expression above, which corresponds t o the RMS
value of the latitude error, is plotted in Pigure C.1. It is seen that, in general,
the RMS latitude error increases with the square root of time. Notice also
that the effect of gyro drift is seen to be a strong function of the celestial
longitude rate.
For a numerical example take the case where the gyro drift rate is 1 arc-min
after 1 hour. Thus the drift rate power spectral density is given by Eq. C-14
as 1 arc-min2/hr3.I n Figure C.2 the RMS latitude error is plotted as a function
of time for four values of east-west velocity:
Stationary case, 1 = 0
590 knots easterly flight a t latitude 35"
590 knots westerly flight a t latitude 35"
Westerly flight a t earth rate
It is seen that for the easterly flight, the RMS latitude error after 8 hours is
about 4 n.m. while for the westerly flight the error is about 13 n.m. Since
the random process under consideration has been assumed to be unbiased,
the RMS values can be interpreted as l o values.
More general methods are available to determine the system's response
t o statistical forcing functions,59but these involve a computer solution of
the complete set of differential equations. The methods illustrated herein
are mainly useful for simplified cases where the system weighting function
can be analytically determined, although analog computer derived solutions
are available via simulation methods.46
References
24 1
REFERENCES
Fagin, S. L., "A Unified Approach to the Error Analysis of Augmented Dynamica.11~
Exact Inertial Navigation Systems," IEEE Transactions on Aerospace and
Navigational Electronics, December 1964.
Farrell, J. L., "Performance of Strapdown Inertial Attitude Reference Systems,"
Journal of Spacecraft and Rockets, Vol. 3, No. 9, 1966.
Farrell, J. L., "Analytic Platforms in Cruising Aircraft," Journal of Aircraft,
Vol. 4, No. 1, 1967.
Fernandez, 3%.and G. Macomber, Inertial Guidance Engineering, Prentice-Hall,
1962,
Frey, E. J. a n d R. B. Harlan, "Airborne Gravimetry Program," M.I.T. Measurement Systems Laboratory, RE-54, 1969.
Garren, Jr., J. F., J. R.Kelly, and R . W. Sommer, "VTOL Flight Investigation to
Develop a Decelerating Instrument Approach Capability," 1969 SAE National
Meeting.
Gelb, A. and A. Sutherland, "Design Approach for Reducing Gyro-Induced Errors
in Strapdown Inertial Systems," AIAA Paper No. 68-830, 1968.
Geller, E. S., "Inertial System Platform Rotation," Transaclions IEEE, Vol.
AES-4, NO. 4, 1968.
Gianoukos, W. and P. Palmer, "Gyro Test Laboratory Unbalance Equations,"
M.I.T., I L Report GT-130, 1957.
Gilmore, J. P., "A Non-Orthogonal Gyro Configuration," M.I.T. Instrumentation
Laboratory, T-472, 1967.
Guier, W. H. and R. R . Newton, "The Earth's Gravity Field," Johns Hopkins
Applied Physics Laboratory, TG-634, 1964.
Hammon, R. L., "An Application of Random Process Theory to Gyro Drift
Analysis," IR E Transactions on Aeronautical Navigational Electronics, Vol. ANE - 7,
September 1960.
Heiskanen, W. A. and F. A. Vening-Meinesz, The Earth and I t s Gravity Field,
McGraw-Hill, 1958.
Hessian, R., "Analysis of a Transformation Computer Used with a Gimballess
I.M.U.," M.I.T. Instrumentation Laboratory RE-531, 1966.
Hildebrand, F. B., Methods of Applied Mathematics, Prentice-Hall, 1965.
Hutchinson, R. C., "Tumbling Method of Locating Accelerometer Input Axes on
a Platform," M.I.T. Instrument,ation Laboratory, E-962 1960.
Jankowski, P. C., "Hybrid Altimeter Using a Strapdown Inertial Navigation
System," M.I.T. Measurement Systems Laboratory, TE-36, 1970.
Jurenka, F. D. and C. T. Leondes, "Optimum Alignment of a n Inertial Autonavigator," IEEE Transactions on Aerospace and Electronics Systems, Vol. AES-3,
No. 6, 1967.
Kasper, J. F., "Error Propagation in the Coordinate Transformation Matrix for a
Space Stabilized Navigation System," M.I.T. Instrumentation Laboratory,
T-448, 1966.
Kasper, J. F., "Gravity Model Refinement Based on Satellite Tracking Data,"
M .I.T. Instrumentation Laboratory, E 1921, 1966.
Kaula, W. M., "A Review of Geodetic Parameters," NASA Technical Note
D-1847, 1963.
Kayton, M. a n d W. Fried, Avionics Navigation Systems, Wiley, 1969.
Kayton, M., "Coordinate Frames I n Inertial Navigation," M.I.T., Ph.D. Dissertation, 1960.
REFERENCES
243
244
69.
70.
71.
REFERENCES
Using Delta Modulated Instruments," M.I.T. Instrumentation Laboratory,
T-300, 1962.
Williamson, R. P , "A Gyrocompass Error Averaging Technique," M.I.T. Instrumentation Laboratory, T-315, 1962.
Wilmoth, E.D., "An Investigation
of Methods for Determining Gravity Anomalies
from an Aircraft," M.1.T Instrumentation Laboratory, T-185, 1959.
Wrigley, W., W. M. Hollister, and W. G. Denhard, Gyroscopic Theory, Design,
and Instrumentation, M.I.T. Press, 1969.
Index
246
INDEX
Coordinate transformation, 1 3
Coriolis acceleration, 18
Coriolis law, 18, 62
Cramers rule, 206,234
Cross product, 19
Fagin, S. L., 6
Farrell, J. L., 22, 192
Federal Aviation Agency, 75
Fernandez, M., 5 , 6
Fischer, N., 6,240
Force, specific, 31,61,79
Forcing function, free azimuth, 187
generalized, 180
local-level, 12 3, 186
rotating azimuth, 188
space stabilized, 104, 185
strapdown, 189
two-accelerometer local-level, 126
Foucault, L., 128,132,134, 135,136,182
Free azimuth system, 187
Frey, E. J., 58
Fried, W., 74,79
Functions required for navigation, 1
Earth, density, 49
ellipticity, 47
equatorial radius, 46
gravitational field, 49
gravitational potential, 49
gravity field, 5 6
inertial angular velocity, 34
mass, 5 1
polar radius, 46
radius magnitude, 47
Earth-rate mode, 128
Einstein, A., 30
Ensemble of constant functions, 236
Equivalence, principle of, 30
Error analysis, accuracy of, 102
generalized, 156
local level, 114
space stabilized, 8 6
unified approach, 7,15 3
Error angles, 21
Error equations, free azimuth, 187
generalized, 156, 180
local level, 123, 186
rotating azimuth, 187-188
Galilei, Galileo, 30
Garren, J. F., 81
Gelb, A., 69, 190
Geller, E. S., 188
Geoid, 57
Geometric inertial systems, 3
Gianoukos, W., 74
Gilmore, J. P., 38,76
Gravitation, acceleration, 32
computation, 83, 169
field, 49,53,55
potential, 49
of reference ellipsoid, 5 1, 60
universal constant, 50
weighting factor, 10,83, 113, 182,197
Gravity, anomaly, 57,59
computation, 113
of earth, 5 6
magnitude, 60
Guier, W. H., 52
Gyrocompass, 203
acceleration coupled, 203
errors, 204
signal flow diagram, 206
INDEX
247
248
INDEX
skew symmetric, 15
square root, 22
transformation, 13
transition, 232
transpose, 13
Measured quantities, definition, 15
Momentum, angular, 68
Navigation error, 94,172
Navigation errors, due to, accelerometer bias,
141-144,233-234
altimeter, 195-197
gyro drift, 106,107, 131-140
initial condition, 145, 147-152,233
system alignment, 145,147-152,233
Navigation weighting factor, 1 0 , 8 5 , 9 5 , 114,
121
Newton, I., 30,44, 79,218
rotational law, 66
Newton, R. R., 52
Noncommutivity errors, 192
Nonstationary statistics, 235
Notation, conventions, 12
state space, 229
O'Donnell, C. F,, 6 , l l , 82
Ogata, K. O., 12
Orthogonality constraint, explicit, 22
implicit, 2 1
Palmer, P., 74
Palsson, T., 199,209
Parkinson, B., 6
Pennypacker, J. C., 194
Perturbation analysis, 20
Pitch axis, 34
Pitman, G. R., 6 , 6 5 , 8 0 , 198
Platform, inertial, 7 1, 82
angular velocity, 89, 111, 158
error angles, 99
Position vector, estimator, 9, 1 0 , 83, 166169
geocentric, 44, 170
Quantization error, 194
Radius of curvature, meridional, 6 3
prime, 6 3
Random walk, 238
Rea, F., 6,240
Redundancy, system, 75
Reference ellipsoid, 45
eccentricity, 46
ellipticity, 47
Reference frame, computed, 11
Reliability, gyro, 75
Roll axis, 34
Rotating azimuth system, 187
Round-off error, computer, 194
Routh's criteria, 207
Russell, B ., 11
Ryan, T. J., 198,209
Schmidt, G. T., 198
Schuler, M., 102, 182
Sciama, D. W., 30
Searcy, J. B., 5 7 , 5 8 , 9 8
Self-corrective strapdown alignment, 210
Semi-analytic inertial systems, 3
Shaping filter, 238
Skew-symmetric matrices, 15
Small angle rotations, 16
Smith, M. A., 217
Sommer, R. W., 81
Space, absolute, 30
Space integrator, 79
Space rate mode, 134
Space stabilized INS, 79
acceleration computation, 82
altitide computation, 86
description, 79
error analysis, 86
gravitational field computation, 8 3
latitude computation, 85
longitude computation, 86
mechanization diagram, 87
mechanization equations, 8 1
platform commands, 82
velocity computation, 85
Specific force, analytic expressions, 61,79,109
computation, 82, 112, 156
Standard deviation, 240
State space notation, 229
Stokes, G. G., 57
Strapdown I.N.S., 4,188
additional error sources, 190
alignment, 198,209-216
error analysis, 18 9 , 2 11
Sutherland, A., 69, 190
Symbology, 24
INDEX
Tangential acceleration, 18
Tetrad configuration, 76
Thompson, J., 69
Transformation, non-orthogonal, 23
coordinate, 1 3
instrument-platform, 4 3
perturbations, 20
similarity, 14
Transition matrix, 229
for short sampling times, 232
Triad configuration, 76
Triple products, 19, 20
Trueblood, R. B., 75
Truncation error, 193
TRW systems, Inc., 75
Two-accelerometer local-level system, 123
error equations, 124
error equation solutions, 128
mechanization diagram, 125
mechanization equations, 124
signal ffow diagram, 130
Two-degree-of-freedom gyros, 2, 65
Unger, F., 69
249