INERTIAL NAVIGATION

BY

C. F. O ' D O N N E L L *

Part II *
D. P R O B L E M S OF I N E R T I A L G U I D A N C E S Y S T E M S

1. Coordinate Systems
At first glance the choice of a coordinate system, provided the origin
and destination are known in terms of coordinates of t h a t system,
appears inconsequential. In practice the choice has a major bearing on
the system mechanization, especially in the case of the position com-
puter. Consider the two choices of coordinate systems for navigation
on a circular table top shown in Fig. 22. Figure 22a shows a Cartesian

~ X =X o ÷o/ojrAxdT~dt

0 SO oftoj (A°-; bR) d'rdt

+

(b) Polar Coordinat,es, Table Top Navigation

FIG. 22.

coordinate system, origin at the table center, with position measured in

terms of x and y coordinates. Figure 22b shows a polar system, again
a Autonetics, Division of North American Aviation, Inc., Downey, Calif.
Part I appeared in this JOURNALfor October, 1958.
373
374 C . F . O'DONNELL [J. F. I.

with the origin at the center of the table, and position measured in
terms of R and 0, a distance and an angle. In m a n y ways the choice
of the polar coordinate system appears more reasonable for it matches
the b o u n d a r y conditions of t h e surface on which we are navigating
more elegantly t h a n do the Cartesian coordinates. However, in navi-
gator mechanization if the accelerometer platform is aligned in the
Cartesian system, one sensitive axis in the x direction, the other sensi-
tive axis in the y direction, then

Ax--2
A~-- i~
and position is given simply by

X=Xo+ fo' foo'A~drdt

Y = Yo+ fo' fo'A,,drd,.

On the other hand, in the polar coordinate system with one sensitive
axis in the R direction and the other in the 0 direction, acceleration com-
ponents are given by
A~ = R -- O~R
Ao = OR + 2~R
and t h e determination of position requires not only double integration
of these accelerations, b u t the subtraction of c o n t a m i n a t i n g terms, and,
for the angle 0, division by a variable length.

R = Ro +
f0 fo" (AI~ + O*R)drdt

Navigation is being carried o u t on the same surface in both cases b u t

t h e c o m p u t i n g problem is m u c h more complex for the polar coordinate
system t h a n for the Cartesian coordinate system. T h e comparison is
shown in block diagram form in Fig. 23.
There is t h e additional problem, illustrated in Fig. 22b, of a variable
azimuth rotation rate in the polar coordinate system. Here a flight
passing close to the pole at the origin can require a swing in azimuth of
almost 180 ° in a very short period of time. This not only taxes the
capacity of the computer, b u t in case the gyro package is being torqued
to rotate with the accelerometer platform, it increases the m a x i m u m
Nov., I958.] INERTIAL NAVIGATION 375

precession rate gyro and torquer must handle accurately. This prob-
lem of a variable azimuth torque is not present in the Cartesian coordi-
nate system, therefore, in this particular example it can be seen that
computer mechanization, torquing rate capability of the reference
gyros, and accelerometer platform hardware are all strongly influenced
by the choice of position coordinates.
r
X" Platfor~ I Co.purer

'Il iol .x)

I .., )

a) Block Schematic, I n e r t i a l Navigator, C a r t e s i a n Coordinates.

Platform ¢¢m~uter

Mu/tlply
d

II Accelera~L on

b) Block 8c~metic, Inertial HavlgaLor, Polar Coordinates

Fm. 23.

In general then there are two problem areas requiring investigation

before a choice of coordinate systems is made. First there is the diffi-
culty of determining which portion of the acceleration measured by the
physical instruments is associated with a change in the position coordi-
nate of interest, that is, x and y for Cartesian coordinates, R and 8 for
polar coordinates. Secondly there is the problem of maintaining the
correct orientation of the accelerometer platform a n d / o r the gyro
reference package.
A number of possible coordinate systems are shown in Fig. 24.
376 C.F. O'DoNNELL [J. F. I.

The most widely known is t h a t shown in Fig. 24a, a geodetic latitude-

longitude system. Here the north and south poles are singular points.
At the north pole itself longitude loses all meaning as a position coordi-
nate and becomes only a direction in which to point. In this case the
coordinate system poles coincide with the poles of the spin axis of the
Earth.
Actually it is not necessary to have this coincidence, as shown in
Fig. 24b. It is possible to place the poles of the navigation coordinate
system in such a position t h a t the equatorial plane of the coordinate
system used makes an angle a with the equatorial plane of the Earth.
N

ste System

dl T ~ e n ~ Plane e) Cyllnflrlc~l Coordlna~e8

F1o. 24. Coordinate systems.

The purpose of this would be to avoid the problems of a rapid azimuth

rotation of the accelerometer platform when approaching the Earth's
north or south pole. The coordinate poles could be placed, for ex-
ample, on the equator in an area in which the vehicle using the inertial
navigator would not normally operate.
Another possibility would be the use of a single pole system shown
in Fig. 24c. In this case only one of the Earth's actual poles becomes a
singular point. It is perhaps easiest to visualize this cordinate system
as being formed by a set of planes fanned out in two directions at right
angles to each other about the south pole. This forms a true orthogonal
Nov., I958.] INERTIAL NAVIGATION 377

set of coordinates and allows flight at any point over the Earth other
than the single pole area without difficulties in computation or genera-
tion of azimuth rates.
The coordinate systems so far considered have all been of the local
level type. It is also possible to employ coordinate systems for which
the platform will be level either only at a point, as in the case of the
tangent plane system shown in Fig. 24d or along a particular great-
circle path, such as the cylindrical coordinate system shown in Fig. 24e.
In both of these systems the problem of excessive azimuth rate does not
exist. However, a new problem has been introduced, that of computing
the component of g being sensed by the accelerometers as a function of
position and either allowing for this in the computer or supplying a
compensating torque to the accelerometer to null out its effect.
One factor which has nothing to do with the difficulty of computing
or torquing in azimuth is the problem of obtaining origin and destina-
tion position in the coordinates chosen. Unless geodetic latitude and
longitude are chosen for the inertial guidance system, auxiliary calcula-
tions must be performed to convert from latitude-longitude to the
coordinates used to determine origin and destination locations.

FI6. 25. Gyrocompass alignment, single gyro.

2. Alignment
Throughout this paper continual mention is made of the fact that
the accelerometer platform must be correctly aligned. During system
operation alignment is maintained through feedback loops existing
between the platform and the computer. However, there remains the
problem of establishing the initial alignment of the system. One
possibility is the use of an optical system employing a theodolite, and
an external reference for aligning the platform both in azimuth and
level. An alternate possibility is to make use of the accelerometers and
gravity to level the platform, with the optical system and an external
reference being employed only for azimuth alignment. If it is required
378 C . F . O'DONNELL [J. F. I.

t h a t the system be completely self-contained, an azimuth reference can

be obtained from the E a r t h rate vector f~ while gravity is used to es-
tablish the local vertical.
In the traditional gyrocompassing method, the platform would be
held level using the gravity vector as a reference for the platform ac-
celerometers and a gyroscope lined up so t h a t its spin axis was approxi-
mately north-south (Fig. 25). T h e n as the gyro is restrained to move
in a locally level plane and as, through the E a r t h ' s rotation, this plane
is moving with respect to inertial space, the total angular rate of the
gyro is as follows :

= (~ cos 4~cos 0 + s)i,~2 sin $ cos0iv + (6 + ~2sin 0)L

where s is t h e rotor spin rate, $ the misalignment angle, and 0 the lati-
tude. Its angular m o m e n t u m is:

/~ = I ( ~ cos 4~cos 0 + s)T, -- A~ sin 4~cos 0iu + A (6 + ~2sin 0)T,

where I is the rotor axial m o m e n t of inertia and A is its transverse

m o m e n t of inertia. T h e n allowing for the fact t h a t the gyro is being
torqued only to maintain its spin axis in a local level plane and t h a t in
general :

o-/- + ,~'x~ =

where ~' -- ~ - six, and ~ is the restraining t o r q u e ; and solving the

resultant differential equations, neglecting t e r m s in ~2~, gives the ex-
pression for determination of $ as

Is~ cos 0 .
~; + A san ¢ = 0.

For small values of ~, sin ~ = ~. This is the same form of equation as

t h a t governing error propagation in inertial navigators. T h e u n d a m p e d

harmonic oscillations in this case have a period r = 2~- ~ c o s 0"

If no disturbing torques are present the platform under the control of
this gyro oscillates about the true n o r t h direction with a peak a m p l i t u d e
equal to the original misalignment angle. Either a fixed reference can
be employed, the mid-point of these excursions noted, and the platform
be brought to a stop at this point, or electrical d a m p i n g can be inserted
in the system to d a m p o u t this oscillation leaving the platform cor-
rectly aligned.
I t should be noted t h a t r increases without limit as 0 approaches 90 °.
T h e system fails at the poles and becomes increasingly sluggish as the
Nov., 1958.] INERTIAL NAVIGATION 379

poles are approached. This is a basic limitation on gyrocompass

alignment.
An alternative approach to the use of the Earth rate vector as an
azimuth reference comes through considering the platform as an angular
velocity resolver. In this case the platform is aligned approximately in
azimuth, and leveled accurately using gravity as a reference. T h e
c o m p u t e r applies E a r t h rates about the x and z axes computed from the
known position of the inertial navigator. If the platform is correctly
aligned, it will remain fixed in orientation with respect to the E a r t h ' s
surface. However, if as shown in Fig. 26a, there is an initial azimuth

Drift ~te/.,~

%
Rate ~ %

%
b) Miselisnment Drift Rate.
a) Platfo

%
I %-

u 6 12 18 24
~, ~, in hours
c) Error Propaga~ca

FIG. 26. Inertial navigator gyrocompass alignment.

misalignment angle ¢, then, as shown in Fig. 26b, the rate supplied to

the platform does not coincide with Earth rate. Therefore, there will
be an error in the platform rate ~ . This means t h a t the platform will
drift with respect to the Earth's surface at this rate.
T a k e the simple case of a platform misaligned in azimuth on the
equator with its x axis pointing approximately north, its y axis pointing
approximately west, and its z axis approximately vertically upward.
Now define a set of angles ¢,, ¢~, and ¢~ to be the small misalignment
angles of the platform axes from the position they would occupy for a
perfect platform correctly aligned. These angles are assumed to
 380 C.F. O'DONNELL [J. F. I.

be sufficiently small so the cosine of any angle equals 1 and the sine
equals the radian measure of the angle itself. With this restriction the
angles can be treated as vectors while the platform is rotating without
the necessity of using a set of Euler angles to represent the platform
misorientation. For this example the initial values of the q~'s are
~0 = 0, ~0 = 0, and ~,0 = ~,(0).
The equations for the ~ angle propagation can be readily set up
realizing t h a t the platform can be regarded as a three-dimensional angu-
lar rate resolver. For this example, with the platform correctly
oriented, full Earth rate, ~2, should be present about the x axis while the
y and z axis components of rotation rate should be zero. In spite of
the platform's initial misorientation the x-axis rate should still be ~2,
to a first order approximation, but if the platform is to remain sta-
tionary with respect to the Earth the y-axis rate should now become
- ~2sin ~z (0), or considering our small angle approximation, - ~ = (0).
The computer which is supplying rates to the platform knows present
position and assumes t h a t the platform is correctly oriented. There-
fore, this spurious y-axis rate, - ~2~=(0), is not supplied, which means
t h a t a drift is taking place about the y axis of + ~2q~,(0).
As soon as this drift occurs, because it is about the y-level axis, the
accelerometer on the x axis senses a component of gravity and delivers
false indications of change in latitude to the computer. This has two
effects: (1) the false rate of change in latitude, ~0, is supplied as a
torquing signal to the y axis of the platform and (2) the false change in
latitude, 80, produces a change in the Earth rates supplied by the com-
puter to the x and z axes of the platform. This is the word picture of
w h a t is happening in the physical system. To complete the description
of the error propagation, the equations for rates of change of the ~
angles are

4, = - ~(~0 + ¢~).
The change in indicated latitude, M, is due entirely, in this simple
example, to drift about the y axis and therefore due to the double
g
integration of the instantaneous angular accelerations - ~, a" Taking
the Laplace transformation and making the obvious substitutions gives
the following form for the equations:
s~= = 0

s~+ ~g ~ - ~ = = 0
 Nov.,I~8.] INERTIAL NAVIGATION 381

T a k i n g the d e t e r m i n a n t of the left hand side of these equations gives

(s~ +a-g) (s~+~22)

; set -g = w02.
S a

Solving for 60, the latitude error, under those conditions in which the
sole driving function is the initial value of ~b, = ¢, (0) gives
- co02f~.(0) ( si~f2t sin co0t)
~O(t) - (ojo 2 _ ~2~) - oJo "

Now c00 has a value of 1.23 × 10 -3 radians/sec while ~2 has a value of

approximately 7 × 10 -5 radians/sec. Therefore the first order form
for 0 is
60(t) = - ¢z(0) sin ~t - - - sin wot.
~00
This shows t h a t for our simple case the principal c o m p o n e n t of error
does indeed have a 24-hr. period although there is an 84-min. period
oscillation superimposed on the principal value (Fig. 26c).
Naturally when the autonavigator is in motion the expressions for
long period oscillations are more complex.
The azimuth misalignment angle is also shown in Fig. 26c. From
this it is clear t h a t one possible m e t h o d of alignment is : set the system
up, align it approximately in azimuth using, say, a magnetic compass as
a reference, and wait for 6 hours. At this time discontinue the align-
m e n t mode as the azimuth error is 0. Another possibility is to make
use of the latitude error in the position computer. Knowing t h a t the
system is at rest with respect to the Earth, then any change in latitude
appearing in the c o m p u t e r can be taken as evidence of an azimuth
misalignment. Measurement of the rate of growth of this error as a
function of time leads to a direct measurement of the magnitude of this
azimuth error. T h e misalignment can be corrected either by step
input or through the use of some automatic feedback system incor-
porating damping. However, although it did not appear in the equa-
tions for Cz in this ~case (on the equator cos 0 = 1) cos 0 does enter the
picture, so this system of azimuth alignment fails at the poles.
In this way using both the Earth's gravitation and the Earth's
rotation as reference, combined with a knowledge of present position,
it is possible to make the alignment of the inertial navigator a com-
pletely self-contained operation.

3. Stabilization

In our system pictures to date, we have shown physical connections

only between the gyro package and the accelerometer platform. In
382 C.F. O'DoNNELL [J. F. I.

an actual vehicle, additional gimbals are required to isolate these sys-

tems from vehicle motion. A possible gimbal system is a 3-axis
2-gimbal system with roll being the outer axis, pitch the intermediate

Vehicle Frame- - - ~ ~ Roll Axis

/fl \
Fl~°l~ts~ P~C~lfe°~eter

a) 4-Axls3-Oh~al Platform Suspension, Pl~ View

,,Servo
~mpllfier

6ear
Drive

Torque,
Due to -T
Platform
8ervo Acceleration

~ I~ae to Mass
Unbalanced Unt~lance
Platform

b} Single-Axls8ta~lli~tion

CoLmtar Torque-~
Rotor & Pl&tform ~ Gyro Rotor %
f Dip Angle / PreeesslonAngle F Servo S l g n a l \

,1 I _
G3To Pick-off, & Ser~o Motor
I G3TO; ~ecvo Amplifier & Gear Train
I
t
O ) SinsleAxis St~ilizing Control Loop

F~G. 27. Platform stabilization.

Nov., 1958.] INERTIAL NAVIGATION 383

axis, and azimuth the inner axis. T h e roll axis is connected directly to
the vehicle frame while the accelerometers and gyros are m o u n t e d on
the inner table. If additional gimballing is required between the
accelerometers and the gyros, as for example in a geometric system,
then these would have to be added with a resultant increase in dimen-
sions and weight of the external gimbals.
If complete maneuverability is required, one additional gimbal
m u s t be added, (Fig. 27a) making it a 4-axis 3-gimbal system, in order
to avoid the problem of gimbal lock. Usually this axis will be a re-
d u n d a n t roll axis or a r e d u n d a n t pitch axis depending on the m o u n t i n g
employed, with a limited degree of freedom.
The mechanization for stabilization is illustrated in Fig. 27b. Here
stabilization about one axis is being considered, the platform as shown
is obviously unbalanced. An acceleration in the direction indicated
would result in a torque on the platform tending to cause it to rotate.
This is sensed by the control gyroscope. The torque causes a small
precession of the gyro rotor. The resulting pick-off angle generates an
error signal. This is fed through an amplifie r to a torque motor
m o u n t e d between gimbal and control axis for the platform. The sign
of the error signal is such t h a t the torque applied to the platform
through the torque motor is in the opposite direction to the torque due
to mass unbalance. If the system is correctly adjusted, then virtually
equal but opposite torques will be applied to the platform from these
two sources, mass unbalance and the torque motor, with the result
t h a t the platform will remain stationary.
The error sensing and feedback loop for this single axis stabilization
is shown in Fig. 27c. As can be seen, the relation between the dis-
turbing torque T~ and the resultant platform error angle ~ is given for
G(s) = 1 by
sK1
4~u(s) = Tv (s)
S + KIK2K~(.4

where K3 is the amplifier stage gain, K1 and K2 are constants relating

mechanical properties of the control gyro, and K , is the conversion
factor from servo amplifier signal through servo motor and gear train
to counter torque.
T h e system as shown is stable, with errors d a m p e d exponentially.
However, it should be noted t h a t there is already one integration be-
tween the disturbing torque and the error signal, giving a 90 ° phase lag.
The characteristics of practical system elements m u s t be considered,
such as back-lash in the gear train between the torque m o t o r and the
control axis, transfer functions of the various amplifiers and transmis-
sion paths in the system. It is possible to obtain sufficient phase shift
at some frequencies of oscillatory disturbing torques to cause the
384 C.F. O'DoNNELL [J. F. I.

system to amplify these oscillations leading to "tumbling" of the plat-

form. For this reason, the design of the platform torquer system re-
quires a detailed servo analysis. It must be tailored to the platform to
be controlled, considering the environment in which it is to exist.
Skillful use of practical components in building up G(s) leads to a
workable system in most cases.
The complete problem of platform stabilization is somewhat more
complex than that shown for the single axis. Error signals sensed by
gyros on the platform must be resolved through an azimuth rotation
angle ~ to the roll and pitch axis of the gimbal system. This is illus-
trated in Fig. 28a. As the vehicle heading changes ~ will usually

X-Gyco ConLrol Axis

Azimuth Resolve["

Pitch
y-Gyro PitCh Servo M~tor
Control Axis - -

!'~oli Servo Mother

~ ~ V e i~] c .le Fc~ne

a) Resolution Problem for Platfo.~m St~bki£zatLon

T~ + ~ • Sine-Coslne

/ k i \K / k," "x/
/ z \ I \ /
i ~ T~ sin Wh J_ T~ cos ¢

-T R shl ~,

\ , / I. t / l ' . / ~ - - ~ " I = } /\

. . . . . . - - 7

b) Electrical &: MechanLcal Resolutiom of Level Stab{lizing Torques

FIG. 28. Two-axis platform stabilization.

change. For example, in a latitude-longitude system the x-gyro control

axis will always point north, whereas the roll gimbal axis is along the
center-line of the vehicle, so the angle between the x-gyro axis and the
roll gimbal axis depends on present position and vehicle heading. The
outputs of both the x and y gyros then must be resolved through the
angle ~bto determine the error signal being amplified and applied to the
roll and pitch gimbal torquers. In block form this is shown in Fig. 28b,
Nov., I958.] I N E R T I A L NAVIGATION 385

where the electrical part of the operation is shown in solid lines with
the mechanical part in dotted lines.
As an additional illustration of the difficulties in platform stabiliza-
tion I would like to mention the problem of cross coupling and its
effect on stabilizing gyros. 3 If oscillatory inputs ~ and ~y about the
x and y axes, respectively, are in phase there will be a resultant average
torque experienced by the gyro controlling about the x axis. This
causes the gyro to drift, producing a system error even t h o u g h the gyro
itself is working properly. In addition the stabilization system is
keeping the average value of ~, the gyro precession rate, equal to 0
as it should.
E. SUBSYSTEMS
1. Inertial Components
a. Gyroscopes
Gyroscopes employed as directional references for inertial naviga-
tion systems m u s t of necessity be several orders of m a g n i t u d e better
t h a n the gyros previously used for attitude references in aircraft. A
drift rate of l ° per hour, for example, while very satisfactory for an
attitude reference in a low speed aircraft, would contribute an error
of 60 nautical miles per hour in an inertial navigation system. This
is not a reasonable degree of accuracy. R a n d o m drift rates of the order
of 0.02 ° per hour have been quoted for unclassified H I G gyros. 4
M a n y of the current inertial systems are based on the use of three
single-degree-of-freedom gyros as directional references. T h e m e t h o d
of use of a gyro as a control for a single axis is shown in Fig. 29. Figure
29a shows in pictorial form the gyroscope itself. A rotor a spins under
torque from a drive m o t o r b in an inner gimbal ring d. This gimbal in
t u r n is supported by the o u t p u t axis bearings j which will be of a very
low coercion type. Any precession of the rotor with respect to the
case g is sensed by the pick-off b which produces a signal sent to the
amplifier and servo control motor for the controlled platform axis.
T h e complete control path is shown in Fig. 29b. As can be seen
from this diagram, if the direction of an external torque, T, is reversed,
T1, then the gyro precession rate, ~, changes direction. The pick-off
is so arranged t h a t the direction of precession of the gyro rotor with
respect to the case is sensed as well as the magnitude of this angle.
Therefore, a counter-balancing torque can be generated through the
servo motor, correct both in magnitude and sign, to maintain correct
platform orientation.
8 This is covered in detail in a paper by R. H. Cannon published in the December issue of
the A.S.M.E., 1957.
4 Philip Klass, in Aviation Week, see reference 4 in the Selected Bibliography appended to
this paper.
386 C . F . O'DoNNELL [J. F. I,

Details of construction of any particular gyroscope for the most

part are classified, but some general remarks about desirable charac-
teristics can be made. For one thing the angular momentum of the
rotor should be high in order to keep the gyro drift rate, ~, small in
the face of the inevitable error torques. In addition this high angular
momentum produces maximum restoring force on the platform in the
presence of an external disturbing torque. This requirement is less
stringent when high performance platform servos are used, but even
here for high-frequency disturbances much of the stabilizing torque is
supplied by the gyro.
Output A x i s

• a - Rotor
T
L° r q u e c . _ ~ Pick-off Rotation rate = S
eads " - . ~ [ i~ [ h l / - ~ Leads MI about spin axis = I
, I- ~ MI a b o u t output axis = A

e e - Spin bearings

f - Flot&tion fluid
g - Outer case
Spin Axis ~ g h - Pick-off
i - Torquer
j - O U t p u t axis bearings

Platform - -

Motor Leads
\ \ \ \ ~ "\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ K \\\\\\\\\\\\\\~\\\\~
a] Single Axis Control Gyro ~/~ConLrol Gyro
~ ' ~g'--'---'~--~--- Pick-o f f

z - Se~o I I~ ~ L '/H

- h
&
U --l~ U
b) Control Loop - Single Axis Control Gyro

FIG. 29. Gyroscope for single-axis stabilization.

High angular momentum can be obtained either by using a large

mass in the rotor or a high spin rate. If the mass is large, the problems
of structural rigidity, heating due to higher power input requirements,
and bearing deformation under accelerations all require careful atten-
tion, while if the spin rate is very high, the problem of long-life spin
bearings becomes acute. Some reasonable compromise is usually
made between these conflicting methods of obtaining the necessary
angular momentum.
The output axis bearing must be of a low coercion type, for if the
gyroscope is to be used to control about only one input axis, it will be so
arranged t h a t the torques about this output axis are forced to 0. Any
Nov., 1958.] INERTIAL NAVIGATION 387
residual torque causes precession of the rotor, which is then interpreted
as being due to an external torque applied to the platform. A com-
pensating torque will be generated through the servo amplifier and
torquer motor causing platform drift.
In order to put the least possible load on the bearing, the gyro rotor
and inner gimbal are often surrounded by a sealed case and floated in
a flotation fluid as shown in Fig. 29a. If the displaced mass of flotation
fluid is equal to the mass of the gyro wheel plus flotation case, then
neutral buoyancy has been achieved and the load on the output axis
bearing has been reduced to a minimum. An additional feature of the
flotation fluid is t h a t if its viscosity is sufficiently high it provides some
natural damping for oscillations between the gyro float and the gyro
outer case.
While the gyroscope shown in Fig. 29a could not control platform
rotation about its spin axis it could control platform rotation about two
axes perpendicular to each other and to the spin axis, with the addition
of another gimbal. Used in this manner two two-axis gyros are suffi-
cient to stabilize a platform.
In the arrangement shown only one axis is controlled. If a torque
occurs about the output axis the case moves freely with respect to the
gyro wheel. This movement is sensed by the pick-off and a torque is
applied through the gimbal servo motor to the platform. There is no
torque being applied externally to the platform to counter, so under the
influence of this torque the gyro precesses (Fig. 30). In this fashion,

Plat
form
Oyro Rotor Precession / ~gl e
the j~
Servo S ~ e . ~ %
~_~
Servo Torque T o ~ To~
Angle
Ro~,or PrecessJ.on

+ x ,~ I ",,,o~
ezo 6 G ~sJ x, ~-' F- ~°---

I
'I I
L J

FIG. 30. Followingof azimuth command angle by level control gyro.

when the platform moves in azimuth under control of the azimuth gyro,
the level gyro automatically follows this rotation, while still controlling
the platform in level.
There are a number of sources of gyro drift present in the gyroscope
shown in Fig. 29a. Any action which causes a torque to be developed
between the case and the gyro wheel can result in gyro and platform
drift. For example, the lead-in wires to the gyro motor m a y exert
small but significant torques causing appreciable drift rates. If the
rotor is not perfectly symmetrical with respect to the center of buoyancy
then any acceleration acting on this mass unbalance causes a torque to
act on the rotor with resultant gyro drift. Convection currents in the
flotation fluid also cause drift-producing torques between the outer case
and the float.
388 C . F . O'DoNNELL [J. F. I.

In addition to mass unbalance there is the problem of anisoelastic

deformation of the spin bearing under accelerations causing the center
of mass of the gyro wheel to become displaced from the direct line of
action of the acceleration and the center of buoyancy, giving another
source of disturbing gyro torques and consequent gyro drift.
These are b u t a few of the m a n y possible sources of extraneous
torques. F r o m this brief description it can be seen t h a t the design of
a gyroscope to meet high performance specifications is a very difficult
engineering task.

b. Accelerometers
Accelerometers in general rely on the fact t h a t a force is necessary
to change the linear m o m e n t u m of any mass. T h e mass is so arranged
t h a t any force acting along a particular axis, its sensitive or i n p u t axis,
is measurable. Knowing t h e force and the mass, acceleration can be
calculated from Force = mass X acceleration. A possible accelerom-
eter configuration is shown in Fig. 31. Here a proof mass is suspended
from a pivot point. With no acceleration occurring along its i n p u t axis
the pick-off is held in a neutral position. W h e n acceleration occurs, the
proof mass tends to swing to the right or left, depending on the sign
of the acceleration, and an o u t p u t signal is generated, amplified, and
fed back through the torquer coils to maintain the proof mass in its
central position. T h e m a g n i t u d e of the torquing current required is a
direct measure of the acceleration taking place. A single integration
of this current gives change in velocity, with the second integration
giving change in position. There are a n u m b e r of accelerometers in
existence, m a n y of which vary only in the m e t h o d of obtaining the force
to oppose m o t i o n of the proof mass under acceleration. These instru-
m e n t s are true accelerometers with their natural o u t p u t s being directly
proportional to acceleration.
A different class of accelerometers can also be obtained in which the
natural o u t p u t is directly proportional to velocity, the first .integral
of acceleration. As has been stated in the preceding section, mass un-
balance makes a gyroscope sensitive to accelerations. A proof mass
can be placed on the inner gimbal in such a fashion t h a t only accelera-
tion along one axis causes the gyro rotor to precess, with torques due to
accelerations along other axes being automatically counteracted by the
system. Further, it has been stated t h a t a torque applied to a gyroscope
determines the precession rate of the rotor. Therefore, the total
precession angle is equal to the integral of the applied torque, which is
in t u r n directly proportional to the acceleration occurring along the
sensitive axis. In this case then, a measure of the precession angle of
this gyroscope gives the accumulated velocity of the system.
Consider a different i n s t r u m e n t in which the m o t i o n of the proof
Nov., 1958.] INERTIAL NAVIGATION 389
mass had been opposed by a viscous drag. T h e displacement of the
proof mass from its initial position is a measure of system velocity in
the direction of its sensitive axis. It is also possible to produce a
doubly integrating accelerometer with a natural o u t p u t equal to the
total position change of the system.
In general, the problems associated with accelerometers are easily
determined. The scale factor relating usable o u t p u t to acceleration
m u s t be linear over very wide ranges of acceleration, and constant in
time. T h e physical construction of the instrument m u s t not introduce
false accelerations into the system.
As an example consider the accelerometer shown in Fig. 31. Torquer
linearity m u s t be maintained for an electromagnetic torquer over a
~.~--- PivoL PofnL

(~nange in Change in
Positron Velocity

FIG. 31. A null-type acce|erometer.

wide range of accelerations; or calibration of the input to the computer

is required, providing an additional complication in operational use of
the system. Also there is the matter of maintaining stability of this
torquer scale factor over wide ranges of operating environmental
conditions. In this instrument there would be problems such as
friction in the pivot bearings, biases due to physical placement of the
pick-offs, and shift in this placement under vibration and acceleration,
and all of the ills of physical instruments. In addition, with this type
of accelerometer, the problems of drift and bias settings for the neces-
sary integrators used to convert its output to velocity and position
changes have to be solved.
2. Computer Characteristics Important to Inertial Guidance
Before examining the characteristics of computers used in inertial
navigation systems it is advisable to examine the equations to be
mechanized. Their contents will determine the desirable c o m p u t e r
characteristics.
390 C.F. O'DONNELL [J. F. I.

In general the motion of a vehicle can be visualized easily by follow-

ing the tip of the radius vector ~ drawn from the origin of a coordinate
system, (in our case usually the center of the Earth) and rotating at
an angular rate ~ (Fig. 32). If the length of 1~ is constant, vehicle

/ L it

Fie. 32. Vector diagram # = 1~ + & X ~.

velocity equals the product of angular rate ~ times the length of/~ times
the sine of the angle between them. In vector notation ~ = ~ X /~.
If ~ = 0 and 1~ is changing in length then V = 1~. In general then
= ]~ + ~ X 1~. A simple extension gives the acceleration ~i =
+ ~ X ~. Substituting for 17 gives the resultant expression for

~ = .~ + 2,s x ~ + # x .~ + ,s x ,~ x _~. (I)

~1. AngleBetween Local
V e r t i c a l and Radius Vector

fP% 2

////--~ Equator, 0 d~egLatitude

Meridians
of
Longitude, A
~ Parallels Of Latitude, 8

Fie. 33. The Earth as an ellipsoid of revolution.

NOV., I958.] INERTIAL NAVIGATION 39I

We can expect t h a t the Earth rate must be considered when navigat-

ing on the Earth. Indeed it appears in Eq. 1 as part of 3, the other
part being plane's rate = ~ . Are we justified in stopping with the
Earth's spin? W h y not consider the rotation of the Earth about the
sun? A moment's thought reminds us that the Earth moves around the
sun at satellite velocity. This means that centripetal acceleration and
mass attraction are in balance and so can be ignored.
One final factor which must be considered if accurate inertial naviga-
tion is attempted, is the Earth's ellipticity. The Earth is usually
represented by an ellipsoid of revolution, allowing for the fact t h a t the
polar diameter is some 23 nautical miles shorter than the equatorial
diameter. The Earth on which we navigate, and the coordinate system
we normally use are illustrated in Fig. 33. For an inertial navigator
with its accelerometer platform held locally level, and its x axis pointing
east-west at all times, the acceleration sensed along this axis is
A~ = A cos OR~ - A sin OR~ -- 20(~ + A) cos Oad sin 20
+ 2A(f~ + 3-) cos 0 + 40(~2 + A) sin Oad cos 20
- 20(f~ + A) sin OR, -- 20(f~ + A) cos OR~
where
R~ ~ -adsin20

R~ ~ ( 1 - d s i n 2 0 + - h a ) .

One thing is quite plain. The problem of extracting the ~ term from
this expression for A, is not trivial. Another apparent fact is that a large
number of derivatives must be taken, and that sines and cosines must
be generated in quantities in solving the navigation problem. These
factors should be borne in mind in determining the suitability of a
particular computer for real-time inertial navigation.
No attempt will be made here to go into the details of computer
mechanization. Only those characteristics of analog and digital
computers which have a bearing on their use in inertial guidance
systems are presented for a basis of comparison.

a. A n a l o g Computers
1. Most instruments naturally produce analog outputs, therefore
the tie-in problems between analog computers and inertial guidance
instruments are usually simple. For example, the outputs of the pick-
offs and the inputs to the torquers in the accelerometer and gyro shown
in this paper are analog in form.
2. M a n y trigonometric functions are mechanized in the solution of
inertial guidance problems. In an analog computer, the use of sine-
392 C.F. O'DONNELL [J. F. I.

cosine potentiometers to produce these functions offers a simple solution

to their formation. In addition resolvers can be used readily to resolve
functions through angles corresponding to physical angles in the system.
3. It is difficult, although not impossible, to obtain an accuracy
greater than 1 part in 1000 from an analog computer. Indeed, under
m a n y conditions it is difficult to obtain even this accuracy.
4. For precision in analog computation you pay with precision in
the physical components used for computing. This poses a difficult
fabrication problem.
5. Analog computer components can be made with much lower
moments of inertia than the physical system being controlled. There-
fore, real-time computation for control of physical systems is almost
always possible using an analog computer.

b. Digital Computers
1. A digital computer usually requires analog-to-digital converters
when used with inertial guidance instruments.
2. Very high computational accuracy can be obtained readily
without requiring corresponding accuracy in the fabrication of com-
ponent parts.
3. It may be difficult to keep up with a real-time control problem.
4. Trigonometric functions are generated by calculating terms of a
series or some similar device unless digital differential analyzer tech-
niques are employed; therefore, the calculation of these functions,
highly prevalent in the solution of inertial guidance problems, can be
quite time consuming.
5. Incremental or whole number inputs may be acceptable although
this depends on the particular computer.
6. For control, the over-all system becomes a sampled-data system
with the digital computer in the control loop; therefore, stability de-
termination becomes a more complex problem than for an all-analog
system.
7. A basic arithmetic unit must be supplied without regard to the
total amount of computation to be done. Therefore, for very simple
computing problems a digital computer may be at a disadvantage by
comparison with an analog computer in weight, volume, and simplicity.
F. ANALYSISOF ERROR CONTRIBUTIONS

i. B ~ s
Using the concept of an inertial guidance channel as a position
servo, the effect of a bias error in the accelerometer of K ft/sec 2 is easily
determined. The error equations are
Nov., 1958.] INERTIAL NAVIGATION 393

K/~
e , (s) = s (s~ + coo~)

a~O~(t) = ~ (1 -- cos coot).

coo

1 f t / s e c 2" a~O~(t) = 6,600 (1 -- cos coot) ft. T h e resultant

F o r K = 100
error is shown in Fig. 34 with both the position error and the platform

o I I I 1 -- I

Pesltion K/
L/-x /--'X_/--
~rror ' ~c , , , , ]
a t-:,y

Viacfccrn, A// 2 ~ ~

Time,[~2nuLes
FIG. 34. Effect of accelerometer bias shift on position error.

angle error sketched in. It can be seen t h a t the position error is of the
form (1 - cos coot), and the resultant platform error angle follows the
same form.

2. Gyro Drift
Using a c o n s t a n t gyro drift rate, ,, and ignoring the angular rate
coupling which t a k e s place over a long term in the actual guidance
platform, the relation b e t w e e n drift rate and position error is determined
as follows :
,tg
~ + c o 0 ~ O , = _ a_
394 C . F . O'DONNELL [J. F. I.

or

- - eg/a .
= + o o2) '
SO

a~Ov(t) =- ae(t -- --o~ols i n ~ ° t ) "

After 6 hours a gyro drift rate of 0.1 ° / h r produces a 36-nautical mile
position error. T h e resultant errors for both position and platform
angle are shown in Fig. 35.

i
l .8, + s a + ~ *
•
~ n % +0 S e+' ~~ 0t l l,

e t , G~ro Drift ~gle |

kr~,le = • t

Poefl.t.lon ~ 0 y . . . , .

- / Time, M i n u t e s

FIG. 35. Effectof gyro drift on platformand positionerror.

There is an interesting difference between the results for this con-

stant gyro drift rate and for a bias error in the accelerometer. Here
the platform angle oscillates sinusoidally about a 0 value, although the
position error is building up with time. T h e reason is t h a t gyro drift
is an active driving source in the platform system. On the average
this drift is cancelled by the incorrect torquing of the platform due
to the position error. This leaves only the 84-min. period c o m p o n e n t
of error affecting the platform angle. If the platform is being used as a
vertical reference, then it is possible to obtain good accuracy in the
indicated vertical in spite of large errors in c o m p u t e d position. T h e
0.1 °/hr. gyro drift rate for example causes peak platform oscillations of
only 1.4 min. of arc.
 Nov., 1958.] I N E R T I A L NAVIGATION 395

3. Impulse
T h e effect of an impulse acceleration with an area k ft. per second
applied to the s y s t e m can be calculated as follows: ~#~ -5 o~02~0~ = k~(t__J).,
6~

where ~(t) is Dirac delta function; or ~O~(s) = k/a • so a~O~(t)

$2 -5 6002 y
k
= - - sin c00t. For a scale factor error of 0.1 per cent in an accelerometer
000
used to measure an a b r u p t change in velocity of 1000 ft./sec., the re-
sultant position error a~Oy(t) = 800 sin o~0tft. with t = 0 at the time of
velocity change. Results are shown graphically in Fig. 36. It should

, •. . . . . . . . . . . . . . . . . . . . . I

Actual
Velocity ~ 0+ I I [

in
Acceleration f ~" i I
Measurc~nent 0

Position ~'~ [##" ~ ./ ~.

Error ~--T 0

- Time, Minutes
FIG. 36. Position error due to pseudo-impulse acceleration error.

be observed t h a t due to the long period of the system, disturbances

occurring over a 3- to 5-min. time interval can actually be t r e a t e d as
impulses occurring a t t h e center of this interval with v e r y little error.

4. Noise
A series of disturbing impulses of r a n d o m m a g n i t u d e and sign ap-
plied to the guidance s y s t e m generate a series of sine w a v e s with random
phase differences. As has been stated previously disturbances of up to
5 m i n u t e s d u r a t i o n affect the s y s t e m almost as impulses• T h e p r o b l e m
of determining s y s t e m position error for such an i n p u t then resembles
396 C.F. O'DoNNELL [J. F. I.

t h e " r a n d o m w a l k " problem in probability theory. Here n steps are

t a k e n along a line from an origin, the rms value of each step being K ft.,
with equal probability t h a t each step is forward or backwards. T h e
rms value of displacement from the origin after n steps, for n sufficiently
large, is a p p r o x i m a t e l y K,J-n ft.
Similarly for t h e inertial guidance s y s t e m if the rms value of errors
due to individual disturbing "impulses" is K1 feet, t h e n after n of these
impulses, for n sufficiently large, t h e rms value of position error should
be K14-n ft. On the assumption t h a t these noise impulses are u n i f o r m l y
distributed in time, the r e s u l t a n t position error increases as K~@.
This result is in a g r e e m e n t with t h e form of error build-up found when
" w h i t e " noise, t h a t is noise with a zero correlation time, is applied as a
driving function to an u n d a m p e d oscillatory system.
T h e preceding p a r a g r a p h s are presented as a plausibility a r g u m e n t
in favor of t h e long-time error build-up of an inertial guidance s y s t e m
actually possessing "square root of t i m e " characteristic. In practice
gyro drift, for example, is never constant, for the e n v i r o n m e n t affecting
it is c o n s t a n t l y changing. T h e same can be said of bias error in ac-
celerometers and of all t h e o t h e r errors going to m a k e up t h e total s y s t e m
error. While these inputs certainly are not "white noise" types, still
t h e correlation times are often sufficiently short so t h a t over a period of
several hours error would increase as the square root of time.

5. System Error from Component Error

Given the errors due to t h e various components, t h e resulting system
error is usually calculated b y combining these errors in an rms fashion.
Care has to be t a k e n in m a k i n g these combinations to consider the dis-
tribution of each error. For most sources a normal distribution is
reasonable. In some cases where an upper bound is placed on t h e error
by m a n u f a c t u r i n g inspection a t r u n c a t e d normal or r e c t a n g u l a r distri-
bution m a y be a valid one to assume. In most cases t h e over-all s y s t e m
error can be safely assumed to have a normal distribution, for w h e n
three or more uncorrelated error sources of comparable m a g n i t u d e are
combined the resultant error distribution is a close a p p r o x i m a t i o n to
the normal.
T w o p a r a m e t e r s often q u o t e d as measures of s y s t e m p e r f o r m a n c e
can be derived from the rms value of system error. One is the circular
probable error ; t h a t is, t h e radius of a circle inside which 50 per c e n t of
t h e s y s t e m errors would fall. T h e relation between the rms errors in
each channel of the n a v i g a t o r and this circular probable error is given
approximately by
C E P -- 0.589E(Range Error)tins -t- (Lateral E r r o r ) m ~ ,
a slight overestimate. T h e o t h e r p a r a m e t e r often quoted is a "so-
called" m a x i m u m error ; an error which is greater t h a n 95 per cent of t h e
Nov., I958.1 INERTIAL NAVIGATION 397

position errors occurring. F o r a normal distribution of errors, this is

twice t h e rms error previously calculated.
F o r example, assume the following distribution of c o m p o n e n t errors
for an inertial guidance s y s t e m used in a M a c h 1 vehicle on a 1-hr.
great-circle flight.
TABLE I.
(rms values used throughout unless otherwise noted)
Range Error, Lateral Error,
Component Error Source nautical miles nautical miles
Range Accelerometer
Scale Factor 1% 1.3
Bias Shift 0.015 ft/sec 2 1.7

Lateral Accelerometer
Scale Factor 1% m m m

Bias Shift 0.015 ft/sec 2 1.7

Level Gyros
0.1°/hr. for each axis 6 6

rms error 6.3 6.2

CEP = 0.59 (6.3 -{- 6.2) n. miles

= 7.5 n. miles

T h e " m a x i m u m " errors are 12.6 n. miles in range and 12.9 n. miles
in the lateral channel. In this example the effect of the assumed gyro
drift rate almost completely s w a m p s o u t the errors from o t h e r sources,
although these b y themselves are not negligible.
Calculation of errors for inertial guidance used in ballistic missiles
follows a s o m e w h a t similar course. H o w e v e r the guidance s y s t e m is in
operation for only a brief period so the long-time error p r o p a g a t i o n in
the system is n o t of interest. In addition the acceleration e n v i r o n m e n t
is v e r y severe. H o w e v e r , given instrument characteristics and the
flight profile, guidance errors at power cut-off can be determined. T h e
m a n n e r in which these errors affect the i m p a c t point can also be calcu-
lated. Given this information navigation errors can be c o m p u t e d in a
straight-forward fashion.
F o r example, consider a ballistic mail delivery service from Los
Angeles to H o n o l u l u - - a n airline distance of a b o u t 2500 miles.
T h e n for an azimuth misalignment error, ¢,°, of 2 milliradians rms,
the resultant lateral error is given a p p r o x i m a t e l y b y Cz X Range, in
2
this case ~ X 2500 miles, or 5 miles.
If the average horizontal velocity is 5000 miles/hr., then flight time
398 C.F. O'DONNELL [J. F. I.

is ½ hr. An error of 0.2 per cent rms in horizontal velocity determina-

2
tion then leads to a range error of 5000 X 1 - ~ X ½ = 5 miles in range.
If these are the only significant errors, then

system C E P becomes = 0.59(5 + 5) miles

= 6 miles

With this system time saved from post-office Los Angeles to post-office
Honolulu over more routine forms of delivery is difficult to calculate.

G. A I D E D I N E R T I A L S Y S T E M S

It is possible to combine information from other navigation devices

with an inertial guidance system to produce an output better in some
ways than the information available from either of the original sources.
Methods of combining this information with an inertial guidance system
follow easily if each channel of the guidance system is considered as a
positional servo. External information for example might be available
in the form of position information or velocity information. These
possibilities are considered separately.

E~r [

,J ,/ ,/
(:~eckpoint Jl Checkpoint #2
a) Use of Isolated Position Checkpoh',t.s

o,o
_ _ ° a "_Y_,. . . . . . . . . . (From
Auxiliary
Source I
• t
b} Use of Continuous Poslticn Cheek Information

FIG. 37. Use of auxiliary position information for inertial navigators.

Nov., 1958.] INERTIAL NAVIGATION 399

1. P o s i t i o n I n f o r m a t i o n
a. Discrete
If position information is available only at intervals then it can be
employed to reduce the position error of the inertial guidance system
to conform with these discrete position checks. The result of using
such information is shown in Fig. 37a. It is observed t h a t the position
error build-up after a checkpoint is more rapid t h a n the initial build-up.
This is due to the velocity error which has built up during the initial
period of operation, T, and has not been reduced by the use of a single
position check.

b. Continuous P o s i t i o n I n f o r m a t i o n
If a system such as Loran is available to give a continuous check on
position for some parts of the vehicle's journey, it is possible to combine
the o u t p u t of the navigator with this position information in a m a n n e r
shown in Fig. 37b. Using the revised feedback, shown in Fig. 37b, the
natural period of the system is decreased and the magnitude of position
errors resulting from a n u m b e r of driving sources has been reduced.
1
The system transfer function has become s2 + K / a + O~o~' so for K = 9g
an accelerometer bias change of 0.015 ft/sec ~ now gives an error of 1000
(1 -- cos ~ 0 t ) ft.
The errors due to gyro drift are also reduced.

-- (~0~ + et)g -- K~Ov = a~O~ and set K = kg

or
a~O, + 60~(k + 1)g = - ~tg
or
~0~ + (k + 1 ) ~ 0 ~ 0 , = - eta00~
or
=
( k- +
,
l) [t 1 sin4-k-+1000t]

1
)< (Position Error in Unaided Inertial System)
k+l

From the results of the preceding analysis there m a y be some ques-

tion as to why an inertial navigator would be used at all with this posi-
tion information available. The answer will depend on the charac-
teristics of the position signal being compared with the inertial navigator
o u t p u t . In the cases outlined so far the aiding position information was
assumed to be perfect. In practice this rarely happens. For example
the position information m a y be accurate on the average, b u t vary
400 C . F . O'DONNELL [J. F. I.

suddenly from m o m e n t to moment. Differentiating this information

provides only a poor source of instantaneous velocity information.
By first using the smoothing provided by the inertial system a signal
which can be differentiated to give accurate instantaneous velocity
can be generated. In addition the inertial system can extrapolate
beyond areas of Loran or similar range-limited coverage, or fill in gaps
in this coverage.

2. Velocity Damping
If an external source of velocity information is available, such as
airspeed or Doppler radar, then a comparison can be made between this
source and the velocity supplied by the inertial guidance system. The
resultant error signal, multiplied by some suitable transfer function,
can be fed back to the input of each inertial navigator channel, as
shown in Fig. 38. Here in its simplest form G(s) might be equal to

S2X (S)

~(S)+~_~
I
I

L
I
Acceleration

From Auxiliary Velocity

. . . . -o,-7sT7 ~%~7- ~° <@

FIG. 38. Velocity damping of inertial guidance.

-- k. In t h a t case a simple second order system is formed with the

1
transfer function s ~ + ks + COo~" Disturbances are damped out to their
average value by the factor e-~t/2.
One disadvantage of this system is t h a t a constant reference velocity
error results in a constant position error after the oscillatory term has
been damped out. This can be eliminated by the use of a third order
kls
system where G(s) becomes The system transfer function
s+k~"
s+k~
becomes s3 + (kl + k2)s ~ + O~o2S+ k,o~o2" In this third order system
position error due to a constant reference velocity error is damped to 0.
However, sudden shifts in this error cause transient position errors
which take time to damp out.
Nov., I958.] INERTIAL NAVIGATION 4OI

It is tempting to consider using an external measurement of ground

speed, perhaps Doppler radar capable of good average velocity accuracy,
to reduce the dependence of the combined systems on gyro drift rate.
However, azimuth gyro drift cannot be controlled in this fashion, and
for extended operating times becomes a dominant error source.
The actual form of G (s) (Fig. 38) used will depend on the character-
istics of the velocity reference signal available for damping. In each
case it is possible to carry out a study showing the optimum form of
G (s) for minimum position error considering the mission to be flown.
While each aided inertial system has some advantages, careful
operational studies will usually uncover defects as well. In practice
there is no generally acceptable substitute for an excellent basic inertial
navigation system.
SELECTED BIBLIOGRAPHY

(1) R. H. CANNON, "Kinematic Drift of Single-Axis Gyroscopes," Am. Soc. Mech. Engrs.,
New York, December 1957, Paper No. 57-A-72.
(2) R. H. CANNONAND D. P. CHANDLER,"Stable Platforms for High-Performance Aircraft,"
Aeronautical Eng. Rev., Vol. 16, p. 42 (1957).
(3) C. S. DRAPER, W. WRIGLEYAND L. R. GROHE, "The Floating Integrating Gyro and Its
Application to Geometrical Stabilization Problems on Moving Bases," Aeronautical
Eng. Rev., Vol. 15, p. 46 (1956).
(4) P. J. KLASS, "Inertial Guidance," Aviation Week Special Reprint, New York, McGraw-
Hill Publishing Company, Inc. (1956).
(5) C. F. O'DONNELL, "How Inertial Navigation Works," Electronic Equipment, Vol. 5,
p. 42 (1957).
(6) W. T. RUSSELL, "Inertial Guidance for Rocket-Propelled Missiles," presented at the
American Rocket Society Semi-Annual Meeting, San Francisco, 10-13 June 1957.
(7) M. SCHULER, "Die StGring yon Pendel und Kreiselapparaten durch die Beschleunigung
der Fahrzeuges," Physikalische Zeitschrift, Vol. 24, p. 344 (1923).
(8) J. M. SLATER, "Choice of Coordinate Systems in Inertial Navigation," Navigation,
Vol. 5, p. 58 (1956).
(9) J. M. SEATER,"Gyroscopes for Inertial Navigators," The American Society of Mechanical
Engineers Paper No. 57-SA-39, n.d. Presented at ASME Meeting 9-13 June 1957.
(10) J. M. SEATER, "Measurement and Integration of Acceleration in Inertial Navigation,"
The American Society of Mechanical Engineers Paper No. 56-A-160, n.d. Presented
at ASME Meeting 25-30 November 1956.
(11) J. M. SEATERAND D. B. DUNCAN,"Inertial Navigation," Aeronautical Eng. Rev., Vol. 15,
p. 49 (1956).
(12) HENRYP. STEIER, "M-H Tackles Inertial Component Challenge," Am. Aviation, Vol. 20,
p. 53 (1957).
(13) W. R. WEEMS, "An Introduction to the Study of Gyroscopic Instruments," MIT Dept.
of Aero. Eng. (January 1948).
(14) W. WRIGLEY, R. B. WOODBURYAND J. HOVORKA, "Application of Inertial Guidance
Principles," Automatic Control, Vol. 7, p. 22 (1957).
(15) W. WRIGLEY, R. B. WOODBURYAND J. HOVORKA, "Inertial Guidance," IAS Preprint
No. 698. Presented at 25th annual meeting, January 28-31, 1957 (MIT).
(16) L. E. GOODMANAND A. R. ROBINSON, "Thermal Drift of Floated Gyroscopes," Jr. Appl.
Mech., Trans. ASME, Vol. 24, p. 506 (1957).
(17) W. T. RUSSELL, "Inertial Guidance for Rocket-Propelled Missiles," Paper presented at
ARS Semi-Annual Meeting, San Francisco, Calif., 10-13 June 1957.
402 C.F. O'DoNNELL [J. F. I.

(18) B. T. PLYMALEAND R. GOODSTEIN,"Nutation of a Free Gyro Subjected to an Impulse,"

J. Appl. Mech., Trans. ASME, Vol. 77, pp. 365-366 (1955).
(19) R. E. ROBERSON, "Principles of Inertial Control of Satellite Attitude," Paper to be
presented at the 9th International Congress of Astronautics, Amsterdam, 1958.
(20) R. E, ROBERSON, "Impact Points of Ballistic Rockets," Jet Propulsion, Vol. 27, p. 1256
(1957).
(21) R.E. RosERSON, "Oblateness Correction to Impact Points of Ballistic Rockets," to appear,
JOUR. FRANKLININST., Vol. 266, December, 1958.
(22) F. STEVENS AND F. W. LYNCH, "Aided Inertial Guidance Systems," Western Aviation,
Vol. 37, p. 6 (1957).
(23) J. M. SLATER, "Gyroscopes for Inertial Navigators," Mech. Eng., Vol. 79, p. 832 (1957).
(24) D. B. DUNCAN,"Analysis of an Inertial Guidance System," Jet Propulsion,, Vol. 28, p. 111
(1958).