ON THE GIENWALEQUATIONS

OF INERTIALNAVSGATION

PMM ~01.28, ta 2, 1964, pp.242-257

V. D. ANDRBEV
(Moscow)

(ReceivedFebruary 2, 1963)

Derived and Investigatedare the equationsof Ideal (unperturbed)and per-

turbed operation of a generalizedsystem of inertial navigationwhich deter-
mines the coordinatesof a moving object from the measurementsof three
linear accelerometersand three angular velocity meters measuring absolute
angular velocity. The instrumentsare located along the axes of an ortho-
gonal trihedron,the origin of which coincideswith a certain point in the
moving object.
In contrast to [l and 23, the orientationof the trihedronaxes Is arbi-
trary.
The equations of ideal operation take into account the non-centralnature
of the Earth's gravitationalfield. The equationsof perturbed operation
Include instrumenterror effects.
Rrror equationsare derived and Investigatednot only for small but also
for the finite values of the variables of the perturbed operating conditions.
The particular cases are the equationsand the results of investigation
of the particular control systems consideredin [l to 73.

1, Let us introduce the right-handedorthogona?systems of coordinates

G&lt and Qlc%llC * The coordinatesystem O,5nf Is an inertial system for
which Newton's laws of dynamics are valid by definition. The choice of the
location of point & and orientationof the axes r;nC are not subject to
other conditions. The origin of the system O,snC coincideswith the
Esrth's center. The orientationof the axes of the O,gnC system is invari-
ent with respect to the inertial Sy6tem. Without loss of generality,their
orientationcan be consideredIdentical.

Let us introducealso the systems of coordinates OsnC and Oxyr with

the origin at a Certain point of the moving object (not necessarilyat
the center of mass). 'The orientationof the 5~qC system of coordinates
coincideswith the correspondingorientationof the O&nC, O,$nC systems
of coordinates. The orientationof the Oxyz system is arbitrary.

297
298 V.D. Andreev

Finally, let us Introduce the system of coordinates O,xyz , the origin

of which Is at the Earth's center, while the orientation of the axes Is
Identical to those of the trlhedron Oxyz .

The problem which Is to be solved by the Inertial system is the determl-

nation of the Cartesian coordinates 5, q, C of the point 0 in the object
within the coordinate system O,~TJ< , and the parameters defining the orlen-
tatlon of the object relative to the xv< axes.

Let along the edges of the Oxyz trihedron, there be installed three
accelerometers and three sensors of the projection of the absolute angular
velocity of the Oryx trlhedron upon its axes.

We will denote the accelerometer measurements by nl, n,, n,, the pro-
jections of the absolute angular velocity measurements by m,, my, mr, and
introduce the vectors

n = n, x + 4 y + n,z, m=m,x+m,y+m,z=w (1.1)

where I) Is the absolute angular velocity of the trlhedron 0ry.z , and x,
r', 8 are the unit vectors of the corresponding axes.

Let us define n , assuming unit accelerometer point masses located at

the point 0 .

If r,,, Is the radius vector of the point 0 In an inertial system of

coordinates, then the condition for relative equlllbrium of the sensing mass
of &he accelerometer is of the form
#r,,,/dt" = F(r,z) + f (l-2)
where F(reon) Is the geometric sum of the sensing mass attraction forces due
to the combined multitude of the celestial bodies, while i Is the sum of
the forces acting on the sensing masses through their suspensions.

In an accelerometer the sensing mass Is usually suspended elastically

[I and 23. me magnitude of the elastic deformation of the suspension Is
proportional to tne force ? and Is the output of the accelerometer*. There-
for , assuming the proportionality coefficient unity, we have
n = ctaroz/cEta
- F(roz) (l-3)
Note that the differentiation In Formula (1.3) 1s carried out In the sys-
tem of coordinates Op$qC .
If r denotes the radius vector of the point 0 relative to 0,, and
r,, Is the radius vector of the point O1 relative to Oar then by taking
into account that
ro2 = r + r01, d2ro,/cZt2= F(rol) (1.4)

l Here the suspension Is assumed tc be a spring while In some practical

accelerometers the elastic forces may also be different, e.g. electro-
magnetic.
 aenera1 equations
or inertia1 nwigat1cn 299

and by introducingthe special designation g(r) for the Earth attraction

force on a unit mass located at 0 , we get
n = dar/dta - g(r) 4 FI(O) - Fl(r) u -5)
In (1.5) the B,(r) Is determinednow by the attractionfield at the point
0 of all the celestialbodles with the exceptionof the Earth.

Let us consider,for the time being, that the object motion (point 0)
occurs at a small distance above the Earth's surface (comparedwith its
radius, for example). Then the difference

F, (0)- F, (r) (W
of the attractionforces at points 0 and O1 becomes negligibly small even
for the nearest celestialbodies includingthe Moon and the Sun. Therefore,
wlth great accuracy
n = d2r t dt2 - g (r) (1.7)

Since the system of coordinates O1<nC translatesrelative to O,gnC,

the differentiationin the formula (1.7) can be consideredcarried out in
the O,cn; system .

2. Let us construct the equationsfor ideal operation,1.e. the equations

for unperturbedfunctioningof the inertial system.

Projectingon the O,xyz axes

dr .
dr=r +wxr,

where the local derivativesdenoted by dots are with respect to the O,xuz
system of coordinates, With the aid of (2.1) from (1.7), and by replacing
a! by m In accordancewith (1.11, we get by Integratingin the &xgz
system
t
dr
dt= In -mx -$-+g(r)ldt+$, r= \(-$-mxr)dt+r'
0 0 (2.2)

With the aid of the computersthe relatlonshps(2.2) permit determination

of the Cartesian coordinates $3 y9 = of the point 0 in the &X~Z coordi-
nate system fof:the quantities n,, ny, n, ; % J my, m, and the initial
values r", g, provided, of course, that Q~(x,u/,z), o,(~,y,z) and
Bz(T..Y,X) are known, which will,be the case If, for example, the Earth's
gravitationalfield Is central (more correctly,spherical).

In order to pass to the Cartesian coordinates 5, n, 6 of the point 0

in the system OlfnC , It is necessary to model the equation

t= S(e x m) dt + to ceea (2.3)

0
300 V.D. Andrew

along with (2.2) which permits, for m,, m,, mi and the mutual disposition
of the systems of coordinates 01 zn?6 and u,xy.~ at the Initial time, the
determination of the direction cosines between the unit vectors {, 11, C
and X, y, E and Equatlons

E = t.r (EfiU (2.4)

which effect the passage from the coordinates x, ut 2 to the coordinates
5, ri> c In (2.3) and (2.4) and the sequel, the symbol (:nc) placed near
the formula denotes cyclic Interchange of the variables and the indices.

It is obvious that the orientation of an obJect in space is given by its

position relative to the trihedron 0x2s which requires an additional
IneaSUrement either of the position of the trihedron OX&? relative to the
ob jeer;, or the angular velocity of rotation of the. set Oxy.? relative to
the object and the solution of the equations similar to (2.3).

If the Earth’s gravitational field Is assumed spherical, i.e.

g tr) = - -& g (r) (2.5)

then Equations (2.2) and (2.3) are solved’independently. In the opposite
case, the gravitational field can be given only In the Earth’s coordinate
system or in a coordinate system wherein the motion of the Earth is known,
for example,in the O,t;ni; coordinate system. In this case the given func-
tions will be gt (g, q, 6, I), g? (h, 11, j, t), and g;(g, ?j, 5, tj . In order to
determine g,, Q,. and gz it is first required to obtain the solution of
Equation (2.3), and then solve simultaneously (2.2) and (2.4).

It 1s understood that knowing the Cartesian coordinates j, n, C it is

possible to obtaln any other ccol,dinates, in the general case curvilinear
and nonorthogonal coordinates x1 I x2, %> which may be dependent on :, n,
c and time t If the surfaces of equal values of the coordinates x1, x2,
will change their position relative to the trihedron Cl <n: with time.
%e latter will take place, fcr instance, if the coordinates x1 f x2, x3 are
rigidly attached to the Earth.
In order to pass to the coordinates Q, wa, n3 , it is obviously neces-
sary to specify three relationships of the type

airs, rl, 5; x1, x2, x3, t1 = 0 (i = i, 2, 3) (3.6)

which in the whole performance region of the lnertlal system must satisfy
the usual conditions of unique correspondence of the r;, n, 6 and x,, x*,
x3 coordinates, i.e. the determinants of the Jacobi functions ei, cz, :3
of 5, n, 6 and x1 , x2, wJ must be nonzero.

It is worth to note that the construction of the equations for ideal

performance in their Integral form (the form in which they are simulated by
the computer of the Inertial navigation system [Z]) is not entirely unique.
They may be constructed in several equivalent but different forms.

For example, in place of Equation (2.2) one may take the equivalent equa-
tion t t
(II - XXIx (I’ -+ XII x r) + g (r)) dt -i- (i)” + w” x r03 dt -t r” (1.7)
r=
 oeneral equations or inertial navigation 301

Equations (2.2) and (2.7) presuppose Integration along those axes in

Oxyz (O,xyz),where n,, ny, n, and m,, m,, m, are measured, and obtaining
the Cartesian coordinates r, z/,o.
Another construction 1s essentially different. Utllizlng (2.x), n can
be projected on the axes of the set UlsnC with integration carried out
along the axes of this set and then obtaining the Cartesian coordinates 5,
n> 6 * One can project n along the axes of any other orthogonal system
#,f'n'C'(with the origin at the geocenter O1 ) the motion of which 1s lcnown
relative to O,fnc , the axes of such system may be, for example, rigidly
attached to the Zarth. Finally, one may project n along the normals to
the coordinate surfaces n, = const , carry out the integration along these
normals and obtain as the result of integration the coordinates x1, x2, x3 *
A possible intermediate method for constructing the equations for ideal
performance is to carry out the first integration along one set of directions
(not necessarily coinciding with the directions of the n components measure-
ments), and the second integration along another set of directions.
Equations (2.3) can also be expressed in other forms, for example, by
introduction of ENer angles, the Oldlng Rodrlgues parameters or the Cayley -
Klein parameters [a and 93.

All variations for constructing the equations of ideal performance differ

insignificantly in the number of required computer operations if one con-
siders the complete equations for ideal performance, an drbltrary orientation
of the trlhedron Oxyz , and an arbitrary motion of the object. The possl-
billty of slmpllfylng the equations for ideal performance Is dependent on the
choice and the maintenance of a SpeclaI orlentatlon of the 0xy.z trlhedron,
on the Imposition on the motion of the object of definite limitations, and
on the neglecting of certain terms In the equations for Ideal performance,
the inclusion of which leads to the accuracy beyond that provided by the
inertial system.

The coordinate set Ox&e can be rigidly attached to the object. Then
the complete system of equations for ideal performance is utilized. The
solution of the system (2.3) at the same time also determines the orientation
of the object In space.
The coordinate set 0xy.s may be fixed in space [23, for example, its
axes may be parall.el to rn< . Then the integration yields immediately the
coordinates <, n. C and Equations (2.3) fall out. Such orientation.of
the set Oxyt-. c&be ensured by a stabilized platform.
If the axes of the set Oxye are parallel to the axes 5', tl*r6's the
orientation of which in the system O,<T& Is a known function of time, then
the Cartesian coordinates c’* n', C' are obtained bv direct integration.
At the same time the position of the set Cxyz relative to the aies of O<nC
must repeat in time the given orlentatlon of the system OIE('n'C'
__ ._ relative
to O,Slj6 .
The coordinate set Dxy~ can be oriented also with regard to the co-
ordinates n1, %r x3 of the point Q of the object as determlned by the
inertial system Itself. For example, one of its surfaces can be a tangent
surface to the area of constant value of some coordinate, for Instance, We,
then one axis of the set OXUZ , such as Ox , is normal to this surface. In
this case the integration along the axis Ox will yield lmmedlately the
coordinate nl.. If the coordinates xl, na, x3 are orthogonal, then it is
possible to locate the axes Oy and 08 along the normals to the surfaces
x2 f const and x3 = const and determlne n2,k3 by Integration. In the
case of nonorthogonal set of coordinates it Is possible to pass to the non-
orthogonal trnlhedron 0xy.z .
Examples of the orientation of an orthogonal set Oxyz which account for
the present position of the object are: the realization of an tracking
Darboux trihedron on a sphere of radius P , concentric with the Earth, If
 V.D. Andrew

the coordinates n3 are spherical coordinates (geocentric or geo-

desic), or a trac~&gn~~ihedron on the surface h = const if the defining
coordinates are the latitude, longitude, and height h above sea level.
The simplest example of expedient transformation to a nonorthogonal tri-
hedron is the case when the defining coordinates are: the distance r from
the center of the Earth O1 and the angles n, and n2 between r and the
In this case, having located the axis 02 along p,
should be located in the planes of&~ and 00,n
case the axes Or, Oy and 02 are normal to the
surfaces
the axeS ;t = const , n2 = const and r = const and the angle between
and Oy is not a right angle and debends upon the coordinates
Ml> wp *
If the object moves along the surface xX = const then the equations
for ideal performance are simplified by the fact that’the part related to
the definition of x1 is eliminated so that no accelerometer along the nor-
mal to the surface R, = const should be needed. Analogous simplifications
are possible when the surface is not a coordinate but is given by the re-
latlonship 0, (x1, X2, KS) = 0 . Example of such simplified systems with
two accelerometers (or equivalent systems with two accelerometers) are the
control systems presented in [l to 53.
Finally, if the object moves along the line u1 (x1, Q, Q, ) = 0 , u, (w, ,
a=, Xc) = 0 , then the second accelerometer may be removed from the system
with the remaining one being oriented along the tangent of this line.
Of co&se, the enumerated simplifications are conditional on the impo-
sition of constraints upon the motion of the object. In particular, however,
the elimination of one or even two accelerometers Is possible in the absence
of such constraints if one or two coordinates are determined not by the
inertial system but by means of other sources of information. For example,
the distance to the Earth’s center may be determined by means of a radio
altimeter [ 21.
We will note also that along with the omission of some negligible parts
in the complete equations for Ideal performance, the simplification of these
equations may also be effected by forming part of the terms not as functions
of the present performance of the object’s motion, but as functions of its
programed values, i.e. as functions of time.

The equations for ideal performance of the inertial system considered

above, were formed with the assumption that the motion occurs at such prox-
imity to the Earth’s surface that in (1.5) the difference (1.6) caa be neg-
lected. It is easy to see that this limiting assumption need not be made.

Let there be n celestial bodies, the attraction of which should be

evaluated, takin: the difference (1.6) into account.
We will denote by r, the radius-vector of the center of mass for the
t-th celestial body relative to the Earth’s center of mass O1. Then the
radius -vector r,rl’ of the point 0 relative to the center of the f,-th body
is equal to
,.@I=
2 r - r4 (2.8)
If one considers that the masses of the considered celestial bodies and
their motions relative to the Earth are known (the object, obvious1
not perturb the motion of the Earth or that of the celestial bodies
that their attraction fields are central, then [lo]

Fl (0) - Fx (r) L-Z (2.9)

where m, is the mass of the t-th celestial body, and y a gravitational

constant.
Taking into account (2-q), the first equation in (2.2) is replaced by
Equation
 I%mw&l cqu4ticns of 1ntrt141 tuvl&4tlon 303

The projections OS bt on the axes of Otxys in (2.16) are determined by

the projections on the axes of Oltnf , which are known as functions of time,
and by Formulas (2.3)
!Phe consideration of noncentrality in the fields of attraction of the
considered celestial bodies does not introduce principal dlfflcultles In the
construction of the equations for Ideal performance but does make these equa-
tions substantially more unwieldy, since the determination of the attraction
fields requires introduction OS n systems of coordinates each of which Is
rigidly attached to the t-th celestiai body. In this case, it becomes neces-
sary that the motion of each celestial body about its center of mass with
respect to the system of coordinates O,sslC (or OlmC) be kmwn.

3. Let us derive the error equations, i.e. the equations describing the
perturbed operation of the inertial system when the initial conditions of
the equations for ideal performance are given inaccurately and the elements
of the s&tern have instrument errors, The error equations determine the sta-
bility of performance of the inertial system and the dependence of its accu-
racy upon the magnitude of errors in setting up the initial conditionsand
upon the Instrument errors of the elements.

As the instrument errors we will take the basic ones: Am and An [2].
It can be shown that any other errors can always be reduced to the equivalent
basic ones.

From (2.2), assuming now

I-’ = I + 6r, m’ = ti) + Am, n’=n+ An (3.4)

we obtain t
-dr’ =: ‘* in’ -m’ x -$- + g (r’)l dt + (-$)O
dt \e
0 (3.2)

f’ =I: dt + (f’)O
0
Subtracting from (3.2) the equations for ideal performance (2.2) and
nOting that m = PI we obtain, using the notation (3.1), the integral error
equations .
--AmxFs -(w+Am)xg$,

+ g (r + 6r) - g (r)]dt+ fF (3.3)

br= \($- Am x r - (o + Am) x br) dt + 6r”

6

Differentiating in the same system of coordinates in which the integrals

are taken, i.e. in the Oxya system, we obtain the differential equation
304 V.D. Andreev

6r"+ Z(w + Am) x 6r'+ (co+ Am) x [(w+ Am) x 6~1+

f (w'+ Am’) x Sr - g (r + dr) + g (r) = An - 2Am x r’ - Am’ x r -
- Am x (o xr) - (co -1 Am) x (Am x r) (X(1)

with the initial conditions

6r (0) = fir”
6r’ (0) =: (fir’)” -A- (So’ - Am”) :,: (r’ $ 6r”) (3.5)

For the given quantities w, Am, An, r projected on the Cxyz axes and
the Initial Values 8r9, 6~” (dr’)” , the equation (3.4) yield5 the er:ors in
determination of the Cartesian coordinates x,y,z by the inertial syctem.
Equation (3.4) is exact. If In it the products of the projections Am and
br are neglected, there results Equation

8r”+ 20 x 6r’ + to x (0 x 6r) + w’ x 8r -g (r + 6r) + g (r) =

=An-2Amxr*-Am.xr-Amx(~>:r)-~x(AR~xr) (3.6)
with the initial conditions

6r (0) -1 W, 6r’ (0) = (Sr.)’ + (~uJ” - Am”) x r’ (3.7)

For Am = 0 , i.e. when the perturbations are merely due to the accelero-
meter and the initial conditions errors, Equation (3.4) Is equivalent to
(3.6).
In (3.6) W, r, Am, dr are given in the projections on the O,xya axes
and w is the absolute angular velocity of rotation for the O,xys trih?d-
ron. Therefore (3.6) can also be written as

$+g(r)--g( r + 6r) I= An - 4m .c -$- - -& (Am x r) (3.8)

‘*ihere 6r, r, Am, An are now determined by the projections on the axes of
the O,<nC coordinate set. Note that the homogeneous equation (3.8) cor-
responds exactly not only to (3.6) but also to (3.4) where the products of
the Am and bF projections are retalned.

Let us turn to the second group of equations for the inertial system per-
formance, i.e. to Equations (2.3) and (2.4).

We have from (2.3)

St* + (W + Am) x Sg = 5 x Am fE*,Cf (3.9)

Or, with the same eliminations as in passing from (3.4) to 3.6)

St’ + w x SE = 5 x Am fZrli> (3.10)

ort’lnally, analogous to (3.8)

gs;“-EX Am (4iiF) (3.11)

 305

where as in (3.8) ai, 6q, SC and Am are defined in the projections on the
o1 5rlC axes. The initial conditions for Equations (3.9), (3.10) and (3.11)
are obvious.

From (2.4) we have (4%) (3.42)

8E = at-r

Denoting by by, the full error, and by 6r, the error defined by the
second group of equations (3.9) and (3.12) or (3.10) and (3.lS), we get

6r, = 6r + dr, (3.13)

where 6r2 = (6&r) 5 -f-(6rl.r)rl -f-@WC (3.14)

and br is the solution of the error equations of the first group (3.4) or
(3.6).
In concluding the deviation of the error equations let us pass from the
vector equations (3,6), (3.10), (3.14) and (3.13) to scalar ones. If the
variation In the noncentricallty of the Earth's gravitational field Is neg-
lected and only the linear terms retained in the expansion of the difference
g fr) - g (r + 8r) 2 then
g (r) - g (r + Sr) = 6 + g (7) = g trt3r - 3r (1r + Brj - r)l (3.15)

Projecting now (3.6) on the axes ryt and taking into account (3.15), we
.
get the first group of error equations in the form
(3.16)
Bx” + (+ - coy2- mz2) 6x + (w, WY - 0,‘) 6y --

- ZW, 6,2_4*
-f- (0, 0, -f- oy’) 65 + 264&k-- $- z (i r + 6r / - T) =

= An, - 2 (Am,z’ - An,y’) - (An’,z - Ammz

I;) - (s~x(Am,y + Am, Z) -
- Am, (w, y + o I z) + 2~ (W!, Am, -+ CO
z Am,) (q/z)

The initial conditions of this group of equations are

8x (0) = 8x0,!
6~‘ (0) = (bx*)’ + (60,” - Am,‘) c3 - (&mzo- Amro)~o

Let us find the equations for the projections on the axes xF,z of bxa,
by,, 62, of the vector 6c, given by the relationships (3.10) and (3.14).

Introducing the Cable of direction cosines

-----
X Y 2
f. aI1 aI2 a13
(3.18)
; an1 a22 a23
I: i a31' a32 / a33
and observing that Equations (2.3) are equivalent to three systems of Scalar
differential equations [q]

ai, + o,ai, - OzUiz = 0 (i = 1,2,3) (123.

.WZ) (3.19)
36 V.D. Andrew

we obtain in place of (3.10) (3.20)

hail’ + CC+,
6ai, - &jai, = Amz~is- Aq,ai, (i = 1,2,3) WB. au)

Introducing the notation eX, ey and O1 for the quantities

(3.21)
and neglecting the squares of the variations of direction cosines a,, we get
from (3.19) and (3.20) the relationships

0,’ + o& - o&, = Am, w/z) (3.22)

But e., er and 13~ are, according to (3.21), the components of small
rotation 0 along the axes xyz . Therefore, (3.22) can be written in the
form of Equation de
fX+a,x0=Am i dt= d”)
(3.23)
with the initial condition e(O) = 8” determined by (3.X).

Now It is obvious that 6r, = 8 X lp, wherefrom

6x, = e,z - 0, y (WZ) (3.24)

The full error, according to (3.13), Is

arc,= a2+Ptxz (XrrG (3.25)

where bx, by and 6.~ are defined bjr (3.16), and 6xa, 6~~ and bt, are
given by (3.22) and (3.24).

If in computing the difference (3.15) one considers the variation in the

noncentrality of the Earth’s gravitational field, then it is necessary to
retain the quadratic terms in the expansion of the spherical Part of the
gravitational field since they are of the same order of magnitude as the
linear part of the expansion forthe correction of nonsphericity (at least
for the Earth’s fleld,in any case).
Introducing the notation

g (r) = - f g (4 -t s (r) (3.26)

where
s(r) = E, (z, ?/, 2) x + Ey (2, y* 4 p + E, (27 y, 4 2

is a vector function of r characterizing the nonsphericity of the attrac-

tion field, we get

6r g (rl
g(r) -g (r + W= - 7 g fr! + ~(rfsr)[*5(,r+sri--~-_-6~~_.
- (grad F$X ) x - (grad cUBrl) y - (grad e$r,) z (3.27)

Since (3.27) contains not only 6~ but also b&, then in this case the
first group of the error equations is not separated from the second group.
Similarly, the equations for ideal performance are not separated in consider-
ing the noncentralitJr of the attraction field.
 oQncra1 aQaQt1ona or inartial nQvl@1on 30-t

It is easy to Include In the first roup of error equationsthe attraction

differencesof other (other than Barthf celestialbodies at the Earth's cen-
ter and at the point 0 of the present posltlon of the object ln accordance
with the equations for Ideal performance (2.10).
In this case bri - bfio, and the error equationsof both groups are also
related.

4. Let u8 indicate now certain general propertiesof the error equations

for the investigatedgeneralizedcontrol system of Inertial navigation,and
also show how to obtain from them the error equationsfor particular systems
consideredIn [l to 53.

The error equations (3.16), (3.22), (3.24) and (3.25) pemnit a group of
transformationsdeterminedby the arbitrary rotation of the O1xyz (OXW)
trlhedron In space. This property of the equationsfollows from the arblt-
rary orientationof the coordinateset (trlhedron) Oxyz and the arbitrary
P) . It can also be proved directly, analogous to the proof In [2]. There-
fore the analysis of the error equationsfor the Inertialnavigation system
for any orientationof the set Oxya (&xy#) can be carried out relative
to another set of coordinates(trlhedron)suitably selected. (Analogously
as It was done In [2] relative to the rotation of the accompanyingtrlhedron
in azimuth).

Such a trlhedron can be, for example, O,g?jC fixed ln space. In this
Case the error equationsare obtained from (3.16), (3.22), (3.24) Md (3.25)
If one assumes piI 0. In place of (3.16) we obtain

sg..+$sE-+5(1r+Br+r) =
(E?U
= Ant- 2 (Am,g’ - Arnql’) - Am,‘5 + 6% rl (4.1)

which follows also from (3.8), and In place of (3.22), (3.24) and (3.25) we
get
O< = AmE, SE, =0,5 --8rrl, SE, = at + SE, (Es) (4.2)

In Equations (4.1) and (4.2) Ant, An,,, An<, Ame, Am,,, Amc
are respectively,the projectionsof An and Am on the 5qC axes . They
can be obtained from An., An,, An,, Am,, Am,, Am, if a,, (t) are known.

Equations (4.1) and (3.8) permlt an Interestinganalogy. They are analo-

gous to the variationalequationsof the motion of a particle of unlt mass
In the Earth's field of attractionwhen the motion of the point IE perturbed
by the forces appearing on the r t-hand side of Equations (4.1) and (3.8).
In particular,the equations (4.1Yh and (3.8) are analogous to the equations
of motion of a particle In a satellitecab!_n [ll]. The Indicatedanalogy
permits, In the latter case, the applicationto the analysis of the first
group of error equatlons.forinertial navigationof the well developed
methods of celestialmechanics.
Ii the attractionof celestialbodies other than the Earth Is also con-
sidered, then the correspondingerror equationwill be analogous to the Per-
turbed motion of a particle ln the field of attractionof n bodies.
308 V.D. Amlrcev

Equations (3.161, (3.22), (3.24) and (3.25) are the error equations for
an arbitrary control system. They are Valid, apparently, for the case when
the orientation of the set Oxyr Is a given function of time, l.e,. does not
depend on the coordinates determined by the lnertlal system, as well as when
the orientation of the trihedron Is given as a function of coordinates de-
termined by the inertial system. It is worth noting that In the latter case
the position of the Oxl/r trlhedron is perturbed by the errors of the co-
ordinate determination.

The homogeneous equations of the first group (3.16) are exact. They de-
scribe the perturbed operating conditions of the inertial system not only
for small but also for large perturbations.

Equations (3.22) and (3.24) of the second group are the equations for
small deviations, since in passing from (3.20)~~ (3.22), the squares of
ba,, (t) were dropped. In order to obtain Equations (3.22) and (3.24) for
large perturbations it 1s necessary to utiline the theory of finite and not
small rotations [93 in transforming the exact equations (3.10) and (3.20).

Note that the second group of error equations for the Inertial system is
of the same form as was obtained In [2], and for given dIJ (t) car be lnte-
grated by quadratures, which follow from (4.2).

Let us obtain from Equations (3.16) the equations of perturbed operation

for the particular Inertial navigation systems investigated in [l to 51.

Assuming in (3.16)

and noting that

r2 = Ea2 -t 9,” + Ce2 (4.4)

therefore, within the accuracy of terms up to second order of magnitude

1f + lir j - r = %, SE+.’ + q* %,’ 4 E,K,’ (4.2)

we obtain Equatlons (4.14) In the paper (23 for the Inertial system where
the integration takes place along the directions fixed in absolute space.

Mrectlng the axis 02 of the set Oxyz In the unperturbed state along
I , and noting that In this case

z=y=o, z=r (1.6)

and within an accuracy of terms up to second order of magnitude
6x=7$3, dy= -ra, dz===Qr, ir-f-&I--r=62 (4.7)

we obtain Equations (3.9) in [2] for the system with three accelerometers
located along the axes of the Darboux trlhedron on the sphere surrounding
the Earth.
 oenera1 equationsor 1nert1a1 navlgat1on 309

The first two equations in (3.16) in this case become the small oscillation
equations for the Schuler pendulum [:5and 123, tro-gyroscopevertical [4],
gyro horizon compass [3], and the system consideredIn [l] If In (4.5) one
lets
62 = Cir= 0 (5.8)
ln [l to 53, the small oscillationequationsfor systems near the position
of relative equilibriumare given. ;&quations (3.16) are exact. They immedi-
ately yield the equations for perturbed motion for arbitrary and not neces-
sarily small deviations.

Let, for example, the trihedron Oxys In the unperturbedposition be the

Darboux trihedron on a sphere [l to 51 surroundingthe Earth. Let us denote
it by OX,Y,Z~ retaining the notation xl/a for the perturbed position.Let
the perturbed position of the set xya relative to the unperturbedbe
characterizedby two angles * a and B according to the direction cosines
table X ?I 2,

X0 co5 p 0 sin p
sin a siri p COSd - sin a Cos /I (4.9)
YO

GJ - cos asin p sin a cos a cos p

The projectionsof the absolute angular velocity of the set 0~~2 upon
its axes are expressed by means of the projections O+, WY,,wzO of the abso-
lute angular velooity of the set ~x,y,z, on the axes xoyOr, and the angles
a and fJ as follows:
0, = wk.Cos p + mu0sin a sin p - 0;” cos a sin /3 -/- a’ cos /3
% = oy, cos a + wzO sin a + p’ (4.10)
w, = Ox, sin p - ouo sin a cos p + oz, cos a cos /I -/- a’sin p

For the system with two accelrrometersand the motion of the point 0 on
a sphere of constant radius F it folloiisfrom (4.9) that

6~ = - fcos ~1sin p, 6y = rsin a, 6.2= r (COSa COS p - 1) (4.11)

Substituting(4.10) and f4.U) Into the first two equationsin (3.16) we
obtain after obvious groupings and simplificationsl *

p" + a’” sin $ cos p + 2~' (w,,sin 3 cos /3- (4.12)

- 0,.sin a co.3 f3 i_ 0,” cm a co? p> f q/,’ (cos a - cos p) +

+ 02, sin a + 0,“’ sin a sin p + (wg2 - uzoQcos a cos p - w~,~)sin p cos a +
+ oG4 (cos p - cos a) sin p - oy,2 sin2 a sin p co.9 p -f-
+ oxDouo (sin” p - co2 p) sin a +
+ o,,OzI (co9 p cos a - cos p - sin2 p Cos a) +
jo,o;.(:!cosacosp-1)sinasinp=O

* The third rotation is IZIeSSential. It can be referred to the position

of the x,y,z, set.
** !i!he
equations are homogeneous.
 a” COS j3 - 2ct’p’ sin p -i- Zp’ (- 03%.sin TJ;-/- 0, sin a co.5 fl - (4.12)
- 0~~ OOSa OOS p) + 0%’ (OOS /3 - CDS O) + Ova’ Sin a sin p - cont.
- wz,’ Cos a sin B + (o02-- oq2 - wv,"(l - co9 a cos p) -
- COS a COS p) Sin
6.&O' a - wx,ow co5 a Sin p - 0%. o,, sin a Sin p +
+~,~,(~in2acas~-~t3s2acos~+cosa)=0, 02,2=g/f

EQuations (4.10) coincide exactly with Che equations of the Schuler Pendu-
lum E5 and 123 Mt? suspension paint of which moves on the sphere of radius r.
Indeed, the pendulum equations, as projected on the axes rgz, me of the
form
Hz - w, Hg = -
ri;‘+ w,_, IF,, Hu’ + o, Hz - a, H, = jr;: (4.13)
For the Schuler p0~d~l~ [5 and 12J, the Projections of angular mOmentm

are
&=, = RZiFti,, Hv = ~~~~~~ HI = 0 (4.14)
The components of P tin the x~L(o~~ are

F,, = - mr (@r*’ + 0%x,a,,) t F,&- TM‘k%,’- @y,%,)

I;;, = ml- h: + qJ - mg (4%)
The substitution of (4.14), (4,15) and (4.10) into (4.13) gives kunedi-
atefy Equations (4.12). Equations (4.12) are ale0 the equations of perturbed
motion of systems If, 3 and 43.

For constant Q-Q*’Q&Y @Lo Equations (4-12) possess a first integral, In

order to obtain It, it Is sufkcient to multiplk the first equation in’(4.12)
by 8’1 the second equation by a’cosg and to add them. Integrationof the
sum yields
p-- 20,~COSacas
v = (a'cos8)2-t. p+

-t- %, a COSa Ct COS’ p + Ox.* (sins @+ 2 OOS CX00s f3) +

+ 0v,2 (Sins a CO@ p -+ 2 cos a cos f3) - ~~~~~, sin u sin p ~0s B + (UB)
-t 2Q&,aG (cos a sin fi co9 @--- sin p) -j
-k 2@Va@ze(sin a c0s 8 - sin a c0s u ~09 Pf = const

!&e LiaPun0v stabillity condition for the solution (4.12) follows from
(4.15) as
%l~-Ws--_~-o,2;>il
r, ?A (4.17)

It was obtained earlier in [6 and ‘73 from consideration of the eqZ%ati#nS

reducible to (4.12).
Condition f’e.17) is a sufficient condition. The papers 16 and 131 show
that condition (4.17) can be considered as a necessary stability conditkin
If fully dissipative forces are assumed in the system. In this ConneCtkn,
it is necessary to note that dissipative forces An inertial systems lead to
the occurence of velocity deviations, and the systems of inertial navigation
in the absence of velocity correction, tend to be designed so as to avoid
dissipation. Therefore, arbitrary introduction of dissipative forces into
the system requires great care In the fnveStig&tiOn of stability.

In order to obtain the equations for perturbed Performance in a system

with two accelerometers and lar$e deviations for I” = rft) when P is de-
termined from auxiliary sources not connected with the operation of the
Inertial system, the r should be considereda given function of time in
(4.11). If r Is computed as a function of two coordinatesdeterminedby
the inertial system, then in writing the equationsof perturbedperformance
the variation of this function should be taken [2-j.

In order to obtain the equationsfor perturbedperformancefor large

deviationsof the system [2] with the variables a, 8 and br , it is neces-
sary in (3.16) to substitute
6x = - (r + 6~) cos 6 sin f3, By = (r + 6r) sin a (4.18)
82 = (r+ 8r)(cosa 60sp -1) + 8r
in place of (4.11).

5. Let us investigatethe stabilityof the inertial system performance

for the case when the unperturbed &-axis of the set Oxys is directed
along r , and w,, U,Q,UJ,and r are constant. From (3.16)
jr +drI -r = 62 + 0 (62) (5.1)
Retaining now on the right-handside of (5.1) only bz , we note that
Equations (3.16) become linear with constant coefficients,the character-
istic equation of which is a cubic with respect to the square of the unknown
¶ =p=
d + 2q2 (@x2 + ay + 6&y + q i- 300" -j36$ (0," +
+ q/2- 20,2) + (W,a + og2+ o,z)sl -
- oo” (q,* - co,2- ol;” - oz2) (20~+ ox2+ wy2- 20,~)
= 0 (5,2)
For stability (non asymptotic)Equation (5.2) must have, as is known,
negative or zero roots, and to multiply roots of the characteristicequation
of the system (3.16) should correspondlinear, elementarydivisors of the
characteristicmatrix.

In order to investigatethe stabilityof the system (3.16),in the present

case, one can utilize the fact that it can be consideredas describingthe
motion of a particle of unit mass under the action of potentialand gyro-
scopic forces,

The expressionfor the potential function can be written

u = - $ i(w,2- COY2- 0,s)(Izs+ (oo"- 0,s - 0,s)6ys -

- (20,z
+ ciL&2
+ ov~~8za
+ 2co,o,8dy
+ (5.3)
+ 20,w,8282+ 2opr 8y8zJ
The following expressionscan be consideredas gyroscopicforces:
- 2w,82‘ fm@
20,8y' (5.4)
since the matrix of coefficientsof these forces is antisymmetric[lb].
The system (3.16) has In this case the energy integral
312 V. D. Andreev

ax.2 + by.3 + 62’2 - 2u = IJonst (5.5)

which can be obtained directly if Equatlons (3.16) are added, multiplied
respectively by 6x', 6y', bz', and integrated

WE If the gyroscoplc forces (5.4) are rejected

2
then there will remain only the potential iorces.
(f/J For stability of equilibrium under the action of
I
only potential forces, the potential function
must have a maximum at the equilibrium point,
,.,Cj (2/
Since the potential function (5.3)i:;a quadratic
(3) form, the conditions for a maximum are the Syl-
~ R’
vester conditions for positive-definiteness of
4
of a quadratic form. In the present case they
Fig. 1
are the inequalities

oo2 - 0, 2_ NY2- wz2> 0, 20: - 20,2 + c!J,2+ coy3< 0 (5.6)

In Fig. 1 O2 = ~1,' i wla, the straight lines 1 and 2 corresponding to

the equations (uo2 - iui2 - a2 -= 0 and 2~,* + 0’ - Z&I*” = 0 are plotted.
The figure shows that the regions defined by (5.6) do not intersect, anr.
therefore, the potential function has no maximum.

Since In the present case the potential function is homogeneous of second

degree then, according to the known theorem of Liapunov [15], the instability
follows from the absence of the maximum without the necessity of considering
the terms of higher orders.

Let us return to the gyroscopic forces (5.4).

In the regions (1) and (3) (see Fig. 1) where the de&ret. of instability*
of the conservative system is odd, the gyrcscopic forcer, according to the
Thomson-Tait theorem [16], cannct stabilize the equilibrium.

In the region (2), where the degree cc‘inrtability is even, the pocsibi-
llty of stabilization by gyroscopic forces remains in principle. This stabi-
llzation, as is known [Pj], has a temporary character and is destroyed by
the forces of full dissipation.

The stabilization by gyroscopic forces results if, for example,

OX2 -to," =o$ UZ? = e2 (5.i)

*where e2 is a sufficiently smail quantity.
It can be easily shown that the polynomial (5.2) satisfies in this case
the Hurwitz conditions. The discriminant A of the cubic equation obtained
from (5.2) by substitution of the variable

1, =
4 ‘I + 219 (0 02-i- &2) (5.8)

* The number of negative stability coefficients of PoincarP [15].

 General equations of inertial navi~atlon 313

is negative
A= - &)pm g-a< 0 (5.9)
All roots of the characteristic equation are therefore simple and purely
imaginary.

6. In conclusion, let us investigate the stability of a system with three

accelerometers when the orientation of the set 0ny.z Is arbitrary, and 1~1
for u(r) is given, in addition, from a source outside the inertial system.

In this case it is necessary to make the following assumption in the error

equations (3.16)
1r+tq --r=O (6.1)
and the error in ITI should be referred to the right-hand side parts in
(3.16).

It follows from (3.16), (3.8) and (4.1) that the homogeneous error equa-
tions become
(4%)
(6.2)
From (6.2) it follows that for r = const when w02 = g/r is constant,
the perturbed motion of the system is stable for any U,(C), w,(t), wu,(t).

In this case for given all(t) the solution of (3.16) follows i~ediate~y
from (6.2).

For constant u),,wy, w, the stability can be detected also without re-
ference to equations (3.8) and (4.1).
In the present case the condition for the maximum of the potential function
is reduced to one inequality
woz- 0 I:
2- 0 s-
U
OLZ> 0 (6.3)
Outside of (6.3) the degree of instability is even and the equilibrium is
stabilized by gyroscopic forces. The latter is easily proved by reviewing
the characteristic equation which, if written in terms of the square of the
unknown, is
9s+ (;:(Q
#)s
-+ "w 2)q' -;-q (:30
o4 + 0 4) $ IO o? (0 o? - 0 2): = 0 (6.4)
where for simplicity the notation
n 0
6)-= o,- -i C$,
'+ OZS (6.5)
iIs: been introduced.
The polynomial (6.4) satisfies the Hurwitz conditions,.since always
(So, .+ 20') (30,4 + 04) - mo3 (wo? - d)-? > 0 (f-5.6)

The discl,iminant 6 of the cubic eqUation

y3 + 3hy -+ L!r = 0 (6.7)

which is obtained from (6.4) by the change of variable

is nonpositive 4
A = _ - oO' d(4~,‘- ,z):! < 0
27 (6.9)
314 V.D. Andrew

If 0 =./= 0, 4002 -&a# 0, and a02 -@‘f& then (6.4) has three different
real roots. Consequently, the characteristic equation has three pairs of
different purely imaginary roots.
For i~#O,ho~ *oaf 0, and oc2=os Equation (6.4) has along with two
real negative roots, also a zero root, and the characteristic equation has
a multiple zero root.
If w=O, then (6.4) has a triple root q1,2,a= -@02, and the character-
istic equation, respectively, a pair of Imaginary roots of the same multi-
plicity.
Finally, for 400~ --os= 0 Equation (6.4) has a multiple root ga#=--O~~,
and the characterlstld‘equation a pair of imaginary multiple roots.
It can be shown that when the roots of the characteristic equation are
multi le, the elementary divisors of the characteristic matrix of the system
(3.167 In the present case remain linear.

The author is indebted to A.Iu. Ishiinskii for review of the manuscript

and useful remarks.

BIBLIOGRAPRY

1. Ishlinskii, A.Iu., Ob uravneniiakh zadachi opredeleniia mestopolozheniia

ob”ekta posredstvom giroskopov I izmeritelei uskorenii (On the equa-
tions for the problem of determining the position of an object by means
of gyroscopes and accelerometers). p,y~ Vol.21, t@ 6, 1957.
2. Andreev, V.D., K teorli lnertsial’nykh sistem avtonomnogo opredelenila
koordinat dvlzhushchegosla ob”ekta (On the theory of Inertial systems
for the self-contained determination of the coordinates of a moving’
object). PMM Vol.28, W 1, 1964.
3. Ishlinskii, A.Iu., K teoril glrogorizontkompasa (To the theory of the
horizon g~ocompass). P# vo1.20, @ 4, 1956.
4. Ishlinskil, A.Iu., Teoriia dvukhgiroskopicheskoi vertlkali (Theory of
the two-gyroscope vertical). p,#/ Vo1.21, W 2, 1957.
5. Andreev, V.D., Ob odnom sluchae ma1y.k.hkolebanll fizlcheskogo maiatnika
s podvizhnol tochkol opory (On a case of small oscillations of a physi-
cal pendulum with a movable suspension paint). PM,?4 Vo1.22, I 6, 1958.
6. Merkin, D.R., Ob ustoichivosti dvizhenlia glroramy (On the stability of
motion of a gyroscope frame). p#y v01.25, t@ 6, 1961.
7. Zhbanov Iu.K., K lssledovanliu svobodnykh kolebanii v slsteme avtonom-
nogo opredetenlla koordinat dvizhushchegosla ob”ekta (On the investi-
gation of free oscillations In the system of autonomous determination
of moving body coordinates). pMMVo1.24, W 6, 1960.
8. Appell, P., Teoretlcheskala mekhanika (Theoretical Mechanics). Fizmatgiz
~01.2, 1960.
9. Lur’e, A.I., Analiticheskaia mekhanika (Analytical Mechanics). Fizmatgiz,
1961.
10. Duboshin, G.N., Teorlia pritiazheniia (Theory of Attraction). Fizmatgiz,
1961.
11. Lur’e, A.I., Svobodnoe padenle material’noi tochkl v kabine sputnika
(Free fallofa material point In a satellite cabin). PKM vol.27, @ 1,
1963.
12. Ishlinskii, A.Iu., Ob otnositl’nom ravnovesii fizicheskogo maiatnika s
podvlshnoi tochkol opory (On the relative equilibrium of a physical
pendulum with a moving point of support). PM V01.20, i-E3, 1956.
13. Koshliakov, V.N., Ob ustoichivosti girogorlzontkompasa pri nallchil
dissipatlvnykh sll (On the stability of a horizon gyrocompass with
dissipative forces). PMX ~01.26, NP 3, 1962.
14. MerMn, D.R., Giroskopicheakie sistemy (Gyroscopic ~Qr,"ystems).
Oustekbizdat, 1956.
15. Chetaev, N-Q., Ustoichivost’ dvizheniia (Stability of Motion],
C+ostekhizdat, 1955.
16. Thomson, W. and Telt, P,, Treatise on Natural Philosophy, Vol.1, Cam-
bridge University Press, 1879.

Translated by VAC.