AGARDAG6331

- ' -111
AQARD-AQ-331
LT
n
9
3a
AGARD
ADVISORY GROUP FOR AEROSPACE RESEARCH & DEVELOPMENT
7 RUE ANCELLE, 92200 NEUILLY-SUR-SEINE,FRANCE
AGARDograph 331
Aerospace Navigation Systems

(Les syst2mes de navigation ac5rospatiaux)
'.*
"
e
T FOR DESTRUCTION
This AGARDograph has been sponsored by the Guidance and Conrr~lPanel of AGARD.
NORTH ATLANTIC TREATY ORWIZATION
Published June 1995
AGARD-AG-331
ADVISORYGROUPFORAEROSPACERESEARCH &DEVELOPMENT
7 RUE ANCELLE, 92200 NEUILLY-SUR-SEINE, FRANCE
AGARDograph 331

(Les syst6mes de navigation aerospatiaux)
by
J. Niemela, M. Kumar, B. Stieler, H. Buell, D. Boozer, R. Langley,
D. Liang, S. Levine, D. Moore, S . Lewantowicz, R. Paschall, N. Carlson,
R. Greenspan, D. DiFilippo, S . Ausman, C. Bedoya, J. Miller and C. Hunt
Edited by
John Niemela, Programme Director
This AGARDograph has been sponsored by the Guidance and Control Panel of AGARD.
North Atlantic Treaty Organization

Organisation du Traite de I'Atlantique Nord
The Mission of AGARD
According to its Charter, the mission of AGARD is to bring together the leading personalities of the NATO nations in the
fields of science and technology relating to aerospace for the following purposes:
- Recommending
effective ways for the member nations to use their research and development capabilities for the
common benefit of the NATO community;
- Providing scientific and technical advice and assistance to the Military Committee in the field of aerospace research
and development (with particular regard to its military application);
- Continuously stimulating advances in the aerospace sciences relevant to strengthening the common defence posture;
- Improving the co-operation among member nations in aerospace research and development;
- Exchange of
scientific and technical information;
- Providing assistance to member nations for the purpose of increasing their scientific and technical potential;
- Rendering
scientific and technical assistance, as requested, to other NATO bodies and to member nations in
connection with research and development problems in the aerospace field.
The highest authority within AGARD is the National Delegates Board consisting of officially appointed senior
representatives from each member nation. The mission of AGARD is carried out through the Panels which are composed of
experts appointed by the National Delegates, the Consultant and Exchange Programme and the Aerospace Applications
Studies Programme. The results of AGARD work are reported to the member nations and the NATO Authorities through the
AGARD series of publications of which this is one.
Participation in AGARD activities is by invitation only and is normally limited to citizens of the NATO nations.
The content of this publication has been reproduced

directly from material supplied by AGARD or the authors.
Published June
1995
Copyright 0 AGARD 1995

All Rights Reserved
ISBN
92-836-1018-0
Printed by Canada Communication Group

45 Sacrt-Caur Blvd., Hull (Qutbec), Canada K I A OS7
ii
Preface
The need for an up to date, comprehensive treatise on aerospace navigation systems has been recognized by the Guidance,
Control and Navigation community. Such a document will serve as a lasting reference as did material prepared in the late
60s and early 70s. These earlier documents, though remarkably prescient in their technical forecasts, have been overtaken
by very rapidly advancing technology.
This new document should be of interest to those who are involved in the integration of navigation equipment aboard an
aerospace vehicle and who may not be knowledgeable about, but would like to have an experts perspective on the
capabilities and limitations of the various navigation sensors, integration techniques and concepts. In addition, it is expected
that navigation specialists will have an interest in broadening their understanding of aerospace navigation. Great care has
been given to the selection of references to previously published material.
The editor would like to acknowledge Dr. George Schmidt of the Charles Stark Draper Laboratory, who conceived of this
AGARDograph and Ir Pieter Ph van den Broek of Delft University of Technology, who provided many constructive
suggestions in the course of its preparation.
JOHN NIEMELA
U.S. Army Member
AGARD Guidance and Control Panel
Prkface
La communautC Guidage, Pilotage et Navigation reconnait la nCcessitC dun trait6 modeme et complet sur les systhmes de
navigation akrospatiaux. Un tel document servirait de rCfCrence permanente, & linstar des textes Ctablis pendant les annCes
60 et 70. Ces documents, dune prescience remarquable dans leurs prkvisions techniques, ont toutefois CtC vite dCpassCs par
des technologies dont lavancke sest rCvk1Ce trks rapide.
Ce nouveau document devrait intCresser tous ceux qui sont impliquCs dans IintCgration des Cquipements de navigation dans
les vChicules drospatiaux et qui ne connaitraient peut-Etre pas, mais qui aimeraient connaitre, le point de vue dun expert sur
les capacitCs et les limitations des diffkrents senseurs de navigation, des techniques et des concepts dintegration. En outre, il
est privisible que les spCcialistes en navigation voudront Clargir le champ de leurs connaissances dans le domaine de la
navigation akrospatiale. Un soin particulier a CtC apportC & la stlection des rCfCrences qui sont faites aux textes dCj&publiCs.
Le rCdacteur tient i remercier le D George Schmidt du Charles Stark Draper Laboratory, qui a conqu cette AGARDographie,
ainsi que lIr Pieter Ph van den Broek de 1UniversitC de technologie de Delft pour les nombreux conseils positifs quil a bien
voulu foumir lors de 1Claboration du manuscrit.
JOHN NIEMELA
U.S. Army Member
AGARD Guidance and Control Panel
iii
Guidance and Control Panel

Deputy Chairman: Mr. J.K. Ramage
Chairman: Mr. S. Leek
Chief,
Flight Control Systems Branch
WLFIGS, Bldg 146
2210 Eighth St, Ste 11
Wright-Patterson AFB, OH 45433-7521
United States
PB 121
British Aerospace Defence Ltd
PO Box 19
Six Hills Way, Stevenage
Herts SGI 2DA
United Kingdom
TECHNICAL PROGRAMME
Programme Director and Editor: Dr. John Niemela
Department of the Army
Electronics Systems Division
A'ITN: AMSEL-RDC2-TS
Fort Monmouth, NJ 07703-5603
PANEL EXECUTIVE
Lieutenant-Colonel M. Mouhamad, FAF
Mail from USA and Canada:
AGARD-NATO
Attn: GCP Executive
PSC I 16 AE 09777
Mail from Europe:

AGARD-OTAN
Attn: GCP Executive
7, rue Ancelle
F-92200 Neuilly-sur-Seine
France
Tel: 33(1) 47 38 57 80
Telex: 610176 (France)
Telefax: 33 (1) 47 38 67 20/57 99
ACKNOWLEDGEMENTDtEMERCIEMENTS
The Programme Director and the Guidance and Control Panel wish to express their appreciation to all authors who
contributed to this AGARDograph and made its publication possible.
Le directeur du programme et la Commission guidage et pilotage tiennent 2 remercier tous les auteurs qui contribukrent ?
lai
rkalisation et la publication de cette AGARDographie.
iv
Contents
Page
PrefaceJPrCface
iii
Guidance and Control Panel
iv
SECTION I
1
Introduction
by Dr. J. NIEMELA
SECTION II
Coordinate Frames
by M. KUMAR
I
SECTION III- NAVIGATION TECHNOLOGY

Introduction
by Dr. J. NIEMELA
43
Inertial Navigation
by B. STIELER
45
Doppler Radar Navigation

by H. BUELL
132
Terrain Referenced Navigation

by D.D. BOOZER
152
Satellite Navigation
by R.B. LANGLEY
158
An Overview of OMEGA Radio Navigation System

by Dr. D.F. LIANG
177
AGARDograph on Advanced Astroinertial Navigation Systems

by S. LEVINE
187
Magnetic Heading References

by D.L. MOORE
200
SECTION IV - SYSTEMS ANALYSIS, DESIGN AND SYNTHESIS METHODOLOGIES
I
~
Introduction
by Dr. D.F. LIANG
207
An Overview of a Generic Multi-Sensor Integrated Navigation System Design

by Dr. D.F. LIANG
210
Deep Integration of GPS, INS, SAR and Other Sensor Information

by Col. Z.H. LEWANTOWICZ and Maj. R.N. PASCHALL
231
Federated Filter for Fault-Tolerant Integrated Navigation

by N.A. CARLSON
265
GPSDnertial Integration Overview

by R.L. GREENSPAN
281
Design Considerations for a Suboptimal Kalman Filter

by D.J. DIFILIPPO
295
Vertical Channel Design Considerations

by J.S. AUSMAN
312
SECTION V - REPRESENTATIVEVEHICLE IMPLEMENTATIONS

Introduction
by Dr. J. NIEMELA
329
Aerospace Navigation Systems Requirements for Fixed Wing Aircraft

by C.A. BEDOYA
330
Rotary Wing Aircraft

by Dr. J. NIEMELA and Dr. D.F. LIANG
381
Spacecraft Navigation Requirements

by J.L. MILLER
390
SECTION VI
Test Methodology
by C.L. HUNT
407
vi
SECTION I
INTRODUCTION
DR. JOHN NIEMELA
U.S. ARMY COMMUNICATIONS - ELECTRONICS COMMAND
RESEARCH AND DEVELOPMENT CENTER
FORT MONMOUTH, NJ 07703-5000
US A
1.
MOTIVATION AND OVERVIEW
The need for an up to date,

comprehensive treatise on
aerospace navigation systems
has been recognized by the
navigation community. Such a
document will serve as a
lasting reference as did
previous material prepared in
the late 6 0 ' s and early 70's.
[l] These earlier documents,
though remarkably prescient in
their technical forecasts,
have been overtaken by very
rapidly advancing technology.
It is anticipated that the
target reader of this document
will be an individual who has
responsibility for the
integration of navigation
equipment aboard an aerospace
vehicle. He may not have, but
is desirous of obtaining, an
expert's perspective on the
capabilities and limitations
of the various navigation
sensor and integration
techniques. In addition, it
is expected that navigation
specialists will have an
interest in broadening their
understanding of aerospace
navigation. Great care has
been given to the use of
references to previously
published material.
The AGARDograph is organized
into six sections with this
section providing the
motivation for establishing
the requirements to assure

that the development of an
aerospace navigation system
will meet its operational
requirement. The relationship
between system characteristics
and the mission scenario has
been often vaguely defined and
typically requires much
attention prior to
development. Included is a
discussion of the methods to
rigorously specify the
technical requirements to meet
the operational requirement.
The second section of this
document reviews navigation
coordinate frames with a
discussion of inertial,
terrestrial and geodetic
coordinate systems.
The third section of the
document describes the
navigation sensor technologies
that are employed in modern
aerospace navigation systems.
Emphasis is placed on the
physical principles of
operation rather than specific
or unique implementations.
The fourth section of this
document addresses system
analysis and synthesis
methods. Examination and
trade-off of the technology,
for each sensing, processing
and control and display
element, generates alternative
preliminary designs. This
section discusses the
identification and
partitioning of the navigation

system functions and
definition of the interface
between functional elements
and the rest of the aerospace
system.
A vital aspect in the
development of an integrated
navigation system is the
processing algorithm that
combines the navigation sensor
data. Also in this section is
a methodology for Kalman
filter development that
includes error models,
covariance and Monte Carlo
simulation techniques with
means to verify the error
models.
The fifth section of the
document describes
representative state-of-theart navigation system
implementations in fixed wing,
rotary wing aircraft and
spacecraft.
The sixth and concluding
section of the document
describes the various test
methods employed to verify the
performance of aerospace
navigation systems. This
includes instrumentation
techniques with emphasis on
standard statistical methods
used by NATO countries. This
discussion provides closure
with the first section on
requirements by relating the
means to verify that the
navigation performance meets
the technical and operational
requirement.
In general, modern aerospace
navigation systems are
comprised of more than one and
in many instances several
independent subsystems. This
considerably complicates the

process of translating the
operational requirement to a
technical requirement. The
importance of a concise
statement of the technical
requirements on which to base
the development of a system
design is of paramount
importance. For this reason,
the following discussion
outlines the critical steps in
development of the technical
requirement that is based on
an explicit or implicit
operational requirement. The
methodology discussed draws in
part on the experience gained
in formulation of the
EFA - [2].
2. OPERATIONAL AND TECHNICAL
REQUIREMENTS FOR AEROSPACE
NAVIGATION SYSTEMS
2.1
OPERATIONAL REOUIREMENTS
The need for a navigation

capability aboard aerospace
vehicles is dependent on many
qualitative and quantitive
factors. For the design to
fulfill the need requires that
the operational requirement,
which is often stated in terms
of a mission scenario and an
aerospace vehicle capability,
be faithfully translated into
a technical requirement. The
technical requirement can then
be stated in terms of
parameters on which a design
can be accomplished.
In virtually all modern
aerospace vehicles the
navigation system must provide
more functionality than the
classical service of providing
direction and distance to a
succession of waypoints. In
many instances the navigation
system must provide data to

other on-board systems so that
they may accomplish their
function to include flight
control, weapon direction,
envelope limiting, approach
and landing, to name a few.
Similarly, in modern aerospace
system designs, the navigation
system is often dependent on
information and processing
assets residing outside the
confines of the navigation
system. In these highly
integrated designs it is
difficult, if not impossible,
to circumscribe the navigation
system as part of the
aerospace system.
Consequently, the above
mentioned navigation
functionality must be provided
with appropriate levels of
performance, to include
accuracy and reliability.
The complexity of these modern
aerospace system designs makes
traceability from the
operational requirement to the
technical requirement to the
navigation system design of
paramount importance. For
this to be accomplished
requires that the operational
requirement be described in
sufficient detail,
particularly scenario and
aerospace vehicle
characteristics that impact
the technical requirement of a
navigation system. Below are
listed some of the principle
considerations and factors.
Detailed description of
nominal mission (altitude,
velocity, dynamics) by mission
segment.
Accuracy for each
mission segment
Reliability
Availability
O
Size
Weight
Power
ECM Environment
2.2
Technical Reauirement
Requirements for aerospace

platforms often give overall
descriptions of the functions
needed to carry out the basic
missions of the platform but
leave the requirements for the
navigation system rather
vague. In the past it has
been adequate to specify that
the platform shall carry, for
instance, an inertial
navigator with a given general
performance. However, the
introduction of satellite
navigation systems whose
performance may be
substantially improved by
integration with dead
reckoning systems has changed
the way in which navigation
system requirements are to be
defined.
Where the sensors to be used
in a navigation system are
likely to be affected by the
maneuvers of the vehicle,
their location, and activities
of hostile parties, are now
required to be specified in
much more detail than
previously. It is to assist
in the preparation of the
technical requirements that
this portion of the paper is
offered.
2.2.1 Characteristics of
Inteqrated Navisation Systems
In general, an integrated
system will consist of a
number of subsystems, linked
together by means of
interfaces and software that
are located in processors
which may be separate devices
or integral parts of one or
more of the sensors. Each of
the subsystems may be capable
of independent operation under
some conditions but will
benefit by making use of
information derived from other
subsystems. In the primary
mode of operation all
subsystems will be
contributing information to
the integrating computer and
the resulting total system
will have greater accuracy
than any individual subsystem.
An important aspect of the
integration is that many of
the deterministic errors of
the subsystems may be
calibrated during the
integration process by cross
comparison of the subsystem
outputs. This has the effect
that when one or more
subsystems become nonoperative, e.g:, due to
terrain screening, the other
systems may continue at a
higher level of accuracy than
would otherwise be possible.
The accuracy will degrade
slowly, depending on the
variability of the calibrated
error sources, rather than
suddenly as would occur with
an uncalibrated sensor. When
operation of the missing
sensor is restored, the system
will once again return to full
operation.
Thus an integrated navigation

system has the property of
robust operation with graceful
degradation in the event of
sensor non-operation and the
ability to recover full
capability should the nonoperating sensors become
operational.
Relation to the
ODerational Requirement
2.2.2
Generally, an operational
requirement will give general
descriptions of the scenarios
of operation and the desired
mission capabilities of an
aerospace platform. An
integrated navigation system
will have various modes of
operation which will react in
various ways to the particular
environment in which they
operate. It is therefore
necessary that the operational
requirement state prospective
missions in some detail, and
to define the requirements for
the navigation system for each
stage of the mission. The
technical requirement must
translate the several mission
capabilities and environments
into technical capabilities
and parameters on which a
system design can be
developed.
For instance, the initial part
of a sortie may be carried out
under benign conditions during
which time the integrated
system is able to calibrate
the sub-system error sources.
Knowledge of the duration of
this period is important in
determining the accuracy to
which the errors may be
calibrated. The sortie may
then continue with a low level
penetration in rugged terrain
during which intermittent

acquisition of satellites may
occur. The duration and
conditions of this phase and
the required navigation
accuracy need to be specified.
Following the penetration

phase, the attack phase may be
in less rugged terrain but
with enemy jamming in
operation. The level of
jamming needs to be specified
in order to determine the
optimum level of rate aiding
for the satellite system.
'
I
~
During each mission segment,

evasive or attack maneuvers
may be carried out. These
will have an effect on the
INS, particularly if it is a
strapped down system, and the
likely maneuver levels and
allowable overall system
accuracy should be specified.
2.2.3
Desiqn Constraints and
Deqrees of Freedom
The requirements for the

system accuracy should be
stated and the times and
conditions under which these
are to be met, defined.
However, such technical
requirements should not define
the details of the integrated
system mechanization. With
these systems, there is a
considerable trade-off of
complexity, accuracy,
integrity and cost which
should be the prerogative of
the system designers to
determine, within the
constraints laid down by the
technical requirement.
Implicit in the process is the
need for close liaison between
the drafters of the technical
requirement and system
designers.
2.2.4
Affordability
An important aspect of all

system developments is that of
affordability. Most new
requirements call for "stateof-the-artuuperformance which
is not readily met and could
be very costly. There is a
need to specify the allowable
cost for a system designed to
meet the requirements, and
those areas in which
performance (in the general
sense) may be traded for cost,
and any areas in which such
trade-offs are not permitted.
2.2.5
Inteqritv Aspects
The requirements on the

navigation subsystem which are
necessary to enable the total
aircraft system to meet the
mission success requirements
must be specified. The
specification should state any
constraints on the methods
which are to be used to detect
and handle the loss or
degradation of data from
individual sensors. This may
be necessary to meet flight
safety or operational
requirements.
The specification should
distinguish between the
methods being used to ensure
adequate integrity for the
flight safety requirements and
those needed to meet the
mission success requirements.
2.2.6 Factors Contained in a
Technical Requirement
The technical requirement for

an integrated navigation
system is frequently developed
and integrated into technical
requirement for an aerospace
vehicle. In this form, or as
a stand-alone document, a
typical technical requirement
for an integrated navigation

system should address the
following points:
O
The system reliability
and maintainability
requirements.
The functionality and

performance *required from
the system at each stage of
flight.
The maximum acceptable

system size, weight, power and
cooling requirements.
The characteristics of
each flight segment in terms
of:
The maximum cost of the

system, both in development,
acquisition and life cycle
cost terms, and areas where
cost trade-offs cannot be
permitted.
- The duration of the

segment
- The occurrence of the
segment in relationship to
other segment (e.g. the number
of times it is likely to
occur, will it always follow
other segments, etc. )
- The aircraft linear and
angular acceleration and
velocity conditions likely to
occur in each segment
The nature of external

conditions at each segment
(e.g. altitude, .terrain,
jamming conditions, etc. )
- The integrity
requirements at each segment,
including the allowable
degradation in function
performance
3.
SUMMARY
The purpose of an integrated

navigation system is twofold to provide information to the
operator and data to other
systems of an aerospace
vehicle with sufficient
accuracy to accomplish the
mission. Implicit is the
function of positioning and
generation of guidance
information to a destination
defined within a specific
coordinate frame. The next
section of this document
discusses navigation
coordinate frames used for
computational mechanization
and display to the aerospace
vehicle crew.
The means to rigorously specify navigation accuracy has been

given much attention by navigation specialists. The results
of the NATO technical community deliberation is contained in
Ref [ 3 ] .
Kayton, M.
John Wiley
C
C
Fried, W.R., Avionics Navigation System,

Sons, New York 1969.
Couperthwaite, W.J. 61 Stokes, r.F., Suggestion concerning

the context of the Navigation Sub-system Section of the EFA
Weapon system Design and Performance Specification, Royal
Aircraft Establishment, Farnborough, Haute, UK, March 1987
NATO Standardization Agreement 4278, Method of Expressing
Navigation Accuracies, 17 August 1993
1
l
SECTION I1
COORDINATE FRAMES
Muneendra Kumar
Defense Mappinq Aqency
8613 Lee Hiqhway
Fairfax, VA 22031, USA.
INTRODUCTION
Accurate navigation requires an understanding of the three coordinate frames

and systems which define positions in
space. The first is the inertial frame in
which the earth revolves around the sun
annually (Section 2). This annual motion
is affected by the general precession and
astronomic nutation and it takes place in
the Earth Centered Inertial (ECI) frame or
Conventional Inertial System (CIS).
Second is the reference frame which
is defined by the daily rotation of the
earth around its polar axis (Section 4).
This Instantaneous Terrestrial System
(ITS) frame requires knowledge of the
sidereal time relationship with the CIS.
Third frame is the geodetic or Earth
Centered Earth Fixed (ECEF) coordinate
system defining the three-dimensional
positions on the earth's surface or in its
adjoining space. This frame is also known
as Conventional Terrestrial System (CTS)
and requires the knowledge of the earth's
polar motion, the gravity field, and its
size and shape (Section 5).
the coordinate frames or systems are

right-handed and orthogonal, and positive
rotation is clockwise when viewed from the
origin towards the positive axis. Further,
the three rotation matrices R x , R , and R
represent the positive rotationd about (he
orthogonal axes X,Y, and Z respectively.
2
In the mean ECI system of epoch (or

CIS), the 2,-axis lies along the earth's
instantaneous axis of rotation positive
towards the mean celestial pole of epoch
(Po), the XI-axis is positive towards the
vernal equinox of epoch ( y o ) and lies in
the plane of celestial equator at right
angle to the Z -axis, and the Y,-axis is
perpendicular CO the 2, and X, axes and
contained in the equatorial plane. These
axes complete the right-handed orthogonal
coordinate frame (Figure 1). The epoch
52000.0 is defined at noon on 1 January
2000 (Section 3).
This inertial system is based on the
Fundamental Katloq 5 (FK 5) system ( 9 ) .
3
The ECI (or CIS) transformation to the
ECEF (or CTS) frame makes use of the new
theories of precession [1,2], astronomic
nutation [ 3 , 4 , 5 ) , change to a new 52000.0
Standard Time Epoch (21, the new definition of Universal Time as defined and
adopted by the International Astronomical
Union (IAU) [6,7], and the latest ECEF
frame, viz., World Geodetic System (WGS)
1984 [e].
In the discussion that follows, all
EARTH CENTERED INERTIAL (ECI) SYSTEM
TIME AND EPOCHS
The two time systems of interest here,

the Sidereal Tlme (ST) and Universal (UT)
or solar time, are both based on earth's
diurnal rotation [lo]. The sidereal Tlme
is determined by observing the transits of
stars across the observer's meridian.
However, since the meridian is involved,
one must consider the effect of polar
motion on the meridian's position. Mean
solar time is associated with a mean or
"fictitious" sun that moves along the
celestial equator with a uniform sidereal

motion. The motion is approximately equal
to the mean rate of the true sun's annual
motion along the ecliptic. If the hour
angle of the mean sun is referred to the
zero meridian (1 = 0') as defined by the
International Earth Rotation Service
(IERS), the resulting time is known as the
Universal Time (UT).
Universal time and sidereal time are
both affected by the irregularities of the
earth's rotation. These irregularities
take the form of polar motion, i.e., the
variations in the position of the earth's
axis of rotation with respect to earth's
crust and the variations in the angular
rotation ( U ) of the earth about its
rotational axis.
Since January 1 1972, the time scale
distributed by most of the broadcast time
services is Coordinated Universal Time
(UTC) [ll]. This redefined time differs
from the International Atomic Time (TAI)
by an integral number of seconds. TA1 is
the most precise and accurate time scale
as determined by the Bureau International
de Poids et Mesures (BIPM), Paris, from
the analysis of data from the atomic time
standards of many countries worldwide.
Universal time, as determined with its
irregularities, is known as UTO. The time
UT1, which is independent of the position
of the observer, is derived by removing
the effect of polar motion from the UTO.
The leap second or one second time step is
introduced periodically to maintain UTC
time within 0.9 second of UT1. This
correction is normally made at the end of
June or December.
3.1 Sidereal Time (ST)
Mean Sidereal Time (MST) is directly
obtained from the apparent right ascension
of transiting stars 1121. On any given
date, a star's computed apparent position
is a function not only of its catalogued
coordinates and proper motion, but also of
the adopted constants and theories of the
general precession, astronomic nutation,

aberration, etc., [ 7 ] .
It is to be noted that the change from
the old FK4 to the current FK5 star system
and adoption of new astronomical constants
do produce a complex and subtle change on
the values of universal time UT1 which is
indirectly derived from the transit time
of stars.
3.2 Universal Time (UT)
The UTl, which forms the basis for the
worldwide system of time, is related to
MST by the equation for Greenwich Mean
Sidereal Time (GMST) of O h UT1 (Table 1).
The sidereal time equation or matrix
[B] in Section 4.4 gives specific values
for the angular rotation velocity ( U ) of
the earth and the ratio of solar time to
sidereal time as 0.997269566329084 (or its
inverse as 1.0023737909350795) at epoch
J2000.0 [7].
UT1 defined as above maintains time
continuity in value and rate of ST at the
change of epoch to FK5 system, i.e., on
1 January 1984.0.
3 . 3 Unit of Time (T)
The unit of time (T) in the formulas

for astronomical precession and nutation
is the Julian Century of 36525 days. The
conventional relationship between the
Julian Epochs (JE) and Julian Ephemeris
Date (JED) is [2]:
JE = 2000.0 + JJED - 2451545.0)
365.25
(1)
It is to be noted here that the Julian

Day (JD) begins at noon ( 12h).
3.4 Changing Epochs
The relationship between the discontiepoch (BE) and the new
JE is expressed as [2]:
nued Besselian time
BE(1950.0) = 1900.O+(JED-2415020.3)(2)
365.24219878
The correspondence between the six
different epochs of BEs and JEs and the
matrix (M) to transform precessing position and velocity components from BE to
JE, as computed through equation ( 2 ) , are
also available in [2].
4
INSTANTANEOUS TERRESTRIAL SYSTEM (ITS1
4 . 1 General
The transformation of position components from the CIS (or mean ECI system of
epoch) to the true ITS (or ECEF) system
(Figure 1) requires formation of three
matrices, viz., the general precession
matrix [D], the astronomic nutation matrix
[C], and the sidereal time matrix [B]. The
starting epoch J2000.0 is defined at noon
time on 1 January 2000 (Section 3.3).
Txl
1x1
For transformation of three velocity

components, rate of change of the matrices
in equation ( 3 ) is to be considered:
as the product matrix [D] in Table 3 .

After the precession matrix [D] has
been applied, the new coordinate frame
X,(X,,Y,,Z ) represents the mean inertial
system oP date (Figure 2).
As the matrix [D] is orthogonal,
the vector equations to transform are:
=
[Dl
XI
[DIT X,
The precession, consisting of three

rotation angles (90-(, 0, 90-z), is given
(5)
(6)
4 . 3 Astronomic Nutation Matrix [ C ]
The astronomic nutation is resolved in

two components, nutation in ecliptic
longitude ( A + ) and nutation in obliquity
( A e ) . It transforms the coordinates from
the mean inertial of date to the true
inertial system of date (Figure 3 ) .
The X,(X,,Y ,Z,) axes (Figure 2) are
transformed with three nutation angles or
rotations ( e , A + , e ) into the X,(X,,Y,,Z,)
axes (Figure 3 ) . These angles are also
depicted symbolically in Figure 4 and the
equations and related coefficients for
computing them are listed in Tables 4 and
5. The matrix [C], orthogonal like [D], is
in Table 6.
The Z,-axis now points towards the
Celestial Ephemeris Pole (CEP) of date and
its orthogonal X, and Y, axes lie in the
true equatorial plane of date. In vector
form:
x,
[Cl
x,
(7)
X,
[C]' x,
(8)
4 . 2 General Precession Matrix [D]
The motion of general precession in

space can be specified by three angles C,
2, and 0 (Table 2). The precession matrix
[D] transforms the mean CIS coordinate
system X (Xl,Yl,Z1)
of epoch J2000.0 (or
ECI) of bigure 1 to the orthogonal mean
coordinate system X,(X,,Y,,Z,)
x,
x*
4 . 4 Sidereal Time Matrix [B]
The ST represents a positive rotation

about the Z,-axis by an angle A , the
longitude of the zero meridian (which used
to be the Greenwich Meridian but now as
defined by IERS) from the true vernal
equinox of date. This sidereal time matrix
[B] or its rotational transformation
accounts for both the earth's rotation

towards the east and precession of the
mean equinox towards the west. Table 1
lists the equation to compute A .
The matrices [B] and [B'] are given in
Table 7 (Section 4.1).
The matrix [B] transforms the coordinates from the true inertial system of
date X (X3,Y3,Z,)to the true ECEF system
X 4 ( X 4 , $,,Z,) or the Instantaneous Terrestrial System (ITS). Figure 5 shows symbolically only the Z,-axis of the new transformed orthogonal 'system. In vector form:
x4
[el x,
x4
[AIT X,
The X, coordinate frame (CTS) is also

fixed to the earth and rotates with it.
The first CTS, defined for global use, was
the World Geodetic System (WGS) 1960,
which in turn has evolved through WGS 66
and WGS 7 2 to the latest and more accurate
version, viz., WGS 84 [8].
6
WORLD GEODETIC SYSTEM 1984
(9)
6.1 General
(10)
The WGS 84 coordinate frame (depicted

without the suffixes) is defined for the
epoch 1984.0 (Figure 7 ) . The CTS implied
in the WGS 84 is defined by the Bureau
International de l'Heure or BIH which is
now known as IERS.
The X, coordinate frame (ITS) is fixed

to the earth and thus rotates with it.
5
vector form:
CONVENTIONAL TERRESTRIAL SYSTEM (CTS)
5.1 General
6.2 Realization
The Z -axis of the ITS frame is still

affected by the polar motion or relative
motion of the earth's instantaneous axis
of rotation with respect to its crust.
When corrected for this polar motion, the
true ITS system transforms to a mean
geodetic ECEF or CTS system [8].
The CTS/ECEF frame in the WGS 84 was

realized by modifying the Navy Navigation
Satellite System (NNSS) or TRANSIT Doppler
reference frame in origin and scale, and
also rotating it about the Z,-axis to
bring its reference meridian to the BIH
defined zero meridian [8].
5.2 Polar Motion
The WGS 84 coordinate system (Figure

7) represents a standard earth rotating
with a constant velocity ( U ) around an
average astronomical pole (CTP) for the
epoch 1984.0.
The polar motion affecting the X,

frame consists of two rotation angles
(Figure 6):
- A negative rotation of angle yp
about the X,-axis.
- A negative rotation of angle xp

about the ',-axis.
The angles x and y are small and
this allows the hse of 'approximations for
the sine and cosine functions of these
angles. The resulting matrix [A] for the
polar motion is shown in Table 8. In
In addition to defining the basic

geodetic reference frame, the WGS 84
defines the earth's geometric figure,
models the gravity field and the related
geoid, and establishes the transformation
constants relating local geodetic datums
to a global reference system (Section 8).
This provides a global reference for
worldwide Mapping, Charting, and Geodetic
(MCLG) and navigational products and
applications.
6.3 Geometry of the Earth

In MChG and navigational applications, three different surfaces are
involved which are depicted in Figure 8:
- The earth's actual physical topographical surface.

- The geometric or mathematically
represented earth's surface, the reference
ellipsoid (Section 6.3.1)
- The equipotential surface defined in

the earth's gravity field, viz., the geoid
(Section 6.3.2; Figure 9 ) .
6.3.1 The WGS 84 Reference Ellipsoid
As the geometric approximation of the
earth's shape, WGS 84 reference ellipsoid
is based on the Geodetic Reference System
(GRS) 1980 [13], as defined by the International Union of Geodesy and Geophysics
(IUGG). The GRS 80 definition consists of
a geocentric equipotential ellipsoid of
revolution defined by the semimajor axis
(a), the earth's gravitational constant
(GM), the normalized second degree zonal
gravitational harmonic coefficient (C20),
and the angular rotational velocity ( 0 ) of
the earth.
The GRS 80 ellipsoid, tied through

the gravity and other data sets [14], was
used to best represent the earth's geoid
by minimizing the geoidal heights between
the ellipsoid and the geoid globally.
This global solution, made possible
with the availability of a dense gravity
data base worldwide and intercontinental
geodetic ties available through satellite
observations since the early ~ O ' S , should
not be confused with the local or regional
datums which are in use all over the
world. To simplify the mapping complexity
of using the different position coordinates defined in the local and regional
datums, these datums have been geodetically tied to the WGS 84 through transformation constants (Section 8). However,
due to the non-availability of common

control points, which are essential to
establish transformation constants,
there are limited number of local/regional
geodetic datums which have been tied to
WGS 84 [E].
Table 9 lists the defining or the
fundamental constants of the WGS 84 [E].
Some of the important derived geometric
and physical constants (Section 6.3.2) and
the conversion factors are provided in
Table 10. Further, it is important to
note that the flattening f of the WGS 84
ellipsoid is a derived constant and is
based on the gravitational zonal coefficient Cq0 (or J2,/J5 of the GRS 80). The
eccentricity e is derived or computed from
flattening f (Equation 14).
The spatial position in the WGS 84
reference frame is .given as rectangular
coordinates X,Y,Z (by dropping the CTS
suffixes in Figure 5 ) or as equivalent
geodetic coordinates viz., latitude 4,
longitude 1, and ellipsoidal height h in
Figure 10. The relationship between the
rectangular and the geodetic coordinates
is in Table 11.
a. WGS 84 Ellipsoidal Geometry Fiaure
10
- - a - - - - shows a meridian section of the
WGS 84 ellipsoid with its semimajor axis a
and semiminor axis b where:
b
a (1-f)
e2
(a2 - b2)/a2
2f - f2
(13)
(15)
The geocentric radius R intersects

the semimajor axis (or the equatorial
plane of the ellipsoid) at angle I$, the
geocentric latitude, where:
R2
tan I$
(1 - e2) tan 4
(17)
b. Ellipsoidal Radii of Curvature The two WGS 84 ellipsoidal radii, which
12
are of fundamental importance in deadreckoning navigation to define the

latitudinal and longitudinal velocities
( t $ ' , n ' ) , are the radius of curvature R in
the plane of meridian (or N-S), and raaius
of curvature R, in the plane of prime
vertical (or E-W):
c. Ellipsoidal Heiqht - The distance

or height of any point P measured along
the ellipsoidal normal or a line perpendicular to the ellipsoid, is known as ellipsoidal height (h) as shown in Figures 8
and 10.
6 . 3 . 2 Gravity Model and Geoid
a. Gravity Model - The earth's total

gravity potential (W) is defined as:
W
(20)
where V is the gravitational potential or

the Earth Gravitational Model (EGM) and
is the centrifugal potential [E]. The EGM
is modeled as a spherical harmonic expansion, complete through degree (n) and
order (m) equal to 180.
The WGS 84 gravity potential (W) then
provides the accurate model of the earth's
gravity field (9) and its variations on or
near its surface. The direction of g is
along the plumb line or the astronomical
vertical [15].
b. Geoid - The general representation
of the geoid (Figure 9 ) can now be defined
in the WGS 8 4 gravity field as the particular equipotential surface (GEOP) as
defined by the equation:
W,,(X,Y,Z)
WO
(21)
where the latest value of WO is listed in

Table 10. The direction of gravity field g
is perpendicular to the geoid at every
point.
For some practical applications, the
geoid, defined as above, is approximated
by the mean sea level (msl) at 1 to 2
meter accuracy level along the coast lines
and over the ocean areas. Under the
continental land masses, the hypothetical
extension of msl is sometimes used as an
approximation for the geoid.
It may be necessary to clarify here
that msl is not an equipotential surface.
By definition, msl is the averaae or mean
of hourly sea level surfaces obGerved at a
point (or tidal benchmark) over a period
of 18.67 years. However, all msl surfaces
so determined in different parts of the
world differ in definition and also do not
belong to a common zero reference.
In a mathematical sense, the geoid is
defined (or realized) as so many meters
(m) above ( + N) or below ( - N) the
reference ellipsoid, the geometric figure
of the earth (Figure 8). Figure 9 depicts
one version of the WGS 84 geoid as a
contour chart with respect to the WGS 84
ellipsoid.
NOTE: More detailed contour charts or
denser grid values of the WGS 84 seoid
with absolute accuracy ranqe of f - 2 to 6 m
are now UNCLASSIFIED and available).
The distance of any point P from the
actual physical surface of the earth to
the geoid (Figure E), measured along the
direction of gravity or plumb line is
called the orthometric height (H) of P. If
P is above the geoid, the H is positive
and, if below the H is negative. 3
practice, the height H is also approximated by elevation above or below the msl.
6 . 4 Relationship Between h, H I and N
The three heights, viz., the geodetic

or ellipsoidal height h (Section 6.3.1),
the orthometric height or elevation HI and
the geoidal height N (Section 6 . 3 . 2 ) can
be related to each other (Figure 8) as:
13
(22)
(23)
then obtained by the equation:
or its equivalent form:

H
Equation (23) is of great practical

importance and it illustrates the use of
geoid heights (Ns) in the determination of
the orthometric heights (Hs) from geodetic
heights (hs), which are directly obtainable from satellite point positioning
solutions of the GPS satellites.
where R and R, are rotation matrices

about the axes X and 2 . The equation (24)
can also be written as:
<
-sin&inA
cos1
0
1](25)
cost$]
6.5 Deflection of the Vertical

cost$sinA sin4
The angle
normal and the
plumb line) is
the vertical (
direction of gravity (or

known as the deflection of
igure 8 ) .
The angle 0, with an average of about

10 arc seconds and a maximum of about 1
arc minute, cannot be neglected in the
modern day navigation when it extends over
time or long distances globally.
7
CTS
0) between the ellipsoidal
7.3 Generalized Local Geodetic System
The axes E and N (In Figure 12)

are obtained by rotating the ENU-system
(Figure 11) by an angle a in azimuth about
the U-axis :
OTHER COORDINATE FRAMES
These are defined by the geocentric

radius R (Equation 16), the geocentric
latitude Ip (Figure 10) which is directly
related to the qeodetic latitude I$ through
equation 17, and the geodetic longitude A .
From Table 11, the radius R can be related
to the geodetic coordinates (4,l,h).
7.1 Geocentric Spherical Coordinates
where rotation R, is about the axis U.
or
(27
J
CTS
7.2 Local Geodetic/Topocentric Systems

or
In Figure 11, the three axes E, N I and

U defines a local geodetic or topocentric
coordinate system. The system origin is at
a point (t$,A,h)with the positive axes E
and N pointing geodetic east and north,
and the axis U pointing upwards along the
ellipsoidal normal or towards positive
geodetic height h respectively.
The coordinates (E,N,U) of P in the
topocentric local geodetic ENU-system are
1 i:]
Cll
c12
13
C2l
c22
23
31
32
33
where the nine elements of the equation

(28) are known as direction cosines. Then,
comparing the corresponding elements of
the two matrices in equations (27) and
14
ship for H is from equation (23).
( 2 8 ) , the following relationships are

obtained :
7.5 Grid Coordinates
c,,
c,,
C,,
c,,
C,,
C,,
c,,
c,,
c,,
=
=
=
=
=
=
=
=
=
-cosasin),
cosacosi
cosQsina
sinasinli
-sinacosi
cosQcosa
cosQcosi
cosQsini
sin6
- sinasinQcosli
- sinasinQsinl
- sinQcosacosl
- cosasinQsinli (29)
The generalized local topocentric

E'N'U-system with its arbitrary azimuth
is conveniently useable worldwide. Then,
the geodetic coordinates ( Q , A ) can be
computed as:
sin Q =
tan li =
tan a =
'33
c32/c31
c13/c23
In many mapping and charting applications, the earth's surface is represented

on a flat map through a projection with
point positions defined in rectangular
grid coordinate system. The grid coordinates (x,y) and geodetic positions ( Q , l i )
are interrelated. The unit of measurement
is grid meter.
The most common map projections are
the Universal Transverse Mercator (UTM),
Transverse Mercator (TRM), Mercator (MER),
Universal Polar Stereographic (UPS), and
Lambert Conformal Conic (LCC) [16,17,18].
The first three projections are based on a
cylindrical, the fourth on a plane, and
the fifth on a conical surface. All these
projections are conformal.
NOTE: In polar regions, the angles a

and li are meaningless or indeterminate and
thus, the equations in (28) with direction
cosines C,, are to be used.
The UTM projection in secant version

with 0.9996 as the scale factor along the
central meridian [16] is the most popular
over non-polar areas (See Section 7.6)
7.4 Horizon Coordinates
7.6 Military Grid Reference system (MGRS)
In the horizon or local level system,

the G-axis is along the direction of the
local plumb line (or gravity) and points
upwards, the axis E" points towards the
local east, and the axis N" is orthogonal
to the axes G and E and points the local
astronomical (true) north (Figure 13).
The axes E" and N" both lie in the local
horizon plane.
The Military Grid Reference System

(MGRS) is used to extend the local grid
system of coordinates to a large regional
application [17,19]. MGRS is only used
with the Universal Transverse Mercator
(UTM) and Universal Polar Stereographic
(UPS) map projections and their associated
grids.
The relationship between the local

geodetic (ENU) and local level (E"N"G)
coordinate system is defined as:
For MGRS numbering scheme, the UTM

projection extends from latitude 80' S
to latitude 84' NI and the UPS projection
covers the remaining two polar caps.
"
li "
li
h
N (meters)
+ T)/cos Q
(31)
OTHER GEODETIC DATUMS AND THEIR

RELATIONSHIP WITH WGS 84
8.1 General
where [,q are the two components of the

deflection of vertical 0 in the meridian
(N-S) and the prime vertical (E-W) planes
(Section 1.6.5) and the height relation-
Currently, numerous local and regional

geodetic datums (coordinate systems) are
in use worldwide. Some of these datums,
covering only small remote islands and

isolated areas, are very poorly defined;
some datums demonstrate large errors and
excessive distortions.
TO map, chart, and navigate in the
complexity of local or multiple datums is
a challenge. one possible solution will be
that all positions required to navigate
are first transformed from their local and
regional coordinate system into a globally
available ECEF coordinate frame, e.g., the
WGS 84 (section 6).
One exception to the above local and
regional geodetic datum complexity is the
newly defined North American Datum (NAD)
1983.
8.2 North American Datum (NAD) 1983
Replacing its predecessor the NAD 27,
the NAD 83 is based on a modern geocentric
concept and also uses the GRS 80
parameters [13] for its definition.
a. Coveraqe Area - NAD 83 essentially
extends over the North American continent
from Alaska to Panama with extension over
Hawaii and other American territories and
some islands in the Pacific and Atlantic.
b. Relationship with WGS 84 - There is
no difference between NAD 83 and WGS 84 as
horizontal reference systems for mapping,
charting, and navigation at 1 to 2 meter
level.
8.3 Other Local/Regional Datums
The relationships between other local
and/or regional geodetic datums with WGS
84 are very complex and also not available
in all cases.
available. The accuracies of these datum

transformation constants, except in a few
cases, are very poor for geodetic applications. However, the available datum transformation constants in [E] are of adequate
accuracies to support mapping, charting,
and navigational applications.
All other local and regional datums
and global reference frames, not listed in
[E], can not be transformed to WGS 84 at
this time due to lack of required common
control data.
8.3.2 Transformation Methods
In avionics or navigation, the nongeodetic transformation methods, which can
be used with necessary accuracy, are :
a. For Non-Polar Areas - In Table 12,
the Standard Molodensky equations, which
transforms the two geodetic coordinates
(4,I) of any local and regional datum to
WGS 84, are given. These equations need
the use of the ellipsoidal and transformation constants listed in Appendices
A, B, and C of [E]. In the absence of a
consistent world height system, the
present practical method assumes that all
orthometric or msl heights are nonvariant
during such datum transformations.
The Molodensky equations do not give
satifactory results for AI in the polar
areas, i.e., from 89' N or S latitude to
the respective pole.
b. For Polar Areas - Near the polar
areas between 89O N or S latitude to their
respective poles the local and regional
datum coordinates should be transformed in
three steps:
8.3.1 Availability of Constants

The transformation constants for the
local and regional geodetic datums tied to
WGS 84 are available in [E], wherein the
number of such datums would be changing as
more and more required data sets become
(Step# 2) (AX,AY,AZ)
16
on IAU 1976 System of Astronomical

____
Constants", Astronomy and Astro and
physics, Vol. 73, 1979.
In this procedure, the coordinate

conversion equations for steps 1 and 2 are
tabulated in Table 11 while the shifts or
transformation constants are in [8].
3.
Seidelmann, P.K.; "1980 IAU Theory of

Nutation -The Final Report of the IAU
Working Group on Nutation", Celestial
Mechanics, Vol. 27, No. 1, May 1982.
4.
Wahr, J.M.; "The Tidal Motion of a

Rotating, Elliptical, Elastic,and
Oceanless Earth", Ph.D. Dissertation,
University of Colorado, Boulder, CO,
80309, 1979.
5.
Wahr, J.M.; "The Forced Nutations of

an Elliptical, Rotating, Elastic, and
Oceanless Earth" , Geophy
6.
Aoki, S., Guinot, B., Kaplan, G.H.,

Kinoshita, H., McCarthy D., and P.K.
Seidelmann; "The New Definition of
Universal Time", Astronomy and
Astrophysics, Vol. 105, 1982.
b. Foot - The international foot is to

used within the Department of Defense [20]
while the US survey foot is used by the
civil users; the difference between the
two units is significant (Table 10).
7.
Kaplan, G.H. ; "The IAU Resolutions on

Astronomical Constants, Time Scales,
and the Fundamental Reference Frame"
US Naval Observatory Circular, No.
163, Washington, DC, 20392, 1981.
9.3 Geodetic Constants
8.
The old practice of using any geodetic

constants in mapping, charting, and navigaional applications should be discontinued. To ensure correct mapping, charting,
and navigation, use of WGS 84 constants is
mandatory.
Defense Mapping Agency; "Department of

Defense World Geodetic System 1984,
Its Definition and Relationships with
Local Geodetic Systems", DMA TR
8350.2, Second Edition, Fairfax, VA,
22031, 1 September 1991.
9.
Fricke, W.; "On the Determination of

the Equinox and Equator of the New
Fundamental Reference Coordinate
System, the FKS", Celestial MechaICs, Vol. 22, No. 2, August 1980.
UNITS OF LENGTH AND OTHER CONSTANTS
9.1 General
In the present era of aviation over
long distances, it is very important to
use correct units. Unit measures of meter
and foot can be significantly different
(Table 10) and thus should be used with
caution.
9.2
Units of Length
a. Meter - The length of international

meter is linked with the velocity of light
and thus has been a constantly changing
unit over the years. In some country, a
leqal meter is in use and its value may
differ significantly (about 1 in 75000 in
Namibia) from the unit definition listed
in Table 10.
10 REFERENCES
1.
2.
Lieske, J., Lederle, T., Fricke, W.,

and B. Morando; "Expressions for the
Precession Quantities Based upon the
IAU 1976 System of Astronomical Constants", Astronomy and Astrophysics,
Vol. 58, 1977.
10. Moritz, H., and 1.1. Mueller;

"Earth Rotation - Theory and
Observation", Ungar Publishing
Company, New York, NY, 1987.
Lieske, J.; "Precession Matrix Based
11. US Government Printing Office; "The
17
Astronomical Almanac for Year 1986",

Washington, DC, 1985.
12. Mueller, 1.1.; "Spherical and
Practical Astronomy as Applied to
Geodesy", Frederick Ungar Publishinq
Co., New York, NY.
~~
13. Moritz, H.; "Geodetic Reference

System 1980", Bulletin Geodesique,
Vol. 54, No. 3, Paris, France, 1980.
14. Defense Mapping Agency; "Supplement to

Department of Defense World Geodetic
System 1984 DMA Technical Report:
Part I - Methods, Techniques, and
Data Used in WGS 8 4 Development",
TR 8350.2-A, Fairfax, VA, 22031,
1987.
15. Heiskanen, W.A., and H. Moritz;

"Physical Geodesy", W. H Freeman and
Co., San Francisco and London, 1967.
16. US Geoloaical Survev: "MaD Proiections

- A Workng Manual"; Pro>essi&nal
Paper 1395, Reston, VA, 1987.
17. Defense Mapping Agency; "Datums,
Ellipsoids, Grids, and Grid
Reference Systems", DMA TM 8358.1,
Fairfax, VA, 22031, 1991.
I,
18. Defense Mapping Agency; "The

Universal Grids : Universal
Transverse Mercator (UTM) and
Universal Polar Stereographic
(UPS)", DMA TM 8358.2,-FairfaxI
VA, 22031, 1989.
19. North Atlantic Treaty Organization
(NATO); "Geodetic Datums, Ellipsoids, Grids , Grid References",
STANAG 2211, Fairfax, VA, 22031,
1991.
20. Defense Mapping Agency; "Geodetic and

Geophysical Sign Convention and
Fundamental Constants", DMA INST
8000.1, Fairfax, VA, 22031, 1991.
18
Figure 1. The Celestial Sphere of Epoch.
North Ecliptic
Pole (NEP)
Winter
Solistice
z1
North Celestial Pole (NCP)

of epoch
Celestial Equator
19
Figure 2. The Celestial Sphere with Precessing Equinox.
5, e , z
7 toQ
0
Q to P
= Precession Parameters
= 9OO-L
= Angle Between Equators
- (90' + Z)
20
Figure 3. The Nutating True Celestial Sphere of Epoch.
21
Figure 4. The Astronomic Nutation
I I Z 2 Axis
II CEP
I4
I
I
I
/Or
z;p
Mean EqJator
Date
True Equator
<
of
Date
(Ito CEP)
Ap = Nutation in Right
Ascension
AV = Nutation in Declination
tanAp= tanAV c o g
sin AV = sin A y sinE
Nutation Parameters E, AV,
E:
= Mean Obliquity of Ecliptic

A y = Nutation in Longitude
E = E
A E = True Obliquity of Ecliptic
A E = Nutation in Obliquity
E
22
Figure 5. The Sidereal Time Transformation.
11 CEP or Z 4Axis
11 CEP or Z
Axis
h = O
at
Oh UT1
h =
oo
at
Timet
A =Ho+ AH+o*(t-'At)
o * = 0 8+
a*= Rotation Rate in Precessing Reference Frame

a' = Earth's Inertial Rotation Rate
m = Rate of Precession in Right Ascension
23
Figure 6. The Polar Motion Transformation.
CTP = Conventional Terrestrial Pole

CEP = Celestial Ephemeris Pole
24
Figure 7. The WGS 84 Coordinate System Definition.

Orgin
= Earth's center of mass
z-Axis = The direction of the Conventional Terrestrial Pole (CTP) for

polar motion, as defined by the Bureau International de
I' Heure (BIH) on the basis of the coordinates adopted for
the BIH stations.
x- Axis = Intersection of the WGS 84 Reference Meridian Plane and
the plane of the CTP's Equator, the Reference Meridian being
the Zero Meridian defined by the BIH on the basis of the
coordinates adopted for the BIH stations.
Y- Axis = Completes a right-handed, earth centered, earth fixed (ECEF)

orthogonal coordinate system, measured in the plane of the CTP
Equartor, 90' East of the x-Axis.
1
BIH-Defined CTP (1984.0)
WGS 84
BIH-Defined
Zero
Meridian
(1984.0)
Earth's Center
- of Mass
WGS 84
Analogous to the BIN Defined Conventional Terrestrial System (CTS) , or BTS, 1984.0.
25
Figure 8. The Relationship between the Earth's Actual,

Ellipsoidal, and Geoidal Surfaces.
I'
Separation
-N
26
?
5!
::
8
n
0
.
P
n
0
0
n
n
0
P
0
n
0
n
N
L
P
0
n
n
n
n
n
n
n
n
..
n
n
n
t
n
0
n
n
N
n
n
n
c
n
.
n
. . . . .
. .
21
Figure 10. The Geocentric Spherical Coordinates (R,W,A).
Z
4
28
Figure 11. The Local Geodetic Coordinates (E, N, U).

(Topocentric)
U
Geodetic
{ meridian of P
29
Figure 12. The Generalized Local Coordinates (E',
N', U).
a =alpha
E'
30
Figure 13. The Horizon Coordinates (E",NI: H).
A'
= Astronomic
Latitude
= Astronomic
Longitude
E"
31
Table 1.
The Sidereal Time Equations **
A = Ho +AH + w*(t-At)
H, = 241 10.54841 + 8640184.8640184.812866 Tu

Seconds of Time
+ 0.093104 T,'-
6.2 x
= Greenwich Mean Sidereal Time at 0 UT1 of JED
Where,
T =du/36525
U
d = JED - 2451545
U
d + 0.5, 1.52.5, ...

U
AH = arctan (cose tanA\lr)
= (Apparent Minus Mean) Sidereal Time

E
= True Obliquity
t
= Time Within Day (UTC)
At = UTC -UTI
w* = Rotation Rate in Precessing Reference Frame

= w'+ rn = 7.2921 158553 x
+ 4.3 x
Tu (Radians/Second)
w' = Earth's Inertial Rotation Rate
= 7.2921 151467 x
(Radians/Second)
rn = Rate of Precession in Right Ascension
= 7.086 x 10-12+ 4.3 x10-15Tu (Radians/Second)

**
See Table 7 also.
Ti
32
Table 2.
The Precession Equations *
g"
= 2306.2181 T + 0.30188 T
Z"
= 2306.2181 T + 1.09468 T 2+ 0.018203 T3
e"
= 2004.3109 T - 0.42665 T2- 0.041833 T3]
2+
0.017998 T
Where,
= [ JED - 2451545 ] /36525

= Julian Centuries from Epoch
J2000.0 (2000 Jan 1.5)
JED = Julian Ephemeris Date

1. See Reference [2,7] .
Note:
2. These three time equations are obtained by setting

T = 0 in equation (7) of reference (21, where
T has a different meanina than here.
Table 3.
The Precession Transformation Matrix
[Dl = R [- (90+2)] Rx (e) R [90-g]
z
[Dl =
cosz cose cosg
sinz sing
- cosz cos0 sing - sinz cosg
- cosz
sinz cos8 cosg
+ cosz sing
- sinz cos0 sing + cosz cos6
- sinz sine
sine cosg
- sine sing
cose
sine
33
Table 4.
The Astronomic Nutation Equations And Arguments
- 46.8150
T - 0.00059 T2+ 0.001813 T3 Arc Seconds

= Mean Obliquity of Ecliptic
= E,
= 23O 26' 21.448"

= 84381.448"
E
+ AE
= E
= True Obliquity of Ecliptic
= [ JED - 24515451/36525
= d/36525
= Julian Centuries from Epoch J2000.0
106
106
=x (A + B . T ) Sin(alik' + a2i.!j''
i=1
F +a
3i
4i
as, n)
I
AE = Nutation in Obliquity
106
=z
i=
1
106
=E
i=l
(Ci + D,T) cos(a

I
li
k' +
a .F
31
a .D
41
a,,n)
I
34
Table 4 (Contd.).
The Astronomic Nutation Equations and Arguments
e
el
= 485866.733 + (1325r+ 715922.633)T

= Mean Anomaly of Moon
31.310T
0.064T
(9gr + 1292581.244)T - 0.577T - 0.012T

= Mean Anomaly of Sun
= 1287099.804
F =
=
D =
=
R =
=
ArcSeconds
Arc Seconds
Arc Seconds
335778.877 + (1342r + 295263.137)T - 13.257T + 0.01 1T
(Mean Longitude of Moon) - R
r
2
3
1072261.307 + (1236 + 1105601.328)T - 6.891T + 0.019T ArcSeconds
Mean Elongation of Moon From Sun
r
2
3
450160.280 - (5 + 482890.539)T + 7.455T + 0.008T
Arc Seconds
Longitude of Ascending Node of Lunar Mean Orbit on
Ecliptic Measured From Mean Equinox of Date
T = (JED - 2451545)/36525
lr = 1296000"
35
Table 5.
1980 IAU Theory of Nutation
Series for Nutations in Longitude (AV) and Obliquity (AE)
(See Table 4)
i a i a 2 a 3 a4 a5
1 0 0 0 0
2 0 0 0 0
3 - 2 0 2 0
4 2 0 - 2 0
5 - 2 0 2 0
1 - 1 0 - 1
6
7 0 - 2 2 - 2
8 2 0 - 2 0
2 - 2
9 0 0
1 0 0 1 0 0
0 1 2 - 2
11
12 0 -1 2 -2
13 0 0 2 - 2
2 0 0 - 2
14
2 - 2
15 0 0
1 6 0 2 0 0
1 7 0 1 0 0
0 2 2 - 2
10
19 0 - 1 0 0
2 0 - 2 0 0 2
2 -2
21 0 -1
22 2 0 0 - 2
23 0 1 2 - 2
1 0 0 - 1
24
25 2 1 0 - 2
26 0 0 - 2 2
27 0 1 -2 2
2 8 0 1 0 0
2 9 - 1 0 0 1
30 0 1 2 - 2
31
0 0 2 0
3 2 1 0 0 0
0 0
2 0
33
3 4 1 0 2 0
35 1 0 0 - 2
36-1 0
2 0
3 7 0 0 0 2
3 0 1 0 0 Q
39-1 0 0 0
4 0 - 1 0
2 2
4 1 1 0 2 0
4 2 0 0 2 2
4 3 2 0 0 0
1 0
2 - 2
44
4 5 2 0 2 0
4 6 0 0 2 0
47-1 0 2 0
4 8 - 1 0
0 2
1 0
0 - 2
49
50-1 0 2 2
1 1 0 - 2
51
5 2 0 1 2 0
0 - 1 2 0
53
1
2
1
0
2
0
1
1
2
0
2
2
1
0
0
0
1
2
1
1
1
1
1
0
0
1
0
2
1
0
2
0
1
2
0
2
0
1
1
2
1
2
0
2
2
0
1
1
1
1
0
2
2
Units: A
i
~
a i a 2 a 3 a 4 a5
-171996 -174.2 92025 0.9

54
1
2062
0.2 -095 0.5
55
1
46
0.0
-24 0.0
56
2
11
0.0
0 0.0
5
7
-3
0.0
1 0.0
5
0
-3
0.0
0 0.0
1
59
-2
0.0
1 0.0 ~ 6 0 0
1
0.0
0 0.0
61
1
-13107
-1.6 5736 -3.1
6
2
1426
54 -0.1
-3.4
63
0
-517
1.2
224 -0.6
64
1
217
-0.5
-95 0.3
6
5
129
0.1
-70 0.0
1
66
48
0.0
1 0.0
67
1
-22
0.0
0 0.0
60
1
17
-0.1
0 0.0
69
-1
-15
0.0
9 0.0
70
-2
-16
0.1
7 0.0
71
3
-12
0.0
6 0.0
0
72
-6
0.0
3 0.0
73
1
0.0
-5
3 0.0
-1
74
4
0.0
-2 0.0
75
2
4
0.0
-2 0.0
76
1
-4
0.0
0 0.0
3
77
1
0.0
0 0.0
7
0
1
0.0
0 0.0
-1
79
-1
0.0
0 0.0
00
1
1
0.0
0 0.0
81
-2
0.0
1
0 0.0
-1
02
-1
0.0
0 0.0
03
2
-2274
-0.2
977 -0.5
04
1
712
0.1
-7 0.0
1
05
-306
-0.4
200 0.0
06
-2
-301
0.0
129 -0.1
-1
07
-158
0.0
-1 0.0
00
1
123
0.0
-53 0.0
89
2
63
0.0
-2 0.0
90
2
0.1
63
-33 0.0
91
1
-58
-0.1
32 0.0
92
0
0.0
-59
26 0.0
3
93
-51
0.0
27 0.0
94
1
0.0
-30
16 0.0
95
0
29
0.0
-1 0.0
96
-1
29
0.0
-12 0.0
97
0
0.0
-31
13 0.0
98
0
0.0
26
-1 0.0
99
0
0.0
21
-10 0.0
16
0.0
-0 0.0
-13
0.0
7 0.0
-10
0.0
5 0.0
-7
0.0
0 0.0
7
0.0
-3 0.0
0.0
3 0.0
-7
2 = 0.0001"; B = D 0.0001'
0
0
0
2
0
2
0
2
0
2
0
0
- 1
0
2
0
2
1
0
0
-2
0
0
0
1
0
0
2
- 1
2
-1
2
0
0
0
0
0
- 1
1
0
0
0
0
0
0
0
0
Q
0
1
0
0
0
-1
0
0
0
0
0
0
1
-1
0
0
1
2
2
- 2
2
2
- 2
-2
0
0
- 2
2
0
2
1
1
1
1
0
1
0
1
0
0
0
2
0
0
0
2
2
-0
6
6
-6
-7
6
-5
5
-5
-4
4
-4
0.0
0.0
3
0
0.0
0.0
0.0
-3
0.0
0.0
0.0
3
3
-3
3
0
3
0
0
0
0
0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
1
1
1
1
-1
1
-1
1
0
-1
-1
0.0
0.0
-3
3
-3
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
2 - 2
2
2
2
4
4
0
0
-2
2 - 2
2
2
0
2
4 - 2
2 - 2
2 - 2
2
0
0
2
-2
0
2 - 1
0
2
2
1
2
2
-3
-2
-3
-3
2
-2
2
-2
2
2
1
-1
1
-2
-1
1
-1
-1
1
0.0
1
2
1
2
2
0
1
1
1
2
0
1
-1
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
2
0
2
2
2 2
2
0 2
2
-2 1
0
0 1
0
0 2
0
0 0
2
1
2
0
0 2
0 - 4 0
2
2 2
2
4 2
0
-1
1
1
-1
1
1
-1
-1
-1
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
-1
1
0
-1
1
1
0
0
-1
0
0
0
0
0
0
0
0
0
0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
36
Table 6.
The Astronomic Nutation Matrix
- sinby
- sinAy cosi
cosAy
sinby
sinE sinAy
COSE cosAy cosE
sinE cosby cosE
- COSE sinE
sine sing
COSE cosby cosE
sine COSAY, sinE
Table 7.
The Sidereal Time Transformation Matrix **
[BI =
61=
cosA
sinA
- sinA
cosA
-a' sinA
-0'
Cosh
** See Table 1 also.
a' COSA
--o'sinA
sinE
sinE cosE
+ COSE COSE
31
Table 8.
The Polar Motion Transformation Matrix
[Since xpand yp are small angles, it is technically permissible to use approximate

transformation matrix]
xp = Angular displacement of CEP from mean terrestrial pole measured

along Zero Meridian (positive south)
yp = Angular displacement of CEP from mean terrestrial pole measured
normal to Zero Meridian (positive west)
38
r(
aa
'
40
01
I
(D
0
4
0
0
0
0
In
4
\o
\o
In
4
+I
+I
+I
*E
4
r(
+I
+I
+I
0
-4
r(
In
a
a
'p
c,
.I.
-4
a
a
rl
In
rl
In
0
0
(v
rl
(U
fi
(v
-4
U
ul
k
0
w
OD
rn
t,
a,
m
*
0
NI
5k
0
\o
OD
rn
0,
-4
k
7
c,
m
Q,
rl
a,
U
a : :
>
c
c3
k
Q,
c,
Q,
-4
c r l
oa,a
E d E:
w
0
a
C O ,
ow k
-4
0 a,
a
k
a
PI
t3
E:
-4
7
t,
II
E+>
39
Table 10
Derived Parameters and Conversion Factors
Derived Constant
Notation
Value
Geometric
Flattening
1/298.257223563
(.00335281066474)
Semi-Minor Axis
6356752.3142 m
First Eccentricity
Squared
e'
0.00669437999013
Physica1
Ellipsoidal Gravity
or Geoidal Constant
U, or
WO
Normal Gravity at the

Equator(on Ellipsoid)
ye
Normal Gravity at the

Poles (on Ellipsoid)
y,
Mean Normal Gravity
ym
62636860.8497
m2 s-'
9.7803267714 m s-'
9.8321863685 m s-'
9.7976446561 m s-'
Conversion Factor
1
1
1
1
Meter (m)
Meter (m)
Int'l Foot
US Survey Foot
1 Int'l Nautical Mile

1 Int'l Statute Mile
Velocity of Light
(In vacuum)
=
=
=
=
=
=
=
3.280833333 US Survey Feet

3.28083989501 Int'l Feet
0.3048 m (Exact)
1200/3937 m (Exact)
0.30480060960 m
1852 m (Exact)
1609.344 m (Exact)
5280 Int'l Feet (Exact)
= 299792458
s-l
40
Table 11
Relationship Between Rectangular and Geodetic Coordinates
A.
Geodetic to Rectangular
(RN + h) cos4 cos1
{RN (1 - e) + h} sin4
(RN + h) cos4 sin1
where,
B.
Rectangular to Geodetic
tan- (Y/x)
NOTE: To solve for geodetic ( 4 ,h), first put h, = 0

and compute 4,, then use this value of 4,to compute h,, and using
h, iterate this sequence. In most cases, one or two iterations
would produce required accuracy (for navigational applications)
for 4 and h.
41
Table 12
The Standard Molodensky Transformation Equations
- Local Geodetic Datum to WGS 8 4 THE MOLODENSKY EQUATIONS:
The corrections (A$,Al,Ah) for transforming local datum
geodetic coordinates (4,l,h) to WGS 8 4 are:
AX siq$ cosl - AY sin4 sinl +AZ cos$
+ Aa(R,e sin$ cos$)/a + Af { R, (a/b)
+ R, (b/a) } sin$ cosl$ ]
[ (R, + h) sin 1" 1-
A$
= [-
AA.
= [ - AX sinl + AY cosl 3
[ (R, + h) cos$ sin 1"
Ah
I-'
[AX cos $ cosl + AY cos$ sin1

+ AZ sin$ - Aa ($/R)
+ Af (b/a) R, sin $\
where,
AX,AY,AZ are
and WGS
4,l,h (to be
are the
NOTE:
the datum shifts between local geodetic datum

84 [ 8 ] .
used on the right hand side of the equations)

local geodetic datum coordinates.
1. All A-quantities are formed by subtracting local

geodetic datum quantities from WGS 8 4 quantities.
ellipsoidalheights (h) are not available for

local datums, the Ah correction is not applicable
to WGS 8 4 .
when transforming -
2 . As
3 . For transformation from WGS 8 4 to local datums,
the above A-quantities and the coordinates on the

right hand side of the equations would require an
appropriate change.
43
SECTION I11
NAVIGATION TECHNOLOGY
INTRODUCTION
DR. JOHN NIEMELA
The purpose of this section to

describe the predominant
navigation sensor technologies
employed in modern aerospace
navigation systems. A series
of papers are presented which
discuss in terms of physical
principles and mathematical
development the capabilities
and limitations of each
technology. Particular
emphasis is given to the error
sources and characteristics.
Understanding of these errors
is of great importance for a
solid appreciation of the
following section of this
document on System Analysis
and Synthesis.
Navigation systems can be
viewed as three broad
categories: self-contained,
externally referenced and
hybrid. The former category,
self-contained, is frequently
referred to as dead reckoning.
A principal attribute of selfcontained navigation systems
is that they are not dependent
on external radio navigation
aids that may be vulnerable in
a tactical environment. In
dead reckoning system,
position is determined by
integrating sensed
acceleration (in inertial
system) or velocities (in
Doppler systems). Typical
sensor measurements are
translational acceleration for
inertial systems and
translational velocity for
Doppler systems. In the
process, measurement errors
are integrated as well. As a
consequence, the accuracy of
computed position decreases

with time and distance
traveled, respectively for
these two navigation
technologies. Such systems
are quite accurate for the
initial time or distance
interval of operation but
generally require a manual or
automatic position updating
mechanism in the course of the
mission. Hence a means is
sought in most aerospace
navigation systems to bound
these time/distance growing
errors. Externally referenced
navigation systems o f f e r such
means.
Attention has been given to
those navigation technologies
that sense the aerospace
vehicle's dynamic state. Such
systems are often termed
inertial navigation systems
(INS) attitude and heading
reference systems (AHRS),
heading references and
vertical gyros. To capture
the underlying technology of
these systems and not be
encumbered by the wide variety
of configurations possible,
two complementary navigation
technologies are discussed:
inertial navigation systems
are magnetic heading
references.
Externally referenced
navigation systems, however,
offer highly complementary
characteristics for aerospace
navigation application,
particularly when integrated
with self contained systems.
Radio navigation aids, the
44
most widely used externally

referenced navigation system
rely on the constancy of an
electromagnetic wave
propagation velocity and
directivity of its propagation
to determine distance and
direction to remote known
locations. Examples include
non-directional radio beacons
(NRB), VHF Omnidirectional
Range (VOR), distance
measuring equipment (DME),
long range navigation (LORAN),
DECCA, OMEGA and satellite
navigation (GPS and GLONASS).
This document provides a
discussion of only two of
these navigation technologies,
OMEGA and satellite
navigation. The technical
literature is rich with
descriptions of the remainder
of these external radio
navigation aids, most of which
are expected to be phased out
of operation.
Two other types of externally
referenced navigation systems
are discussed in some detail
due to characteristics that
make them suitable for
military application: terrain
referenced navigation systems
and astroinertial navigation
systems. Though each is in
fact a hybrid navigation
system - comprised of both
self-contained and externally
referenced elements, a unique
attribute common to each
deserves special attention:
Their external reference is
not susceptible to being
denied by enemy ECM
techniques.
The integration of two or more
different types of navigation
systems, typically a
combination of self-contained
and externally referenced
navigation systems, results in
a hybrid navigation system.

The sensors which comprise
such hybrid navigation systems
typically have complementary
error spectrums and/or data
availability characteristics
which yield a more accurate
and reliable navigation
system. Due to their
importance for military
aerospace navigation
applications, a complete
section of the document has
been dedicated to the system
analysis, design and synthesis
of such systems.
45
Inertial Navigation
B. Stieler
DLR, Institute for Flight Guidance
Postfach 3267
W-3300 Braunschweig
GERMANY
1. Introduction
The development of this technology is truly exciting. It took place primarily in Germany, the United States and the former Soviet
Union [Gi 71, So 76, Ma 901.
The gyrocompass indicating true north on a moving base as on ships can be regarded as the beginning of inertial navigation. The
first seaworthy instrument was built by Anschutz with contributions made by Max Schuler and Albert Einstein and installed on the
fast steamer "Imperator" in 1913 [Sc 621.
At the end of World War I the allies had in the Treaty of Versailles imposed restrictions to Germany for the maximum size of
ships to be built. The so-called "vest-pocket battleships" were much more exposed to heavy seas than their bigger counterparts.
These restrictions promoted in this country gun stabilization and inertial technology in general, which culminated at the end of
World War 11 in a functioning air-supported gyrocompass with electronic Schuler tuning for the "One-Man Submarines", in the V2
guidance system and a true concept for an inertial navigation system (INS) [Gi 711. After the war the development of this
technology was taken over by the superpowers, the United States and the former Soviet Union [Ma 901.
The functional diagrams for a platform INS and for a so-called "strapdown INS (SDS)" are shown in Figs. 1.1 and 1.2. "Strapdown"
comes from the fact that the sensors of the SDS are strapped down to the vehicle. Both figures show us that the inertial
navigation system (INS) provides all information about the kinematics of a vehicle, namely attitude and heading, ground speed and
position, and also angular rate and acceleration independent of any sources of reference from outside. No question about its role
for military aviation, marine navigation and for missiles! A high inertial technology is nowadays a trademark for military
independence of many countries. Inertial navigation is also widespread in use in civil aviation and in space flight. Any large civil
aircraft is equipped with tvm or three inertial navigation reference systems. The advent of the "Global Navigation Satellite System
(GNSS)" will in principle not change this situation in the time to come, especially under the consideration that a solution for its
undisturbed availability in times of strained political situations for the countries running the system is not yet in sight.
Since the whole chain of information on the vehicle motion is contained in the INS output, these systems are also used as
measuring instruments in flight tests and in other cases where accurate angle, velocity and position measurements have to be
carried out in a difficult dynamic environment. Inertial sensors and systems are used above, on and under the ground. Examples
are land surveying, borehole measurements, pipeline inspection in the Arctic and inspection of the tracks for highspeed trains and
automobile tests, to name only a few.
Recent achievements in the high accuracy flight tests of "Microwave Landing Systems (MU)"[Hu 741 and the high accuracy
landing guidance of aircraft [Ja 901 are based on outputs from inertial systems which, together with outputs from other sensors,
were combined into optimal information on attitude, velocity and position.
This draws attention to one aspect which, among others, should be one of the conclusions from this contribution. Like all technical
instruments an INS has its specific error behaviour which is, however, of different nature compared to those of satellite and radio
aids (e.g. GNSS, VOR/DME, TACAN, radar, laser tracker, Doppler radar, U).An optimum of information can be obtained by
using information from different sources. In this respect mathematicians have provided technicians with a fairly efficient tool in
the form of the Kalman filter algorithms [Ka 601. The above mentioned achievements in precision navigation and flight testing
have been obtained through the contribution of engineers in the form of hardware and of mathematicians in the form of software.
The advantage of blending INS data with those of different sources lies in the fact that the levels of information on angular rate,
acceleration, attitude, heading, ground speed and position are connected in the system in a mathematically strict sense as to be
seen from Figs. 1.1 and 1.2. Aiding the system by means of external references on the velocity level (e.g. Doppler radar) or
position level (e.g. GNSS, VOR/DME), also increases the accuracy for the other levels. In Fig. 1.3 it is shown that the Kalman
filter estimates the systematic INS errors down to the levels of gyro drift and accelerometer bias, i.e. based on the external
measurements, an in- flight INS calibration is carried out by this algorithm. After a sufficient calibration time periods of missing
external measurements are bridged with only slowly growing system errors. This is the reason why systems for high-accuracy
simultaneous measurements of velocity and position or attitude and velocity, or even the three states together, are always
centered around an INS. It is also advantageous to use an INS in those cases when the high-accuracy measurement of position is a
main purpose of the test as indicated above with the example of MLS testing.
This chapter is arranged in the following train of thoughts. The directional reference in an INS is explained with the mechanical
gyro and stabilized platform as examples. It is symbolized by the weathercock in the functional diagram for a platform INS in Fig.
1.1. The characteristics of the gyrostabilized platform as directional reference serve to visualize the characteristics of the
"analytic platform" in the navigational computer of modern strapdown systems. Again this is symbolized by a weathercock in Fig.
1.2, the functional diagram for a strapdown system. Directional references and their error characteristics are discussed in Section
2.
Accelerometers as sensors for measuring the translational motion are fairly simple instruments in principle, but the formula for
their output signal on the rotating earth is lengthy and it is the basis for programming the navigational computer of an INS. These
aspects are treated in Section 3.
The integration of the accelerometer signal to ground speed and position and the control or computation of the directional
reference is subject of the navigational computer. The interlinking of all signals within an INS causes error characteristics more
benign than we would expect from our school learning. This is subject of Section 4. Also the INS for worldwide navigation, common
features and differences of all mechanizations are discussed. The goal is to derive the error model appropriate for integrating the
INS with the other sensors and systems discussed in this book.
The main part of this chapter contains only general outline which are essential from the system point of view. Special features as
coordinate systems for inertial navigation supplementing the chapter "Navigation Coordinate Systems", digital data processing Of
inertial signals especially in strapdown systems, and optical gyros are treated in the appendices which are named correspondingly
with C, D and 0.
An excellent textbook about the material covered in this chapter was written by the author's teacher and friend at MIT, the kite
Kenneth R Britting, Sc. D [Br 711.
46
---------Navigation Computer
h
h
f rm
altimeter
horizontal and
p o i it ion
(:$
groundspeed
attitude,
heading
Effects of Feedback Loops:
Foucault modulated Schuler o s c i l l a t i o n ,
T=
04 min
24 hour o s c i l l a t i o n
Fig. 1 . 1 Functional Diagram of an Inertial Navigation System in Platform Mechanization
llaviqation Computer
Effects of Feedback Loops:
Foucault modulated Schuler oscillation. T = 84 m i "
49
\f;l
acceleration
24 hour oscillation
F i g . 1 . 4 . 8 Functional Oiaqram of an I n e r t i a l Navigation System (Strapdown System)
Fig. 1.2 Functional Diagram of an Inertial Navigation System in Strapdown Mechanization
47
INS
Optimal'.Estimates on
Angular Rate
Gyro Drift
Acceleration
Accelerometer Bias
P
Ground S eed
>
Kalman
Attitude-, Heading Error
Filter
Ground Soeed Error

I
1
Position
Position Error
Position
Reference
INS Aiding - In-flight INS Calibration
Fig. 1.3 Aiding an Inertial Navigation System on the Position Level with the Use of a Kalman Filter
Literature Chapter 1
[Br 711 Britting, K. R : "Inertial Navigation Systems Analysis."

Wiley-Interscience, New York, 1971
[Gi 71) Gievers, J.: "Erinnerungen an Kreiselgerate."
Jahrbuch 1971 der DGLR page 263 - 291
[Hu 74) Hurrass, R, Stieler, B.: "Zum Einsatz des hybriden FlugmeDsystems bei der Vermessung eines Mikrowellenlandesystems."
Proceedings of the DGON Symposium Gyrotechnology, Braunschweig 1976
[Ja 901 Jakob, Th.: "Integrated System for Automatic Landing Using Differential GPS and Inertial Measurement Unit."
Proceedings of Second International Symposium on Precise Positioning with the Global Positioning System, Ottawa, Canada 1990
[Ka 601 Kalman, RE.: "A New Approach to Linear Filtering and Prediction Problems".
Journal of Basic Engineering (ASME), Vol. 82 D, pp. 35-45, 1960
[Ma 901 MacKenzie, D.: "Inventing Accuracy - A Historical Sociology of Nuclear Missile Guidance."
MIT Press, Cambridge, Mass., USA, 1990
[Sc 621 Schuler, M.: "Die geschichtliche Entwicklung des Kreiselkompasses in Deutschland." VDI-Zeitschrift, Vol. 104, No. 11, pp.
469 - 508, 1962
[So 761 Sorg, H.: "From Serson to Draper - Two Centuries of Gyroscopic Development."
Proceedings of the International Navigational Congress 1976, Boston, Mass., USA
48
2. Directional References
2.1 T h e Mechanical Gyro a s Directional Reference
Mechanical gyros and gyro-stabilized platforms are the classical directional references in moving vehicles. Once their functioning
and their error behaviour is well understood, it is not a major step to see the common features and differences to strapdown
systems whose directional reference is called "analytic platform".
We will look at the mechanical gyro and platform in this chapter from the system point of view only, i.e. exclude major design
specifics, high frequency characteristics and other details. The reader interested in these details is referred to the literature v r
69, De 70, St 821.
The rotating wheel is the sensing element of the mechanical gyro. To describe its characteristics mathematically, we define the
angular momentum vector in the gyro element-fixed coordinate system shown in Fig. 2.1:
This vector is dominated by H, , the rotor's angular momentum about the spin axis, whose magnitude is estimated from:
(2. 2)
H = 1.w
with I = rotor's moment of inertia about the spin axis, which for a ring and the measures of Fig. 2. 2 is:
(2. 3)
I = (1/8)-(m, dt - mi d: ) = 416 g cm2
With the rotor's angular rate about the spin axis:

(2. 4)
w = 2 n f and f
= 400 Hz, for platform gyros
we obtain for the rotor's angular momentum magnitude:
(2. -5)
H = 1.046. lo6 g cm2 s-l

= 1.046. lo6 dyn cm s
1.046 IO-' N m s.
Gyros mounted on a platform of an inertial navigation system (INS) have an angular momentum of this magnitude, since they
stabilize the platform and rotate with respect to inertial space very slowly about their sensitive axes, i.e. the axes perpendicular to
the spin axis. Gyros in a strapdown INS are hardmounted to the aircraft and and have to be tor ed about these axes
corresponding to the aircraft maneuver. Since the electric power for torquing is proportional to H , their rotor angular
momentum is lower by 1 order of magnitude.
Y '
Because of the high magnitude of the rotor's angular rate and momentum about its spin axis, we can neglect the maneuver
dependent components and approximate the total angular momentum vector by:
(2. 6)
= (0
H)T.
According to the Theorem of Coriolis [Wr 691 Newton's Law reads:
El
wig
x .H
main contributor
"technical gyro
equation"
causing interior
gyro dynamics
"nutation" for instance.
The z-component of this equation is of no interest, since it expresses: motor torque E friction torque.
In the low frequency range the gyro characteristics are described solely by the "Technical Gyro Equation":
(2.8a)
M E w x H ,
and in components:
49
G y r o Element G i m b a l
Rotor
Rotor Drive
Spin Axis
Figure 2.1 Mechanical Gyro Notation
Figure 2.2 Gyro Rotor Cross Section
50
The Technical Gyro Equation can be interpreted by the "Rule of Spinvector Alignment" illustrated in Fig. 2.3.
'Rule of Spinvector Alignment"
T h e gyro r e a c t s t o a disturbance vector M or o in such a way t h a t i t s angular momentum vector
will align itself with t h e disturbance vector M or U following t h e s h o r t e s t path
H applied:
w applied:
D i r e c t i o n o f response angular r a t e :
H vector i s "hunting" the i n p u t
torque vector! caused by mass m
D i r e c t i o n o f response toraue:
a c t s so as t o take the H vector by the shortest
way towards the input aiigular r a t e vector
Figure 2.3 Illustration of the "Rule of Spin Vector Alignment"

The setup in Fig. 2.3 right can in principle be dirctly used as a directional reference. This is the case for lower quality cockpit
instruments and for vertical gyros (VG) and directional gyros (DG)) in air and ground vehicles and missiles. We will come back to
these instruments.
For the derivation of the block diagram of Fig. 2.4a for such an instrument we introduce in Eq. 2.8b:
(2.9)
6 ig
with Big = the spin vector's angle with respect to the inertial reference. For small angles Oig can be regarded as a vector.
The torques on the gyro rotor comprise:
- the command torque M' a lied through the gyro torquer mounted at the gimbal axes and
- the disturbance torque MfPdiscussed below.
Instead of the torques we introduce the corresponding angular rates due to the "Technical Gyro Equation":
(2.lOa,b,c)
M d = d x H ,
M' = o' x H,
M = M ' + M d
with w ' = command rate vector and d = gyro drift vector. Bearing this in mind we may derive from Eqs. 2.8 and 2.9:
(2.11)
dig
0'
+ d.
The corresponding block diagram of the gyro as directional reference with respect to inertial space is shown in Fig. 2.4b.
The inertial-fixed reference direction is now replaced by the reference direction in question (index r). For navigational purposes i t
is the earth-fixed reference direction (index r = n) with its components x = north (N), y = east (E) and z = down (D) (s.
Appendix C, Fig. Cl). For missile control the reference direction often is the line of sight (LOS, index r = I). For the
corresponding block diagram we split up oIg:
(2.12)
with
olg
Laer
T'T
org
earth rate
transport rate
misalignment rate =
(s. Appendix C, Eqs. C 7 - C 9 and C 19).
We thus obtain for the rate of change of the misalignment of the gyro spin axis with respect to the reference direction:
(2.13)
+ d -
mir.
51
'
*
acceleratio
-b
I
I
attitude,
heading
Figure 2.4 Block Diagrams of the Mechanical Gyro a s Directional Reference with Respect to Inertial Space (a,b),
with respect to North, East and Down (c) and Functional Diagram of the Directional Reference on a Moving Base (d).
52
The block diagram illustrating this equation, i.e. the use of a PO as directional reference on the earth is shown in Fig. 2.4~.We
have combined @ I r = U'" + onrwith ole = earth rate and o = reference direction rate with respect to the earth.. The small
has its components with respect to north, east and down ( S . Appendix C, Fig. C 4).
angle misalignment vector e = (eN cE
The gyroscopic directional reference with respect to the navigational frame on a moving base is finally illustrated in Fig. 2.4 d.
The integrator with its 3 input signal vectors is the representation of the gyro - or the gyro-stabilized platform as we will in the
next section. For proper functioning (eng-+0) this input has to be kept close to zero as indicated in this figure. It also shows as
dotted lines the mechanical linking of the gyro to the base via the gimbal axes. Synchros mounted to the gimbal axes allow to
measure attitude and heading of the base (aircraft, missile) with respect to this reference direction.
With the torquing signal o t in Fig. 2 . 4 ~being zero, d t cain cause the gyro spin vector to deviate from its initial direction. This is
also true for U' = d = 0 and for a geostationary gyro with,its spin vector initially slightly misaligned with respect to the earth axis
and as shown in Fig.2.5. The component: of earth rate ole in t h e gyro coordinate frame (index g) cause this sensor to carry out
the motion expressed through Eq. 2.13: E = - oICand illzstrated in this figure. We will come back to the analysis of this motion
further below when we discuss the gyro-stabilized blatform.
I
Figure 2.5 Motion of the Geostationary Gyro on the Rotating Earth
For inexpensive directional references of lower quality as the vertical gyro (VG) and the directional gyro (DG) the excursion of
the misalignment angle with respect to the reference plumb line or to the reference magnetic north is measured by additional
sensors and is driven close to zero by means of control loops to the gyro torquer as indicated in Fig. 2.6. Fig. 2 . 4 ~may be used
for the derivation of the block diagrams in Fig. 2.7 where instead of the currents fed into the torquers the corresponding
command rate vectors U' are shown. In the case of the vertical gyro (VG) bubble levels sense the horizontal misalignment
components ch. In the case of the directional gyro (DG) a magnetic compass (flux valve, flux gate) measures eD. Though the long
term directional reference is provided by the additional sensors, they cannot be used without a gyro. The reference direction
sensors are too much affected by the aircraft maneuvers. The gyro's task is to avarage the high frequency components of the
sensor's measurements. The following torquing rule applies for these instruments:
(2.14a)
of
- e.
In the case of an INS the misalignment angles are kept close to zero by the following torquing rule:
(2.14b)
Ut
i.e. by the computed s u m of earth rate and transport rate (s. Section 4.2).
The so-called "Attitude and Heading Reference System (AHRS)"is settled in its accuracy as directional reference between the
VG/DG and the INS. They are mechanized like an INS but equipped with inertial sensors of lower quality. For high quality
attitude and heading indication they are aided by external sensors; in their most simple realization they are slewed to magnetic
north by means of a flux valve, for instance, as it is done in the DG.We will come back to the AHRS in Section 6. Table 2.1 shows
the achievable accuracy for attitude and heading, i.e. the uncertainty range of the misalignment angles eN,E,Dfor all three
applications.
VG
horizontal misalignment
(plumb-line error)
vertical misalignment
(heading error)
eE
fdcgI
DG
.4l)
AHRS
.25
3+6a
1t6u
INS
.os
.4
') unaccelerated flight; 6u = deviation t variation; deviation = difference between magnetic north and true north variation =
difference between indicated north and magnetic north (s. Chapter "Magnetic Heading References").
Table 2.1 Misalignment Errors of Three Types of Direction References
53
synchros
rotor
bubble
levels
heading
synchro
torquers
Principles of
remote magnetic
compass
DirectionaI Gyro
Vertical Gyro
Figure 2.6 The Two-Degree-of-Freedom (TDF) Gyro for U s e a s Directional Instruments
vertical gyro
V h
.
accelero-meter
K,
directional gyro
(index h = horizontal)
Figure 2.7 Control Loops for the Directional Gyro (CG) and the Vertical Gyro (VG)
54
2.2 T h e Gyro-Stabilized Platform a s Directional Reference
As already mentioned, the VG's and DG's as fairly simple directional references are characterized by the fact that the gyro rotor
has a high angular freedom with respect to the case - the angle between the rotor and the case being the aircraft's attitude or
heading angle, for instance. Mechanical gyros for platform stabilization or angular rate measurements, shown in their principles in
Fig. 2.8, have a very low angular freedom. The "DynamicallyTuned Gyro (DTG)" [St 821 is the modern version of such a sensor.
As to be seen from Fig. 2.8, one "Two-Degree-of-Freedom Gyro (TDF gyro)" can be used for stabilizing two axes.
The mechanical gyro for platform stabilization is a nullsensor. The angle Elcg = 0 (pickoff angle) between the rotor and the case
(subscript c) is closely held to zero by means of a control 1.00~for properly slewing t h e case and platform. For the derivation of
the block diagram for the TDF gyro in Fig. 2.9 we split up w'g in the Technical Gyro Equation 2.8 :
(2.15)
= @ic +
+ 6cg
and solve it for the components of 6. We introduce wiC = o for the input rate of the case with respect to inertial space. Due to
the transfer function to be seen in Fig. 2.9:
the TDF gyro is often called "free rotor gyro (FRG)". We have put in this relationship p =
o dt.
As to be seen from Fig. 2.10, the control loops for stabilizing the two axes of a platform connect the gyro pickoff output signals 0
via electronic networks F to the servo motors SM mounted on the gimbal axes parallel to the pickoffs. Since both axes of the TDF
gyro can be regarded as uncoupled (s. Fig. 2.9), the block diagram for only one axis is shown in Fig. 2.11. We will use this for
discussing the characteristics of the gyro-stabilized platform.
Three major subsystems are to be seen in this block diagram: the gyro, the electronics and the platform. The input into the latter
one is the output current from the electronics. It is converted into a torque in the servo motor with the gain StM:The torquer
time lag has been neglected. Disturbance torques M Y acting on t h e platform are due to friction in the gimbal bearings or due to
mass unbalance of the platform, for instance. Both torques move the platform against its inertia. The platform output angular
rate o ' p with respect to inertial space is sensed by the gyro.
From the system's point of view we are only interested in the low frequency response of the gyro-stabilized platform, for which the
inputs into the integrators of the block diagram must be zero.
At first we assume a disturbance torque M Y is acting on the platform and no command or disturbance torques M', Md are acting
on the gyro. Since for quasistationary conditions the input into the gyro as integrator must remain zero, this disturbance torque is
compensated by a servomotor torque i; St', with the current generated by the control loop.
With similar arguments we can see, that a in the low frequency range command or disturbance torques acting on the gyro rotor
are compensated by a gyroscopic torque 0; H, i.e. by a corresponding rotation of the platform. Making use of Eq. 2.10, we obtain
as Performance Equation of t h e Gyro-Stabilized Platform:
(2.17)
o'p
0'
which, in principle, is identical to Eq. 2.11 for the free gyro. This is not surprising, since the stabilization loop slews the platform
with respect to the rotors of the 3 orthogonal gyros and t h e gyro-stabilized platform can be regarded as an ideally supported 3axis gyro. Figs. 2.4 b to d can thus also be used for the representation of t h e gyro-stabilized platform in a block diagram or a
functional diagram.
Stabilization of a 3-axis platform requires the use of 3 SDF gyros, as shown in Fig. 2.12 or 2 TDF gyros. In the latter case one
gyro axis is redundant and often is used for fault detection, i.e. for checking whether the platform truly moves according to the
command rate vector applied to the 3 gyro axes.
Fig. 2.12 shows 2 platform mechanizations, the "three- gimbal platform (TGP)" and the "four-gimbal platform (FGP)". The latter
is a prerequisite for flight maneuvers, when the aircraft attitude passes a 90 pitch angle.
The definition of the Euler angles for roll, pitch and yaw for describing the aircraft's attitude and heading is given in Appendix C ,
Fig. C 2. The TGP is mechanized in a way, that these angles are directly measured at the gimbal axes via the resolvers RR, PR
and YR. From this drawing and Appendix C, Fig. C 2 it can be deduced also, that at the pitch angle 0 = 90 the roll and yaw
axes are parallel and the TGP has lost one degree of freedom. This state is called "gimbal lock".
The FGP is a TGP mounted in an additional outer gimbal OG', but with its roll and pitch axes transposed. The 3 innermost
gimbals are gyro- stabilized as in the TGP, but the outer roll servo motor is controlled by the angle between the inner gimbal IG
and the former outer gimbal OG. Its task is to keep the angle $' = 0 in normal flight. If the aircraft passes the critical region
where gimbal lock w u l d occur with the TGP, the control loop from the inner roll angle +' to the outer servo motor changes its
sign and becomes unstable. This causes the outer gimbal OG' to seek another equilibrium which is 180 away from the initial one.
The FGP thus trades the "gimbal lock" problem for the "gimbal flip" problem which requires a servo loop fast enough so that the
gyros do not hit their stops during the maneuver of the aircraft. In the Litton LN-3 Inertial Navigation System (INS) - mounted in
the Starfighter and the Phantom Fighters - the gimbal flip requires less than .3 s to achieve 90 of the required 180 change. For
more details see [St 821.
2.3 Misalignment Kinematics of an Earth-Referenced Gyro-Stabilized Platform
Since Eq. 2.17 is identical in principle with the characteristics of the free gyro, we can use the block diagram Fig 2 . 4 ~and Eq. 2.13
for the earth-referenced gyro-stabilized platform, too, and we can write:
(2.18)
E,
0'
d - cain
P
Case
Case
Torque
Generator
Torque Generator
Signal
Generator
Torque Generator
\\
Elastic Restrajnt
Gyro Element Gimbal
Damper
Signal
Generator
Signal Generator
Outer Gimbal
Figure 2.8 The Single-Degree-of-Freedom (SDF) and Two-Degree-of-Freedom Gyros in Principle
0,
>
'"
>
-1
-
ox
>
ex >
M,'
Figure 2.9 Block Diagram of the Two-Degree-of-Freedom Gyro, also Called Free Rotor Gyro
56
Figure 2.10 Stabilization of a Two-Axis Platform with a Two-Degree-of-Freedom Gyro
Figure 2.11 Block Diagram for the Platform Stabilization by Means of a Two-Degree-of-Freedom Gyro
(one axis shown only)
57
NA, EA, VA
North, East, Vertical

Accelerometer
NG, EG,VG
North, East, Vertical

Gyro
RR, PR , YR = Roll, Pitch, Yaw Resolver
t
Yaw Axis
SM
= Gimbal Servo Motors
= Platform
OG
= Outer Gimbal
IG
= Inner Gimbal
Roll A x i s
Yaw Axis
Figure 2.12 Schematic View of a Three-Gimbal Platform (TGP) and a Four-Gimbal Platform (FGP)
58
The vectors e' and d do not need the index p, since it is quite obvious that they are acting in the platform coordinate frame. The
term wln is known in the navigational coordinate frame only. With the platform misalignment angles E = (eN e,
we
may transform it into the platform frame (s. Appendix C, Eq. C 19 and C 20):
(2.19)
win
x wfp.
win t E,
For keeping E , to zero, i.e. the gyro-stabilized platform aligned with north, east and down, the right hand side of Eq. 2.18 must
vanish, i.e. the gyros must in theory be torqued with a'* - d. In practice this is never the case exactly and one has to live with
deficiencies as indicated in Fig. 2.4 d by the asterix. Th&e are firstly daily variations of the gyro drift. Secondly one has to rely on
values for earth rate and transport rate compensation computed in the navigational computer and applied to the gyros as
feedback signals in the so-called. "Schuler feedback loop" (s. Section 4.2.2). They are corrupted by velocity and position errors
resulting in in the error term 6 ~ " .The platform slewing is also affected by the gyro scalefactor error IC and gyro reference axes
alignment errors (aii = misalignment of the ith sensor reference with respect to the jtb platform axis). The latter are comprised
in the following matrut:
(2.20)
We thus obtain for the true platform slewing rate:
at =
(2.21)
(I t K ) - ( a i e
:
)
'a
= (I
K ) . ( u " t 6ai"), E win t 6a:
+Kahn.
The gyro errors due to scalefactor and input axis alignment errors (elements of K) and to drift d in Eq. 2.18 are combined to:
K . ( a i C t men), t d
K - a ' t d,
whereby Kwen generates roughly speaking groundspeed-dependent and Kw ie + d time-dependent platform misalignment errors.
The effect oPK(ale t wen), is more dominant during easterly than westerlyhights.
(2.22a,b)
6wg =
=
The variations of the platform misalignments with respect to north, east and down over time thus are governed by the following
differential vector equation:
(2.23)
6,
e, x wfp tgw;
t hag.
Assuming the geostationary case (ain= ai'), perfect platform slewing (bin
= 0) and gyro bias drift only (608 = D), we obtain
the following relationship:
(2.24)
de t
0'
- wiC t D = R
cos4
cD cos4
- sin4
E,
t eN
e,
sin4
sin4
cos4
wL
D,
which is shown in Fig. 2.13 as block diagram. All 3 platform axes are coupled within an oscillation network. Oscillations are excited
by initial misalignment angles, for instance.
The following differential equation can be derived for the input of the east-west axis:
(2.25)
..
R cos4 S e E (-R cos$) dt
R sin4
/eE
(-R sin#) dt
t e . ~ 2(cos24
t sin241 = o ,
E
E
wllich shows that the frequency of this oscillating network is equal to earth rate ( U = Q). This should not surprise us if we
remember that a gyro and thus the gyro-stabilized platform keeps its attitude with respect to inertial space, for zero input driving
functions. On the rotating earth it carries out a motion in the negative sense of the earth rate, as indicated in Fig. 2.5.
(2.26)
2.4 T h e Analytic Platform as Directional R e f e r e n c e in Strapdown Systems
In strapdown computers the reference direction is computed in form of the Crib matrix - a 3x3 matrix for vector transformation
from the body (index b) to the navigational (index n) frame, with its axes north (x = N), east (y = E) and down (z = D, s.
Appendix C, Eq. ClO). With the computed transformation matrix C, the acceleration f,measured in the body-fixed coordinate
system can be transformed to the navigational coordinate system accorkng to:
(2.27)
f,
c,, .
f,
The C,, matrix thus is the directional reference for strapdown inertial navigation. It can be regarded as the "Analytic Platform"
as indicated by the weathercocks in Fig. 1.2 in comparison with Fig. 1.1. There is no difference in principle between the
acceleration and the computational process of inertial navigation following the weathercocks, i.e the directional references in
strapdown and platform systems - regardless of the aircraft maneuver, the computed vertical acceleration will always point
downwards in a strapdown system, too! From Appendix C, Eq. C11 we also see that the C,, matrix allows to compute attitude and
heading of the vehicle as already indicated in Fig. 1.2.
These considerations give an indication of the allowable overall drift of the reference direction computation. Its numerical
integration must be of "inertial quality", i.e. its uncertainty < < .01 deg/h. More details about the problems for computing this
matrix are discussed in Appendix D.
Now we turn only to the principles of the C,, matrk computation. It can be computed from the matrix differential equation:
59
1
Qsi ncp
-Qsincp
+Rsincp
Figure 2.13 Kinematics of the Earth-Referenced Geostationary Gyro-Stabilized Platform
60
(2.28)
c,,.
C,,
R"b,
where
This skew spmetric R matrix can be regarded as the 6q x operator, i.e its matrix-vector product is identical to the vector cross
product. R: is the result of the aircraft angular rate akb measured by the gyros in the body frame b and the Schuler feedback
:
a computed from earth and transport rate in the navigational frame n:
(2.30)
RP
n;b
- c , , .n;-c, ,.
This relationship will be used in the error analysis further below. In practice the Cnbmatrix is mostly derived from the product of
the 2 matrices:
c,,
(2.31)
= Cni.Cib
CT, ' C i b ,
i.e. the transformation matrices from the navigational to the inertial frame and from the inertial to the body frame. Again it
should be mentioned that the former changes with earth and transport rate, i.e. very slow with regard to time and the latter with
the aircraft maneuver, i.e. very fast. Its computation is most demanding as regards computer speed and algorithms. This is
discussed in more detail in Appendix D.
Assuming the first order integration algorithm would deliver sufficient accuracy, we may write for the time increment t i
tn
Cib (t,)
(2.32)
= Cib(tn-l)
+ J Cib(r)-a t ( r ) d r
Ci, (tn-l )
(I
A*:),
'n-1
with the angular increment matrix
(2.33)
A+:
Rib
b
At
:.
-A#z
$z:
0
-A$y A # x
]I!:-
ib
A#;
X.
The elements of A9Lb in this first order algorithm are the angle increments A # p of the mechanical or optical gyros. They deliver
a pulse train proportional to the input angular rate component wlb, whereby each pulse has an identical weight A#p (s.
Appendices D and 0).The pulse train is counted in an up/down counter and read out at sampling times TSto deliver the integral
of the measured angular rate over the sampling interval. Written as a vector it is:
I, tr'
A # i b = Saib dt.
(2.34)
'n
For the moment we have assumed aib to be without an error.
The integration of C in in Eq. 2.31 is carried out similarly but does not cause major numerical problems, since the elements of Rin
contain the sum of earth rate and transport rate only. The latter is the so- called "Schuler feedback" discussed in Chapter 4.2.
2.5 Misalignment Kinematics of t h e Directional Reference in Platform and Strapdown Systems
Up to now we have discussed the computational problems only. For any gyroscopic measurement of uibdeficiencies 6aibdue to
scalefactor errors and drift (s. Appendix 0 for optical gyros or the cited references for mechanical gyros) and due to the
digitizing of analog signals (s. Appendix D) have to be taken into account:
(2.3Sa,b)
aib*= aib + 6 a i b = (I
6oib = 6 o g = K a i b + d,
+ K)*(aib + d)
(I + K ) . o i b
with the scalefactor and input axes misalignment matrix K as shown in Eq. 2.20 and d = gyro drift vector. The superscript
refers more distinctly to a gyro error.
'g"
After integration, i.e. on the level of the directional reference, K causes maneuver-dependent errors and d time-dependent
errors.
The rest of this section is devoted to the discussion of the common features of the gyro-stabilized and the analytic platforms with
respect to their error characteristics. Once this has been achieved, we have found an essential basis for the common features of
the navigational system error propagation within platform and strapdown systems.
The subject of the following train of thoughts is the error propagation of the C,, matrix, whose computation is based on Eq. 2.28.
Instead of the exact matrix C,, we introduce the erroneous one:
(2.36)
where C
,
.
,
Ci
,=
C
,
.
;
= (I
- E)
(s. Appendix C,
C,, = (I
- E). C,,,
Eq. C 20) comprises the small angle deviation elements cN, E,.+ and
tD
between the
61
computed (index n*) and the true (index n) navigational reference coordinate frame. The time derivative of C;, can either be
computed from C:, = (I - E) . C,,:
(2.37a)
C:,
E . C,,
( I - E)*C,,
E . C,,
( I - E).Cn,*
QEb,
or it can be computed according to Eq. 2.28:

(2.37b)
C:,
CO',
Robb*
(I
E).C,,-
Robb*,
QT* and
the
into the
The angular rate matrix QEbr is according to Eq. 2.30 computed from the difference of, the gyro measurements
Due to velocity and position errors the latter has an error of 6Qf. The transformation of Q*:
Schuler feedback
body frame is based on the computed Cf matrix:
Qr'.
(2.38)
QF*
c;, .
C,;
(Q?
(I
6QF)
(
E)
Cib
QF +
)
:
2
5
6
( I - E)
. Cnw
The gyro measurement matrix Q*:

= Q: + 6Qg is already given in the body frame. According to Eq. 2.35 it is in error by drift,
scalefactor error and misalignment. All are comprised in the skew symmetric matrix:
(2.39)
6Qg
-7'
- 6 ~
6wX
y"]
Subtracting Eq. 2.37a and 2.3% and using the relationship C,;
equation:
Qb-
Cbn= Q , we obtain the following matrix differential
This can be written as a vector differential equation:

(2.40b)
L,
E,
cd;
+ sed; +
Cn,*6cd~
This equation is identical in principle to Eq. (2.23) for the gyro-stabilized platform. It differs insofar only as the gyro drift affects
the analytic directional reference computation via the C,, matrix; in a strapdown system the gyros are "strapped down" to the
vehicle, i.e. they are moving with respect to north, east and down according to the aircraft maneuver. This is visualized in Fig.
2.14; in a 360 deg turn the effects of constant sensor errors will thus partially cancel out. During this maneuver the error
propagation of the strapdown system is similar to the one of Carousel INS. This is a platform system with continuous rotation of
the horizontal sensors about the vertical axis for avaraging out the effect of their errors.
It is interesting to note that the directional accuracy of the strapdown system does not depend directly upon the aircraft maneuver only indirectly through the measurement errors d which contain terms due to gyro scalefactor error and input axis misalignment.
Since the C,, matrix is the only term in Eq. 2.40b hinting at the gyro axes coordinate frame, we may generalize its use for all
inertial system mechanizations navigating with respect to the navigational coordinate frame. The index b is then replaced by the
gyro measurement axes coordinate frame m:
(2.41~)
E,
win
6cdF
cn;6wg
and we may write:

- for the "North Indicating System (NIS)", i.e. the platform system with its axes pointing north, east and down:
m = n, i.e. C,,
I,
(the system error dynamics derived in Section 4.2.2 are characteristic for all the subsequent system mechanizations);
- for the platform system measuring in the wander azimuth coordinate frame:
m = a;
(in Section 4.3 it is shown that this mechanization avoids the singularities of the NIS at the geographic poles);
- for the platform system measuring in a coordinate frame which is continuously rotating about the vertical axis (Carousel
System):
m = c
(by means of the "carouseling" the effect of the horizontal sensor errors upon the navigational accuracy should be reduced);
- for the platform system measuring in the inertial coordinate frame:
m = i;
(electrostatic gyros are operated untorqued, i.e. the platform is space-stabilized),
and finally for the strapdown system:

m = b
(this is the modern INS mechanization).
62
/
-'
EN
/
,T\,
I
I
I
\
'\
'\
Figure 2.14 Horizontal Misalignment Error Growth During a Straight Flight with Subsequent Turn in a Strapdown System
Due to a Constant Drift
[De 701 Denhard W.G. (Ed): " Inertial Component Testing: Philosophy and Methods"
AGARDograph 128, AGARD 1970
[St 821 Stieler B., Winter H.: "Gyroscopic Instruments and their Application to Flight Testing"
AGARDograph No. 160, Vol. 15, 1982
wr 691 Wrigley W., Hollister W. M., Denhard W.G.: "Gyroscopic Theory, Design, and Instrumentation"
The M.I.T. Press Cambridge, Mass. USA, 1969
63
3. The Measurement of the Translational Motion
- Accelerometers and Acceleration with Respect to Ground

3.1 T h e Pendulous Accelerometer and i t s Output Signal
The principle of a conventional accelerometer is shown in Fig. 3.1. A proof mass is suspended in a case and confined to a zero
position with the help of a spring or a rebalance loop as described below. In general, damping is added to give the spring/mass
system a proper dynamic transfer function. Aircraft accelerations act upon the accelerometer case and cause the mass to react
with a displacement with respect to the zero position so that the resulting spring force F compensates the acting acceleration. The
displacement of the mass with respect to the case is then proportional to F.
For convenience we introduce the specific spring force f = F/m and call it the accelerometer output signal as indicated in Fig. 3.1.
Not only f but also gravitation G is acting upon the proof mass. Both cause its acceleration with respect to inertial space due to
Newton's second law:
(3.la,b)
f + G
d2R
-
where R is the radius vector from the earth's center of gravity which is the origin of the inertial coordinate system (index i), to
the proof mass' center of gravity.
In some cases it is convenient to combine the right hand side of Eq. 3.lb into a vector a or
acceleration or of gravitation is of prime importance. The specific force vector is then:
(3.2)
or
f = a
f =
g' when either the effect
g'.
The type of accelerometer that is presently used in most operational inertial navigation systems is the restrained pendulum
accelerometer [QF]. Fig. 3.2 shows the construction principle of such a device. A pendulous mass is suspended and restrained to
a zero position by a control loop. Optical, capacitive or inductive pickoffs detect a deflection of the pendulum which is forced back
to its zero position by means of this control loop with an input current into the torquer. This current i that is necessary to
compensate an acceleration and to bring the pendulum back to its null position is then a measure of the specific force f. It is
converted into a voltage U by means of a precision resistor.
The true output signal of an accelerometer contains input signal-dependent errors comprised in the scale factor error
signal-independent errors comprised in b:
(3.3)
f'
= (1
IC)
(a
b) = (1 t IC)a
IC
and input
b.
As soon as both parameters can be modelled - as a function of temperature, for instance - they do not affect the measurement
accuracy. What bothers is the stochastic variation, i.e. the day-to-day repeatability, for instance. This is a true indication of
quality. Threshold is another one.
3.2 T h e Acceleration with Respect to Ground
The second time derivative of the radius vector in Eq. 3.1 is the acceleration of the point P of measurement with respect to the
earth's center of gravity which is the origin 0 of the inertial frame (index i = Earth Centered Inertial Frame (ECI), s. Chapter
"Navigation Coordinate Frames")). It can be expressed in any other reference coordinate frame (index r), for instance the line of
sight (LOS) coordinate frame for missile application or the navigational frame (index n) for navigation with respect to the earth
through application of the Theorem of Coriolis:
d R
d R
d t
(3.4)
dt
(3.Sab)
d2 R
d2 R
z
li xlr
+
Ii
with
-1 d Rt2
T
vertical
d2
a i r x R.
zir
;1
R
2 airx
d R
-
Ix R
2 win x
d t
d R
d win
d t
horizontal
d t
1'
Coiiolis
uir x (a" x
cain x (ainx R)
R)
1'
centrifugal accelerations.
The angular rate uir or ai' are the respective s u m s of earth rate plus transport rate (s. Appendix C, Eq. C 7 to C 9).
Eq. 3.5 simplifies considerably if we introduce the velocity of the point P of measurement with respect to the surface of the
reference ellipsoid of the earth:
(3.6)
which, expressed in the n-frame coordinates, is:
64
el
dt2 i
= acceleration w i t h respect
t o i n e r t i a l frame
f = specific force t o
b a l a n c e p r o o f mass
(spring force)
G = gravitation
Figure 3.1 Forces and Output Signal in an Accelerometer
S I GNAL
0
POWER
0
Figure 3.2 Principles of the Pendulous Accelerometer with Caging Loop
65
I2 1
(RN + h)
[ ( R E + h)
VN
(3.7)
Vn
d
COS
-h
The resulting acceleration of P is:

d2 R
d t 2
v + (2
WiC
v +
Wen)
Oie x
( ai'
R).
The last term on the right hand side is compounded with the gravitation G in Eq. 3.1 to give the gravity g as shown in Fig. 3.3:
(3.9)
uie x (mi'
R)
= G
R.Q2/2
- sin 24
0
1 + cos 24
deflection of the
B
.
RR2
s i n 2 p S f - 6 arc min
Figure 3.3 Gravitation Vector G and Gravity Vector g

So we obtain for the ground acceleration:
(3.10a,b)
(2 w
Wen)
v +
+ g
or in components:
-2 vE P sin
2 V, P sin
-2 v, R' cos
4 + h 6
4 - 2 h R' cos 4
4 - v, 6
where these terms must be compensated in the computer:

The following abbrevations have been used:
(3.10d)
R' =
n +
i/2
and the Coriolis acceleration vector:

(3.10e)
= (2
WiC
Wen
v.
-1
[ -":1,
1'
66
A good example for understanding Coriolis acceleration is to think of a car driving with constant speed V, from the equator to
the northpole. Though not accelerating with respect to the ground an east-west accelerometer will sense an acceleration because
the earths tangential velocity of fl. R = 1668 km/h has to be slowed down to zero at the northpole.
3.3 T h e Accelerometer Output on a Misaligned Directional Reference
In platform systems the acceleration is measured in the platform coordinate frame f = C ,f,, with C the transformation
matrix containing the platform misalignment. According to Appendix C, Eq. C 20 it iscompofed of the smae angles c , , ~ , ~ .In the
strapdown system the acceleration is measured in the body frame and then transformed into the navigational frame n a the C &
matrix which is in error with respect to the accurate one by the small misalignment matrix C,:,. Both misalignment matrices are
identical (s. J2q. C 20). With the same arguments as in Section 2.5 we can make for all inertial navigation systems the following
assumption for the acceleration in the navigational coordinate frame:
(3.11)
6V, =
E,
a,
6 cn + 6 g,
C,;
bm ,
where the index m stands for the coordinate frame, in which the accelerometers are measuring (s. Section 2.5).
[QF]: QA-1000 Series, QA- 1200-Series and QA-1300 Series Q-Flex Accelerometers.
Sundstrand Data Control, Inc. Document 012-0293-001
67
4. Inertial Navigation
Integration of the DirectJonal Reference with
the Translational Motion Measurement
4 . 1 Introduction
In Section 4.2 we will begin the discussion on inertial navigation systems (INS) and their error characteristics based on the
mechanization indicated in Fig 1.1. The platform shown here is always kept horizontal and pointing north regardless of t h e
vehicle's maneuver. Therefore an inertial navigation system (INS)with this characteristic is called "North Indicating System (NIS)"
in the following. Special emphasis is also laid on the vertical channel in which as compared to the horizontal channels the physical
law of mass attraction between two bodies (earth and proof mass of the accelerometer) leads to an instability of the error
growthz. The aiding of the vertical channel with the barometric altitude will be discussed.
The NIS can certainly not be used for passing the geographic poles where all directions are pointing south or north. For worldwide
inertial navigation another mechanization is required. It is the so-called "Wander Azimuth INS" discussed in Section 4.3. Some
words will be said in Section 4.4 about the INS with space-stabilized coordinate frames used on earth and in space. As pointed out
in Section 2.3, the functioning of the mechanical platform can be taken over by an analytic platform computed in the navigation
computer. The data processing following this analytic platform computation is identical to that of the mechanical platform system
as indicated in Fig. 1.2 for the NIS as an example. Section 4.4 deals with this modern INS mechanization, commonly known as
"strapdown systems (SDS)". The characteristics of the error dynamics for all inertial navigation systems constitute the topic of
Section 4.6. The error model derived therefrom can be used in a Kalman filter for the integration of any platform or strapdown
INS navigating in terms of longitude and latitude with external sensors and systems.
As autonomous inertial navigation systems are expensive and other navigational information often lies at the wayside and can be
blended with the inertial information, accuracy requirements on the inertial side can be reduced thus leading also to a reduction of
costs. Inertial systems of this kind are often referred to as "attitude and heading reference systems (AHRS)".They will be
treated in Section 4.7.
4.2
Inertial Navigation with Sensors Aligned t o North, E a s t and Down

- t h e North Indicating System (MS)
4 . 2 . 1 Functioning of t h e N I S
Fig. 1.1 shows the diagram of an N I S which we use to obtain an understanding of the inertial navigation system (INS) functioning.
The platform represented as a round disc in Fig. 1.1 is suspended in the vehicle (airplane, missile, ship) with three degrees of
rotational freedom. For simplification this figure shows only the vertical axis. The rotational freedom as well as the other
indicated signals are understood to have 2 or 3 orthogonal components; 2 components for horizontal acceleration, velocity and
position, 3 components for the angular rate.
On the platform one gyro can be seen - in place of the three gyros - which, with the control loop from gyro pickoff (P) via the
electronics to the servo-motors at the gimbals, ensures the stabilization of the platform.
As we have seen in Section 2.2, the gyro;stabilized platform used as a directional reference with respect to the navigational
frame, follows the performance equation: E = 0' + d - m l n , regardless of the vehicle's maneuver. To keep the misalignment
L )
T < < 1 deg in magnitude, the condition for torquing the 3 gyros 0' =
- d has to be met, i.e.
angle vector E =
with w L n =U ' ' + m'ys.
Appen& C, Eqs. C 8 and C 9) it has to be equal to the s u m of earth rate plus transport rate and drift as
indicated in Fig. 1.1.
The computation of the earth rate and transport rate vectors are dependent on the computed ground speed and position, i.e. on
the measured acceleration vector f'. Eq. 3.10 is the basis for their computation.
Since with the NIS the geographic coordinate frame is stored in the platform, the aircraft attitude and azimuth can directly be
measured at t h e gimbal axes (s. Fig. 1.1).
As already mentioned above, pure inertial navigation can only be carried out in the horizontal plane and Fig 1.1 is valid for this
case only. Navigation in the vertical plane has to be aided by the barometric altitude, as will be discussed in Section 4.2.3.
4 . 2 . 2 T h e E r r o r Dynamics in t h e Horizontal Channels of t h e NIS
4 . 2 . 2 . 1 T h e E r r o r Dynamics of a Single Axis Inertial Navigation System
The signal flow of Fig. 1 . 1 for the two horizontal channels from the accelerometer, via the integrators and the division by the
representative radius of curvature of the earth, to the torquers of the gyros are called Schuler loops. They are the electronic
implementation of a two-axis mathematical pendulum whose length is equal to the radius R of the earth, having the frequency and
period of
(4.2.la,b)
ws =
I/g/R
= 1.235. los3 rad/s,
TS = 84.4 min
which are called "Schuler frequency" and "Schuler period".

A pendulum tuned to this frequency always indicates the vertical on a moving vehicle, once it has been aligned to it prior to the
start. Neither a gyrocompass (s. Section 5) nor an INS tuned to this frequency will b e excited to disturbance oscillations by the
horizontal acceleration. These rules were revealed by Schuler in his publication of 1923 [Sc 231. The Schuler period also governs
the motion of a stone traversing a hole drilled through the earth regardless in direction - friction obviously neglected [Ma 781. It
governs also the motion of a satellite surrounding the earth when it is computed from the orbital radius and gravitation. For a
"roof top" orbit it is again 84 minutes. We will come back to the satelite period in Section 4.4.
68
Neither a mathematical pendulum nor a physical pendulum can b e implemented with this frequency. For the latter the following
relationship would have to be satisfied:
i2/ r = R
(4.2.2)
Assuming i = radius of inertia = 2.5 m, the pivot-to-center-of mass separation of r = 1 pm would have to be manufactured. In
practice the Schuler tuning condition can only be met with the aid of a gyroscope - as in a gyrocompass or the INS.
The simplified single axis INS shown in Fig. 4.2.1 may serve as a basis for the following discussion.
The platform (P) is mounted on a vehicle heading north and gyro-stabilized about its east-west axis (perpendicular to the plane of
the drawing) by means of the servo loop from the gyro (G) signal generator (S) to the servomotor (SM) of the platform. The
accelerometer (A) has its sensitive axis in the direction of motion and its output signal f is coupled to the torquer (T) of the gyro
via the integrator and the amplifier - 1/R V is the north velocity and e the angular deviation with respect to the vertical.
The system's error dynamics are described by the linearized state space equations:
0
with D = gyro and B = accelerometer bias assumed to be constant in the mean. The state vector is:
x
(4.2.4)
6V
(6.5
B)T
The w-terms comprise random Gaussian distributed noise with zero mean. Instead of w; and w&, we have introduced wD and we
to remind of the fact that these terms are caused by gyro and accelerometer noise.
Equation 4.2.3 is of the following form:
x
(4.2.5)
P.x
w.
Its solution has the general form:

II
x(t,)
(4.2.6)
*(to, tII-l). x(tII-l) +
J *(t,,T).W(T)
dT,
'0-1
= state transition matrix. For constant coefficients of the system matrix
with +(tII,
domain:
*(s) = ( s I
(4.2.7)
F the transition matrix is in the Laplace
P)-',
1;
with s = Laplace variable. For zero random driving functions (w = 0) this leads to the following solution of Eq. (4.2.3) in the time
domain:
(4.2.8)
6V
II
1
si/w
R(l - CO)
0
CO
R w si
0 -si/(R w)
CO
0
0
0
0
0
0
(t) =
D
B
R(At - si/w)
R(l - CO)
(1
si/w
1
0
co>/w*
-(lsi!wco),g]
01
We have used here the following abbrevations:

(4.2.9 a,b,c)
At = t
to.
si = sin os At,
CO
cos os At.
Horizontal platform misalignment e, velocity error 6V and position error 6s = R. 6@ due to east-west gyro drift D and northsouth accelerometer bias B are plotted in Fig. 4.2.2. All position errors begin with zero slope, indicating that for short periods of
time the NIS is very accurate.
Distinctly to be seen is the Schuler oscillation on the 3 system levels. All NIS errors are bounded, except for the effect of gyro
drift D on the position error. It confirms the importance of the gyro on the long-term system accuracy. The slope, i.e. the mean
velocity error provides the 'Rule of Thumb for Inertial Navigation':
(4.2.10)
6v
R.D
=)
1 km/h navigational error per D = .01 deg/h
Physically this rule becomes understandable if we consider that t h e gyros keep the reference coordinate system for navigation,
stored in the platform and that the gyro drift causes an analogous drifting of it (as a reminder: 1 degree of longitude = 111 km
at the equator).
It is interesting to note that the effect of accelerometer bias B upon the NIS position error is not as severe as one could expect.
The misalignment error and the position error are alike. The mean error is limited to:
(4.2.11)
e = 6S/R = 6$ = B/g.
69
Fig
4.2.1 Signal
Flow Diagram and Block
Diagram
for the Single Axis Inertial Navigation System
IQ
Ellecl of
levellirq Gyro Orill IO-MlO
Acceletmeler Bios I B * IO'gl
h )
on Alldude
IKnI
0.34
40
20
I I m i n l 60
80
u.4
on Velocily
6V
,' B1
r$
an Position
-closed ktuler - ioop

--- open khuler - loop or novigolion on plow eorlh wilh
-.-
6S, Rbq
meon'value
Fig. 4.2.2 Errors of a Single Axis Inertial Navigation System
71
This means that the accelerometer bias is compensated by a component of gravity due to the mean platform tilt.
For periods that are short with respect to the Schuler period, one can introduce the approximations in Eq. (4.2.8):
(4.2.12a,b)
si
CO
sin as At G us At - (as At)3/6

cos us At Z
1
- (as At)'/2.
This leads to the curves marked in dotted lines in Fig. 4.2.2, which for a longer time period are also valid for the set-up of Fig.
4.2.1 without Schuler loop or for inertial navigation on a flat earth.
Another effect is worth mentioning, the effect of sensor scalefactor errors on the NIS performance. Let us begin with the gyro
scalefactor error Kg ( s . Eq. 2.22). It causes an error in the platform slewing 60' = K g ( o l e + wen) and thus has a similar effect
as the gyro drift. It is bigger for easterly flights when O" and men have the same sign than for westerly flights. The corresponding
drift component about the north-south axis is then:
(4.2.13)
d' =
I C ~a' Z
,002 deg/h
for
I C ~ 9
100 ppm and V = 1000 km/h at the equator.
The assessment of the accelerometer scalefactor error's ( K ~ )effect on system performance yields again a surprising result. It is
active only when the vehicle is accelerating, but not during cruise. Assuming that the take-off acceleration is very short compared
to the Schuler period, we may regard the velocity error 6V = wa.V as an initial velocity error 6V(t,). From Eq. 4.2.15 we obtain
as position error:
(4.2.14a)
6S(t) =
I C ~
(V/us) sin us t,
i.e. it is zero in the mean! Only for short intervals of time compared to the Schuler period of 84 min, say 10 min, or for inertial
navigation on a flat earth the position error is as expected:
(4.2.14b)
6s
d.S.
In summary we may conclude that inertial navigation is only feasable because our Lord has made the earth spherical. The time
exponent of the position error growth is lower by 2 as compared to our expectations from school physics!
It is not surprising that this was not recognized at the beginning of the inertial navigation development which took place in
Germany before and during World War 11. Outsiders did not believe in the feasability of inertial navigation due to their
interpretation of the position error growth with school physics as a basis. The so-called "law of the third power" to which they
t3. It was Reisch who proclaimed inertial navigation as the
referred says that the position error due to gyro drift grows
"principle of plumb line rotation" shown in Fig. 4.2.3 [He SO]. It indicates that position changes can be measured by the inclination
of a space-stabilized platform with respect to a plumb line reference. Gyro drift causes a position error growing linearly with time
in this navigation system which lacks the plumb line reference on a moving vehicle. The INS in Fig. 4.2.1 has the plumb line
reference in form of the Schuler feedback. From today's point of view this was Reisch's valuable contribution to this development,
but one cannot follow him in his demand to be the inventor of inertial navigation as proclaimed again in [Hi 921. This honour is
certainly due to Boykow [Bo 351. The fairly benign INS error characteristics do not depend on the mechanization proposed by
Reisch and before him by Boykow. It governs any INS mechanization as shown in Sections 4.4 and 4.6 and is due to the shape of
the earth.
Figure 4.2.3 Reisch's "Principle of Plumb Line Rotation" for Position Measurement from the
Inclination Measurement of a Space Stabilized Platform
72
So far we have discussed t h e effect of deterministic sensor errors on the NIS performance only. In modern sensors, especially in
the ring laser gyro (RLG), the deterministic errors are fairly small and the effect of stochastic errors on system performance
becomes more dominant (s. Appendix 0). It is also of great importance for system modelling in a Kalman filter, and this will be
the subject of our next discussion.. Tools for estimating their effect are the Monte Carlo simulation on a digital computer and the
covariance analysis for predicting their f 1 U boundaries. The results of the former, shown in Fig. 4.2.4, are based on a simplified
version of Eq 4.2.6:
(4.2.15)
x(tJ
*(tnltn-lIfx(tn-1)
W(fn-J>
with the following state transition matrix for a first order integration:
(4.2.16)
*(tn ,tn-
I t P(tn- 1) -At.
In this simulation gyro noise was investigated with a random walk coefficient of rD = .01 d e g / f i .
From the different sample functions plotted in Fig. 4.2.4 the R M S values were calculated for a fixed time and plotted as dotted
flu-curves. In order to compare the simulation with the theory, we carry out a covariance analysis using the formular [Ge 741:
deterministic propagation of
covariance, i.e. propagation of
uncertainty in knowledge of
system error state between
tn-l and to.
where
added uncertainty due to

integration of
sensor and system noise.
The covariance matrix q specifying the sensor and system noise, has on its main diagonal the squares of the corresponding
random walk coefficients and otherwise zeros. For noise the corresponding elements read on the accelerometer and gyro level (in
Eq. 4.2.3 wB =k 0 and wD +: 0):
(4.2.18a,b)
q22 = (re/60)2
with rB [(m/s)/l/;;-]
and
and rD [deg/
q33 =
(1-~/60)~
fi].
The lo-values for the system uncertainty is the square root of the corresponding element on the main diagonal of the resulting
covariance matrix, for instance:
(4.2.19)
06s = P I 1
Using the state transition matrix from Eq. 4.2.8, we obtain the following result for the system uncertainty propagation due to gyro
noise, i.e. a random walk on the misalignment level:
(4.2.20a,b,c)
uZE
u2hv =
u2hs =
rZD
[t/2 t 1/(4ws) .si2],
r2D R g . [t/2 - 1/(4ws). si21,
r2DR~ . [3. t/2 - 2/w;si
+ 1/(4wS)-si2],
f
with the abbrevations:

(4.2.21a,b)
si = sin(wst) and si2 = sin(2wst).
The results of the computed f 1 U values are plotted in Fig. 4.2.4 as dot-dash curves. They correspond fairly well with the
simulation if it is kept in mind that the +lo bands enclose 68% of the 8 Monte Carlo runs.
In a Kalman filter Eqs. 4.2.15 and 4.2.16 are used for the prediction of the system error state based on the last optimal estimate
(x(tn-l) = %(tn-l)).Since the random driving function w(t) cannot be predicted but its mean is zero per assumption, w(tn.l) has to
be assumed to be zero for the prediction.
4.2.2.2 T h e Error Dynamics of a T h r e e - A x i s Inertial Navigation System
f o r a Short P e r i o d of T i m e ( < 6 hours)
Fig. 4.2.5 shows in simplified form the block diagrams of the error models for a three-axis NIS in the lower half and the
barometric aided altitude channel in the upper half. On the left side the sensor errors are listed and on the right side the system
output errors.
As compared to the complete error model of an MS [Br 711, which will be discussed in Section 4.2.2.3 (s. Fig. 4.2.7), the SOcalled 24-hour oscillatio? caused by the cpupling of the platform angles N,E,D via the components of the earth rate plus t h e
transport rate A (with A = R + X ) or 4 , respectively, has been cut down to the coupling of the azimuth misalignment E,, into
the rotation about the east-west axis. For the purpose of discussion this is justified because on the one hand cD is in general one
order of magnitude higher than the-horizqntal misalignment angles E, E. .On the other hand, for flight periods of up to 2 hours we
can assume with good accuracy: sin A t A t and cos A t S 1 (error in'this approximation < 10 %).
For an estimation of the errors in the MS channels we assume that the slewing of the vertical gyro as well as the compensation of
the Coriolis acceleration can occur without errors and there remain two Schuler loops as in Fig. 4.2.1. In addition to these Schuler
loops the influences of the azimuth error cD as well as of the vertical gyro drift are indicated.
On the one hand the azimuth deviation E~ causes acceleration errors in the horizontal channels which are orthogonal to the
respective components VN,Eof the horizontal acceleration of the vehicle. This is indicated in Fig. 4.2.6a for an acceleration of V,.
13
2.88
epsikrcnin)
-2.ee
5.88
dU (n/s)
-5 .ee
Figure 4.2.4 Errors and f l u Bands of a Single Axis Inertial Navi a 'on System
Due to a Stochastic Gyro Drift (Random Walk of .01 d e g w )
74
I
Sensor
System E r r o r Model
Errors
I
Model Atmosphere
Altimeter
Vertical Accelerometer
6hb
I
I
--2T
-+
I Altimeter
Error
63
I Altitude
I Error
8"
I
I
I Longitude
East Accelerometer
Error
North Gyro
Vertical Gyro
I1 Latitude
Error
I
East Gyro
North
cc e\erometer
Figure 4.2.5 Simplified Block Diagram for the 3-Axis Error Model of a North Indicating INS
Including Vertical Channel
Figure 4.2.6 View upon the Platform Misaligned about t h e Vertical Axis
75
Assuming that the acceleration of the vehicle from start to the constant cruise velocity is short compared to the Schuler period,
the time integral of this cross-track acceleration error is equal to an initial velocity error:
6VN = 6V,(O)
(4.2.23a,b)
~ ~ ( 0. )V,,
6VE(0) = 6V(O)
ED(0)
v,.
On the other. hand, the azimuth error

has an effect comparable to a north-south or east-west gyro drift through the coupling
of the rates Q and A . cos I$ into the respective orthogonal platform axes:
DL
(4.2.24a,b)
D, -
(E,
+ D,. t ) . 4;
Db = DE
t (E,
+ D,.
t). A
.COS
I$
The relationships compiled in Eq. 4.2.25 were obtained in a way similar to the one for the single axis inertial navigation system in
Section 4.2.1; the state space equations of the two uncoupled horizontal channels were supplemented by Eqs. 4.2.23 and 4.2.24 and
cos4 was assumed to be constant.
(4.2.25)
T
si/(Rw)
0
si/(Ro)
-( 1-CO)
CO
0
0
0
0
CO
si/(Ro)
-Ro si
CO
-si/( R o )
0
0
(1 - CO)
R o si
CO
0
0
0
0
-Ri
(AA n/o si)cosQ
-A$
R(A- R co)cos$
0
t - si/@
AA
COS$
(R/o)cosI$ si
1
0
0
t/2-
A /W2)COSQ( 1-CO)
0
0
(l-co)/g
0
We have again made use of the abbrevations in Eq. 4.2.9, i.e. w = us,si = sin wSt and CO = cos wSt. The readers attention is
also drawn to the fact that the vector-matrix arrangement in this equation alleviates the identification of state vector
dependencies.
Of the results we want to discuss only the position error equations and cast a glance at the main sources. The initial position
error (6I$(O) and 6A(O)) can be assumed to be zero, as on any major aerodrome the geographic longitude and latitude are
indicated at the aircraft parking positions. The initial velocity error can also be neglected if the switching of the NIS into the
navigational phase takes place on a stationary aircraft, V(0) is then accurately known.
In comparison to Eq. (4.2.8) essentially only the effects of t h e azimuth alignment error E , and of the vertical gyro drift D, have
been added. It is interesting to note that the deviation with time of the cross-track-error is equal to that of a common dead
reckoning navigational system in the east-west channel (6h(t) cos 4 = - ~ ~ ( 0AI$).
) . In the north-south channel this is true for
short times only when 6Q(t) = ~ ~ ( 0 [AA
) . - (Q/ws). sin w,At]. cos Q Z E , . Ah. cos Q. After approximately 10 minutes, earth
rate coupling becomes effective.
It is this difference in t h e error characteristics that allows in-flight gyrocompassing of a misaligned MS based on a Doppler radar
for measuring the body-fixed ground speed components (s. Chapter Doppler Navigation) and the misaligned platform as heading
reference. We will come back to this topic in Section 6.
If we start from the assumption that before the start the platform has aligned itself in the north direction and in the horizontal
plane as a result of a gyrocompassing procedure, we shall under ideal conditions have the following relationships between the
alignment and sensor errors (s. Section 5):
(4.2.26a,b)
E,,~(O)
~ ~ ( 0z)
f BE,,/g
DE/(Q cos
the numbers being valid for B =

reduced in this case to:
(4.2.27a,b)
6Q(t)
6A(t)
(Z
Q)
(Z
. 1 mrad fi 20 arc sec)
1 mrad
3.4 arc min),
g and D = .01 deg/h as well as Q = 45 deg. The position errors in Eqs. (4.2.25) are
E,(O).S~/R - D, -[AA.t/2
(A/@) cos $I . (1 - cos wSs-t)]
-cD(O).SN/R - D,. [t - sin us t/ws],
76
S,, being the distance flown. For periods that are short in comparison with the Schuler period (6 min approximately for a 10 %
erior) these relationships will be reduced to:
(4.2.28a,b)
6$(t)
~ A ( ~ ) . c o4s E
~ ~ ( 0S,/R
) .
E~(O).S,/R,
which means, that after fully completing self-alignment, the navigation error of the NIS under ideal conditions initially consists
exclusively of the cross-track error, and is identical with that of a normal dead reckoning navigational system with ideal
velocimeters!
Before we discuss in Section 4.6 the NIS horizontal errors over a longer period of time ( > 6h), the INS vertical channel and other
INS mechanizations should be described briefly. Then we are in a better position to understand common features and differences
in their error models as a basis for Kalman filter design.
4.2.3 T h e Error Model f o r t h e V e r t i c a l Channel o f a n Inertial Navigation System
The decoupling of the vertical from the horizontal channels is justified, as will be shown in Section 4.6. Inertial navigation in the
vertical channel consists of a mere double integration with gravity compensation. The latter includes the decrease of gravity as a
function of height due to the physical law of mass attraction between two bodies - the earth and the accelerometer's proof mass.
The block diagram for the vertical channel's error model is shown in Fig. 4.2.5, top, in thin lines. Due to the positive feedback of
the gravity dependence upon height the autonomous inertial navigation in the vertical is unstable with the eigenvalues:
(4.2.29)
A,,2 = k
p = 1/2g/R.
f
They describe the following error growth:

(4.2.30)
6h(t) = 6h(O).cosh At
- B,/A2-(1
- cosh At),
where B, is the accelerometer bias. This error growth is plotted in Fig. 4.2.7.
For navigating over a longer period of time, the vertical channel has to be aided by external measurements, as indicated in Figs.
4.2.5 and 4.2.8 for use in a civil aircraft. The difference between the baroinertial height h and the barometric height hb is fed
serves to
back to the baroinertial velocity and the acceleration via the gains K, and q.The shunting integrator with input gain
compensate quasiconstant acceleration errors due to sensor bias and gravity compensation errors. The gains quoted in this figure
are valid for the Litton LTN 90. Due to the fairly weak barometric coupling, the baroinertial height follows in the low frequency
domain the barometric height and in the high frequency domain the purely inertial height. The most important sideeffect of this
baroinertial coupling is the generation of a signal for the vertical velocity. It can hardly be derived from pure barometric
measurements.
Within a flying aircraft the static barometric pressure measurement is a function of
- the magnitude of the air flow, i.e. the true air speed (TAS) and the Mach number,
the direction of the air flow, i.e. the angles of attack and sideslip,
- and the flaps and throttle settings wu S O ] .

An optimal measurement position for the static pressure at the fuselage is selected during the design phase of the aircraft by
means of windtunnel tests and by calibration flights, the results of which are "optimal" for a certain flight condition only. For other
flight conditions an error model including calibration parameters is layed down in the "air data computer (ADC)" [NM 841.
The static pressure is thus a fairly accurate altitude reference during cruising speed in the subsonic range. It is true that the
aircraft will thus not follow a certain geometric altitude, but the isobars of the atmosphere. This is a sound basis for air traffic
control, since the controller knows that all aircraft will fly in their respective flight levels (FL) which he has assigned them to (FL
= barometric mean sea level altitude [ft]/lOOO above 3000 ft above mean sea level (MSL)).
The calibration of the static pressure for instationary flight conditions and the generation of a reliable reference for the vertical
speed and the height is difficult and limited in its accuracy [Au 911. Out of this reason a separate chapter is devoted to this topic
in this book (s. Chapter "Vertical Channel Design Considerations").
For flight-test with accuracy requirements in the 1 m range it has also to be kept in mind that additional errors are due to the
conversion of the analog pressure measurement into a digital signal and due to the conversion of this digital signal into a
barometric altitude.
The former is affected by the volume of the pressure transducer, for instance, which causes the time lag in the order of
magnitude of 1s [Re 851. The digital data handling within the ADC includes a lowpass filtering with a time constant of the same
order of magnitude [NM 841 again. These time lags T affect the barometric computation during ascent and descent with 6h Z
T-h.
The latter, i.e. the conversion of the pressure into the barometric altitude, depends upon t h e ground pressure and temperature as
well as the outside temperature.
The ground parameters are requested by the pilot from the controller and entered into the ADC when approaching an airport.
Errors in these parameters cause additional barometric altitude bias and scalefactor variations.
If all error parameters mentioned above remained constant during the flight with respect to the runway, they could be modelled as
a bias in an estimation process based on external measurements. Unfortunately this is quite often not the case as the flight test
results in Fig. 4.2.9 demonstrate. They were obtained with the twin engine turboprop aircraft Do 228 of DLR with static pressure
holes in the fuselage. The plotted differences between the barometric altitude obtained from the Nord Micro ADC and the
reference obtained from the DLR "Avionic Flight Evaluation System (AFES)", i.e. a flight path based on microwave radar, laser
tracker and INS, shows a very problematic time characteristic during the approach. Neither can this be removed by feeding the
barometric altitude into the inertial system as indicated in Fig. 4.2.8, nor can this be estimated correctly by a Kalman filter with
IO'
Ah
PI
1-
Ii
x)
Figure 4.2.7 Unstable Error Growth of the Unaided Inertial Vertical Loop
"
I
c)
baroinertial
s
baro
Figure 4.2.8 Barometric Aiding of the Inertial Vertical Loop
78
2200
2000
1800
1600
+
E!
1400
-2
1200
1000
i3
800
600
400
200
10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95
time [min]
500
400
300
n
I$=
200
100
a)
a)
E!
-3
-100
- 200
-300
- 400
__
0
10 15 20 25 30 35 4 0 45 50 55 60 65 70 75 80 85 90 95
time [min]
Figure 4.2.9 Barometric Altitude and Altitude Error of the Twin Engine Aircraft Do 228
During Several Runway Approaches
79
the bias modelled as a state vector component.
The modelling and calibration of the barometric altitude errors during flight maneuvers is a fairly difficult task and their
compensation or estimation accuracy with a Kalman filter is limited [Wi 751. For obtaining a vertical flightpath reference over a
limited period of time, it is advisable to rely on the inertial altitude aided only by external measurements as satellite navigation or
the radar altimeter. This topic and problems connected when using a conventional INS for this purpose are covered in [St 94).
4.3 Inertial Navigation with S e o s o r s Aligned t o a Waoder Azimuth Coordinate F r a m e
The wander azimuth coordinate frame (subscript a), defined in Appendix C, Section C 4 (s. Fig. C3), is obtained from the
navigational frame through the "wander azimuth rotation U" about the vertical axis. Among different choices for the selection of
the U [St 821 the most common stems from the fact that the a-frame is not slewed about the vertical a i s with transport rate, i.e.:
(4.3.1)
0:
= [V,/(R,
= -
h)] tan #.
Prior to navigation t h e a-frame is aligned to true north, and the vertical in the same fashion as in an N I S (s. Section 5). The aframe remains in these directions when the system stays stationary or moves exactly northward. When it then passes the north
pole, there is not switching about the vertical by 180 deg as in an NIS, and its x-axis is pointing south. When the system moves
eastward, the wander angle o! increases; it decreases again, when it moves westward. Arriving at the initial meridian, the x-axis of
the a-frame is pointing exactly north again. Only after a passage of the pole and returning to the starting point on a different way
the x-axis of the a-frame is pointing in the opposite direction.
The computation of the position is based on the integration of a transformation matrix (s. App. C, Eq. C 17):
(4.3.2)
t
Cca(t) = cca(o) t ]Cca(T).
nY(T) dT
[ :;
'31
cl,
'131
'22
'23
'32
'33
'
where
with the components of w r from Appendix C, Eq. C 15.

The matrix Ccacan also be interpreted as being obtained from the rotations of the wander azimuth coordinate frame with respect
to the earth-futed frame via the angles A, # and o! (s. Appendix C, Eq. C17).
Equating the elements of (4.3.2 and C17), the geographic position and true north can be computed using:
(4.3.4a to c) sin# = -C13 ,
tank = -C23/C33 ,
tano! = -C12/Cll.
Although singularities do exist in the computation of longitude X and the wander angle o! at the geographic poles, they do not
exist for the computation of the matrix C c (t), so that inertial navigation is assured on a polar flight and A as well as 0: are again
correctly computed shortly after passage of a pole.
According to [Ka 691 the integration of Eq. 4.3.6 has to be executed with 26-bit words and the integration of acceleration with 17bit words, if the corresponding drift is to be kept below 0.001 deg/h.
4.4 Inertial Navigation with Sensors Aligned t o an Ioertial Coordinate F r a m e
Inertial navigation systems aligned to a space-stabilized coordinate frame (index "i") are used for terrestrial navigation using
platform systems equipped with electrostatically supported gyros (ESG); those gyros remain untorqued in the navigation mode.
These systems are also used for space navigation with platform and strapdown systems.
Figs. 4.4.la,b show twu possible configurations for inertial navigation with the acceleration measurement in the i-frame. It is quite
obvious that the system in Fig. 4.4.la navigates with respect to the earth after transforming the acceleration from the i-frame
into the n-frame with the help of the transformation matrix in Appendix C, Eq. C 3b. The navigational computation is identical
with that of the N S . The Autonetics ESGM (electrostatically supported gyro monitor) system for marine application works in this
fashion.
With the help of the transformation matrix C i = CinCna(s. Appendix C, Eqs. C 3b and C 17) navigation could also be carried
out in the wander azimuth frame, discussed in t%eprewous section.
The configuration in Fig. 4.4.lb is the one for space navigation with strapdown and platform systems. Navigation is carried out
within the inertial frame. This kind of navigation seems very simple on first sight due to the mere double integration of
(4.4.1)
t2
= fi
Gi(Ri).
The difficulty lies in the exact modelling of the earth's mass attraction Gi(Ri) for which we can find assumptions in Chapter
"Navigation Coordinate Frames".
On first sight again it is difficult to see that, when geostationary, this INS is governed in its interior error dynamics by the Schuler
oscillation and the instability of the vertical channel similar to the conventional NIS. Fig. 4.4.2 may serve to explain this. It shows
80
the spherical earth and the equipotential surfaces of gravitation:

(4.4.2)
Gi(Ri) = Go
!!S!-zi
R3
with Go and ,R
, the values at the surface of the earth. When the stationary INS computes a position error into the y-direction, for
instance, the resulting computed but not acting horizontal y-component of G tries to pull this error back. The computed Gcomponent due to an x-position error causes the instability discussed in Section 4.2.3. So we obtain as simplified error equation
for the stationary INS
A=
Platform
3 Accelerometers
G = 3 Gyros
Computer
I
I
win
--n
W
uie
-n
Computer
I !*
b)
U,
Y
R_i
I
I
Figure 4.4.1 Two Concepts for Inertial Navigation in the Space-Stabilized Coordinate Frame
where E is the respective misalignment angle including the effect of gyro drift and b is the respective accelerometer error. This
equation shows that the horizontal channels are governed by the Schuler frequency (s. Eq. 4.2.1a) and the vertical by the
instability known already from Section 4.2.3 (s. Eqs. 4.2.29 and 4.2.30). Fig. 4.4.2 visualizes these effects.
The Honeywell ESGN (electrostatically suspended gyro navigator) for marine application and the Honeywell GEANS (gimballed
ESG aircraft navigation system) - both are space stabiliced platform systems - navigate in the inertial coordinate frame. The
position components
are transformed to geocentric latitude, longitude and altitude with the help of the transformation
matrix CO,in Eq. C 3b ofkppendix C.
The next question is connected with the error dynamics of the space-stabilized INS in orbit. The main role of such a system in a
satellite is thrust control durin orbital changes and attitude determination. For a satellite in orbit the forces due to gravitation G
and centrifu al acceleration w gR are in equilibrium - thus its orbital frequency w is equal to the Schuler frequency for this orbit
w s = ( G / R ) k l t ( s . Eq. 4.2.1). This may give already an indication that the INS error dynamics will be governed by the local
81
0,50
-t
0,75
1,O
,42
0
Figure 4.4.2 Physics of the Schuler Feedback and the Vertical Instability in a Space-Stabilized Inertial Navigation System
."Ti
x,
orbit
earth
k
Figure 4.4.3 Physics of the Schuler Oscillation of a Space-Stabilized Inertial Navigation System in Orbit
82
Schuler frequency us,too, as in all o!her INS mechanizations. This is indeed the case as may be shown for an INS whose computed
coordinate frame is used in the satellite as vertical reference with respect to the earth. The local vertical coordinate frame (index
v) has its x-axis pointing up parallel to local vertical, its y-axis pointing tangentially to the orbit righthanded in Fig. 4.4.3, and its zaxis pointing normal to the orbital plane. Making use of the block diagram in Fig. 2 . 4 ~for the kinematics of a computed reference
direction with respect to the true direction and of the transformation matrix in Eq. C 20 of Appendix C between both, we arrive at
the following state space equation of the misalignment angles:
with the following solution for an initial misalignment angle c,(O):

(4.4.5a,b)
cos uSt,
E,(t) = c,(O)
(t) =
(0) sin oSt.
These results correspond in principle to the ones of the NIS on ground in Eq. 4.2.25.
4.5 Inertial Navigation with Body-Fixed S e n s o r s - Strapdown S y s t e m s
Strapdown systems (SDS) have beeen treated in this section in one breath with platform systems, since the platform as inertial
measurement unit (MU) including the command rates for platform slewing can be fully replaced by a strapdown IMU plus
computer for the corresponding transformation matrix computation. The latter is nowadays the preferred INS mechanization due
to the obvious reasons concerning cost and reliability. Throughout this chapter the platform system is used primarily for
visualization of system characteristics.
Besides the advantages mentioned above, strapdown systems offer additional ones:
- they deliver measurement signals not only for navigation and guidance, but also for flight control, namely angular rate and
acceleration in the body-fixed coordinate frame (s. Pig. 1.1 and 1.2);
- they are much more flexible in their adaptation to different system concepts, since no restrictions exist for the choice of the
IMU coordinate frame. This means that a strapdown M U not hard-mounted to the vehicle can carry out its measurement in any
coordinate frame and the SDS Will still navigate correctly. Here are some examples:
For submarines the INS is the only navigational means for underwater missions. The accuracy requirements for these missions are
with 1 Nh4/day allowable upper position error growth fairly difficult to achieve. Solutions developped by two competitive companies
are based upon ring laser gyro (RLG) strapdown MUS mounted again on a rotable platform.
In the model developed by Litton, the RLGs and the accelerometers are rotated with alternating directions about the vertical
platform axis, thus raising the 3 FUG'S input above the lock-in level (rate bias) and providing conditions for avaraging out the
effect of sensor errors with respect to the navigational coordinate frame (carouselling) [KO 901.
In the model developed by Honeywell, a two-axis indexer (roll on outside, azimuth on the inside) is employed for rotating the IMU
with dithered RLGs periodically 180 degrees about the roll and the azimuth axes, thus providing again the conditions for avaraging
out the effect of constant sensor errors upon the navigational error [Le 871.
In another proposal it was shown how to use a strapdown M U for gyro-stabilizing an instrument and at the same time using it for
inertial navigation [St 801.
Conventional SDSs used in civil and military aircraft are truely "strapped down" to the vehicle. They are equipped with RLGs in
general and use the wander angle coordinate frame for inertial navigation. As mentioned in Section 4.3, their output signals are
identical with the one navigating in the navigational coordinate frame (North Indicating System (MS)), but they do not have any
problems in passing the geographic poles.
The reader more interested in SDSs is referred to Section 1 for a brief opposition of platform and strapdown systems. Sections
2.4 and 2.5 give an introduction into the core of the SDS - the analytic platform as directional reference and its error behavior.
The problems of SDS signal processing are outlined in Appendix D. Appendix 0 describes the optical gyros as sensors which are
responsible for the widespread use of the SDS. Sections 3 and 4 are applicable to platform and strapdown systems as well.
4.6 Error Model f o r all Inertial Navigation S y s t e m Mechanizations
Navigating io T e r m s of Longitude and Latitude
The derivation of the vector error equations for the directional reference of a strapdown system in Section 2.5 and the
acceleration measurement in Section 3.6 have enabled us to generalize these results for any INS mechanization system with
respect to the earth. The idea behind this generalization is similar to the statement made in the previous section about the
flexibility of a SDS for adaptation to different system concepts - no restrictions exist for the choice of the IMU coordinate frame
(index m), i.e. a strapdown IMU not hard-mounted to the vehicle can carry out its measurements in any m-coordinate frame and
the SDS will still navigate like the INS developed in this sense. The error propagation of this SDS will thus be the same as that of
the INS.
The vector error equations for the directional reference and the acceleration measurement in Eqs. 2 . 3 6 ~and 3.12 read:
(4.6.la,b)
with
En
= en x w r
(4.6.2)
eo
= (eN
+ 60:
t 6 ~ : ~ ,
Vn =
en x a n - 6cn t 68,
6fn,
misalignment angles of the platform or the transformation matrix

about the north, east and down axes;
83
(4.6.3)
gin
,ie
(4.6.4a)
6~:"'
cn,. 6 w g
men = vector sum of earth rate
transport rate (s. App. C, Eqs. C8 andC9);
= error of [MU-slewing (platform slewing) or IMU angular rate measurement;
Cnm = matrix for vector transformation from the sensor measurement (index m)
(4.6.4b)
to the navigational (index n) coordinate frames;

m = n for the north indicating platform system, i.e. Cnm = I,
m = a for the platform system slewed to the wander azimuth coordinate frame,
i.e. Cnm = Cna (s. Eq. C 13),
m = c for the platform system continuously rotated about the vertical axis
with 1 revolution per minute, for instance (Carousel System)
for avaring out the effect of horizontal sensor errors upon navigational accuracy,
m = i for platform systems measuring in the inertial coordinate frame,
i.e. Cnm = Cni (s. Eq. C 3b),
m = b for strapdown systems mounted to the vehicle,
(4.6.4~)
6gS = K u m
im
i.e. Cnm = Cn,
(s. App. C, Eq. CIO);
= gyro errors (s. Eq. 2.22),
6 0 g I= d, since K is hardly observable, in general this approximation is valid,

(4.6.44)
~ =) angular
~
rate of [MU-slewing (platform-slewing = cat applied to the
(ox my w
gyros, s. Eq. 2.21 for the case m=n) or IMU angular rate measurement;
(4.6.4e)
x;:
ry
y;
;:
a=
scalefactor error plus input axes misalignment matrix (s. Eq. 2.20),
zy
Qzx
(4.6.4f)
(dx
dy
(4.6.5)
vn
(VN
VE
V,)T
(4.6.6a)
an =
(ax
ay
aZ)T =
(4.6.6b)
fm =
(fx
fy
fZlT=
(4.6.7)
cn
(4.6.8)
gn = g($). (5
(4.6.9a)
6fo = Cnm6fa = acceleration measurement error in the n-frame;
(4.6.9b)
6fa = Kam
dZ)T = drift vector;

=
(VN
v,
-h)T;
vehicle acceleration
measured accelerion fn = Camfm;
accelerometer output;
Coriolis acceleration (s. Eq. 3.10e);

-I)
with 5 and
r)
deflection of the vertical (s. Chapter "Navigation Coordinate Frames");
b = acceleration measurement error due to scalefactor error plus input axis

misalignment (s. Eq. 4.6.4e) and accelerometer bias,
6fa
b, since K is hardly observable,
6 f a = 0 , in the NIS (m = n), since it is not observable with respect to
E ~ .
These relationships show u s that for INS error modelling its mechanization enters the algorithm only through the transformation
matrix C . The internal coupling between the state vector components is the same for all of them and we can use the "North
Indicatini3ystem (NIS)" as basis for further discussion.
Fig. 4.6.1 shows without simplifications the block diagram for the erroneous signal flow in an NIS. The lower feedback for slewing
the vertical platform axis is proportional to tan$ , indicating that inertial navigation with this kind of mechanization is impossible
for polar flights. In the discussion of this section we will stay in the allowable latitude ranges (e.g. $ < f 88.5 deg for the
Ferranti FE 500).
For flight times above 6 minutes, the drifts of the horizontal gyros become significant. The Coriolis acceleration errors and the
earth rate couplings are of importance after approximately 2 hours.
If the N I S is standing on the ground, the dynamics of its errors for long periods are described
711, Eq. (7-49)):
(4.6.1 1)
= os f
wj
R sin $, see Figures
the following frequencies (s. (Br
1.1, 1.2 and 4.6.1 feedbacks A and
a, see Figures 1.1 and 4.6.1 feedbacks C and also Figure 2.14.
The first two frequencies characterize the Foucault-modulated Schuler oscillation, i.e. the oscillation of a freely swinging pendulum
of length R on the rotating earth. The oscillation plane of this pendulum stays, as it is well known, inertially fixed, i.e. it rotates
with respect to the earth with the negative vertical component of the earth rate.
The third frequency characterizes the 24-hour oscillation, i.e. the motion as indicated in Fig. 2.13, which is carried out by a
misaligned gyro-stabilized platform without a Schuler loop on the rotating earth.
84
o w
m
:
m
W
U
.>
.>
W
II
-w
U-
cum
U >
nu
3
:c
m
Lzg
II
-01
*.
_I
I--
O L
u a
1 YI
4
1
I
I
I
I
I
I
Figure 4.6.1 Block Diagram for the Signal Flow in a North Indicating System (NIS) Subject to Errors
85
The linearized error Eq. 4.6.12 for the M S is arranged in a way to alleviate the insight into the relationship between left and
right hand side.
(4.6.12)
[ 66
6);
63,
6V,
I/R
-6 /R
-2Asin$
l/(Rcos$)
-(Rcos$ +
/cos$)
-!&in$
2Q'sin$
-1/R
(VNtan$-h )/R
1/R
-tan$/R
Asin$
-4
Aces$
-g'
g'
+E
-vN
-Asin$
Acos$
wQ
WA
WbN
WbE
WdN
WdE
wdD
In this equation'we have made use of the following abbrevations:

(4.6.13 a to
I$
0)
V,/R
R + A
V,
+ ZR
+ A/2
' V ~ S+ ~h~Q ~
V,/(RCOS~
GE
& + h
V, -
sin# - hcos4)
g + h . - 2VER'C0S4 - Q v,
A6h/R
6h/R
g' =
AA' =
A&; =
6h/R
- A C O S ~6h/R
A s i n 4 6h/R
- E$ 6h + (VE A sin4 + h g ) 6h/R
-2R'cos4 6h - 2); (VNsin4 - hcos4) 6h/R
uncorrelated noise.
The third last column on the right hand side of Eq. 4.6.12 contains terms proportional to 6h and ah. They are separated from the
horizontal components of position and velocity errors, since their contribution is fairly often negligible so that the M S horizontal
and vertical axes can b e separated, as shown with the following numerical example.
The altitude error 6h causes-an error in the platform slewing corresponding to a gyro drift d' and both 6h and 6h a corresponding
accelerometer error b'. For A G 4 f R or V G 1000 km/h at 4 = 45 deg we obtain:
(4.6.14a,b,c)
d'
3 . los3 deg/h
and
b'
b'
3 . lo-' g
3.
per 6h = 1000 m.
per 6h = 10 m/s.
Eq. 4.6.12 is of the type 4.2.5 and can be used in a Kalman filter for modelling the errors of any INS, i.e. the north indicating
system (NIS), the wander azimuth platform system, the space stabilized platform system employing electrostatic gyros and the
strapdown system (SDS). This statement is true as long as "north and south" are still meaningful, i. e. the regions of the
geographic poles are exempted. The state vector consisting in this equation of the position errors (64, 6A), the velocity errors
(6V,, 6V,) and the attitude and heading errors (E,, eE, eo) has to be supplemented by the model of the sensor errors, i.e. the
accelerometer errors (6f ,6f ) and the gyro errors (6wx,6,w ,6wz) in the coordinate frame of the measurement axes. In case they
are modelled as bias tern% &th additional uncorrelated n o d , Eq. 4.2.3 for the single axis INS may serve as an example. It should
also be emphasized that the transformation matrix Cn, between the sensor measurement axes and the axes of the navigational
coordinate frame is properly considered as pointed out above. For a system in flight all components of the so-called system matrix
P are maneuver-dependent coefficients. The state transition matrix for predicting the INS errors thus can only be computed online.
For using the error model 4.6.12 some more practical hints. If an NIS has to be modelled, the accelerometer error BN,Eshould
not be included in the state vector, because it is not observable. These sensor errors cannot be distinguished from a platform
misalignment, i.e. the state vector components .zEN 5
11 comprise them, too. In all other cases when the accelerometer axes are
not always held parallel to the axes of the nawgational coordinate frame, the accelerometer biases are observable and worth
modelling.
At the end of an M S selfalignment mode discussed in Section 5 , the accelerometer errors B, are compensated by a platform tilt
about the horizontal axes and the east-west gyro drift DE is compensated by an azimufE offset eo. These conditions remain
en the N I S is switched from the alignment to the navigational mode, but still is geostationary. So theoretically these sensor
errors do not directly cause a position error growth at the beginning of the navigational mode. They do it indirectly only due to
86
the cross-track error which according to Eq. 4.2.28 depends upon the azimuth misalignment tD which again depends upon DE at
the end of the alignment mode.
The situation is different for the north-south gyro drift. At the end of the alignment mode it is compensated by a platform tilt (s.
Fig. 5.4) unless the gain
is shunted by an integrator (s. Fig. 5.4). In the former case the east-west position error increases
immediately after switching into the navigational mode.
All items mentioned above are also valid for the geostationary strapdown system (SDS). But when this system is moved, the
equilibrium between sensor error projection in the north-south, east-west and down directions and the misalignment angles is
disturbed as indicated in Fig. 2.14. Due to this maneuver-dependency all sensor errors are observable in a Kalman filter and are
worth modelling.
The theory discussed above is confirmed in the measurements of a stationary Litton LN-3A platform INS plotted in Fig. 4.6.2 for
the velocity and position, and the Schuler oscillations as well as the initial part of a 24-hour oscillation are distinctly to be seen.
The INS in-flight system errors differ from the stationary system errors by the maneuver-dependency of several contributors, as
to be seen from Fig. 4.6.3.
There are firstly the cross-track velocity (6VNE E, f tD.VEN) and position errors (6SN, E f cD.AS ) as outlined in the
previous section. They are common to all inertial' navigation sysfems and are superposed to the Schuler oscf&ons.
There is secondly the fact that with strapdown systems the sensor errors are effective within the body-fixed axes, i.e. they act
within the navigational coordinate frame via the maneuver-dependent Cn,matrix (s. Eqs. 2 . 4 1 ~and 3.11 as well as Fig. 2.14).
There is thirdly the fact that the sensor errors themselves can be maneuver-dependent as within mechanical gyros, for instance
(s. [St. 821).
In order to estimate the accuracy of a geostationary INS over more than 2 hours time of flight for a certain class of sensor and
initial alignment errors comprised as 1-0 values, Eq. 4.6.12 was solved on a digital computer and its results have been plotted in
Fig. 4.6.4 as 1-U INS errors over a 6-hour and a 36-hour period. They describe the two-dimensional Gaussian distribution of the
horizontal position error (s. Fig. 4.6.5 and 4.6.6). From the effect of the various uncorrelated error sources on the position error
the following can be stated:
-As V = 0, the cross-track error does not become significant.
-For short periods of up to 1/2 h in the north-south channel or 2 h in the east-west channel the influences of the initial
misalignment and of the accelerometer bias are identical, which can be explained by the fact that initially the azimuth
misalignment does not become significant.
-In the north-south channel the initial misalignment causes a mean position error increasing linearly with time - according to Eq.
(4.2.24b) the azimuth misalignment has the effect of an east-west gyro drift.
-The gyro drift causes a mean position error increasing linearly with time, which is due mainly to the horizontal gyros. According
to Eq. (4.2.25) the vertical gyro drift of a stationary NIS after a time of
(4.6.15)
At E 4/(R cos$) > > 2h,
causes the same mean position error as that produced by an east-west gyro drift of the same amount. However at longer times
the effect of the vertical gyro drift becomes dominant.
The two components of the horizontal velocity error or the horizontal position error have Gaussian distribution as indicated in Fig.
4.6.5. Horizontal sections through this distribution are error ellipses as described by the exponent of p(x) in Fig. 4.6.5 and as
shown in Fig. 4.6.6 for the position. The projection of the ellipses on the north-south and east-west axes are the 1-U values
obtained from Eq. 4.6.6 for SNE. For system specification these 1-U values in the two axes are not so convenient; the
manufacturer often gives the oveiall specification in form of the CEP or SEP (circular error probable, spherical error probable)
value which in a two-dimensional or three-dimensional case is the radius of a circle or sphere comprising the errors with 50 %
probability unless otherwise mentioned (e.g. CEP,, or SEP,, corresponds to 95 % probability) [STA 1. The CEP can be
computed from the corresponding elements of the covariance matrix as pointed out in [Ka 691. In the example of Fig. 4.6.6 we
obtain, with u N = 5.62 km, uE ;3.81 km and p = - 0.8, the 50 %
' and 90 % probability circles CEP = 5.0 km. For equal U-values
in all three dimensions we obtain for instance: SEP = 1.5382 U.
87
0.01
-fN
s *]
[m/
-0.01
5
6V
53
-5
time [h]
time [h]
10
-10
Figure 4.6.2 Velocity and Position of the Stationary Litton LN-3A Platform Inertial Navigation System
88
1
E:
.r(
'0
c8
e,
5:
30
20
10
5@
40
60
70
FO
70
90
80
100
time [min]
0.e
0.6
0.4
0.2
0.0
-0.2
-0.4
-0.6
-0.8
- 1 .o
IO
50
20
50
40
a0
90
100
time [rnin]
200
-1looI
'
10
'
20
'
30
'
'
40
.
A
.
-
' . '
53
60
70
80
90
.
100
time [min]
Figure 4.6.3 Typical Flight-Test Results of the Litton LTN-90 Strapdown Inertial Navigation System
89
10.0
(ij Effect of i n i t i a l platforn misaiignment
8.0 @ Effect of g y r o drift

@ Effect of acceleroneter bias
6.0 @ Total error
6%
L O O
Kfl
2.0
0.
time [ h ]
10.0
8.0
6.0
6%
4.0
Kfl
2.0
0.
[hl
time
12
2L
time [ h ]
0.
90.C
60.
bSE
Kfl
30.
1 I
_-
----
---
0.
12
Uncertainties for initial misalignment angles:

for gyro drift:
for accelerometer bias:
36
1
i
-
time
[hl
u ( c ~ , ~ ) = 0.1 mrad = 20 arc sec,

u(eD)
= 1 mrad = 3.4arc min,
0.01 deg/h,
O(B,.,~) = 0.0001 g.
u(DN,E,D) =
Figure 4.6.4 1-0 Position Error of a Stationary North Indicating System
36
90
E ( _ x ) = X = mean value of vector

E [( _x - j ) ( _ x
)T
] =P = corvariance matrix of vector

-
Figure 4.6.5 Two-Dimensional Gaussian Distribution
6SNkm
5,62*
2
-0,8.3,81 5,62
-0,8 ' 3,81- 5,62
3,812
Figure 4.6.6 Position Error Ellipsis and Circular Error Probable of a North Indicating System
91
[Au 91 J Ausmann, J. S.: "Baro-Inertial Loop for the USAF Standard RLG I N . "
Navigation: Journal of The Institute of Navigation, Vol. 38, No. 2, Summer 1991, pp. 205
- 220
[Br 711 Britting, K: "Inertial Navigation System Analysis"

Wiley Interscience 1971
[Bo 35 J Boykow, J. M.: "Einrichtung zum Messen von Wegstrecken."
Deutsches Patent 1935
[Ge 741 Gelb, A (Editor): "Applied Optimal Estimation."
The M.I.T. Press, Cambridge, Mass., USA, 1974
[He 801 Heilbron, H., Klein, G.: " Ein Ruckblick auf die Entstehung der Tragheitsnavigation."
DGON Zeitschrift Ortung und Navigation, 1/1980, pp 36 - 60.
[Hi 921 Hilscher, G.: "Flug ohne Sterne."
Olynthus Verlags Anstalt, Vaduz, 1992
[Ka 691 Kayton, M., Fried, W. R: "Avionics Navigation Systems".
J. Wiley and Sons, Inc. , New York, 1969
[KO901 Kohl, KW.: "A new High Accuracy Ship's Inertial Navigation System PL4l MK4."
Proceedings Symposium Gyro Technology 1990, DGON and Universitaet Stuttgart, 1990
[Le 871 Levinson, E., Majure, R: "MARLIN - The next Generation Marine Inertial Navigator."
[Ma 781 Magnus, K: "Kreisel als vielseitige Hilsrnittel in Luft- und Raumfahrt."
Zeitschrift fur Flugwissenschaften und Weltraurnforschung 1978, Vol. 2, pp 217 - 227.
[NM 841 "Digital Air Data Computer, Part No. 12238-3, Maintenanca Manual with Illustrated Part List, 34-10-10"
Nord Micro, Oct. 15/84
[Re 85J Redeker, A, Voersmann, P.: "Precise Vertical Speed Reconstruction Based on Vertical Acceleration and Barometric
Altitude."
Z. Flugwiss. Weltraurnforsch. 9. 1985, Heft 4
[STA ] " Standardization Agreement; Subject Method of Expressing Navigation Accuracies."
North Atlantic Treaty Organization, STANAG 4278
[St 80J Stieler, B.: "Fahrzeug mit einer Plattforrn fuer eine Ziellinienstabilisierung in Verbindung mit einem Inertialsystem."
Deutsch Patentschrift DE 3019743 C2, Anmeldetag 32.5.1980
[St 821 Stieler, B., Winter, H.: "Gyroscopic Instruments and their Application to Flight Testing."
AGARD-AG-160-Vol.15, 1982
[St 941 Stieler, B.: "Generation of an Altitude Reference Using a Conventional INS.
DLR-Mitt. 94-xx, 1994
[Wi 75 J Winter, H.: "Messung der Vertikalbewegung eines Flugzeugs."
Deutsche Luft- und Raumfahrt, Forschungsbericht 75-4 1
[ w u 80 J Wuest, W.: "Pressure and Flow Measurement."
AGARD-AG-160-Vol.11, 1980
92
5. Initial Alignment and Calibration of Inertial Navigation Systems

5.1 Introduction
The initial self-alignment loop or gyrocompassing loop of a stationary inertial platform is in principle the electronic equivalent of a
mechanical gyrocompass used as a heading reference on ships.
The gyrocompass is in principle a damped gyropendulum with a horizontal spin axis and a separation between center of gravity
and center of support of about 1 mm. In the geostationary case this is a northseeking instrument, the input signals being the
rotation of the earth and gravity. It has a settling time in the order of 4 hours.
In the following we will briefly review the principles of the gyrocompassing loop of an inertial platform and the accuracy achievable
for the initial alignment.
5.2 The G y r o c o m p a s s i n g Loop of a n I n e r t i a l Navigation S y s t e m
In Section 2.1 we have discussed the effect earth rotation has upon a free gyro whose spin axis is slightly misaligned with respect
to the earth axis. This gyro will carry out a 24-hour motion around Polaris (s. Fig. 2.5).
Fig. 5.1 shows in principle the mechanical gyrocompass differing from the free gyro in Fig. 2.5 through the pendulosity only. Its
spin axis is nearly horizontal. Once disturbed from its equilibrium, namely true north, it will oscillate in an elliptic cone about this
equilibrium with a frequency adjustable by the pendulosity or it will align itself with true north if it is damped as indicated in this
figure.
Fig. 5.2 indicates how the pendulosity and the damping can be replaced by an accelerometer
and two feedback loops. The
accelerometer input axis is aligned with the spin axis. Its output signal is fed via the gain - K
to the horizontal torquer for
raising the frequency, and via the gain - K, * H to the vertical torquer for raising the damping opthe sensor's oscillation.
Fig. 5.1 The Motion of the Undamped and the Damped Gyrocompass
Fig. 5.2 The Gyrocompass Implemented with a Free Gyro, an Accelerometer and Two Feedback Loops
93
We find the same interconnections in the simplified gyrocompassing loop of an inertial platform whose block diagram is shown in
the lower half of Fig. 5.3. The output of the north-south accelerometer A causes a platform rotation w, and 0 , about the eastwest and the down axes via the gains -KE and - K,. The upper half of tge block diagram shows the levelling loop for the northsouth axis.
One of the simplifications in this diagram concerns the model of the free platform ( S . Fig. 2.13) which has been reduced to the
coupling of the vertical misalignments into the east-west axis via the horizontal component of earth rate Q.cos 4. The coupling
between the horizontal axes via the vertical component Q.sin of earth rate has been neglected, which is acceptable since the
horizontal misalignment angles are smaller by one order of magnitude than the azimuthal misalignment as indicated in Table 2.1
and as will be shown below.
5.3 Equilibrium and Sensor Calibration a t t h e End of t h e Alignment Process
From the block diagram in Fig. 5.3 one can easily derive t h e following relationships for the equilibrium of the platform at the end
of the alignment process when the inputs to the integrators are zero:
- for ideally compensated sensors, i.e. no accelerometer biases (BN,E = 0) and no gyro drifts (D,,E,D= 0) the platform will be
ideally aligned with the vertical and true north:
- the horizontal accelerometer biases (B, E 0) are compensated by corresponding gravity components generated by horizontal
platform misalignment angles (remember the accelerometer output signals are null in the equilibrium phase in this case):
(5.2)
E,,,
= f BE,, / g (=
. I mrad = 20 arc sec for B,,, =
9);
- the east-west gyro drift (DE 0) is compensated by a corresponding component of the horizontal earth rate generated by a
vertical platform misalignment angle in the equilibrium phase:
(5.3)
cD
DE
/ (R.cos 4)
( = 1 mrad = 3.4 arc min for DE = .01 deg/h and
- if in addition the north-south and the vertical gyros are corrupted by the drift D,,,,
equilibrium phase additional control offset signals:
= 45');
the horizontal accelerometers furnish in the
(5.4)
Whilst the accelerometer biases B,, and the east-west gyro drift DE cannot be measured - their effect is compensated by gravity
or earth rate, respectively - the' north-south and the vertical gyro drift D,,, cause an output voltage of the horizontal
accelerometers and thus can be measured.
The additional tilt due to D,, and thus the control offset voltage, can be reduced by setting the loop gain K, high or preferably by
shunting the amplifier of the levelling loop with an integrator as indicated in Fig. 5.4 for the north-south gyro drift. In practice
only this sensor error is compensated in the manner shown. The vertical gyro drift can only be measured by means of the synchro
mounted on the vertical gimbal axis, which requires the absence of any platform motion.
Eq (5.3) is applicable to all north-seeking methods with gyroscopic sensors. These methods are limited to 80 deg latitude,
approximately (= 76.5 deg for Delco Carousel lV platform system). Note that only the initial self-alignment of the INS is limited
to this latitude; the actual navigation function can b e performed at all latitudes (s. Section 4.3).
If we choose the following gains in the gyrocompassing loop of Fig. 5.4:
(5.5 a,b)
1/(2gT),
K,
K,
K,
1/(16gT Rcos
4)
and the gain of the shunt integrator in the levelling loop:

(5.5c)
KL
K,
Qcos
4,
both loops - the north-south levelling loop and the gyrocompassing loop - have the same dynamics, namely four equal roots w i t h
the time constant:
(5.6)
~ = 2 T = 6 0 s
(the figure was selected in laboratoy tests [St 781) and the problems of compensating the north-south gyro drift and of vertical
platform alignment are alike.
If the gyrocompassing alignment loop is implemented in this fashion, the system will carry out the elliptical motion as shown in Fig.
5.1; the semimajor axis of the ellipsis is then proportional to the vertical or azimuthal misalignment and the semiminor axis to the
east-west or horizontal misalignment. With the time constant mentioned above we obtain as ratio of both axes 140, i.e an initial
horizontal misalignment excites a much bigger disturbance motion of the gyrocompassing process than an initial vertical
misalignment. In order to speed up the alignment process, it is advisable to split it up into two modes - the levelling and the
gyrocompassing mode. During the former the horizontal misalignments in both axes are minimized with a levelling loop as
indicated in Figs. 5.3 and 5.4 lasting 1 min approximately. During the subsequent gyrocompassing mode the azimuthal
misalignment is minimized during 8 min approximately. The figures quoted are valid for the Litton LN-3A platform system.
It is shown in [St 781 that for strapdown systems another alignment concept has advantages as compared to the control loops
shown in Figs. 5.3 and 5.4. Only two levelling loops are implemented here, and the steady state signal to clamp both horizontal
axes to gravity are proportional to the north-south system drift D, and the vertical misalignment E,. Both parameters are
estimated with the use of a simple Kalman filter neglecting all system dynamics. The major advantage of this concept lies in the
fact that the estimation of both parameters becomes with time increasingly insensitive to vehicle sway due to wind gusts or
94
Platform Alignment Loops
Sensor Error Model
I K" I
Vo
Drift
Vertical
and oise
Gyrocompassing
Loop
East Gyro
Drift and Noise
I
1-4
1
<
1 1% I
Norkh Accelerometer
Bios and Noise
Figure 5.3 Simplified Block Diagram of the Gyrocompassing and the North Levelling Loops for an Inertial System
East Accelerom eter

Bias and Noise
North Gyro
Drift and Noise
I
I
North - South
Levelling Loop
Vertical G ro
Drift a n d t k s e
East Gyro
Drift o n d Noise
North Accelerometer
Bias and Noise
I
Figure 5.4 Simplified Block Diagram of the Gyrocompassing and the North Levelling b o p s
for an Inertial System Including Integrating Network for North Gyro Drift Compensation
95
baggage loading in an aircraft.
We have not yet mentioned the vertical accelerometer. Its output signal, which should be grevity, can certainly he used for
calibration. For the field-calibration of all sensor biases (DNE and B, E D ) and scale factors (S, E for the gyros, S, ,D for the
accelerometers) the gyrocompassing mode is carried out sev&ra? times with the input axes of each i>;o once in the nor# and once
in the south directions and the input axes of each accelerometer once in the up and once in the down directions, which requires 9
orientations of the platform. The s u m and differences of the currents into the north or south gyro iNs and into the up or down
accelerometer iu,D are evaluated in the following way:
(5.7a,b,c,d)
+ is
D = -S (I,
)/2
B = -S (io
iu)/2
S'
= (i,
sa
(iD -
is )/(2 R cos
4)
iu)/(2g).
5.4 Azimuth Alignment Accuracy Due t o Random Walk Gyro Drift
So far we have discussed the initial alignment accuracy due to systematic sensor errors. We know that such errors can be
compensated once they are known, but the user has to live with a certain day-to-day variation of the systematic errors which
always limit the final alignment accuracy. Modern sensors, especially the ring laser gyros, have an unprecedented low systematic
drift so that the stochastic error known as random walk dominates. One cannot predict the effect of stochastic errors as we have
done for systematic errors. All one can predict is the expected value or the f l u band which encloses 68% of all samples. In
order to derive the f 1 U values for the azimuth alignment due to east-west gyro random drift, we use the integral of Eq. 5.3, i.e.
we divide the f 1u values for all integrated random drift angles through the integral of earth rate. The result is:
(5.8)
u(cD )
=
=
r(dE ) t'/*
P
O
/ ( R cos 4 - t )
r(dE) [deg/l/h ]
/ (R cos 4 . [ G I ) .
This equation is evaluated in Fig. 5.5 for a random walk coefficient of r(d ) = .001 d e g / F which is approximately the value
measured at DLR for the Honeywell GG 1342 RLG (U1 88). According to mjitary requirements the alignment time is limited to
below 4 min with an accuracy below 10 arc min.
CJ ( E ~ /) arc
min
4 Vmin 6
Fig. 5.5 Azimuth Alignment Accuracy Due to East-West Random Walk as a Function of Time
[St 781 Stieler, B., Zenz, H.P.,: "On the Alignment of Platform and Strapdown Systems."
Proceedings of the DGON Symposium Gyro Technology 1978, Stuttgart, Germany
[UI 881 Ullrich, D., Liibeck, E., Wetzig, V.: "Static and Dynamic Measurements with a Ring Laser Gyro."
Proceedings of the DGON Symposium Gyro Technology 1988, Stuttgart, Germany
96
6 . Attitude and Heading Reference Systems (AHRS)

6.1 Introduction
Attitude and Heading Reference Systems (AHRS) will play an increasing role in future aerospace guidance and control due to the
increasing role of the Global Navigation Satellite System (GNSS) and its integration with inertial reference systems of reduced
performance. T h e latter is the keyword for describing the AHRS system structure. As already mentioned in Section 2, the AHRS
is no stand-alone system, but always integrated with other types of sensors or systems in a predefined control loop structure or
with the aid of Kalman filters. The margin between both types of system integration is floating.
The principles of a fairly simple AHRS system were already outlined in Section 2.1 in connection with the Vertical Gyro (VG) and
the Directional Gyro (DG). We will review them in this section briefly and describe some modem system implementations.
6.2 T h e
A H R S with Magnetic Aiding
In Section 2 we have seen that the directional reference of the inertial navigation system (INS) and thus also of the AHRS can be
implemented as a physical or analytic platform. It was shown that its characteristics can be described in both cases by that of a
three axes free gyro with an ideal suspension. The AHRS signal flow is discussed in the following as platform implementation only.
Let u s first discuss the basic functioning of the "attitude and heading reference" within the simplest AHRS application by looking
again to that of the VG and DG in Figs. 2.6 and 2.7.
We recall that in the INS the accelerometer output signals were used in a first step to compute groundspeed and in a second step
to compute the Schuler feedback for platform slewing. This resulted in an ideal vertical reference with an 84-minute Schuler
period of the closed loop, but required gyros of "inertial qualitf (s. Section 4.2.2.1).
The vertical reference in Figs. 2.6 and 2.7 for measuring attitude is based upon bubble levels or accelerometers for sensing the
misalignment about the horizontal axes. Their output signals are directly used for gyro or for platform slewing for nulling the
misalignments as shown in these figures.
Assuming in Fig. 2.8 the gain
& = 0, the gain K,
defines the system's tuning about the horizontal axis. The time constant:
T = l/(g K,) < 10 min

is much lower within the AHRS than within the INS. Due to the stronger coupling of the horizontal gyros to the horizontal
accelerometers, the effect of gyro drift D, upon horizontal misalignment:

eh = D,.
T = D,/(g
K,)
is much smaller than in an INS and the requirements for the gyro performance can be reduced by at least one order of
magnitude. Even this sensitivity to constant or slowly varying gyro drift is reduced by the additional integrator within the control
loop.
These advantages are paid by the increased sensitivity of the misalignment angles to horizontal accelerations. As indicated in this
figure, the most simple means to cope with this effect is the cutoff of the feedback at B certain acceleration level.
The directional reference for measuring heading is based in an AHRS upon a magnetic compass for sensing magnetic north, i.e.
based upon a so-called flux-gate as indicated in Figs. 2.6 and 2.7 (s. Chapter "Magnetic Heading References") or based upon a 3axis strapdown magnetometer. The difference between AHRS north and compass north is used for gyro or for platform slewing to
magnetic north.
The flux gate has its built-in vertical reference which is not as accurate as the AHRS vertical reference. The integration of a 3axis strapdown magnetometer with an AHRS is thus the system concept of higher performance.
Soft and hard iron induced magnetometer errors can fairly well be calibrated with the help of the gyros within the AHRS. For this
purpose the aircraft standing on the airfield is rotated about the vertical axis with respect to magnetic north and the vertical gyro
readings and magnetometer readings are evaluated with special software described in [Ba 85, St 901, for instance.
6.3 T h e A H R S with Doppler Aiding
The Doppler-aided AHRS is of higher performance than the one mentioned above. It has true navigational capability and is a
preferred system for helicopters. It combines the advantages of the INS and Doppler, i.e. accurate attitude heading and
groundspeed with reduced time-dependent position error growth.
The AHRS for this application is in principle identical in its signal flow to an INS (s. Fig. 4.2.5), cut is equipped with sensors of
lower performance. Its Doppler aiding consists of feeding the differences between the inertial velocities and the Doppler velocity
back into the AHRS to correct the system attitude, heading and velocities.
Fig. 6.1 shows the error block diagram of a so-called third-order in-flight gyrocompassing loop, whereby the Doppler velocities
measured in the body-fixed coordinate frame are resolved into the north and east directions by the attitude and heading angles as
measured by he AHRS. This is indicated in the upper right corner of Fig. 6.1. Due to the feedback in the horizontal channels, the
Schuler oscillations of the AHRS are damped and the frequency is augmented: the loops are N-times Schuler tuned and often
critically damped.
fairly simple analysis reveals the ground speed error 6V due to a Doppler velocity error 6vd, to an accelerometer error b and
to a gyro drift d C means "variable in the Laplace domain"):
91
8,
East Accelerometer
Horlh Gyro
I
I Longitude
! Error
1
-
0,
cos q
I
Vertical Gyro
I Lotitude
I Error
1
-
East Gyro
North Accelerometer
, 6Vi - 6VdsinJ,+ E~ V d cos$

INS Azimuth
6V:
Resolver
6VdcosJ, - E~ V d sinq
Figure 6.1 The Doppler-Aided AHRS or the Third-Order In-Flight Gyrocompassing
6V =
(6.3a)
ss
[s2 +
In the high frequency domain (s

(6.3b)
6V H 0
1
+ (w')*(l
H
-)
R K,)]
[sc
+ gz
(sK,
LOOP
gK,) 6?].
the ground speed error is:
and in the low frequency domain it is:

(6.3~)
6V
11 4. R K11
[D
+ K,SVd]
.01 RD
+ 6vd.
The latter is valid for the following gains used in the Singer Kearfott AHRS of the type SFIM 253 A
(6.4 a)
with w
(6.4b)
1
=
RK, z 100
w f
10 w s
closed loop frequency, w s = Schuler frequency (s. Eq. 4.2.la) and:
K,
P 14 w s
.l,
with 6 = damping coefficient of the closed loop.
The Doppler noise is thus damped out and the navigation system error follows the low frequency Doppler error, but reduces the
effect of inertial sensor errors upon navigation accuracy.
Fig. 6.1 also shows that the north velocity difference between the AHRS and the Doppler is fed back to the vertical gyro to
correct the azimuth AHRS misalignment. Though neither the AHRS nor the Doppler radar "knod the direction of true north,
both find it "hand in hand" by means of this feedback signal. This in-flight gyrocompassing becomes feasible, since due to earth
rate the errors of this AHRS caused by an azimuth misalignment differ in the north- south channel from those of a Doppler
system with external heading reference, i.e. a conventional dead reckoning system. In the east-west channel the INS error due to
98
the azimuth misalignment equals the conventional dead reckoning system through the well known crosstrack error, and feeding
back the corresponding velocity difference is of no use. For verification compare in Eq. 4.2.5 64(t) = f(ED) and 6h(t) cos 4 =
f(ED).
Eq. (4.2.28b) finally reveals that also in the north-south channel the INS error is equal to that of the dead reckoning system, if it
is initially aligned using the stationary gyrocompassing procedure. From this we may deduce that Eq. (4.2.26b) also limits the inflight gyrocompassing accuracy under the presumption that the Doppler velocities are measured accurately in the body-fixed
coordinate frame.
Since the AHRS is equipped with lower quality gyros,. this in-flight gyrocompassing is not applied in general and it is always
integrated with a sensor for magnetic north.
6.4 Modern
AHRS Implementations
From the previous sections we may deduce the following principal differences between the modern AHRS and the INS
implementations:
-The AHRS is equipped with sensors of lower performance; fiber optic gyros (FOGS, see Appendix 0) are very promising sensors
at present, micromechanical gyros and accelerometers have the potential to penetrate this application in t h e future [Ma 92, El
911.
-The computation of the analytic platform (see Section 2.4 and Appendix D), i.e. the high frequency data processing is identical in
principle in both systems, the AHRS and the INS.
-In contrast to the INS the AHRS always depends upon external sensors as for heading (magnetic compass), for ground speed
(Doppler radar or air speed indicators) or for position (ground- or satellite-based radio stations) aiding.
For the development of the proper aiding software on the basis of Kalman filtering modern AHRSs can be treated in their error
models like an INS (see Section 4.6), whereby the lower sensor performance is reflected in the initial covariance matrix and the
system noise matrix.
In integrated AHRS-based navigation systems available on the market one can find fixed gain control loops again, as discussed in
the previous sections [Ha 891, which often are the result of a laboratory Kalman filter simulation.
,
[Ba 851 Baeumker M.: "Verfahren zur Kurswinkelbestimmung

- mittels und zur automatischen Kalibration eines in einem
iuftfahizeug fest montierten Dreiachsen-Magnetometers."
Europaische Patentschrift, Veroffentlichungsnummer 0 226 653 B1, Anmeldetag: 20.12.85
[El 911 Elwell, J.: "Progress in Micromechanical Inertial Instruments."
Proceedings Symposium Gyro Technology 1991, DGON and Universitaet Stuttgart 991
[Ha 891 Ilandrich, E. et al: "Fiber Optic Gyro Strapdown System."
[Ma 921 Mattisek.. A... Baeumker, M.: "Integration

of a Fiber Optical Gvro Attitude and Heading Reference System with
...
biffereniiai GPS."
[St 901 Stieler, B.: "Verfahren zur genauen Vermessung raumlicher Winkel, Trajektorien, Konturen und Bewegungsvorgangen
sowie Schwereanomalien mit Kreiseln und Inertialsystemen."
Deutsches Patent Nr. DE 40 29 215 C 2, Anmeldetag: 14.09.1990
99
Appendix C
Coordinate Frames f o r Inertial Navigation
C 1 Introduction
I
This appendix is written by the authors of Chapters "Navigation Coordinate Frames" [Ku 941 and "Inertial Navigation" in order to
harmonize the coordinate frames with the practical requirements of inertial navigation and to introduce notations used in this
chapter. Notes in { ) refer to [Ku 941 and those in ( ) to this chapter.
C2 T h e Inertial, t h e Earth-Fixed and t h e Navigational Coordinate F r a m e s

For inertial navigation, the required reference is the inertial coordinate frame in which the earth revolves around the sun. This
yearly motion takes place in the Earth Centered Inertial {ECI) or the Conventional Inertial System {CIS}coordinate frame (index
i) and is defined/based on the Fundamental Katalog {FK}5 system [Ku 94, Section 21.
For navigation in a geodetic system, the required reference is the Earth Centered Earth Fixed (ECEF) coordinate system or the
Conventional Terrestrial System {CTS) It is called earth-fixed coordinate frame (index e ) in this chapterdefining positions in a
three-dimensional frame on the surface of the earth and its adjoining space [Ku 94, Section 5).
At time zero, the basic Earth Centered Inertial (ECI) (index i) system and the Conventional Terrestrial System {CTS) (index e)
are assumed to be coincident. Currently, the zero epoch is J2000.0, which is defined to start at noon on 1 January 2000 [Ku 94,
Section 31.
The rigorous transformation from ECI to CTS frames are described in [Ku 94, Section 41. However, for the purpose of navigation
and the accuracies involved, we can make certain approximations.
To achieve the above, the matrix [D] for precession, matrix [C] for astronomic nutation, and the matrix [A] for polar motion
defined in Chapter 1 can be omitted. This approximates the relationship between the X1 and XS frames of [Ku 94, Section 41 to:
XS = [B] X1 with [B] according to Table 7 and A according to [Ku 94, Table 11.
In this chapter we use the orientation of the i- and the e- frames as shown in Fig. C1, i.e. with their origin in the center of the
ellipsoid, the x-axis pointing to geodetic north and the y- and z-axes in the equatorial plane. At t = 0 the z-axes of both systems
are pointing to the Greenwich meridian. For inertial navigation we reduce the rotation angle A contained in [B] by A = R -t with:
(C 1)
= 7.2921158553. lop5 rad/s 9 15.041 067
deg/h.
The matrix cooresponding to [B] in [Ku 94, Table 71, for vector transformation from the e- to the i- frame is thus:
The reference navigational coordinate system (index n) for inertial navigation is the local geodetic system defined in [Ku 94,
Section 7.21. Its origin is in the proof mass of the accelerometer or the origin of the inertial measurement unit (IMU). Instead of
the sequence of axes east, north, up {E,N,U) used in [Ku 941 we use north, east and down (N,E,D
x,y,z) in this chapter. The nframe is the result of a first rotation with respect to the e-frame through the geodetic longitude about the xe-axis, a second
rotation through the geodetic latitude $ about the y -axis and a translational motion through the geodetic height h to the nframe origin mentioned above. The matrix correspon&ng to Rx(90-9). Rz(h+90) in [Ku 94, Eq. 241 for vector transformation
from the n- to the e-frame is:
0
-sin$ sinh
sin$ cosh
cosh
sinh
-sin$
cos$ sinh
cos$ cosA
For C i n replace in this relationship the geodetic longitude A by the celestial longitude:
(C 4 )
A = h + R t .
(C 3b)
cosh
sin$ cosh
sinh
-sin$
cos$ sinh
cos$ cosh
The actual coordinate frame for inertial navigation is the horizon or the local level system [Ku 94, Section 7.41. Due to the
"deflection of the vertical" gravity in the n-frame has small horizontal components:
which, when left uncompensated in t h e computer, are nulled during the initial alignment of the inertial navigation system (INS).
T h e true rotation of the actual navigational frame with respect to the e-frame is thus:
Re f e r ence Coord inat e Fra mes

1
, y ,z
inertial
XsXe
frame i
x, , y e , z e e a r t h - f i x e d frame e
x,, y,, z,
navigational
frame n
present point of
measurement
Greenwich
meridian
l o c a l meridian
Earth R a t e a n d
Transport R a t e in
t h e n -Frame
Fig. C 1 The Inertial, the Earth-Fixed and the Navigational Coordinate Frames
101
4'
(C 6a,b)
4 - <,
X' = X
r)
cos
4.
In conventional inertial navigation systems with 1 NM/h position error growth these effects are neglected.
The vector for the s u m of earth rate
with
[Ku
2Section and
= {\)
01and transport rate 07 is in the n-frame:
= {RJ the radii of curvature of the reference ellipsoid in the meridian (N-S) and of prime vertical plane
6.3.11.
C 3 T h e Body-Pixed C o o r d i n a t e F r a m e
The body-fixed coordinate frame (YPR)(index b), i.e. the coordinate frame fixed to the aircraft and relates with the navigational
or the local level system {E"N"G) through attitude and heading
as the yaw angle, P as the pitch angle and R as the roll angle).
In this chapter we use the international standards of flight mechanics [Std 701 (s. Fig. C2) with the Euler angles defined by the
following sequence of rotations from the navigational into the body frame, i.e.:
- the first rotation about the zn and zb "f's (both are parallel in the beginning) through the yaw angle Y,
- the second rotation about the y, axis lyng still in the horizontal plane through the pitch angle 8 and
- the third rotation about the xb axis through the roll angle O.
L"
Figure C 2 The Body-Fixed Frame (Index b), the Local Geodetic or Navigational Frame (Index n)
and the Euler Angles for Aircraft Roll (0), Pitch (e) and Yaw (Y)
The matrix C, for vector transformation from the b- to the n-frame is composed of 3 transformation matrices
(C IOa,b,c)
Cn
= C(O)-C(e).C(Y).
=I
cos0
as''
c o s 0 sin''
-sine
sin+
s i n 0 cos''
cos+
sin''
sin+
s i n e sin''
+cos+
cos''
sin+
cos0
as+
s i n 0 cos''
+sin+
sin''
s i n 0 sin"
-sin+ cos''
cos+
cos+
case
'12
i
'l
l
'22
'31
'32
'13
32'
c33
102
In platform systems the 3 Euler angles are measured at the gimbal axes with angle encoders. In strapdown systems they are
computed from the elements of the C,, matrix:
(C lla,b,c)
0 =
tan-'
tan-'
sin-' C,,
c,,/c,,
c,, /cl,.
C . T h e Wander Azimuth Coordina : F r a m e
The wander azimuth coordinate frame (subscript "a", s. Fig. C 3) or the generalized local geodetic system [Ku 94, Section 7.31 is
rotated about the vertical with respect to the navigational coordinate frame through the "wander angle a".The angular rotation
vector between both frames is:
(CW
= (0
O:
oc)T.
north pole
wander a n g l e -
geographic-
(C 13)
c,,
CT" =
cosa
sina
sina
cosa
103
where Vx,y are the ground speed components in the a-frame:
(C 16)
3,Section
(Q},
=
v
cav,
(G}areHgain
the radii curvature of the reference ellipsoid in the meridian and the prime vertical plane
(V
=
6. .1].
-h)T
o?
T h e matrix C,, for vector transformation from the a- to the e frame is essential for the computation of position A, $ and the
north direction a:
cos$ cosa
-cos$ sina
-sin$ sinX cosa

(C 17)
c,,
C,;Cna
sin$ sinX s i n a
+cosX cosa
-cos$ sinX
-sin$ COSX sina

+sinh cosa
cos$ cosX
+cosX sina
sin$ cosh cosa

tsinh sina
-sin$
Cll
c12
c2,
c22
:;:
c3,
c3,
c33
C 5 The Actual Navigational Coordinate F r a m e

Including E r r o r s o f Platform Slewing o f T r a n s f o r m a t i o n Matrix Computation
The navigational coordinate frame (s. Section C 2) is the reference frame for navigation with respect to the earth. In the North
Indicating Platform INS (NIS) the platform is slewed to this direction (s. Fig. 1.1 and Section 2.2). In strapdown systems it is a
computed reference frame (s. Fig. 1.2 and Section 2.4). In both cases we have to assume a small angle misalignment of the
platform (index p) or the computed navigational coordinate frame (index n*) about all three axes:
i.e. the small angle vector E contains as its components the misalignments about the north, east and down axes as indicated in Fig.
C 4. The small angle transformation matrix C in case of the platform system or C,,. in case of the computed frame can be
nP
written as:
I
(C 20a)
= ( I
= C,,'
Cnp
E),
with I = unity matrix and E = skew symmetric containing the misalignment angles:
0
(C 20b)
E =
.ON -;
-ED
E,
+E
E X .
The right hand side of the relationship indicates that E may also be regarded as the vector cross product of
From both relationships it is obvious that:
( C 20c)
c:p
Cn.,
c:,.
Down,
(I
2,.
E).
1,
Fig. C 4 Misalignment Angles with Respect to the Navigational Frame
E.
104
Literature Appendix C
[Ku 94) Kumar, M.: Chapter "Navigation Coordinate Frames" of this AGARDograph
(Std 701 "Flugmechanik, Begriffe, Benennungen, Zeichen, Grundlagen LN 9300."
Beuth Vertrieb GmbH, Koln, 1970
105
Appendix D
Digital Readout of Inertial Sensors and
Strapdown Algorithms for the 'Analytic Platform'
D1 Introduction
In so-called "strapdown systems" the gyros and accelerometers are rigidly mounted to the vehicle and their output signals are
processed in the computer for attitude, heading, ground speed and osition computation. The sensor delivers analog signals in
general which are digitized and sampled at time intervals n * TS
= sampling time increment) for further processing in the
computer.
(g
Three different possibilities for digitizing the analog readout signal of inertial sensors are discussed in this section under the
aspect of subsequent high accurate integration. The digital integration process is the basis of attitude and heading computation
and inertial navigation. Its accuracy must be adapted to that of the measurement process. For the reference direction computation
with "inertial qualit)' with < .01 deg/h uncertainty, for instance, the integration error should be lower by at least 1 order of
magnitude. This makes it understandable that the following considerations apply primarily to the gyro measurements.
The three digitizing methods discussed are:
- the analog-to-digital converter (ADC), s. Fig. D 1,
- the voltage-to-frequency converter (VFC), s. Fig. D 2,
- and the pulse rebalance loop (PRL), s. Fig. D 3.
Not only the digitizing method is of importance for the subsequent integration process, but also the computation speed, i.e. the
sampling time and closely connected to it the algorithms for integrating the gyro and accelerometer readings into the computed
spatial reference direction, velocity and position. They are called "strapdown algorithms". This aspect will be briefly discussed at
the end of this section.
D2 Analog-to-Digital Conversion ( A D C )
The ADC converts the analog signal into a whole number digital signal with a resolution limited by the "least significant bit
(LSB)". Table D 1 lists some data of a high-accuracy ADC and Fig. D 1 shows the block diagram for a gyro with analog readout
and AD conversion. It also includes the subsequent sampling and the scaling by S', which comprises the gyro and ADC scalefactor.
The output signal U * is a whole number representation of the input signal w , but contains the sensor errors and the ADC errors
listed in Table D 1 plus the noise due to the limited resolution. Because of the roundoff process, information of LSB/2 may get
lost.
Table D 1 Data of the High Accuracy Analog-to-Digital Converter Type DAS 1153 of Analog Devices [AD 901
Resolution n:
Throughput ,Rate:
Nonlinearity
Nonlinearity Temperature coefficient (TC):
Gain T C
Zero T C
Power Supply Sensitiviv.
For ,
o
(D 1)
16 Bits (negative siga with two's complement)

25 kHz min
f 30 ppm full scale range (FSR)
*f 2ppm/K
max
8ppmlK max
f80 pV/K max

f10 ppm FSR/%V
= 500 deg/s the peak value of this quantization error (uncorreleted noise) is:
6w,
= fLSB/2
= f o,/zn
(-
55 deg/h );
The figures in ( ) refer to the hardware examples.

The digital angular rate roundoff noise has zero mean and is uniformly distributed over the range of
deviation is (s. [Be 661,Chapter 7):
(D 2)
u6w = f LSB/I/ 1 2
* LSB/2.
Its standard
= f .289.LSB (= f 3 2 deg/h).
In order to derive the time dependence of the so-called "random walk" angle, i.e. growth of the standard deviation of the integral
of the angular rate roundoff noise over time, we assume a first order integration algorithm whereby the angular rate W * sampled
at time intervals TS delivers the angle increment of A + = a* .Ts for each cycle. The standard deviation of one integrated angle
= u6@*TS.As the A # * s add up to the total angle so do the angle error
error increment AE is obtained in the same way:
increments AES add up to the total error E. Since the ALESare uncorrelated from cycle to cycle, their variances have to be summed
up over the integration time. With n = t/T' the standard deviation of the integrated angle error is:
This result holds for all integrated ADC output signals. It also holds for the first order integration process in any digital
computer, where in critical cases, as with the strapdown algorithms discussed below, the word length has to be properly adapted.
This characteristic is called "random walk", whose uncertainty increases with t1I2. We define as corresponding "random walk
coefficient":
rADC
= rfirst order integration process
= LSB.V-
(=
77.8 deg/ l/h ).
Applied to the ADC-digitized gyro signal, we obtain for an integr on cycle of 50 Hz, which is a common value in strapdown
inertial. navigation systems, a random walk coefficient of .075 deg/ r h . Since this value is intolerable for inertial navigation, the
ADC is not recommendable for this application. This stochastic error is avoided by using one of the other two digitizing methods
106
IMd
U'
.Y
Fig. D 1 Block Diagrams and Signals for Analog (a) and Digital Gyro Readout (b) Using an Analog-to-Digital Converter (ADC)
107
G'
QO
1
VFC
r----------1
I
I
--
' Y '
I
- 1
-ti
Fig. D 2 Block Diagram and Signals for Digital Gyro Readout Using a Voltage-to-Frequency Converter (VFC)
108
QO
G'
AP
Reset
Counter
Pulse Frequency Modulation
4 C'rP
Binary Pulse Width M o d u l a t i o n
IW '
switching of current pulse
A0
synchronized w i l h v o l l a g e
Ap
I Ap
Ap
Fig. D 3 Block Diagram and Signals for Digital Gyro Readout Using a Pulse Rebalance Loop (PRL)
I09
discussed below.
In Fig. D 4 simulation results are plotted proving the theory derived above. An ADC with LSB =
receives as input signal a
sine wave and the difference between the truncated and the exact sine wave was integrated over a time of 300 s. Ten runs were
simulated with different initial conditions for the input signal. The single input sample function shows the computed mean value
and standard deviation as well as the theoretical standard deviation according to Eq. D 2. The ten output sample functions are
plotted together with their
1 U bands computed for fixed times. The theoretical standard deviation according to Eq. D 3 can
thus b e compared with the calculated one - the correspondance is satisfying if it is kept in mind that it is based on ten sample
functions only.
A = 7,78
Asinot
CO
= 6,464 rad/s
T, = I s
$+-
LSB=103
xi for a single run
xa for 10 runs
computed mean and f C
theoretical and computed i-C
J ~ ~
J ~ ~
10.2
. .
300s
300s
-5 10.4
-10-2
Fig. D 4 Simulation Results for the Random Walk Error in the Integrated Output of an Analog-to-Digital Converter
D3 Voltage-to-Frequency Conversion ( W C )
In the VFC the analog output voltage of a gyro is converted into a pulse train, whereby each pulse A P carries the same
information proportional to the input voltage. In the case of a gyro A P it is an angle increment A+p. Its weight is derived from the
ratio OF voltage to frequency
(D 5 4
A P = umax/f,
(D Sb)
A W = wF/,
( = .18 arc sec for ,a
= 500 O/s
and ,f
= 1 MHz).
The frequency in the numerical example equals the high-accuracy synchronous VFC whose data are listed in Table D 2.
Table D 2 Data of a High Accuracy Synchronous Voltage-to-Frequency Converter Type AD652KP/BQ
of Analog Devices [AD 901
Output Frequency
Initial Gain Error:
Gain Temperature Coefficient (TC):
Linearity Error:
OfFset:
Offset:
Response Time:
lMHz
<*.5 %
< *25 ppm/K
<*SO ppm
<*2 mV
TC <25 pV/K
One period of new output frequency plus one clock period
This pulse train is fed to a buffer or reset counter which is read out at the sample time T by the computer and multiplied with
the scalefactor For processing by the integration algorithm. As indicated in Fig. D 2, the sampled angle increment A* is
proportional to a mean angular rate w .
110
The sampled angular increment ACP' is also corrupted by noise, known as "quantization noise" - at each sample time increment
just one pulse increment A W may have got lost. In reality this information is not lost, but saved for the next sampling cycle, as
can be seen from the simplified VFC block diagram in Fig. D 2. The input voltage U is integrated and the output of the integrator
is fed to a comparator. Each time the output voltage of the integrator exceeds a certain threshold, an electric charge increment
AQ of a precisely known area is released to reduce the integrator's net charge. The voltage pulse A P is fed to a reset counter
The roundoff error remains stored in the integrator. The functioning of the VFC in Table D 2 is different, but here again no
information is lost.
D 4 T h e P u l s e R e b a l a n c e Loop (PRL)
A different approach for accurate readout of inertial sensors, the so called "pulse rebalance loop (PRL)" or "incremental caging
loop", is shown in Fig. D 3. A quantizer is placed into the rebalance loop releasing a current i of constant amplitude to flow
through the torque generator in positive or negative directions for the pulse time increment .Ip.The maximum input angular rate
is thus defined by the current level i flowing in one direction only:
but reduced by the fact that the current is not applied continuously but as a flow of electric charges Q, which need time to reach
the maximum level. The resulting torquer scalefactor S' is thus reduced by a certain amount, 10 % for instance.
With the so-called "binary pulse width modulation" - this will be discussed only - the current pulse is applied in the positive and
negative directions during one cycle, whereby the switching of the current direction is controlled by the input signal.
Due to the inductivity of the torquer coil the pulse rebalance torquing frequency is fairly low. A common value is:
(D 7a)
fQ
= 1
kHz.
Each positive and negative pulse is measured with readout counting pulses of much higher frequency:
(D
f' = 256. fQ,
(the figure is taken from [Ra 77]), thus improving the resolution of one pulse to
(D 8 )
A W = o,/f'
(= 7 arc sec for ,a
500
O/s
).
The pulse evaluation is the same as mentioned for the VFC digitizing method.
Compared to the analog rebalance techniques and external digitizing process, the pulse width modulation torquing has the
following advantages:
- the dissipated rebalance energy within the sensor is constant and not proportional to w 2 as with the analog rebalance method
employed with the other two methods; this is an optimal basis for keeping the torquer at constant temperature and thus its
scalefactor at a constant value;
- it works only at two points of
the torquer characteristic line (torque = function of current), which is favorable for linearity;
- the scalefactor stability depends upon the stability of the constant current source only and not upon additional error sources of
the digitizing element (ADC or VFC).
Nonlinearities may arise in pulse-rebalanced sensors from eddy currents varying with the input signal. These effects are
minimized by observing certain rules in the design of the torquer [Ra 771.
Fig. D 5 shows two test results for the stability of a SDF gyro (Ferranti M 2519) and an accelerometer (Litef B 250) plus pulse
rebalance electronics designed at DLR [Ra 77).
The accelerometer readings were taken in 1976 during a period when earthquakes in China may have caused movements in the
foundation of the laboratory. The digital accelerometer readout, compensated for these movements, has a mean slope of 3 .
glday.
The gyro measurements were taken with the axes in the optimal orientation (output axis up). The readings prove that the stability
of t h e sensor in this position plus rebalance loop is better than 0.001 deg/h.
The pulse rebalance method may be used not only with inertial sensors but also with all sensors with compensation readout (e.g.
flow and pressure sensors).
D5 Computation Cycle and A l g o r i t h m s for Spatial D i r e c t i o n , V e l o c i t y and Position
I n t e g r a t i o n from Gyro and A c c e l e r o m e t e r D a t a
Due to the well known sampling theorem, the maximum frequency ,f

sampling frequency:
,f
contained in a signal to be sampled may only be half the
1/(2.TS).
In that case only the original signal can be recovered from the sampled signal. Once this sampling theorem has been violated, the
frequency spectrum of the sampled signal is contaminated by the so-called "aliasing error", which is impossible to correct
afterwards.
Fig. D 6 shows the aliasing error for the two cases that the signal vias digitized by the ADC and by the VFC or PRL method. For
illustration the sampling time has been chosen in both cases to be 20% longer than the sine wave period T. All sampled data
contain t h e aliasing error, i.e. they pretend a signal of much lower frequency, whereby the amplitude of the ADC-digitized signal
111
a ) Test Results f o r L i t e f Accelerometer
B 250
ooo uncompensated t e s t r e s u l t
... e r r o r due t o i n c l i n a t i o n o f foundation
xxx compensated t e s t r e s u l t
b ) T e s t Results f o r F e r r a n t i Gyro M 2519 (OA up)
Figure D 5 Test Results of a Rate Integrating Gyro Type Ferranti M 2519

and of a Pendulous Accelerometer, Type Litef B250 Including a Pulse Rebalance Loop [Ra 771
112
t-l
w*(TS = 1.2T)
Ts
Figure D 6 Aliasing Error Due to Sampling of an Analog Signal without Avaraging (ADC)
and with Avaraging over the Sampling Period (VFC, PIU) when the Sampling Frequency is Too Low
113
is identical to the input amplitude and the VFC and PRL digitized amplitudes are much lower. The latter ones are integrating
digitizing methods!
A common source for the high frequency content in the measurement signal of inertial sensors is the vibration due to the turbines
of an aircraft, for instance, or due to the piston motor of vehicles. In inertial navigation systems (INS) employing ring laser gyros
(RLG), the dithering mechanism of these sensors is another vibration source. The RLG is dithered with a 400 Hz frequency and
an amplitude of a few arc minutes, for instance. Neither for navigation nor for vehicle control the tracking of the high frequency
vibration is of importance, especially since its amplitude is fairly small in general. For inertial navigation another aspect is of
highest importance and because of this the sampling theorem has to be met: the core of the navigation computation is the
integration of the gyro and accelerometer measurements to the spatial direction reference, the velocity and position (s. Fig. 1.2).
Algorithm errors much smaller than the "inertial quality" mentioned above are required.
Let u s first concentrate on the computation of the direction reference, i.e. the computation of the C,, matrix in Eq. 2.21 and Fig.
1.2.
For motions in one plane only, the angle between two directions is the integral of a single gyro measurement - regardless of the
sampling time its accuracy at sampling intervals is limited by the quantization noise only. For spatial motions this is no longer true,
as visualized by Fig. D 7a. The so-called "coning motion" is shown here, i.e. a spatial sinusoidal motion about the two axes x and y
which are phase-shifted by 90. The analysis of the kinematics indicate, that the z-axis senses a continuous rotation, though the
coordinate system stays inertially fixed. If these angular rotations are correctly measured by the gyros and the algorithms for
computing the direction of the coordinate system are correct, the computed direction will stay inertially fixed as required. If the
sampling theorem is violated or the algorithms are of low quality, the reference direction computation will drift.
The so-called "non-commutativity, indicated in Fig. 7 b is another example of the requirements for spatial rotation integration. A
1" rotation about the x-axis and a 2nd about the y-axis bring the coordinate system into a position differing from that after a
reversed sequence of rotation. This shows us that in a vibrating environment the sampling theorem has to be met in a strict
sense, so that the computer can accurately follow the sequence of the spatial rotation.
The problem of strapdown algorithms touched off in Section 2.4 can mathematically be visualized after partially integrating Eq.
2.26 a:
where AOib(7) is the integral of the gyro measurements. The update of the first term on the right hand side of this equation is
not so problematic. It comprises the first order integration algorithm. The integral on the right hand side indicates the
contribution of the higher order terms. It depends not only on the integral of the gyro measurements, but also on the rate of
change of the transformation matrix, indicating that Eq. 2.26 can only be solved in the computer by means of approximations which
should be accurate enough so that:
- there is no drifting of
the computed reference directions,
- the computed transformation matrix is "normal", i.e. its unity vectors remain unity in magnitude and
-
the computed matrix is "orthogonal", i.e. its unity vectors keep their orthogonality.
If Eq. D 10 is chosen for the computation of the C. matrix, only 3 of its elements can be derived independently of 3 gyro
measurements. The other 6 can be computed by the retuirements of normality and orthogonality. Due to the former the quadratic
sum of one column's or r?ds elements must be.unity:
(D l l a )
CZl
Cii
+ c;2 + CZ3
+ Cii + c23i
1,
and due to the latter the dot product of one row and one column must vanish:
(D l l b )
CilCj1 + Ci2Cj2 + Ci3Cj3 = 0
Though the direct integration of the C,, matrix according to Eqs. D 10 and D 11 is not the preferred way for integrating the
gyro data, this matrix is the core of the strapdown algorithms. We have called it "analytic platform" in Section 2.4 because the
acceleration measured in the body frame is transformed with its help into the navigational frame. Neglecting Coriolis and gravity
compensation, the latter is:
(D 12)
= ',b(')
'
v b(t)
The transformation of the accelerometer output increments into the navigational frame and their summing up is an integration
algorithm of first order only neglecting higher order terms due to the integral on the right hand side of this equation. A sampling
frequency adapted to the vibration environment and higher order algorithms are prerequisite for an acceleration transformation
and integration with sufficient accuracy. Otherwise the so-called "sculling errors", resulting from in-phase transformation and
acceleration errors as illustrated in Fig. D 8a, may arise and act like measurement signal errors.
Still another acceleration error source - the so-called "size effect" - has to be taken into account in a vibrating environment. It is
not due to an algorithm error, but due to the fact that the three accelerometers have different distances to the axis of rotation.
This is illustrated with the help of Fig. D 8b. Choosing in Eq. 3.5 the body-fixed coordinate frame (index n = b), neglecting earth
rate and setting R = r, we may derive the following output signals of the north and down accelerometers for a vibration with 8 =
114
a) The Coning Motion
0,=
Y
0,=
an sin Rt
pn cos nt
Z
1
-apn
-2
0,
b) Non Commutativity of Rotations
dY
initial
position
x7
/"
x-axis
y-axis
after 90
1st rotation
about :
after 90
2ndrotation
about :
x+Y
X
/""
y-axis
' y
x-axis
Z
Figure D 7
The Coning Motion and an Illustration for the Non-Commutativity of Rotational Sequences
I15
1, - 4 0 0 H ~
a.
3 . 6 0 ~min
150/S
I7
A n /
w u v
i.p.4'1 J
product in noviga tion

computer causes
rectified
sculling
error
Y
sculling a boat
Acceleration Error Due to Size Effect
Fig. D 8 Acceleration Errors due to Sculling (a) and Size Effect (b)
116
8 sin u t about the east axis:
(D 14a)
(D 14b)
-(6)'-rX
= -
(b
rx
w cos ut)'
x rz = (rX t A?)
w 2 sin wt.
Transformed into the horizontal north-south axis, we obtain:
(D 15)
= a
cos 8 t a
X
(b
sin 8
rx [-cos2wt
U)'
t(l t
Arz/rx) sin'wt].
cos 2wt, i.e. zero in the mean, otherwise it is rectified. It is compensated in a RLG-INS.A survey of
For Arz = 0 we obtain a
computation and compens%ion problems in such a system is to be found in [Ma 821.
Two other ways of transformation matrix computation should briefly be mentioned
so-called "quaternions".
- the integration of the Euler
angles and of the
The rate of change of the Euler angles is governed by the differential equation [Std 701:
(D 16)
[i] [
=
sin+ tan8
cos+ tane
cos*
-sin*
sin*/cose
cos*/cose
1. [ i]
nb
with:
with
TI I '
"Schi --:r" feedback
. Section
gyro measurements.
The integration of the Euler angles and the computation of the C,, matrix according to Eq. 2.24 has the following properties:
Advantage:
Disadvantage:
it remains orthogonal and normal;

Eq. D 16 is highly nonlinear and has a singularity at 3 = f 90,
which corresponds to "gimbal lock" discussed in Section 2.3.
The transformation matrix computation based on the so- called "quaternions" is most commonly used. The orientation of the bodyfixed frame with respect to the navigational coordinate frame can be described by one rotation through an angle 6 about one
axis defined with respect to the reference frame through the 3 direction cosines [Br 55) (s. Fig. D 9). These 4 parameters are
comprised in the 4 quaternions, written as a column matrix:
Fig. D 9 Parameters Describing Quaternions
(D 18)
with
90
(qo
91
'1
%IT
q2
= cos(6/2)
q1 = cosa sin(6/2)'
q2 = COSSsin(6/2)
q, = cos7 sin(6/2).
Though the vector transformation can directly be based on quaternions using special algorithms, it is carried out with the
transformation matrix C,, in the strapdown computer. C,, is composed of the quaternion elements in the following way:
The quaternion update is based on the differential equation:
(D 20)
q = Q-w'/2,
where:
(D 21)
w ' = ( O
wy
ox
wz
IT
and
9,
Q =
-9,
91
9,
92
93
93
-92
-4,
-9,
-93
42
9,
-91
9,
9,
When initially both coordinate frames b and n coincide and the Euler angles (@o, l
U
' ,)oare
, zero, the quaternions are:
(D 23)
( 1
0)T.
For nonzero initial Euler angles the initial quaternions are derived by means of a successive rotation with 'Yo (i = l), 8, (i = 2)
and 4'o (i = 3), using Eqs. D 20 to 23:
(D 24)
q(i) = Q(i-1). Aq(i, i-l),
' , when a = fl = 90 and 7 = 0 it reads, for instance:

where Aq(i, i-1) is taken from Eq. (D 18). For the rotation with U
(D
25)
Aq(1, 0)
(c0s('U0/2)
~in('U,/2>)~.
The transformation matrix computed with uaternions is always orthogonal. The normalisation is based on the fact that the socalled "quaternion norm" has to be unity (9, + q2 + qf t q2 = 19: = I), its approximation reading:
The most severe environment for strapdown algorithms is within a ring laser gyro INS, where the whole inertial measuring unit
( M U ) is vibrating due to the dithering of the RLGs with a few arc minutes amplitude and 400 Hz frequency. The sampling
frequency in these systems has to be 3 to 4 times this dither frequency, i.e. it is above 1 kHz. The transformation matrix
computation and the velocity transformation are in general split up into 2 parts ]Sa 841. The results of a high frequency (>' 1
kHz) precomputation are the corrected angle and velocity increments. At a lower frequency (50 to 100 Hz) the integration of the
transformation matrix and the acceleration transformation are carried out. This can b e seen from Fig. D 10, the functional
diagram of the Litton LTN-90 INS.
References (Mc 68, We 78,Sa 841 give a good insight into the problems of strapdown algorithms.
118
CAC. HEM DATA
Fig. D 10 The Functional Diagram of the Litton LTN-90
L i t e r a t u r e Appendix D
[AD 901 "1989/90 Data Conversion Products Databook"

Analog Devices, , Inc., One Technology Way, P.O. Box 9106, Norwood, Ma. 02062-9106, USA
[Be 661 Bendat, J.S., Piersol, AG.: "Measurement and Analysis of Random Data."
John Wiley & Sons, Inc. New York, USA, 1966
IBr 5 5 ) Brandt, L.: "Vector and Tensor Analysis."
J. Wiley & Sons, Inc. 1955
[Ma 821 Mark, J.G. et al: "Design of RLG Inertial Systems for High Vibration".
CHl820-0/82/0000-0379 $00.75 1982 IEEE
[Mc 681 McKern, RA.: "A Study of Transformation Algorithms for Use in a Digital Computer"
MIT Master Thesis, No. T-493, 1968
[Ra 771 Rahlfs, D.: "On the Accuracy Improvement of Incremental Readout Techniques for Inertial Sensors".
ESA-TT-573, Translation of DFVLR Report No. DLR-FB 77-70, 1977
[Sa 841 Savage, P. G.: "Strapdown System Algorithms"
AGARD Lecture Series No. 133, Advances in Strapdown Inertial Systems, 1984
[Std 701 "Flugmechanik, Begriffe, Benennungen, Zeichen, Grundlagen LN 9300 (Flightmechanics Standards)"
Beuth-Vertrieb GmbH, Koln, 1970
[We 78) Wetzig, V.: "Computation of Attitude and Heading in a Strapdown Computer"
DFVLR Internal Report No. IB 153-78/32, 1978
I19
Appendix 0
Optical Gyros
0 1 Introduction
The so-called "Sagnac Effect" is the basis for optical gyros. It describes the effect of rotation upon 2 light beams travelling in
opposite directions on the same optically closed path. Their phaseshift is proportional to the rotation component about the axis
normal to the plane set up by the beams. This phenomenon was first observed by Harres in 1912 and by Sagnac in 1913 [Ha 12,
Sa 131.
The sensing element of the optical gyro thus is massless and in contrast to the mechanical gyro is not affected by the dynamic
environment. This feature appears to offer enormous advantages in a strapdown system (Section 2 and App. D). Therefore,
optical gyros seem to be the ideal sensors for use in navigation and flight test systems.
Optical gyros in use today are fiber optic gyros (FOG) and ring laser gyros (RLG).In the following sections a brief review of the
physics of the Sagnac ring interferometer and its use in the FOG and RLG is provided. The main sources of errors are also
discussed and examples are given for both of them.
0 2 T h e Sagnac Ring I n t e r f e r o m e t e r (SRI) as Basis f o r Optical Gyros
Around the turn of the century physicists discussed intensely the nature of light. Michelson contributed to the discussions in 1881
with his investigations about the effect of translatory motion of a light source upon the speed of the emitted light. For his
observations he used the interference of lightbeams travelling in the direction of motion and perpendicular to it. They did not
show any measurable effect, thus proving that the speed of light (c) is not affected by the motion of its source, but remains
constant with respect to inertial space. However, for the same reason, the investigations of Sagnac with lightbeams in rotary
motion, see Fig. 0 1, showed an effect [So 781.
The SRI setups in Figs. 0 l a and b consist of a light source (U),
a beam splitter (BS), three fully reflecting mirrors M1, M2 and
M3 in Fig 0 la, or a glass fiber in Fig 0 lb for guiding the t
m light beams I and I1 on a square or circular path back to their
common origin on BS, where part of their power is combined and projected onto the screen (SC). If the optical pathes for both
beams are identical, a photon emitted from BS on paths I or I1 will reach the origin again after time T = L/c, where L is the
length of the path and c the speed of light.
wave
M2
glass
sc
sc
Figure 0 1 Sagnac Ring Interferometer (a) and its Circular Substitute (b)
For derivation of the effect of rotation upon both beams we follow [Ro 851. Any rotation
the axis normal to the plane of both beams affects the travelling length of both beams:
(0 1)
L,,,, =
"' TLII
- positive as shown in
Fig. 0 1 - about
= u1,11. T1,Il~
with T = signal transmission time, v = w ' r = tangential velocity of the ring and
= signal transmission velocity. In another
120
arrangement:
(0 2 )
TI,,, = L/(U,,,,
An observer in the center
;4.
of rotation will measure the signal transmission velocities according to the relativistic addition theorem
as:
(0 3a,b)
U,,,,
(U
v)/(l
u-v/ct)
with co = vacuum light speed. The transmission time difference between both beams is:
(0 4a,b,c)
AT = T,
TI,
2 L v
=
2
CO
(1
--
- S2)
2 L v
2
CO
2 L r
w.
CO
with B = v/co. The time difference can also be interpreted as if the beams had travelled optical paths differing by the length. This
interpretation helps us to understand the ring laser gyro very easily.
It is interesting to note that this formular is valid for any signal travelling in a closed path in opposite directions, i.e. as well for
sound and for light! The technical problem lies in the measurement of the minute time difference - remember c," is in the
denominator! In practice it can.only be measured with light as transmitting signal. When the two light beams are brought to
interference, the time shift causes a noticeable Sagnac phaseshift of
(0 5 )
$s
= 2n uAT =
U
CO
ho
877 A
w = 3.57 arc min for w = 1 rad/s.
CO A0
with v = co/ho = frequency of light and A = Lr/2 = area of the circle. The figures in this equation are valid for the following
assumptions: a circular path of 10 cm diameter, a light source of He-Ne ( = 6.33 lo-' cm) and c = 3 10" cm/s. The numbers
reveal how weak the signal is.
It has been shown in [Po 671 that A is the area enclosed by the beams travelling on any path.
0 3 T h e Fiber Optic Gyro (FOG)
Though the FOG is in its principle closer to the Sagnac ring interferometer than the ring laser gyro discussed in the next section,
its development began a t a time when the latter was commercially already available. This is due to the fact that the proper
technology and readout concepts became available only in the late 1970ies.
Its basic principles are shown in Fig. 0 l b with the output signal according to Eq. 0 5 .
Three principle obstacles had to be overcome in order to make the FOG a serious competitor to existing gyros:
- the low scalefactor according to Eq. 0 5 ,
- the readout insensitivity at low input rates and
- readout nonlinearity as well as ambiguity.
0 3 . 1 S c a l e f a c t o r I n c r e a s e and E r r o r Reduction
With the advent of fibers as optical wave guides the increase of the Sagnac ring interferometer sensitivity became easily
accessible. Each turn of a fiber optic coil increases the sensitivity in Eq. 0 5. This idea was first suggested in [Kr 681. But along
with this benefit went adverse effects, since light was increasingly brought in contact with matter.
The resulting increase of temperature sensitivity and magnetic sensitivity was reduced by the use of polarization preserving fiber
and designing the sensor setup in a reciprocal fashion [U179, U1 801, in which both lightbeams suffer the same fate on their path
from light source to light sink or they follow the same number of reflections and transmissions. This is accomplished by using 2
beam couplers so that both beams are reflected and transmitted twice (s. Fig. 0 2).
Superluminiscend
Diode (SLD)
Polarizer (Pol)
CP 2
Figure 0 2 Functional Diagram of a Reciprocal Fiber Optic Gyro Setup
121
The increasing pathlength of both optical beams increases the effect of backscattering and output noise, since the backscattered
light causes a phaseshift of the light beams in a similar fashion as the the Sagnac effect. Light from a luminiscent diode with very
short correlation length brought a drastic improvement [Pe 821.
03.2 Increasing Sensitivity f o r Low Input R a t e s
Fig. 0 3 may serve to explain the problem. It shows the superposition of the two countertravelling beams and the corresponding
light intensity as a (1 + cos$,)- function of the Sagnac phaseshift $ with zero scalefactor at zero input rate. Only a phaseshift of
f n / 2 would bring the sensor into the desired range of highest senztivity at low input rate. Such a "nonreciprocal" phaseshift $nr,
i.e. a phaseshift acting upon both beams in a different manner, is generated by the phase modulator PM in Fig. 0 2 at one end of
the fiber coil.
The phase modulator PM causes a photon of beam I to be delayed after having travelled through the coil and a synchronous
photon of beam I1 to be delayed before entering the coil. PM is stimulated with a squarewave voltage U
for an individual
phaseshift of n/4 and a period of twice the light travelling time nL/c with n = refraction index of the #er. The resulting
phaseshift $ is the difference of both beams, i.e. equal to f n / 2 . It is superposed to the Sagnac phaseshift and causes a lightintensity-fun%on proportional to (1 + cos($
n/2)] as indicated in Fig. 0 4 left. It is demodulated and results in an output
voltage U
sin$. For low input rates the sensitivity of this "phasemodulated" FOG is thus proportional to CO, as expected. For
high input rates it is both nonlinear and ambiguous. The proper signal evaluation in so-called "phasemodulated FOGS" (s. Fig. 0
Sa ) is then carried out digitally.
Phase modulation can be accomplished by simply stretching the fiber with piezo elements or better by applying an electric field
perpendicular to the fiber. These phase modulators are manufactured in integrated optics based on the use of lithium niobade.
03.3 Scalefactor Linearization with a Control Loop f o r Sagnac Phaseshift Compensation
The scalefactor of the Sagnac ring interferometer can only be linearized through a control loop for nulling the Sagnac phaseshift
4s,i.e. a method must be found for generating a controllable constant nonreciprocal phaseshift between both beams. This can be
accomplished by a controllable nonreciprocal frequency difference 4, = f, between both beams, because its corresponding
phasedifference is $nr = f m - nL/c.
The breakthrough for such a nonreciprocal "frequency" difference came by the proposal to stimulate the phasemodulator with an
increasing voltage ramp as indicated in Fig. 0 4 right, causing a linearly increasing phaseshift in both beams, with the one in beam
I time-delayed by n * L/c [Le 851. The phaseshift difference 4 of both beams is controllable by the ramp's inclination as
indicated above and can be used to compensate the Sagnac phaseshyh $S.
The steps superposed to the ramp serve to sense the Sagnac phaseshift as discussed with the phasemodulated FOG mentioned
above, i.e. their demodulated output is controlled to zero by the ramp's inclination as indicated in this figure.
The voltage ramp for increasing phaseshift has to be set to zero again at a certain limit. This is done when the phaseshift has
nominally reached 2 n as indicated in Fig. 0 4b. At that instant a pulse is released to the computer whose weight is:
It is identical in principle to the one of the ring laser gyro (RLG) discussed in the next section.
Differences between the actual phase release and its nominal value remain stored in the fiber and serve to stabilize this "closedloop" FOG scalefactor.
This closed-loop FOG (s. Fig. 0 Sb) has for the time being the best performance data, namely
scalefactor variation. As compared to the RLG it has the following advantages:
- it is not subject to the so-called "lock-in
5 deg/h drift and 5 500 ppm
effect" of the RLG and no mechanical dither is required to compensate it;
- the use of integrated optics promises that it will become cheaper;

- the solid state laser instead of a gas laser required for the RLG promises a longer lifetime;
- t h e pulse weight and the bandwidth is limited by the speed of
the electronics only.
Though these advantages are of great importance for many applications, it does not seem likely that the FOG with conventional
dimensions (< 10 cm radius) will reach the performance of the RLG. Single pieces have been developped with bigger diameter
and superb performance as regards drift [Sc 921.
0 4 T h e Ring L a s e r Gyro (RLG)
The setup for the ring laser gyro (FUG)in Fig. 0 6 differs in principle from the one for the Sagnac ring interferometer in Fig. 0
1 in the following way. The light source - a gas laser - is mounted into the optical path replacing the exterior light source plus
beam splitter; the optical path layed down by mirrors M1,M2, M3 acts as a resonance cavity for the laser.
In this particular device the optical path is composed of three mirrors only. This is pursued at present by most of the
manufacturers with the exception of Litton, which selected four mirrors. The RLG used by St. Petersburg Electrotechnical
University for high accuracy goniometry has 4 prisms instead of mirrors [Fi 941.
In Fig. 0 6 the mirror (Ml) is partially transmitting and the prism (P) is part of the interference optics.
Like in the linear laser the ring laser is tuned to an oscillation with a wave length equal to an integer of the cavity length L. Due
to the Sagnac effect, the cavity length of both light beams I and I1 will apparently be displaced from L by an amount AL, = c' A t ,
and AL,, = c * At,,, or relative to each other by an amount AL (s. Eq. 0 1). This results in the frequency difference:
122
Field strength of beams I and II :
El, II = a exp j (2 n: vt
A
2rcx rt -$
cp )
+h
Photo diode PD measures : I = (E,+ E,,)*

IO
= - (I
2
+ cos cp,) - U (q,)
Figure 0 3 The Interference of the 2 Light Beams and the Corresponding Light Intensity a s a Function of their Phase Shift
123
Phase-Modulation with
Changing Direction
Digital Ramp plus

Square-Wave Bias
Modulation
lTl+
I
I
Cp nr
Figure 0 4 Two Kinds of Phase Modulation in a Fiber Optic Gyro (FOG)
124
a)
Superluminiscend
Diode (SLD)
- I, sin (cp,
0s-0
U,
+ 6cp)
b)
Super1uminiscend
Diode (SLD)
Polarizer (Pol)
CP 2
Voltage Controlled
Amplifier
0
- -
/-
Demod.2
Control
Electr.2
-+
Reset-Ramp
Gene rator
Reset PulseSA4
Figure 0 5 Functional Diagrams for Fiber Optic Gyros with Two Kinds of Phase Modulation:
Square-Wave Modulation (a) and Digital Ramp Modulation (b)
125
photodiodes 1
plane mirror 3
cathode
Figure 0 6 The Ring Laser Gyro (RLG) in Principle
wandering
Figure 0 7 Interference of the 2 Light Beams of the Ring Laser Gyro and Interference Pattern Readout
I26
Au[Hz]
A
L A
w[deg/s]
= 1,588
for a triangular path with a side length of 10 cm and A = 6.33 10-7 m (He-Ne laser).
This frequency difference causes a wandering of the interference pattern at the screen (SC) proportional to the angular rate with
respect to inertial space. This is indicated in Fig. 0 7.
Two photodiodes mounted on the screen (SC) will count the number of fringes that pass it and allow the definition of the sense of
rotation. Each fringe may then easily be converted into a pulse which indicates the angle increment of the sensor with respect to
inertial space. The number N of the pulses for the total input angle displacement:
defines the RLG nominal scale factor S, which for = 6.33 10-7111 and a triangular RLG of 10 cm side length has the following
magnitude:
=
Lh = 2.25 arc sec/pulse.
N
4 A
In Fig. 0 8 a setup is shown indicating the RLG characteristics. It seems as if the mirrors were passing alongside the space- fixed
light pattern [Sc 661. This setup indicates that the RLG is an integrating gyro with digital output and ideally suited for use as a
flight test sensor and in strapdown navigation systems.
(0 9)
s =
We will discuss in the following some design criteria and error sources common for all RLGs and compare them with those of the
FOGS, if possible.
Figure 0 8 Model for the Functioning of a Ring Laser Gyro

0 4 . 1 Design Criteria Common to all Ring Laser Gyros
Let u s first comment on the selection of the laser - gas laser or solid state laser.
In difference to the FOG h i c h operates at the fixed frequency of its exterior light source, the RLG operates at two frequencies
both of which have to be amplified inpendently in one laser. This can be accomplished in a gas laser only in which due to the
Doppler effect of its atoms the intensity of the amplified light as a function of resonator length or lasing frequency, respectively,
is bell- shaped with a bandwidth of 1500 MHz, approximately corresponding to 1 wave length A.
Gas flow within the laser tube caused by the voltage between anode and cathode (Langmuir flow) and temperature gradients
affects the lasing frequency (Fizeau effect). For its overcoming RLGs are always equipped with two anodes and one cathode, or
vice versa, mounted symmetrically within the tube, as indicated by Fig. 0 6. If only one anode were to be used, the result would be
an apparent rotation of the sensor proportional to a multiple of 100 deg/h [Po 681. The sensor may be biased to compensate for
some of these effects by adjustment of the high voltage (ca. 1500 V) in both branches: The number just quoted indicates that the
voltage should be properly controlled and temperature gradients should b e prevented.
The natural He-Ne laser wave length is A = 633 nm. The exact lasing wave length is adjusted due to the geometric path length,
and we may assume that its mean value for both beams varies linearly within one A-period. Within that path length range the
RLG scalefactor is insensitive to path length variations, since in Eqs. 0 7 to 0 9 the variations of nominator and denominator
cancel!
127
signal at photo diode

AC-filter
Airror2
rl
..~
........................
:.
controller for
minimizing
lock-in threshold
I
I
pathlength
controller
I
\
Figure 0 9 Control Loops for Stabilizing the Beam Geometry of a Ring Laser Gyro
For preventing the path length to swap the A-period and for scale factor stabilization, the RLG is firstly built of material with a
very low thermal expansion coefficient (Cer-Vit, Zerodur) and secondly is equipped with a path length control loop as indicated in
Fig 0 9. The input signal into this loop is the mean intensity of both beams measured at the backside of one of the mirrors (all of
them are partly transmitting). Its control to its relative maximum controls at the same time the path length to a constant value!
This is accomplished by positioning one of the remaining mirrors with a piezo element. The mean light intensity maximum
definition is done by dithering this piezo element with a frequency w and filtering out the 2w- term.
The path lenght control to the maximum of the mean light intensity makes any other temperature control of the sensor
unnecessary. We may thus state two essential advantages of the RLG over other gyroscopic sensors: digital output and
t e m p e r a t u r e insensitivity - at least in theory.
0 4 . 2 Error Source Common to all
Ring L a s e r Gyros - t h e Lock-In Effect
A RLG incorporating the above design features will work satisfactorily for high input rates limited only by the readout electronics.
But the RLG has a lower limit on input rates of a few hundred degrees per hour called "lock-in" range (S. Fig 0 10). Above this
range its actual input-output relationship is approximated by (s. also Eq. 0 7):
(0 IO) A u
4 A
L A
[ 1
(wL/o)2
11'2
w,
with w L = "lock-in rate".

This lock-in effect is experienced by any two weakly coupled oscillators with slightly natural frequencies and thus known in other
fields of physics, too. If this oscillating system is excited near their mean natural frequency, both subsystems will oscillate at the
same frequency. This phenomenon is of benefit in acoustics (tuning of different violins in an orchestra) or electronics (mechanical
and electronic coupling in a quartz watch), for instance yet, it is the main problem in a RLG. The beat frequencies between both
beams and sensed at the output diode will be experienced only, if the oscillation frequencies are distinctly different as to be seen
from Fig. 0 10.
The coupling of both light beams is caused primarily by the scattering of light at the mirror surfaces (s. Fig 0 11 top) and within
the laser tube. It cannot be completely overcome. Its reduction requires on the one hand mirrors with low backscattering and high
reflectivity. It requires on the other hand an additional control loop for the light beam geometry ( s . Fig. 0 9 ) , since wL is the
result of the superposition of the backscattering of all sources within the resonator, especially at the mirrors (S. Fig. 0 11
bottom). Due to the phase-relationship of all backscattered lightsources the superposition of all of them can be driven to a
minimum [Ro 921. For present RLGs of inertial quality and with a beam length of 10 to 15 cm wL is in the order of 100 deg/h.
The control loop for minimizing wL stabilizes this parameter at the same time, thus stabilizing the RLG scalefactor as well (S. Eq.
0 10 and [Ro 921).
The sensitivity threshold mentioned above is still too high for use in an inertial navigational system. Among the different
128
The Lock-In Effect of the Ring Laser Gyro
Lock- In Compensation: Rotational Oither
Input-Output with Dither
typically : wo= 100 i o 300 dep/s
d)
Scale Factor for Dithered
x=y1
Ring Laser Gyro
with dber
0.43 x 1 ~ - 9
-6
1.3~10
0.
2
I
I
without dither
II
I
w' 0
0L ,
W'
=--
U
'
- 300 deg/h
WOu
200.000. deg/h = 5 6 d q / s
Figure 0 10 The Lock-In Effect of the Ring Laser Gyro and Bias Techniques to Overcome it
129
reflected and backscaltered

light energy of beamI:
backscattered energy visible
as red point on
Figure 0 11 Backscattering at one Mirror and within the Resonator

techniques for 'biasing" the RLG, i. e. making it sensitive to low input rates (mechanical dither, mnstant rotation, magnetic
mirrors, Faraday ccU) the mechanical dither is most mmmonly in use. As indicated in Fig. 0 10, the RIxi is rotated within its
a s c pcricdicaUy with a frequency of 400 Hz and an amplitude of a fnv arc minutes approximately. Distinctly to be seen in Fig. 0
I2 is the dither spring of the Honeywcll RLG in the center of the lasing block. For its mmpensation at the output signal this RLG
has a case-fucd prism and readout as indicated in Fig. 0 12 bottom.
Since this bias is applied periodically, the gym will enter tbe lock-in region twice per dither cycle and each beam aill loose the lock
of phasc to some extent The resulting noise accumulates in a so-called 'random walk' after integration of the mcasured rate to
an angle. The uncertainty of this crmr u p r c s c d as Iu band increases with the q u a r e mat of the elapsed time. This
is particularly serious d e n short measurement times are of interest as vith the INS alignment (E. Seetion 5 ) or with the extraction
of the sensor's drift.
Another ermr source with the dither bias is due to the scalefactor nonlinearity at input rates near the madmum dither rate (a.
Fig. 0 IOc). It arises from the fact that at this point the highly nonlinear maximum dither sine wave dives into the lock-in region.
This scalefactor nonlinearity show an almost square law dependence on the lock-in rate amounting to 50 ppm at a lock-in rate of
0.4 dcg/a. For its wcrmming the dither frequency in the order of 400 Hz and a few arc minutes amplitude arc
appm-ately
varied randomly.
Among the other bias techniques mentioned abwc the oncs h a d upon magnetism (magnetic mirrors. Faraday ccU) seem to open
the most elegant way to mpc with the lock-in effect. They arc hampered by the fact that both of them increase backscattering
within the resonator and from this point of view arc not promising for highquality sensors. For the time being dither bias is most
widely spread for inertial hystems of 1 NM/h performance. In spcfial syatem implementations as in inertial systems for
submarines we also find rate bias. These hystems have proven already navigation accuracies in the range of 1 NM/d [KO 901 and
seem to have a gmwth potential [Ro 921.
Corn
rciauy available RLGs have excellent performance data: <0.01 dcg/h drift, < 5 ppm scalefactor variation and < 0.001
d e g / r h random walk [U881.
130
@ Block Mdlon
Due to Dllher
Readoul
Mirror
Fringe Motion
Readout
Prism
Dllher in Laser Output
Ddeciorr
Equlvalenl Readoul
@Mollon Due to Dllher
Figure 0 I 2 The lloneyvell GG 1342 hng Laser Gyro and 11s Optomeehanical Readout Compensation for Dither
131
Literature Appendix 0
[Fi 941 Filatov, Y.U. et al: "Laser Goniometric System for Measurment of Object Angular Position."
Proceedings of the 3rd International IMEKO Symposium on Laser Metrology for Precision Measurement and Inspection in
Industry, Heidelberg, March 1994,VDI-Report No. 1118
[Ha 121 Harres, F.: "Die Geschwindigkeit des Lichtes in bewegten Korpern."
Dissertation Universitat Jena, Erfurt (1912), no longer available
[KO901 Kohl, KW.: "A new High Accuracy Ship's Inertial Navigation System PIA1 MK4."
[Kr 681 Kroy, W., Mehnert, W.E.: "Anordnung zur Messung von Drehgeschwindigkeiten."
DBP Nr. 1 807 247, 1968
[Le 851 Lefevre, H.C. et al: Double Closed-Loop Hybrid Fiber Gyroscope Using Digital Phase Ramp."
Proceedings of the 3rd International Conference on Optical Fiber Sensors(0FS '85), San Diego, PDS 7.1-7.8, 1985
[Po 671 Post, E.J.: "Sagnac-Effekt."
Review of Modern Physics, Vol. 39, No. 2, (1967), pages 475 to 493
[Pe 821 Petermann, K,: "Intensity-Dependent Nonreciprocal Phase Shift in Fiber-optic Gyroscopes for Light Sources with Low
Coherence."
Optics Letters, Vol. 7, No. 12, pp 623-625, 1982
[Po 681 Podgorski, T.J.: "Langmuir Flow Effects in the Laser Gyro"
IEEE Journal of Quantum Electronics, Vol. QE-4, No. 1 pp 11-18, 1968
[Ro 851 Rodloff, R: "Der Laserkreisel - EinfluD der Ringlaser-Geometrie auf die Kreiseleigenschaften."
Laser und Optoelektronik Heft Nr. 2, 1985
[Ro 921 Rodloff, R: "Lasergyro - the Next Generation."
[Ro 931 Rodloff, R et al: "Hochaufloesendes Kreiselsystem zur Prazisions-Winkelmessung und
Deutsche Patentanmeldung P 42 31 935.8, Anmeldetag 18. 09.1992
- Navigation"
[Sa 131 Sagnac, G.: "Comptes Rendus, 157, (1913)

[Sc 921 Schroeder, W. et ai: "Telescope Pointing and Tracking with Optical Gyros."
[Sc 661 Schulz-Du Bois, E.O.: " Alternative Interpretation of Rotation Rate Sensing by Ring Lasers."
IEEE Quantum Electronics. Vol QE-2,pp 299-305, 1966
[So 781 Sommerfeld, A.: "Optik."
Verlag Harri Deutsch, Thun, Frankfurt/M, 1978
[U1 791 Ulrich, R: "Fiber-Ring Interferometer Polarization Analysis."

Optics Letters Vol. 4, pp152-154, 1979
[U1 801 Ulrich, R: "Fiber-optic Rotation Sensing with Low Drift."
Optics Letters Vol. 5, pp 173-175, 1980
[Ur 881 Ullrich, D.: "Static and Dynamic Angular Measurements with a Ring Laser Gyro."
132

Heinz Buell
GEC-Marconi Electronic Systems Corporation
164 Totowa Road
Wayne, NJ 07474-0975
U.S.A.
1. INTRODUCTION
The use of the Doppler principle to measure

velocity has been underway since the early
1950's. The Doppler principle or effect has
been applied to many systems in which
velocity is an important variable. The
discussion herein will concentrate on the use
of a self-contained radar in an airborne
vehicle to measure the velocity of that
vehicle. Fixed-wing aircraft and helicopters
have used Doppler radars successfblly for
many years, and drone aircraft and missiles
are beginning to exploit the low cost and
high reliability of Doppler radars.
The Doppler effect or frequency shift was
first described by the Austrian physicist,
Christian Doppler (1803-1853) for the case
of sound waves. This effect is also
characteristic of light and microwave energy.
In each case a shift in frequency is observed
when the source of radiation is moving
toward or away from an observer of that
radiation. The frequency shift increases
when the source moves toward the observer,
and decreases when the source moves away
from the observer.
Airborne Doppler radars are both the source
(transmitter) and the observer (receiver) of
the microwave energy utilized to measure
vehicle velocity. A typical Doppler radar
transmits a small amount of energy toward
the ground, and measures the Doppler shift
in that portion of the transmitted energy
reflected or back-scattered toward the
receiving antenna. The resultant Doppler
frequency shift is directly proportional to the

vehicle's velocity relative to the ground.
Note that two frequency shifts occur in this
case; one between the energy (source no. 1)
and the ground (receiver no. 1) and another
between the energy reradiated from the
ground (source no. 2) and the receiving
antenna (observer no. 2).
The frequency shift is proportional to that
component of vehicle velocity which is
parallel to the direction of the radiation.
Radiation that is transmitted perpendicular to
the direction of the vehicle's velocity, or
velocity vector, will not experience any
Doppler shift. Radiation transmitted parallel
to the velocity vector will experience the
maximum frequency shift. One beam of
radar energy will therefore provide a
measurement of one component of vehicle
velocity. In general, three or more noncoplanar beams must be used if the three
orthogonal components of vehicle velocity
are to be measured.
Doppler radars to measure velocity of
aircraft have been in production and in use
~ over 20,000
since the early 1 9 5 0 ' ~and
systems have been produced since then by
several manufacturers. Some of these
systems provide only velocity outputs
whereas others also perform navigation,
guidance and control and display fbnctions
by using data obtained from the vehicle's
attitude sensors. Some Doppler radars
measure not only velocity but also the
133
vehicle's altitude above the ground, and

thereby eliminate the need for a separate
radar altimeter.
The size, weight and cost of Doppler
velocity sensors and navigation system has
decreased dramatically since 1950; weight,
for example, has dropped from
approximately 160 Kg for the ANIAPN-8 1
to less than 4.5 Kg for Dopplers now in
production for drone aircraft, and to less
than 5.9 Kg for helicopter Dopplers.
It can be shown that

if V = aircraft's speed relative to the
ground
c = speed of propagation of
electromagnetic energy
f, = transmitted frequency
L = "looking angle'' - (angle between
the line of flight and the
transmitted beam direction)
and fd = the Doppler frequency shift,
then f d = B f,cosL
Performance of a Doppler radar is

determined by a number of basic
system design characteristics. The major
design trade-offs will be discussed, including
the resultant performance variations, and a
typical Doppler design will be described.
The key error contributors will be listed and
summarized; errors of a particular system
can be described in detail by the
manufacturer of that system.
Cost of a Doppler radar depends upon its
performance, and also upon the quantity
being purchased and the rate of production.
In general Doppler velocity sensors are
inexpensive alternatives to high-quality
inertial systems, and in fact, a Doppler radar
and a low-cost inertial system are less
expensive than a high-quality INS.
Note that V cos L is the component of

velocity in the direction of propagation.
The wavelength A. is related to f, and c by
Equation (1) can therefore be written as:
Equations (1) and (3) constitute the basic

expression for measurement of velocity by
applying the Doppler principle to the
propagation of electromagnetic radiation.
LTNE
2.0
2.1
Principles of Operation of a
Doppler Radar
(1)
FLTGMT V
The Dopder Equation
When the Doppler effect is utilized in an

airborne Doppler velocity sensor, the source
is a microwave transmitter located in an
aircraft moving with a velocity V relative to
the earth's surface (see Figure 1).
Figure 1. The Basic Doppler Equation
134
2.2
Determining the Total Velocity

Vector
A single beam such as that shown in Figure 1

could be used to determine the aircraws
speed relative to the ground, provided that
the aircraft speed were horizontal with no
vertical velocity and no sideways "drift"
(assuming that A and L of equation (3) were
known). Normally, however, the aircraws
speed does contain components of vertical
velocity and drift. To determine the total
velocity vector, V, it is therefore necessary
to employ at least three (non-coplanar)
beams. The velocity so determined is relative to a frame of reference attached to the
antenna assembly providing the beams; i.e.,
relative to the aircraft's coordinate system,
(X,Y,2).
A Doppler radar velocity sensor is ideally
suited for use in helicopter hover control,
fire-control and weapons delivery and in
aiding inertial navigation systems (INS) since
these systems require velocity in air-frame
coordinates. Velocity sensors such as
gimballed INS and GPS receivers measure
velocity in geographic coordinates, and true
airspeed sensors measure velocity relative to
the air mass. Doppler radars are thus a
unique source of velocity data for many
functions.
Consider an antenna assembly generating

four beams A, B, C and D as shown in
Figure 2. The Doppler frequency of the
beam is (from
signal received in the
equation (3):
Any three of the equations can be solved for

V,, V,,, and V,, assuming that the Doppler
frequencies, the wavelength and the beam
direction in the x, y, z coordinate system are
known.
The fourth equation provides redundant
information that can be used to test overall
performance of the system. For example, if
V,, V,, and V, as computed from N = 1,2, 3
agree with V,, V,, and V, computed from,
say, N = 1, 2, 4,then the overall operation of
the system is valid. Such ''redundant''
velocity checks are an important part of the
BITE function of modern Dopplers.
I)
X'
Figure 2. Total Velocity Vector Geometry
135
2.3 Antenna Stabilization
For most applications of Doppler velocity

sensors it is necessary to find the horizontal
(VH,V,) and vertical (V,) components of
velocity. The horizontal components, in
particular are needed to perform deadreckoning navigation. The (x,y,z) velocity
components must be transformed through
vehicle pitch and roll to obtain V,, V, and
V,. Pitch and roll are usually obtained from
the vehicle's attitude sensor (vertical gyro or
inertial navigation system). Almost all
modern Doppler radars utilize antennas that
are fixed or body-mounted, and this
transformation is performed in a digital
computer that may be in the Doppler itself,
or in a separate "mission" computer. Some
older Dopplers had antennas that were
mounted on pitch and roll gimbals so that
their outputs of V,, V, and V, were the same
as V,, V, and V,.
Such mechanically stabilized antennas
eliminated the need for "data" stabilization of
V,, V,, and V, and also provided better
performance during maneuvers over water.
These systems were heavier and required
considerably more volume for the gimbal
structure as well as a separate radome. For
these reasons most modern Dopplers use
fixed antennas.
eliminated by mixing or heterodyning the

received signal with a portion of the
transmitted signal. The major remaining
sensitivity to frequency is that of the antenna
beam geometry or (y, a,$), which depends
on the antenna design itself.
Antenna Beam Geometrv

Considerations
2.5
Equation 4 shows that fd depends on beam

geometry, that is, upon (y, a , $) of each
beam. These equations can be combined to
obtain V,, V, and V, as a function of the 12
beam angles. It can be shown that the
velocity errors are given approximately by
EV, = Tan
y o Ey
(5A)
V
X
where y o
Ey
average y angle of the

4 beams, and
average uncertainty of y .
VY
where a,,
average a angle ofthe

4 beams and
2.4
Ea
Transmitter Freauencv
Considerations
Doppler radars have been allocated the

frequency band of 13,325 f 75 MHz. Some
earlier Dopplers operated in the X-band, in
the vicinity of 8,800 MHz but most of these
have been retired from service.
Equation (1) shows that the Doppler
frequency shift depends on the transmitter
frequency f,. This sensitivity is virtually
&, =
vz
where
$0
E$
average uncertainty of U
(C)
Tan
$o%$
average $ angle of the

4 beams and
average uncertainty of $.
The direction of a beam in space can be

defined in terms of two angles; the angle $
136
can be expressed as a function of y and a.

Selection of y and a is one of the
fundamental trade-offs of a Doppler radar
design. A large y and a large a (small JI)
results in beams pointed nearly straight
down, and thus provide a relatively high
returned signal strength. This is particularly
important when flying over water where the
radar back-scattering decreases with
increasingly smooth water. Figure 3 shows
the radar back-scattering versus incidence
angle. (The beam angle 9 during level flight,
is the same as the angle of incidence). It can
be seen from Figure 3 that the back-scattered
energy is higher for land and rough water,
but diminishes sharply with increasing
incidence angle over smooth water.
Extensive operation over water therefore
favors a smaller $ and thus larger y and
a angles.
Equations (5A) and (5B) show that the

sensitivity of EV, and EV, to beam angle
uncertainty (Ey and E a ) increases sharply as
y and a increase. Thus velocity accuracy
favors smaller y and a angles and a larger JI
angle. Doppler radars have been designed
with y and a angles as small as 65", but
were intended primarily for over land and
low altitude flight. Most Doppler radars
have larger y and a angles - of the order of
75" - which results in somewhat greater
sensitivity to beam uncertainties but better
overwater performance. Some Dopplers
have angles as large as 80" but are intended
for operation primarily over water. Beam
angles in the 72" to 75" range are a
reasonable compromise for Dopplers on
aircraft that may experience a variety of
missions.
Equations (SA) and (SB), when evaluated for
beam angles of 73" show a sensitivity of
velocity error of 3.9% per degree of beam
uncertainty. Beam geometry must therefore
be known to better than 0.06 degrees if
velocity accuracy is to be better than 0.25%.
Normal production tolerances make it very
difficult to achieve these beam accuracies;
this problem is overcome by measuring the
actual beam geometry of each antenna on a
radar range and then applying the resultant
"calibration" factors to the velocity outputs.
2.6
Figure 3. Scattering Cross Section vs.

Incidence Angle (y)
Dopnler Velocitv for Navigation
A Doppler dead-reckoning navigation system

integrates the north and east components of
velocity to determine the change in present
position, and adds this change to initial
position to obtain present position. The
horizontal velocity components, V,, and V,,
must therefore be transformed through the
aircraft's heading angle, H, to obtain V, and
V,. The vehicle's heading reference is used
to provide H and can be a magnetic compass
137
or a more sophisticated and costly inertial

navigation system. The transformation is
usually performed in the Doppler itself, when
it is configured as a navigator, or in a
separate mission or navigation computer
when the Doppler is used as a velocity
sensor.
2.7 Types of Transmission Modulation
Four types of transmission systems have

been used in Doppler radar sets: continuous
wave (CW), frequency-modulated CW (FWCW), coherent pulse, and incoherent pulse.
A free-running highly stable CW oscillator is
required to supply the transmitter energy in
CW, FM-CW, and coherent pulse systems.
In each case, the received signal is mixed
with a portion of the stable transmitted signal
(or reference oscillator) to achieve
coherence, and the beat signal is detected.
In an incoherent pulsed system, succeeding
pulses are emitted with random phase (as in
the case wherein a magnetron is used for
transmission). In such systems, the requisite
coherence for recovery of Doppler
information is achieved by mixing two
echoes derived from the same transmitted
pulse (Janus mixing); such a system is
frequently called self-coherent. In an FMCW system, the transmitter is frequency
modulated, as the name implies. The
received signal is mixed with a portion of the
transmitted energy (thereby achieving
coherence) and the Doppler shift in a
selected sideband is used. The first or J,
sideband is frequently used since its
amplitude versus altitude is nearly flat at low
altitudes. This system is relatively
unresponsive to unwanted reflections from
nearby structures such as radomes and
airframe structure and thus is not likely to
experience a false lock-on.
2.8
The Donnler Snectriim
The beamwidth of a beam of radar energy is

inversely proportional to the antenna size.
Aircraft installation constraints, and cost and
weight considerations limit the size of
practical antennas. Doppler radars with
antennas that are 0.3 meters to 0.6 meters
long have beamwidths of 8" to 4". Since the
beam from a microwave antenna cannot be
made to have zero width, it overlaps a
number of equal-frequency shift lines in the
narrow direction. As the beam moves across
the randomly situated scatterers, the result is
a Doppler return consisting of a spectrum of
frequencies, rather than a discrete frequency.
The Doppler shift which must be measured is
the center frequency of this spectrum. Since
the illumination of the target surface is most
intense at the center of the transmitted beam,
and diminishes in power toward the edges,
the result is a spectrum whose amplitude has
an essentially normal (or Gaussian)
distribution, as shown in Figure 4.
The Doppler spectrum of Figure 4 is
idealized, since the actual nature of the
Doppler return is complex and capable of
description only in statistical terms. At any
instant, the Doppler spectrum can be quite
unlike the figure and may present a
highly skewed picture. This can be seen by
considering a microwave signal transmitted
from the antenna and incident upon some
small physical object. The object scatters a
portion of the incident signal in every
direction including the direction back toward
the transmitter. The intensity of the back
scattered signal depends on the size, shape,
orientation, and electrical properties of the
scatterer. In actuality, the reflector involved
is not a single scatterer; a reasonably large.
area of the ground or of the surface of the
sea is illuminated by the transmitted RF
energy. Thus the target contains a large
138
number of randomly positioned, physically

independent scattering centers. The net
return signal available is then the sum of a
large number of waveforms reflected by
many scatterers. Each constituent waveform
has an amplitude and a phase determined by
the corresponding scattering center in the
target area. Since these amplitudes and
phases are randomly distributed quantities,
the return signal can be described adequately
only in statistical terms. The Doppler
spectrum is statistically equivalent to
narrow-band noise; the band center
frequency is the desired mean Doppler
frequency.
The Doppler spectrum width, A fd, at the

half-power points is given by
A fd=z(siny,) A y.
(6)
A y is the round-trip half-power width of the

antenna pattern in the y coordinate.
A narrower spectrum width Af, results in a

smaller instantaneous error in measuring the
band center frequency. Equation (6)
indicates that the spectrum width is reduced
if y o is smaller, and if the beamwidth is
smaller (a larger antenna).
Because of the appreciable spectrum width
resulting from the use of a practical-sited
antenna, the instantaneous frequency of the
Doppler signal is therefore subject to random
fluctuations about its mean value. The
values of speed and distance measured by a
Doppler system are, even with an ideal
instrumentation, subject to error. This error
will be discussed in Para. 3 . 1 .
2.9
Figure 4. The Doppler Spectrum
Measurement o f the Dormler

Freauency Shift
Frequency measurement is accomplished in a

device called a frequency tracker. The
frequency tracker must provide a singlevalued output representative of the spectrum
of frequencies comprising the Doppler
signal. The frequency tracker can be a
frequency discriminator, an autocorrelator,
or an axis-crossing frequency meter. Most
trackers are of the frequency discriminator
type, and measure the centroid of power of
the spectrum of Doppler frequencies. It has
been found that frequency trackers generally
measure this centroid with great accuracy
except possibly for a small bias or constant
offset of about 0.2 K d h r or less. The
effects of this bias offset are most noticeable
in helicopters where velocity is usually low.
139
As vehicle speed increases, the bias offset

has relatively less effect.
acceptable value (or when the

operator chooses to observe Doppler
radar silence), it indicates memory
and switches into its acquisition
mode. When the system is in the
memory mode, either the last
measured velocity, or velocity
computed from true airspeed and the
last measured wind, can be provided
as a substitute output (the last
measured wind is computed, during
normal Doppler operation, from
Doppler ground velocity and air
velocity; normally, the wind is
averaged over some time period to
smooth out gust effects).
In addition to determining the Doppler

frequency shift, the frequency tracker
generally does the following:
a.
b.
It tracks the signals; i.e., it provides a

continuous measure of a frequency
characterizing the input. It does this
as the input frequency changes and in
the presence of background noise.
Time constants of typical frequency
trackers are generally less than 25
m.s. to assure adequate tracking
during maneuvers. When the effects
of beam switching are taken into
account then the effective time
constant is somewhat greater. For
example, a tracker with a 20 m.s.
time constant and operating in a
system with a 8 Hz beam switching
rate has an effective time constant of
about 100 m.s.
It acquires the signal when the
system is turned on or after a period
of signal loss. This is usually wholly
automatic in modern systems;
however, approximate velocities can
be provided by external velocity
sensors such as inertial navigation
systems to reduce the acquisition
time.
2.10
Over Water Operation
Doppler system operation over water differs

from operation over land in three important
respects: (a) signal strength over water is
less than over land, (b) the centroid of power
of the returned signal is shifted by the radar
back-scattering properties of water, and (c)
the mean velocity of the scatterers which
give rise to the sea return appears as a
velocity error.
C.
It provides a measure of the

adequacy of the signal-to-noise ratio.
Below some critical value, tracking is
not reliable. In the presence of a
substandard S/N, the operator
receives a warning and certain
functions are performed by other
means.
The back-scattering properties of an

extended target medium can be specified by
the scattering coefficient, as a function of
incidence angle, 9.(The scattering
coefficient is defined as the ratio of power
back-scattered by the target to that which
would be back-scattered by a perfectly
reflecting hemispherically isotropic
scatterer). The scattering cross section per
unit area, a,, is equal to 2 cos 9 times the
scattering coefficient, f(9). (See reference
1, the chapter on "Sea Echo") Figure 3 gives
the scattering cross section (a,) as a function
of incidence angle (9).
d.
When the S/N is below a minimum
The curves of Figure 3 are based on the GPL
140
sea scale (see reference 1) developed for use

in Doppler radar set design, since the
conventional Beaufort and Douglas sea
scales were not sufficiently fine for adequate
use. A comparison of the relationships
among the GPL, Hydrographic Ofice
Publication 606F, Beaufort, and Douglas
scales is shown in Figure 5.
Note that over land, U,, is essentially
constant. Over water, however, uo
decreases rapidly as the incidence angle, 9,
increases; the specific values of uo (9)are
finctions of the sea state.
The nature of the first two phenomena, the
reduction in signal strength over water and
the change in calibration results in an error
proportional to speed. In practice, this shift
results in an error on the order of 1% to 5%
in magnitude. Its exact value depends on
beam shape, beam direction, and on sea
state.
1 1 I 1
CODEPER
CPI,
HYDROCIUI'IIIC BEAUVORT
OFFICE PUlt.60611
DOUGLAS
F
*
l
/
'
12
Figure 5. Approximate Relationships

Among Various Water Surface Condition
Scales
Several techniques have been developed to

minimize the ovenvater shift. (a) The error
varies substantially as the square of the y
beam width, so that narrow beams are
desirable. This technique has the
disadvantage that it increases the antenna
size. (b) The use of a manual landsedswitch to change the system calibration
as a finction of the terrain being traversed.
The over-water calibration value
corresponds to the required calibration over
the most frequently expected sea state. (c) If
the signal power in each slice of the beam at
the same 9 angle had the same Doppler
spectral composition as all other slices, then
the calibration shift would be zero. Certain
beam shaping techniques can result in
approximating this situation. (d) A beamlobing technique can be used to substantially
reduce the over-water calibration error.
Each antenna beam is switched a small
amount in the y direction at a low frequency
(about 20 hz). The return signal then
consists of two Doppler spectra existing
alternately in time at the switching rate.
The cross-over of these two spectra is
invariant with the sea-state since the return
at the crossover point for each spectrum was
derived from the same group of scatterers or
incidence angle. Thus, even though the
return signal spectra are distorted over
water, the crossover point remains
unchanged and the ovenvater shift is
essentially eliminated.
Techniques (b) and (c) are most frequently
used in modern Dopplers.
The Doppler frequency shift provides a
measure of the velocity of the aircraft
relative to the scattering surface. If the
surface is moving, this velocity will include
the surface velocity, thus producing an
erroneous measure of aircraft velocity.
141
Water surface movements result from

current flow and surface wind. Random sea
current flow is generally no larger than 0.7
K d h r . with random direction; such random
velocities tend to average out for long
flights. The Gulf Stream, whose direction
and velocity are generally known, has a
surface speed less than 6 Km/hr. Surface
winds can cause water surface particle
movements on the order of 4 to 8 K d h r . or
higher, depending on the surface wind speed.
In general, if W = the surface wind, and EW
= velocity of the surface particles as a result
ofW, then
EW = 1.28W) (W>2)
Doppler signal-to-noise ratio at the highest

altitude, highest velocity, and over the worst
terrain, as compared to the frequency tracker
sensitivity.
Using the radar equation an expression for
the Doppler S / N can be derived. The
equation is valid for coherent systems using
post-tracker mixing:
Nd
where
s =
(7)
Nd
Most of the effects of water surface motion

can be compensated for in the navigation
computer by manual insertion of known
surface current velocity or surface wind
conditions.
2.11
Doppler signal-to-noise ratio

(Doppler SN).
Ratio of the total Doppler signal
power to the noise power in the
bandwidth of the Doppler spectrum
The bandwidth of the Doppler
spectrum is measured at the 3 dB
points.
Donpler Signal-to-Noise Ratio

P,
A Doppler radar set measures the Doppler

frequency by processing the signal and noise
within the Doppler spectrum bandwidth Af,;
where yo is the mean value, and A y is

round-trip half-power width, of the antenna
pattern in the y coordinate.
There is a minimum acceptable signal-tonoise ratio ( S / N ) , within the Doppler
bandwidth, below which the frequency
tracker can no longer acquire the signal and
determine the Doppler frequency shift within
acceptable limits. This figure defines the
frequency tracker sensitivity, and is a
hnction of the tracker design. The altitude
capability of a Doppler radar can be
expressed as a function of the available
Average transmitted power per beam.
Go = One-way maximum antenna gain

relative to an isotropic radiator.
Note that Go=kA, where
A2
k is a constant, and A = antenna
area.
W,
RF attenuation in the plumbing of the

transmitter and receiver paths,
including duplexing and wave guide
losses.
E = Efficiency factor. (Ratio of available

Doppler S / N to the Doppler S / N
which would be available if all of the
received signal were converted to
Doppler information. This includes
spectrum utilization in pulsed and
142
FM-CW systems, gating

improvements, gating losses, and
noise-fold-over losses.)
f ( $ ) = Scattering coefficient at the given $ angle. (f ($) = 0.5 U, sec $ where
The term E in equation (9) for an

FM-CW Doppler is given by
where M is the Modulation number:
is the scattering radar crosssection per unit area.

U,
h = Transmitter wavelength
F =
Antenna pattern parameter

accounting for loss of power outside
the main lobe 3 dB limits (normally
between 0.5 and 0.67)
L = Attenuation in the atmosphere

NF
=F
W
noise figure
Boltzmann's constant (1.38 x

watt-secondddegree Kelvin)
absolute temperature (290" Kelvin)
=2
m sin 2 7c fm>
C
(12)
where T is the round-trip time and r is the

one-way distance to the ground.
Several of the parameters in equation (9) are
of interest. First, the product Go h2 is
directly proportional to antenna area hence
Doppler S/N is proportional to antenna area.
The Doppler S/N is a direct function of the
(and hence of the
scattering coefficient, f (I))
scattering cross-section, U,)
The Af, bandwidth is given by:
A fd = Doppler bandwidth
Afd=2J sin y o A y
Altitude above the terrain.
Since
(6)
Thus, Doppler S/N is inversely proportional

to the aircraft speed.
From equation (9), it is apparent that the
parameters W, and NF must be kept as low
as possible.
and h=rcos $, where r=range to the terrain,

equation (8) can be written
Frequency tracker sensitivity is expressed as

acquisition sensitivity (the minimum S / N at
which signal can be acquired and tracked),
and drop-out sensitivity (the minimum S/N at
I43
which the system can no longer track, and

hence goes into memory). Typical values are
5 dB acquisition sensitivity and 3 dB dropout sensitivity.
Typical values for S/N ratios over land at
300 meters (1000 fi.) AGL are 25 to 30 dB,
and thus such Dopplers have a margin of 20
to 25 dB. Ovenvater operation can result in
a 10 to 15 dB loss in S/N due to poor
backscattering resulting in less margin.
Dopplers that will frequently fly over water
should therefore have a higher S/N typically 35 - 40 dB over land - to assure
adequate margin over water.
2.12.
Doppler ECM Considerations
Doppler radars transmit very low amounts of

power - generally less than 250 m.w. - in
narrow beams directed nearly straight down.
The low power combined with very low
W signature and
side-lobes results in a low F
thus considerable difficulty in detection by
enemy ECM gear.
3.0
Doppler Radar Accuracy
A number of sources contribute to the total

Doppler radar set error in measurement of
velocity. These errors are either bias or
systematic errors (those which remain
essentially constant during operation) or
random or non-systematic errors (errors
which vary during operation). Some of the
systematic errors can be reduced by
calibration. Errors can also be classified as
either slope type (those which vary as a
percentage at speed) or zero type (those
which are constant). The major sources of
velocity error are discussed below.
The Doppler frequency shift is a result of

back-scattering from a large number of
randomly positioned, physically independent
scattering centers. The return signal is
indistinguishable in its statistical properties
from the signal obtained by passing white
noise through a band-pass filter having a
power transfer characteristic of the same
finctional dependence on frequency as the
Doppler spectrum (white noise is defined as
noise having a uniform power density at all
frequencies of interest). The spectrum, to a
first approximation, has a Gaussian shape,
with a bandwidth proportional to the speed.
This noise-like nature of the Doppler
spectrum gives rise to the Doppler
fluctuation error. The relative fluctuation
error, e, can be expressed as
h.Av.siny,
e,=av= 1
V 2 c o s y 0 VT
= K,/
(13)
VT = K,/ D
where u v is the RMS velocity error,

and a is the relative velocity error.
V
Equation (13) shows that, since V and T
appear only as a product, the relative or
percentage error in velocity is a finction only
of the distance travelled, and is independent
of speed. That this is true, and that the
percentage error varies inversely with the
square root of distance travelled,
can be seen intuitively: as D increases, the
Doppler information is derived from an
increasing number of independent scatterers,
and the amount of refinement of
measurement (smoothing) increases.
3.1.1 Power Spectral Density
3.1
Fluctuation Error
Many applications use Doppler velocities to
144
aid other sensors such as inertial navigation

systems. It is useful, in listing Doppler
requirements for such systems, to speci@the
power spectral density per unit of speed, call
the Doppler Po number.
It can be shown that
The instantaneous difference between the

true ground speed and the measured ground
speed is a random function of time. Suppose
this representation of ground speed error as
a function of time is transformed into the
frequency domain, with amplitudes of the
error given as a function of the frequency
components making up the error. Then the
mean square error density in each frequency
interval can be determined in units of
(knots)2/ (frequency interval). If the
frequency intervals are made sufficiently
small, a continuous distribution is
approached. This distribution is called the
ground speed error power spectral density,
and its units are (knots)2/Hz. If all of the
error frequency components are summed
(integrated) from - 03 to + 00, the result is the
variance (square of the standard deviation )
of the error distribution.
Hence, Po and e, are related by the equation
Since the true ground speed varies during a

mission, the error power spectral density is
normalized'' and defined in units of
(knots)'/Hz/knot of ground speed. The error
power spectral density at any speed can be
determined by multiplying the normalized
error power spectral density function by the
ground speed.
It can be shown that the error power spectral
density is essentially constant in amplitude
from virtually zero frequency up to the
bandwidth of the Doppler radar frequency
tracker, normally greater than ten Hz. Po is
defined as the error power spectral density in
the vicinity of zero frequency, normally
expressed as (kn~ts)~/[(radian/sec/knot]
(using angular frequency).
Po
v2
Po = De,
x
(15)
Note that Po has the dimension

[(knot~)~/(rad/sec)]
knot, or distance.
In a typical fixed-antenna Doppler radar,
each of the three orthogonal velocity outputs
is characterized by an associated Po:
Typical values for Pox,P and Po, are 0.005,

PY
0.005 and 0.002 respectively.
Equations (16) through (18) describe the
effect of Po on long-term velocity or position
error. It can be shown that the equation for
the "instantaneous" velocity error is
(EVx/V)2= x POx./V/2T",
where z is the correlation time of the
velocity error. The correlation time is
effectively the frequency tracker time constant as modified by the beam-switching
effect. Similar equations occur for eVYand
EV,.
3.2
Antenna Geometry Errors and

Calibration
The frequency shift error, Afd, caused by an

error in beam direction or A y is given by
2Vsiny Ay,
Afd = --
(19)
or
=
-Tany.Ay.
(20)
fd
The velocity component V, can be derived

from Equation 4, and is given by
3.3
and all )I
The EV, is given by, approximately;
Equation (22) shows that errors in pitch

stabilization tend to cancel. A positive
(nose-up) error in pitch would increase the
forward beam y angles (y A and y B) and
decrease the aft beam y angles (yC and yD).
Errors in roll stabilization would not affect
V, accuracy. Similar results occur for EV,,
except that the role of pitch and roll angles is
reversed. Equation (22) is particularly
important since it quantifies the dependence
of Doppler performance on the calibration
and stability of the beam geometry. As is the
case with most radars, performance of a
Doppler radar is very dependent upon the
quality of the antenna - both design and
fabrication. Errors in measuring or
calibrating the geometry of each beam result
in the velocity errors defined by Equation
(22).
Transmission Frequency Errors
Equation (21) shows that transmission

frequency directly affects the value of the
measured Doppler frequency. However, the
frequency stability of existing solid-state
microwave sources is so tight that this error
is normally quite small. Moreover, phased
linear array and planar array antennas can be
so designed as to make the Doppler beam
geometry independent of transmission
frequency.
3.4
where it is assumed that all

angles are equal.
145
Errors in Frequencv Measurement
This error is a hnction of the

instrumentation of the frequency tracker.
The error can normally be expressed as a
fraction of the Doppler spectrum width and
hence as a percentage of velocity.
Frequency trackers having errors on the
order of 0.02% to 0.05% are within the
current state of the art. Trackers can be in
analog or digital form. Analog frequency
trackers usually have a bias offset as well due
to the analog components that are used.
Digital trackers usually have a smaller bias
error, and also can be programmed for faster
acquisition times. Digital trackers are now
being used in most new Dopplers.
3.5
Errors in Data Conversion
The measured Doppler frequencies

representing the desired velocity components
must be converted into some other form in
order that the information can be used in a
computer or by other equipments in the
vehicle. This data conversion can deteriorate
the accuracy of the velocity information.
The error has been particularly significant in
the earlier Doppler radars, which used
relatively inaccurate conversion devices.
Airborne digital computers accept Doppler
146
frequency data in digital form with negligible

error, i.e., with an error which is limited only
by the number of binary bits used.
beam can cause significantly larger errors- of

the order of 4% per degree.
3.7
3.6
Errors in Stabilization (or in

Conversion from Vehicle to
Ground Coordinates)
Since the basic Doppler velocity information

is in vehicle coordinates and is generally
required in ground or some other
coordinates system, a conversion from the
former to the latter is usually necessary.
This conversion amounts to stabilization
about the pitch and roll of the vehicle. In
antenna stabilized systems, the Doppler
outputs are in the desired coordinate system,
and the stabilization error is primarily a
fbnction of the errors of the vertical
reference and of the servo or servos
controlling the antenna. In fixed antenna
systems, Doppler velocity is in vehicle
coordinates and is corrected for the pitch and
roll of the vehicle in a computer which
receives these angles from a vertical
reference. In this case, the resulting error
consists of the error of the vertical reference
and the error of the stabilization computer.
The error resulting from pitch and roll
uncertainties is quite small - 0.014% per
degree of uncertainty in pitch. Reduction of
this error to extremely small values can be
achieved through the use of an accurate
vertical reference system and a high precision
servo or stabilization computer. Note that
the sensitivity of velocity errors to pitch and
roll errors is fbndamentally different from
that of errors in beam direction. Pitch and
roll errors affect all four beams equally and
hence there is cancellation of IIIst-orderII
effects. Residual effects are 2nd-order and
of the magnitude given above. Beam
direction errors - if identical in sign and
magnitude - would behave similarly. Beam
direction errors that differ from beam to
Terrain Errors
The error resulting from the backscattering

characteristics of the target area is called
"terrain error''. Over land, this error is
insignificant. Over water, as explained in
para. 2.10, the terrain error can be
appreciable, as a result of changes in
scattering coefficient (for various sea state
conditions), as a hnction of beam incidence
angle.
Water surface droplet motion and water
current motion also cause Doppler velocity
errors; however unlike the calibration shift
error, these errors decrease in percentage as
the vehicle velocity increases, and therefore
tend to become acceptably small for most
fixed wing aircraft. Helicopters, however,
fly at lower speeds and water motion is an
important error source.
3.8
Altitude Hole Errors
The effect of altitude holes for pulsed and

FM-CW systems is to distort the Doppler
spectrum and change its effective centroid of
power. Since this phenomenon is related to
the ovenvater calibration shift error, it can be
seen that techniques that reduce the
ovenvater shift also reduce the altitude hole
errors.
3.9
Installation Errors
Errors in installation of the Doppler radar

antenna relative to the aircraft reference axis
contribute to the total system error. This
reference axis is normally the heading and
attitude reference system(s). The primary
error in horizontal velocity results from
147
misalignment relative to the heading

reference, and the primary error in vertical
velocity is due to misalignment relative to the
vertical reference.
3.10
Radome Errors
Most modern Doppler radars have integral

or built-in radomes and are mounted so that
these radomes are flush with the aircraft's
surface. In these applications, radome errors
are removed during the antenna calibration
process. In a few special cases a separate
radome is used and its effect on beam
geometry, and hence velocity accuracy must
be included in the overall error budget.
3.11
Total Doppler Error
On the assumption that systematic errors

have been removed or reduced by some form
of calibration procedure, the total Doppler
velocity error can be obtained by combining
the Doppler fluctuation error (for a desired
distance or velocity and smoothing time), the
non-systematic components of the other
errors discussed above and the residual
portion of the systematic errors.
Reference 1 contains a detailed discussion of
Doppler radar errors.
4.0
Doppler Radar Design

Configuration
Doppler radars are currently in production in

two basic forms: as a velocity sensor, and as
a self-contained dead-reckoning navigation
system. The latter system configuration
sometimes includes some form of external
velocity aiding from sensors such as True
Airspeed and inertial navigation systems.
Doppler velocity sensors are externally aided
to speed up acquisition time, to enable
computation of wind speed and direction,
and for performance validation or BITE.

4.1
Typical Doppler Velocity Sensor

Designs
Current Doppler radar velocity sensors are

all solid-state designs, and use body-mounted
antennas to minimize size, weight and cost.
An integral radome simplifies installation in
the airframe and eliminates the need for a
separate radome. The antennas generally
consist of some form of a printed circuit
board with radiating elements printed on a
dielectric substrate. The dielectric substrate
is bonded to a metal ground plane thereby
forming a very thin, lightweight and low cost
antenna. One form of such an antenna
utilizes microstrip-based elements to
generate the beams. A typical antenna
generates four beams from a single aperture.
Separate transmit and receive antennas, or
space-duplexing, are used when isolation
between these two functions is critical, such
as Doppler radars intended for operation
over water where higher transmitter power is
used to assure reliable operation. Doppler
radars intended primarily for over land
operation sometimes use a single antenna to
simultaneously transmit and receive RF
energy. Isolation between these two
functions is achieved via RF circulators, and
results in a lower cost design, but reduced
S / N due to transmitter coupling or leakage
into the receiver.
Today's transmitters are all solid-state
devices such as Impatt Diodes, Gunn
Diodes, or Dielectric Resonating Oscillators
(DRO). The latter are usually excited by
Gunn Diodes or FET devices. Typical
power outputs are 50 milliwatts to 250
milliwatts. The higher power units can also
have higher AM and FM noise so that the
signal-to-noise ratio is not necessarily
148
increased in direct proportion to the increase

in transmitter power output.
Solid-state microwave switches are used to
direct the transmitter output to each of the
ports on the antenna to form the desired
beams. Two sets of switches are needed for
space-duplexed antennas, whereas only one
set is needed for a single antenna design.
The microwave switches are usually
combined or integrated with the receiver
fbnction. The latter consists of a device such
as an RF circulator that taps off part of the
transmitter output to serve as an input to an
RF mixer. The other input to the mixer is
the received signal from the ground.
The mixer output is amplified, filtered,
converted to digital form and then sent to a
digital signal processor to extract the
Doppler frequency shift. The mixer output
includes not only the spectrum of Doppler
frequency shifts that is a measure of vehicle
velocity, but also a signal that results from
leakage of transmitter output directly into
the receiver path. Leakage signals can be up
to 40 to 50 dB higher than the spectrum to
be measured, and must first be eliminated to
enable processing of the desired signal. A
Leakage Elimination Filter, or LEF, centered
at zero frequency shift is usually employed to
reduce leakage by 30 to 40 dB. This filter
can be fairly broad for fixed-wing aircraft
Dopplers, but must be extremely narrow (a
few Hertz) for helicopter Dopplers to enable
operation at hover.
The digital signal processor computes the
centroid of power of the Doppler spectrum,
and also the signal-to-noise ratio (S/N) of
this spectrum. A low S/N results in a "notrack'' or memory indication and the signal
processor is switched into its signal
acquisition mode. When the S/N increases
above some predetermined level, a "Track"
state is indicated and normal operation

resumes. During Memory, the Doppler radar
outputs a Memory indication and its velocity
data should be ignored. It is usually
advisable to wait one to two seconds after
Track is indicated before using the velocity
data. Memory is usually indicated when S/N
drops below +3 dB and acquisition or
Normal mode is indicated when S/N is above
+5 dB. Acquisition usually takes one to two
seconds after the S/N of the received signal
increases above 5 dB.
Doppler radars are available in one or two
box configurations. In a 2-box system, the
antenna, transmitter, receiver, and
microwave switches are mounted into one
unit, and the LEF, digital signal processor,
interface, power supply and timer are
mounted in a second electronics unit. A
single-box configuration, of course, contains
all of these elements in one box. Doppler
radars are available with many types of
interfaces although digital is most commonly
specified.
Current Dopplers require less than 30 watts
and often operate from 28VDC, although
some systems use both 28VDC and 115V,
400 Hz A.C. Many current Doppler radars
use frequency-modulated continuous wave
(FM-CW) modulation of the carrier. The
FM is sinusoidal with a modulation index of
approximately one, and at a frequency of 30
KHz, resulting in an altitude hole at
approximately 4600 meters. A second FM,
usually at 40 KHz, is applied alternately with
the first one to assure continuous operation
through and above this altitude hole.
A Built-in-Test mode, or BIT, is
incorporated to test Doppler fbnctional
operation during flight and to provide the
flight crew with system status. A BITE
fbnction unique to Doppler radars is the
149
goodness check provided by the 4th or

redundant beam velocity. This test is used
during normal 4-beam operation, and also to
veri@ proper lock-on during acquisition
mode.
CEI'
4.2

Svstems
The advent of very powefil, low-cost

single-chip computers has enabled the
development of low-cost Doppler navigation
systems that provide present position and
also guidance signals to destination or
waypoint coordinates stored in non-volatile
memory. Left-right steering signals, distance
and time-to-go to these destinations are
typically provided. The Doppler navigation
system is provided with heading, pitch and
roll from the vehicle's heading and attitude
sensors. These sensors in moderate
performance aircraft are usually a magnetic
compass and a vertical gyro. More
sophisticated aircraft may have inertial
sensor-derived heading and attitude that have
greater accuracy, but also at a higher cost.
Figure 6 shows the position error in terms of
CEP as a function of heading error, for 3
values of Doppler velocity error. It can be
seen from this figure that the dominant error
source is actually the heading reference and
not the Doppler radar. Selection of the type
of heading and attitude sensors is a function
of the missions to be performed. In most
cases these three angles are in synchro
format although they are being provided in
digital format in many new aircraft. Doppler
velocity (three-dimensional) is then
transformed through these angles into North,
East and vertical components, and the North
and East components are integrated to
determine change in present position.
Figure 6. Doppler Radar Navigation

Accuracy vs. Heading Error
A Doppler navigation system computes
chanse present position. Computation of
the actual present position therefore requires
that some initial value of present position be
inserted into the system. This is usually
performed just prior to takeoff, or shortly
afterwards when the aircraft flies over a
known checkpoint.
Compensation for the three errors introduced

when flying over water is performed as part
of the navigation routine. Sea current data
are entered manually, usually as magnitude
and direction relative to North. Windinduced water motion is removed by the
computer using the empirically-derived
relationship discussed in Para. 3. Wind data
are obtained by either manually inserting
wind magnitude and direction, or wind data
is automatically computed from the
difference of True Airspeed (if available) and
Doppler velocity. In each case the flight
crew must indicate to the navigation
computer when the aircraft is flying over
water via a "Land/Sea" switch. The use of a
land-sea switch to indicate flight over water
can also be used to apply an approximate
correction for the over-water shift. Another
approach is beam-shaping to automatically
150
compensate for this shift using the

techniques discussed in Para. 2.10. Many
Dopplers are now using this method
successfully, and it will probably become the
standard technique in virtually all future
Dopplers.
Navigation during periods when the Doppler
radar has lost track and is in the "Memory''
mode, is normally performed using the last
valid velocity data prior to loss of track.
This llrememberedl' velocity is used until the
Doppler resumes normal operation, or is
replaced by a manually entered estimate of
velocity by the flight crew if Memory is too
long. Another method is to compute, during
normal Doppler operation, the wind vector
using True Airspeed (TAS) and Doppler
velocity. During Doppler memory, an
average value of the computed wind vector
is added to True Airspeed to arrive at an
estimate of vehicle velocity. Many airspeed
sensors are single-axis and thus do not
provide the cross-axis or transverse
component of airspeed. The effect of this
"missing" component of airspeed is an error
in the change in present position during
maneuvers and particularly during turns
where the transverse component of wind
changes with aircraft heading. Use of True
Airspeed plus computed wind during
Doppler memory will compensate for most
of the error caused by loss of valid velocity
data even when a single-axis TAS sensor is
used.
Display and control of Doppler-derived
navigation and guidance data has been
provided in many cases by a Control/Display
Unit or CDU dedicated to the Doppler
navigator. Increasingly, the Control and
Display function is being provided by a
central or shared unit that is also used for
other avionic systems such as fuel
management and radios. The availability of
map data bases resident in a few high density

memory chips will now enable the CDU to
also display the terrain, major man-made
features, the aircraft's present location, and
the optimum flight path to the destination.
The computational power of modern highspeed single chip computers enables present
position to be provided in many different
mapping coordinate systems. A typical
Doppler navigator provides data in
Latitudekongitude and also in Military Grid
Reference Systems, or MGRS. The latter is
particularly important for Army aircraft since
they must interact with ground forces that
use MGRS for their own position
determination and for artillery pointing and
weapons delivery. Other coordinate systems
can be provided as required by specific
missions.
Current Doppler navigation systems can
store large numbers of destinations or
waypoints, and some systems now permit
entry of sequences of waypoints to form
routes. The flight crew enters only the
predetermined route number and the
ControlDisplay unit provides steering
signals to each waypoint of that route,
including automatic transition to the next
waypoint. The flight crew can store present
position coordinates while over a target, and
recall these later in the flight for review and
for taking further actions.
The CDU can display wind magnitude and
direction, can indicate time-to-go to
destinations, can display groundspeed, trackmode-good and other flight parameters.
When the Doppler navigation system is
operating in conjunction with a magnetic
compass, the local magnetic variation and
the vehicle's magnetic deviation effects can
be entered for subsequent use in converting
magnetic heading to true heading.
151
The Control and Display functions described

above are being performed in a number of
aircraft and helicopters by CDUs dedicated
to Doppler navigation. Advanced and/or
sophisticated aircraft are beginning to use
central ControVDisplay systems in which the
Doppler functions are but one subset of a
large number of functions including radios,
weapons delivery, fire-control and fuel
management. These systems can also
perform the navigation and guidance
calculations, and the Doppler radar is then
used only to provide vehicle velocity. The
cost of digital processors may be low enough
to allow a redundant navigation function to
be performed in the Doppler for back-up
purposes.
References
1.
Myron Kayton and Walter R. Fried,

Avionics Navigation Systems, John
Wiley and Sons, Inc., New York,
1969.
2.
M. I. Skolnick, Introduction to Radar

Systems, McGraw-Hill Book Co.
Inc., New York, 1962.
3.
H. Buell, "Doppler Radars for Low

Cost Medium Accuracy Navigation",
Agard Conference Proceedings No.
176 on Medium Accuracy Low Cost
Navigation, Sandefjord, Norway,
September, 1975.
152
TERRAIN REFERENCED NAVIGATION

Drayton D. Boozer
Sandia National Laboratories
Albuquerque, NM, 87185-5800 USA
ABSTRACT
Terrain referenced navigation is a technique for
improving the accuracy of a navigation system by
correlating a sensed elevation profile of terrain beneath a
vehicle with stored terrain elevation data. Position
estimates are referenced to the terrain data and are
insensitive to position bias errors in the terrain data.
Because of this characteristic, terrain referenced
navigation systems are especially usefbl in applications
that require accurate navigation relative to targets,
obstacles, structures, and other features whose locations
are derived from the same source as the stored elevation
data. Example applications include lowemission terrain
followingkerrainavoidance, target queuing for standoff
weapon terminal sensors, indirect ranging and ground
proximity warning. System navigation accuracy
depends primarily on the ratio of terrain roughness to
terrain data vertical accuracy and secondarilyon
navigation system accuracy, vehicle ground clearance,
ground cover, vehicle maneuvers, and update frequency.
Terrain referenced navigation systems are often
considered for use with terrain masking for covert, lowaltitude ingress into hostile areas. Low probability of
intercept radar altimeters may be used in these
applications. A key issue is the availability and quality
of terrain elevation data. This chapter provides avionics
system developers an overview of terrain referenced
navigation system capabilitiesand characteristics, and
an outlook for future applications.
ACRONYMS
AHRS
Attitude Heading Reference System
CEP
Circular Error Probable
DTED
Digital Terrain Elevation Data
DLMS
Digital Landmass System
DMA
Defense Mapping Agency
GLONASS Global Orbiting Navigation Satellite System
GPS
Global Positioning System
INS
Inertial Navigation System
SITAN
Sandia Inertial Terrain Aided Navigation
TERCOM Terrain Contour Matching
TERPROM Terrain Profile Matching
TFtTA
Terrain FollowingITerrainAvoidance
TRN
Terrain Referenced Navigation
WGS
World Geodetic System
1. INTRODUCTION
Two generic types of navigation systems are often
used in military vehicles; inertial navigation systems
(INS)and doppler/attitudeheading reference systems
(AHRS). MSs are typically used in fixed wing
aircraft and doppler/AHRS in rotary wing aircraft.
Once aligned, an INS operates autonomouslyof

external stimuli, but errors drift with time. INS
stochastic error characteristicsare well understood
and can be accurately modeled. Doppler/AHRS
navigation systems produce a navigation Glution by
integrating doppler velocity measurements
transformed via the AHRSdetermined vehicle
attitudes. Errors drift with distance traveled rather
than time, thus, doppledAHRS systems are often
preferred to INSs for rotary wing aircraft. Both types
of navigation systems can use position updates from
external sourcesto remove accumulated drift. Terrain
referenced navigation (TRN) is a technique for
updating the position of a navigation system by
correlating a sensed elevation profile of terrain
beneath an aircraft with stored digital terrain elevation
data (DTED). TRN updates may occur over preplanned mapped areas, or as they become available
over large mapped areas. For simplicity, an INS-type
navigation system used in an aircraft and updated over
large geographical areas is emphasized in this
presentation. The concept has been extended to
standoff weapons, land vehicles, doppler/AHRS
navigation systems, and limited geographical areas but
these applications will not be discussed here.
2. CONCEPT
TRN consists of sensing a terrain elevation profile
beneath an aircraft and correlating the profile with
stored DTED to produce an estimate of aircraft position.
For tractability, the approximate shape of the trajectory
must be known and the correlation limited to areas of
reasonable size. An INS, usually with barometric
altimeter aiding, provides the approximate trajectory.
TRN systems provide three dimensional position
updates to the navigation system by estimating INS
trajectory errors. A radar or laser altimeter measures
ground clearance and the DTED gives terrain elevation
above mean sea level (MSL). Implementation requires
an INS, an altimeter, DTED, and a flight computer for
executing the TRN algorithm. Figure 1 illustrates the
sensing of the terrain profile along the true ground track
and the terrain profile along the INS ground track.
3. THEORETICAL OVERVIEW
Estimating errors in an INS trajectory by measurements
functionally related to the terrain profile beneath the
aircraft requires application of nonlinear estimation
theory. As shown in Figure 2, the sensed profile
obtained along the true ground track is compared to
DTEDderived profiles for possible horizontal locations
of the aircraft within a position uncertainty region and a
153
fiterror surface generated. The sensed terrain elevation

profile is obtained by subtracting the ground clearance
measurements from aircraft altitude. The terrain
elevation profiles from the DTED are produced by
interpolating the DTED at the appropriate horizontal
position along the translated INS ground track. The
location with the best fit is the estimate of the aircraft's
horizontal position.
Even in the simplest case of an INS with only position
errors, error-free DTED, and an error-free altimeter, the
correlation process may yield multiple solutions due to
identical terrain profiles. Thus there is a fundamental
issue of ambiguous solutions located in the aircraft
initial position uncertainty region. Mean squared
difference is the metric most often used for fiterror. In
the error-free case the minimum of this surface is zero
for both the correct and ambiguous solutions. Errors in
the DTED and/or altimeter cause the d a c e to be
positive at the minimum and increase the possibility of a
position estimate far from the true position. The
possibility of a "false fix" results from the nonlinear
aspect of the estimation problem. Usually TRN
performance specificationsseparate nonlinear (false fix)
and linear (update accuracy) effects. The characteristics
of system errors, the size of the uncertainty region,
terrain characteristics and the aircraft ground track over
the terrain are primary contributorsto the probability of
false fix [l].
The accuracy of the TRN position estimate for a simple

case is derived from application of linear estimation
theory. If one assumes independent profile
measurement errors, the circular error probable (CEP)
horizontal accuracy, where position biases between the
actual and sensed trajectory in each of three dimensions
are estimated, and terrain slopes are independent in the
down-range and cross-range directions, is [2],
CEP = 1.17 On/(hJ O&
(1)
where
CEP is the circular error probable of the update (m)
is the standard deviation of the profile
measurement errors (m)
os is the standard deviation of local terrain slopes at
the measurement locations in both down-range
and cross-range directions (unitless)
N is the number of independent measurements
(3,
osparameterizesboth terrain correlation length and

elevation variation effects. While (1) suggests that CEP
can approach zero for some choice of N given any On ,
INS velocity errors make arbitrarily small error

unachievable in practice. Because of the measurement
geometry, aircraft altitude (vertical position) errors in
DTED coordinates are estimated more accurately than
horizontal position errors. For this case,altitude error
standard deviation is O & N .
Figure 1. TRN Process
154
Sensed
profile/
/
/
TRUE
IkIC
I1
l
.
U
Figure 2. Horizontal Position Estimation

Using horizontal INS position errors modeled as
independent random walks, no altitude error and firstprinciple covariance analysis [3] lead to the expression,
CEP,, = 0.57 SV1I4 (Ads)318
results in a CEPss of 29 m. Because of the assumptions

leading to (2), predictions should be treated as
approximations,a conservative lower bound for TRN
accuracy.
(2)
DTED errors are not explicitly considered in (1) and (2).

Lower accuracy is achieved when the DTED vertical
(elevation) errors have the same correlation length as
CEP,, is the steady state circular error probable of
the
terrain itself. This is conceptually reasonable since
horizontal position updates (m)
in
this
case the correlationprocess must accommodate
is the standard deviation of the profile
6,
errors that undulate statistically like the terrain itself.
measurement errors (m)
is the deterministic local terrain slope at the
The lack of statistical error models for the only broadh
Defense mPPing
available, u-s*
measurement locations in both down-range and
@MA) Digital Landmass System (DLMS) Level 1
cross-range directions (unitless)
Ad
is the distance between profile measurements (m) DTED, is the key reason TRN has not been developed to
a theoretical level comparable to other navigation
S
is aircraft ground speed (ds)
is INS maximum velocity error ( d s )
updating techniques like ground- and satellite-based
6V
ranging systems.
The primary value of (2) is that it shows the sensitivities
of accuracy to implementation parameters. Steady state
4. IMPLEMENTATION CONSIDERATIONS
systems typically consist offour basic
CEP is most sensitive to the ratio onhleast sensitive to
fircraft
elements: an INS,an altimeter, a DTED database, and
SV, and nominally sensitive to the time between profile
an algorithm executed by a flight computer. The
measurements A d s . Using typical values of
function of the INS is to provide a stable threedimensional approximate trajectory whose primary
SV = 1 ds (1 nm/hrclass INS)
errors are horizontal position and altitude bias errors. A
s
= 250ds
1 nm/hrclass INS is a common choice. Either strappedAd = 100m
= .05 (moderately rough terrain)
down or gimbaled INSs can be used. A barometric
h
altimeter is traditionally used to stabilize the vertical
On = 15m
dimension of the INS and can be retained in TRN
where
155
systems. In either case,the navigation system vertical

channel must be accurately modeled since TRN requires
accurate measurement of the terrain profile.
DTED is a static database of terrain elevations over a
prescribed geographical area. It may be produced from
topographic maps of various scales or stereo
photographs of different resolutions. The DTED with
the widest geographic coverage is produced by the U.S.
DMA. DLMS Level 1 DTED have been produced for
about 70% of the world's landmass and are distributed
on compact disk read-only memory (CD-ROM) media
[4]. Level 1 DTED have an elevation value for each 3"
in latitude and longitude within f 50 latitude of the
equator. Larger horizontal post-spacings (in degrees
longitude) are used at latitudes greater than 50 to
maintain approximately constant spatial sampling of the
terrain. Post-spacing is constant at 93 m in latitude but
varies around this value in longitude. Algorithm design
and implementationare simplified and computer
throughput minimized when DTED have constant
horizontal post-spacing. When using Level 1 DTED,
this requires a priori preparation of the flight DTED.
As flight computers become more capable, the trend
toward using Level 1 DTED as produced by DMA
directly in TRN algorithms will likely continue.
The function of the altimeter is to measure nadir ground
clearance, the distance between the aircraft and the
terrain at the same horizontal coordinates as the aircraft.
Radar altimeters measure the distance to the closest
point in the antenna beam. Unless modeled, the
difference between the closest point and the nadir
ground clearance is a measurement error whose
magnitude is a function of aircraft ground clearance and
attitude, antenna beam pattern, and terrain
characteristics. These effects can be eliminated with a
laser altimeter gimbaled to point at nadir. Radar
altimeter antennas used on aircraft typically have broad
beam patterns that accommodate significant aircraft
maneuvers while continuing to measure the closest point
beneath the aircraft. Such altimeters have been used in
most TRN applications. Radar altimeter antennas with
narrow beam patterns require either gimbals or limiting
terrain sensing to specific aircraft pitch and roll attitude
limits. Attitude limits are a function of aircraft ground
clearance. Whenever radar altimeters are used,
modeling of beam pattern effects in the TRN algorithm
will improve performance, especially at higher ground
clearances.
Altimeters typically provide ground clearance
measurements to the flight computer at a rate
sigrufcantly greater than the iteration rate of the TRN
algorithm and spacing along the ground track that is
small with respect to the DTED horizontal post-spacing.

Depending on the algorithm design and DTED source,
measurements may be either averaged or simply used as
needed by the algorithm. Trees, foliage, ice, structures
and other ground cover, as well as soil type, afect the
altimeter measurements. TRN system performance
depends on the altimeter measuring whatever is
contained in the database. For example, if the DTED
contain elevations of tree tops rather than the ground
beneath the trees, the altimeter should range on treetops
for best TRN correlation performance.
TRN algorithms must solve a nonlinear estimation
problem in real time. Practical algorithms require a
number of simplifying assumptions and in most cases
were developed for specltic aircraft avionics and for use
in specific scenarios. Given this situation it is not
surprising that a number of TRN algorithms have been
proposed and developed over the last thuty years. The
TERCOM approach developed for U.S. cruise missiles
[5,6,7] is a batch-processing design that was developed
for accurate, but limited, geographical coverage DTED,
small flight computer processing throughput and
memory, and an accurate INS. The missile flies a
constant heading course over the DTED. TERCOM
provides one position update for each DTED "patch" or
matrix. Thus INS updates are infrequent. TERCOM
was extended to accommodatearbitrary shaped
trajectoriesover wide-area DTED [8]. Algorithms of
this ilk must deal with aircraft maneuvers and lower
quality DTED. Sandia Inertial Terrain Aided
Navigation (SITAN)[9] was initially developed for a
specific weapon application that required vehicle
maneuvering, short flight distances, and relatively small
initial position errors. These requirements led to an
extended Kalman filter, recursive algorithm design that
processes altimeter ground clearance measurements and
produces an estimate of aircraft state every lOOm along
the aircraft ground track. A key idea is use of stochastic
terrain linearization to permit use of a single extended
Kalman filter when horizontal position errors are much
larger than the local terrain correlation length. The
initial application led to "track-mode" SITAN. Later, a
bank of parallel filters was used with decision logic (an
"acquisition mode") to accommodatelarger initial
position errors [lo]. The SPARTAN algorithm [ 111,
originally designed by A. R Runnalls, uses maximum
likelihood estimation in such a way that there is more
measurement smoothing before incorporating
measurement information into the Kalman filter than in
SITAN but more frequent INS updates than in
TERCOM. All current implementations use variations
on one of these three basic approaches, e.g. terrain
profile matching (TERPROM) [121 is based on an
extended TERCOM acquisition mode coupled with a
SITAN track mode.
156
Key implementation issues are the balance between onboard processing versus preflight mission planning and
access to the stored DTED during flight. To minimize
requirements on flight systems, the DTED may be
formatted into a grid with constant horizontal postspacing. Data compression may be used to minimize the
amount of on-board storage at the expense of requiring
in-flight reconstruction of the DTED. Storing only the
DTED on the aircraft that may be used in a given
mission lowers the required DTED storage capacity at
the expense of requiring more pre-mission effort.
Trends are to store the DTED of large areas on the
aircraft and to use the DTED in the format in which it is
supplied by DMA. During flight the DTED of the area
being over flown are extracted from the large capacity
storage device and placed in a buffer where they can be
rapidly accessed by the TRN algorithm. How this is
done is often determined by DTED access requirements
of other system functions like lowemission TFmA. The
flight computer interfaces to the altimeter, INS,and
cockpit displays are not usually affected by TRN.
The primary determinants of incremental flight
computer resources needed to implement TRN beyond
accessing the DTED and placing it in a buffer are the
speed of the aircraft and the size of the largest
horizontal position errors that the system must
accommodate. Higher aircraft speeds q u i r e greater
computer throughput. The required maximum position
error depends primarily on the longest anticipated flight
without updates and INS quality.
5.
SYSTEM CONSIDERATIONS
TRN position estimates are referenced to the stored

terrain data and are insensitive to bias errors in the
terrain elevation data. Because of this characteristic,
TRN systems are especially useful in applications that
require accurate navigation relative to targets, obstacle,
structures, and other features whose locations are
derived from the same sources as the stored terrain data.
Example applications include lowemission TFmA,
target queuing for standoff weapon terminal sensors,
ground proximity warning and indirect ranging
[13,14,15]. Three dimensional position bias errors
made in the DTED production process can be estimated
by a combined satellite navigation (GPS andor
GLONASS) and TRN system. This is especially
important in systems using the DTED for TFfI'A since
DTED bias errors cannot be estimated by satellite
navigation alone. Satellite navigation is performed in
world-wide coordinates and has no reference to errors in
the DTED used for TFfI'A. Thus, even though the
absolute accuracy of a satellite system may be superior to
a TRN system in world-wide coordinates (e.g. World
Geodetic System (WGS)), a TRN system may be more
a m a t e with respect to obstacles and features within the
DTED because of errors made in registering the DTED

to world-wide coordinates. Used alone, TRN systems
can only estimate the total bias in each position
dimension with respect to the INS position but cannot
estimate the constituent parts of the total bias in each
coordinate; DTED position bias errors and INS errors.
To improve underlying INS accuracy, TRN systems

must be used in areas for which terrain of s a c i e n t
roughness and DTED of appropriate accuracy and
quality are available. The area over which the TRN
system searches for updates is an important
considerationsince it determines the TRN system's
ability to correct INS drift error accumulated during
periods of no updating. The larger the search area the
greater the possibility of false fix. A false fix that is
recognized by the TRN algorithm logic is not of
particular concern, but large search areas do require
greater care in design and testing of the TRN algorithm
to ensure that no unrecognized false fixes are used as
INS updates.
The robustness of the TRN system design to DTED error
processes is perhaps the most important consideration
for system developers. As used here, robustness is the
ability of the design to navigate in the presence of a
variety of DTED errors. Because of its broad-area
coverage, Level 1 DTED is used by most TRN systems.
TRN system performance using Level 1 DTED is very
repeatable over the same geographical area but tends to
vary from one geographical region to another. This is
caused in part by terrain and ground cover differences
but primarily by the varying characteristics of DTED
errors. As one gains experience using the DTED in
TRN systems, some areas may need to be reworked or
the TRN algorithm changed to accommodate the errors
discovered. This learning process is inevitable until
world-wide DTED derived from the same source
materials become available.
With very accurate and expensive-to-produceDTED,
TRN system horizontal position accuracies rivaling
those of GPS can be achieved. In TRN systems using
Level 1 DTED over broad areas, accuracies in the range
of 50-200 m CEP are typical for low-flying aircraft.
Since TRN systems require terrain roughness and
operate best at low ground clearance, they are especially
attractive for use in covert attack TFfI'A systems.
Conversely, these are the flight regimens that cause
satellite-based systems the most problems; satellite lineof-sight masking by the airframe or terrain, and signal
jamming. Low probability of intercept radar altimeters
are oAen considered for use in these applications.
Most TRN system implemented to date are add-on's to
existing avionics systems. These systems are termed
157
"loosely coupled" because TRN INS error estimates are

simply added to the output of the standard INS solution
in software. Future systems will likely be more tightly
integrated with real time estimation and correction of
INS inertial sensor parameters.
6. OUTLOOK
The coincidence of commercial and military

requirements in combination with advances in
computer, communications and satellite technology will
ensure availability of very accurate, highquality, worldwide databases of both DTED and feature data early in
the twenty-first century. By using satellite positioning,
the absolute position of terrain and features will be
known to sub-meter accuracy. These databases will be
used pervasively in military aircraft to enable noemission TFRA,very accurate indirect ranging and
ground proximity warning. TRN will be used in
integrated avionics systems to provide navigation
redundancy and to improve position estimation
accuracy. The availability of significantly greater
computer throughput, memory and mass storage will
lead to implementationof more capable TRN
algorithms; performance improvementswill be greatest
at higher ground clearances and over very rough terrain.
Altitude position estimation accuracy will be improved
over that available from satellite-based sensors alone.
This is because altitude error is greater than horizontal
error in satellite-based sensors,TRN is especially
sensitiveto altitude errors, and the DTED will be
accurately positioned with respect to world-wide
coordinates.
Over the next 10-to-20 years, TRN will be integrated
with GPShNS systems to improve lowemission TFRA,
indirect ranging and ground proximity warning
performance when using Level 1 DTED. TRN will
enable conversion of local DTED coordinates into
world-wide coordinates in areas of suflicient terrain
roughness.
Operational use of TRN by the U.S. is currently limited
to the TERCOM system used for cruise missile
guidance. TRN integration with GPSANS in fixed- and
rotary-wing attack aircraft is an area of current
development in Europe, Australia and the U.S. Fielding
of several operational systems during the decade is
anticipated.
REFERENCES
1. HiMchs, P R, "Advanced Terrain Correlation
Techniques", IEEE Position Location and
Navigation Symposium, 1972, p 94. [also in
hyton, Myron, NAVIGATION,
LAND, SEA, A& &
SPACE,IEEE Press, 1990, pp 288-2951
2. Hostetler, L. D., and Andreas, R D., "Nonlinear

Kalman Filtering Techniques for Terrain-Aided
Navigation", IEEE Transactions on Automatic
Control, Vol. AC-28, No. 3, March 1983, p 320.
[also in Sorenson, H. W., KALMAN FILTERING:
THEORY AND APPLICATION, IEEE Press, 1985,
pp 325-333.1
3. Maybeck, P. S., STOCHASTIC
MODELS,
ESTIMATION,
AND CONTROL,
VOL. 1, Academic Press, 1979, p 300.
4. U.S. Defense Mapping Agency, DIGITIZING
THE
Funm, 3rd Edition.
5. Sorrels, C A., U.S. CRUISE
MSSILE
PROGRAMS,
McGraw-Hill, 1983, p 9.
6. Werrell, K. P., THE EVOLUTION
OF THE CRUISE
MSSILE,
Air University Press,1985, p 136.
TECHNOLOGY,
7. Betts, R K., ed., CRUISEMISSILE,
STRATEOY,
POLITICS,
The Brookings Institution,
1981, p 37.
8. Bialecke, E. P. ,and R C. Lewis, "A Digital Terrain
Correlation System for Tactical Aircraft",
IEEWAIAA 5th Digital Avionics Systems Conference
Proceedings, Oct. 31-Nov. 3, 1983, p 14.1.1.
9. Hostetler, L. D., "A Kalman Approach to
Continuous Aiding of Inertial Navigation Systems
Using Terrain Signatures", IEEE Milwaukee
Symposium on Automatic Computation and Control
Proceedings, April 1976, pp 305-309.
lO.Boozer, D. D., Lay M. IC.,Fellerhoff, J. R, "The
AFTyF16 Terrain-Aided Navigation System",
Proceedings of the IEEE National Aerospace and
Electronics Conference Proceedings, May 1985,
pp 351-357.
11.Henley, A. J., "Terrain Aided Navigation Current
Status, Techniques for Flat Terrain and Reference
Data Requirements", IEEE Position Location and
Navigation Symposium Record, March 20-23, 1990,
pp 608415.
12.Robins,A. J., "Recent Developmentsin the
"TERPROM" Integrated Navigation System",
Proceedings of the Institute of Navigation 44th
Annual Meeting, June 21-23, 1988, pp 5846.
13. Beal, C., "World in a box: Air Navigation Leaps
Forward", International Defense Review, Vol. 25,
May 1992, pp 417-422.
14.Priestley,N., "Terrain Referenced Navigation", IEEE
Position Location and Navigation Symposium Record,
Mar 20-23,1990, pp 482-489.
15.Bennetf P. J., "Enhanced Navigation and Displays
from Passive Terrain Referenced Avionics",
Proceedings of the IEEE National Aerospace and
Electronics Conference Proceedings, May 1988,
pp 209-216.
158
SATELLITE NAVIGATION
Richard B. Langley
Geodetic Research Laboratory
Department of Geodesy and Geomatics Engineering
University of New Brunswick
Fredericton, N.B., E3B 5A3
CANADA
1. INTRODUCTION
Satellire navigation provides unprecedented accuracy and
world wide coverage to aerospace vehicle. This section
reviews in some detail the NAVSTAR Global Position
System (GPS) followed by a survey of other satellite
navigation systems:
2. SATELLITE NAVIGATION SYSTEMS
3. NAVSTAR GLOBAL POSITIONING SYSTEM

3.1 Introduction
The Navstar Global Positioning System (GPS) is a satellitebased positioning system currently in operation by the
United States Department of Defense. Work on the system
began in 1973 as a result of the merger of the U.S. Air
Force's 621B Project. Both of these programs had been
established in the mid-1960s to develop a passive navigation
system using measured ranges. GPS consists of three
segments: the satellites, the control system, and the users.
3.2 Satellites
When fully deployed by late 1993 or early 1994, the satellite
segment will consist of a constellation of 21 Block I1 or
second-generation satellites plus 3 in-orbit operating spares.
The satellites are arrayed in 6 orbital planes inclined 55" to the
equator. Each orbit is circular with a nominal altitude of 20
183 km. The corresponding orbital period is 12 sidereal hours,
one half of the earth's period of rotation.
The deployment of the Block I I satellites was preceded by a
program of testing using prototype, or Block I, satellites. The
Block I satellites were launched from Vandenburg Air Force
Base in California using Atlas F rockets. I O of these satellites
were successfully launched between February 1978 and
October 1985. They were placed in nominally circular orbits
with semimajor axes of about 26,560 km. The satellites were
positioned in two orbit planes with inclinations of about 64"to
provide maximum coverage for the main military testing area
for GPS, the Yuma Proving Grounds in Arizona.
Whereas the Block I satellites were launched using an
expendable launch vehicle, it was originally intended to launch
the Block II satellites, up to three at a time, using the Space
Transportation System - the Space Shuttle. But after the
Challenger accident, the decision was made to use an
expendable launch vehicle and a new rocket, the Delta 11, was
developed for this purpose. The Block I1 satellites are
launched from the Cape Canaveral Air Force Station next door
to the Kennedy Space Center.
-- Block IIA
-- Block IIR
-- Block Ill
The satellites achieve their final orbits i n steps. The first and
second stages of the rocket together with the nine solid-fuel
strap-on booster engines put the third stage of the rocket and
the attached satellite into an elliptical orbit with a perigee
height of about 180 km and an apogee height of about 870 km.
(The apogee is the point on the orbit furthest from the earth.)
The third stage, called a Payload Assist Module (PAM), is then
used to increase the apogee height of the orbit so that it
matches the height of the desired final orbit, about 20,200 km,
above the earth's surface. At this point, the satellite is in a

highly elliptical orbit, called a transfer orbit. The PAM is
jettisoned and the satellite's orbit insertion engine is fired to put
the satellite into an approximate final orbit. Small adjustments
are then made using the satellite's thrusters. The orbits of the
Block I1 satellites are nominally circular (maximum
eccentricities are about 0.01) with a semimajor axis of about
26,560 km and an inclination of about 55". The resulting
orbital, periods of the satellites are within a minute or so of
exactly one half of a sidereal day. In order to achieve global,
24 hour a day coverage, four satellites will be placed in each of
6 orbital planes, named A through F. The right ascensions of
the ascending nodes of orbits in adjacent planes are separated
by 60'. This arrangement of the satellite constellation results
in at least six satellites being visible at all times.
3.3 Signals
3.3.1 The Carriers. Each GPS satellite transmits signals
centred on two microwave radio frequencies, 1575.42 MHz,
referred to as Link 1 or simply LI, and 1227.60 MHz, referred
to as L2. These channels lie in a band of frequencies known as
the L band which starts just above the frequencies used by
cellular telephones. Within the L band, the International
Telecommunications Union, the radio regulation arm of the
United Nations, has set aside special sub-bands for satellitebased positioning systems. The LI and L2 frequencies lie
within these bands.
Such high frequencies are used for several reasons. The
signals, as we have said, consist of a number of components.
A bandwidth of about 20 MHz is required to transmit these
components. This bandwidth is equal to the whole FM
broadcast band! So a high, relatively uncluttered part of the
radio spectrum is required for GPS-type signals. The GPS
signals must provide a means for determining not only high
accuracy positions in real-time, but also velocities. Velocities
are determined by measuring the slight shift in the frequency of
the received signals due to the Doppler effect In order to
achieve velocities with centimetre-per-second accuracies,
centimetre wavelength (microwave) signals are required.
A further reason for requiring such high frequencies is to

reduce the effect of the ionosphere. The ionosphere affects the
speed of propagation of radio signals. The range between a
satellite and a receiver derived from measured signal travel
times will therefore be in error. The size of this error gets
smaller as higher frequencies are used. But at the LI frequency
it can still amount to 30 metres for a signal arriving from
directly overhead. For some applications, an error of this size
is tolerable. However there are applications that require much
higher accuracies. This is why GPS satellites transmit on two
frequencies. If measurements made simultaneously on two
well-spaced frequencies are combined, it is possible to remove
almost all of the ionosphere's effect.
Although high frequencies are desirable for the reasons just
given, it is important that they not be too high. For a given
transmitter power, a received satellite signal becomes weaker
the higher the frequency used. The L band frequencies used by
GPS are therefore a good compromise between this so-called
space loss and the perturbing effect of the ionosphere.
159
GPS signals, like most radio signals, start out in the satellites as
pure sinusoidal waves or carriers. But pure sinusoids cannot
be readily used to determine positions in real-time. Although
the phase of a particular cycle of a carrier wave can be
measured very accurately, each cycle in the wave looks like the
next so it is difficult to know exactly how many cycles lie
between the satellite and the receiver. This ambiguity can be
resolved using the differential technique pioneered by
surveyors [Wells and Kleusberg, 19901 but can be time
consuming.
The GPS PRN codes have additional useful properties. When

a receiver is processing the signals from one satellite, it is
important that the signals received simultaneously from other
satellites not interfere. The GPS PRN codes have been
specially chosen to be resistant to such mutual interference.
Also the use of PRN codes results in a signal that is essentially
impervious to unintentional or deliberate jamming from other
radio signals. The possibility of deliberate jamming of GPS
signals is something that the U.S. Department of Defense, the
owners of the system, have to wony about.
In order for a user to obtain positions independently in realtime, the signals must be modulated; that is, the pure sinusoid
must be altered i n a fashion that time delay measurements can
be made. This is achieved by modulating the carriers with
pseudo-random noise (PRN)codes.
At the present time, the C/A-code is modulated onto the LI

carrier whereas the P-code is transmitted on both LI and L2.
This means that only users with dual frequency GPS receivers
can correct the measured ranges for the effect of the
ionosphere. Users of single frequency receivers must resort to
models of the ionosphere which account for only a portion of
the effect. It is access to the lower accuracy C/A-code which is
provided in the GPS Standard Positioning Service (SPS), the
level of service authorized for civilian users. The Precise
Position Service (PPS) provides access to both the C/A-code
and the P-code and is designed (primarily) for military users.
The SPS incorporates a further intentional degradation of
accuracy, called Selective Availability (see 3.3.7).
These PRN codes consist of sequences of binary values (zeros

and ones) that at first sight appear to have been randomly
chosen. But a truly random sequence can only arise from
unpredictable causes which, of course, we would have no
control over and could not duplicate. However, using a
mathematical algorithm or special hardware devices called
tapped feedback registers, we can generate sequences which do
not repeat until after some chosen interval of time. Such
sequences are termed pseudo-random.
The apparent
randomness of these sequences makes them indistinguishable
from certain kinds of noise such as the hiss heard when a radio
is tuned between stations or the "snow" seen on a television
screen. Although noise i n a communications device is
generally unwanted, in this case the noise is very beneficial.
Exactly the same code sequences are independently replicated
i n a GPS receiver. By aligning the replicated sequence with
the received one and knowing the instant of time the signal was
transmitted by the satellite, the travel time, and hence the range
can be computed. Each satellite generates its own unique
codes, so it is easy for a GPS receiver to identify which signal
is coming from which satellite even when signals from several
satellites arrive at its antenna simultaneously.
3.3.2 The C/A-code. Two different PRN codes are transmitted

by each satellite: the C/A or coarse I acquisition code and the P
or precision code. The C/A-code is a sequence of 1,023 binary
digits or chips which is repeated every millisecond. This
means that the chips are generated at a rate of 1.023 million per
second and that a chip has a duration of about 1 microsecond.
Each chip, riding on the carrier wave, travels through space at
the speed of light. We can therefore convert a time interval to
a unit of distance by multiplying it by this speed. So one
microsecond translates to approximately 300 metres. This is
the wavelength of the C/A-code.
Because the C/A-code is repeated every millisecond, a GPS
receiver can quickly lock onto the signal and begin matching
the received code with the one it generates.
3.3.3 The P-code. The precision of a range measurement is

determined in part by the wavelength of the chips in the PRN
code. Higher precisions can be obtained with shorter
wavelengths. To get higher precisions than are afforded by the
CIA-code, GPS satellites also transmit the P-code. The
wavelength of the P-code chips is only 3 0 metres, one-tenth the
wavelength of the C/A-code chips; the rate at which the chips
are generated is correspondingly I O times as fast: 10.23 million
per second. The P-code is an extremely long sequence. The
pattern of chips does not repeat until after 266 days or about
2.35 x IOI4 chips! Each satellite is assigned a unique oneweek segment of this code which is re-initialized at
Saturday/Sunday midnight each week.
3.3.4 The Broadcast Message. In order to convert the

measured ranges between the receiver and the satellites to a
position, the receiver must know where the satellites are. To
do this in real-time requires that the satellites broadcast this
information. Accordingly there is a message superimposed on
both the LI and L2 carriers along with the PRN codes. Each
satellite broadcasts its own message which consists of orbital
information (the ephemeris) to be used in the position
computation, the offset of its clock from GPS system time, and
information on the health of the satellite and the expected
accuracy of the range measurements. The message also
contains almanac data for the other satellites in the GPS
constellation as well as their health status and other
information. The almanac data, a crude description of the
satellite orbit, is used by the receiver to determine where every
satellite is. I t uses this information to quickly acquire the
signals from satellites which are above the horizon but are not
yet being tracked. So once one satellite is being tracked and its
message is decoded, acquisition of the signals from other
satellites is quite rapid.
The broadcast message also contains another very important
piece of information for receivers that track the P-code. As we
mentioned, the P-code segment assigned to each satellite is 7
days long. A GPS receiver with an initially unsynchronized
clock has to search through its generated P-code sequence to
try to match the incoming signal. It would take many hours to
search through just one second of the code, so the receiver
needs some help. It gets this help from a special word in the
message called the hand-over word (HOW) which tells it
where in the P-code to start searching.
The GPS broadcast message is sent at a relatively slow rate of
50 bits per second, taking 12.5 minutes for all the information
in the message to be transmitted. To minimize the delay for a
receiver to obtain an initial position, the ephemeris and satellite
clock offset data are repeated every 30 seconds.
The U A code and P code chip streams are separately combined
with the message bits using modulo 2 addition. This is just the
binary addition that computers and digital electronics do so
well. If the code chip and the message bit have the same value
(both 0 or both 1) the result is 0. If the chip and bit values are
different, the result is 1. The carriers are then modulated by the
code and message composite signal. This is readily done with
the L2 channel as it only carriers the P-code. But the LI
160
C/A-CODE
MESSAGE
L2 CARRIER
Figure 3.1 How the different components of the GPS signal are combined. Note that the various waveforms are
not to scale.
channel has to carry both the P-code and the C/A-code. This is
achieved by a clever technique known as phase quadrature.
The P-code signal is superimposed on the L1 carrier in the
same way as for the L2 carrier. To get the CIA-code signal on
the LI carrier, the unmodulated carrier is tapped off and this
tapped carrier is shifted in phase by 90". This quadrature
with the C/A-code signal and
carrier component is mixed
then combined with the P-code modulated in-phase component
before being transmitted by the spacecraft antenna.
3.3.5 Binary Biphase Modulation. Carrier waves can be

modulated in a number of ways. The amplitude of the carrier
can be varied, the frequency can be varied, or the phase can be
varied. Phase modulation is the approach used for the GPS
signals. Because the PRN codes and the message are binary
streams, there must be two states of the phase modulation.
These two states are the normal state, representing a binary 0,
and the mirror image state, representing a binary 1. The
normal state leaves the carrier unchanged. The mirror image
state results in the unmodulated carrier being multiplied by -1.
Therefore a code transition from 0 to 1 (normal to mirror
image) or from 1 to 0 (mirror image to normal) each involves a
phase reversal or a phase shift of 180". This technique is
known as binary biphase modulation. An interesting property
of binary biphase modulation was exploited by one of the first
commercially available GPS receivers, the MacrometerTM.By
electronically squaring the received signal, all of the
modulation is removed leaving a pure carrier. The phase of the
carrier could then be measured to give the ambiguous range
measurements used by surveyors. Of course, the broadcast
message was lost i n the process and so orbit data had to be
obtained from an alternate source.
The composite GPS signal consists then of carriers modulated
by the PRN CIA and P codes and the broadcast message. The
combining of these different components is illustrated in Figure
3. I . Forgetting for a moment that GPS is a ranging system, we
could consider the satellites to be simply broadcasting a

message in an encoded form. The bits of the message have
been camouflaged by the PRN code chips. The effect of this
camouflaging is to increase the bandwidth of the signal.
Instead of occupying only a fraction of one kiloHertz, the
signal has been spread out over 20 MHz. Inside a GPS
receiver, the code matching operation de-spreads the signal
allowing the message to be recovered. Clearly this can only be
done if the receiver knows the correct codes. The de-spreading
operation conversely spreads out any interfering signal
considerably reducing its effect. This is a common technique,
especially in military circles, for ensuring security and
combatting interference and is known as direct sequence
spread spectrum communication. Spread spectrum signals have
the additional property of limiting the interference from signals
reflected off nearby objects (multipath).
The features which the GPS signal is called upon to provide are
summarized in Table 3.1.
3.3.6 GPS Time. The timing and frequency control for the
carriers, the PRN codes, and the message all comes from an
atomic oscillator on board the satellite. For redundancy each
Navstar GPS Block I1 satellite carriers four oscillators - two
cesium beams and two using rubidum vapour cells. One of the
oscillators is selected by the spacecraft controllers to provide
the frequency and timing requirements for generating the
satellite's signals.
The signals transmitted by the GPS satellites are referenced to
GPS (System) Time. Until June 1990, this was the time kept
by a single atomic clock at one of the monitor stations.
However, the practice now is to obtain GPS Time from a
composite or "paper" clock consisting of all operational
monitor station and satellite clocks.
161
GPS Time is steered over the long run to keep it within about 1
microsecond of UTC ignoring leap seconds. So, unlike UTC,
GPS Time has no leap second jumps. At the integer second
level, GPS Time was equal to UTC in 1980 but at the present
time, due to the leap seconds which have been inserted into
UTC, GPS Time is ahead of UTC by 7 seconds plus a fraction
of a microsecond that vanes day to day.
A particular epoch is identified in GPS Time as the number of
seconds that have elapsed since the previous Saturday / Sunday
midnight. Such a time measure is, of course, ambiguous so
that one must also indicate which week the epoch is in. GPS
weeks are numbered consecutively with week 0 starting on 6
January 1980. Epochs may also be identified by Julian Date
(JD), the number of days and fractional days elapsed since
noon UT on 1 January 4713 B.C., or the Modified Julian Date
(MJD) which is equal to JD minus 2400000.5 JD and MJD
are frequently used by astronomers, navigators, and others for
compactly and unambiguously identifying a particular epoch in
time.
Atomic clocks perform best if they are not continuously

adjusted. So the clocks in the GPS satellites are only
approximately synchronized to GPS Time. The offsets of the
satellite clocks from GPS Time, which can be a millisecond or
so are carefully monitored by the GPS Operational Control
System and the United States Naval Observatory (USNO). For
each satellite clock an offset at an initial epoch, a linear drift
term, and for rubidium clocks a drift rate of change term are
determined. The parameters are uploaded to the corresponding
satellite and subsequently included in its navigation message.
A GPS receiver uses the satellite clock data to convert the
measured pseudoranges from the satellite time scale to GPS
Time. Also included in the satellite message is the offset of
GPS Time with respect to UTC.
When a GPS receiver initially acquires signals, its clock will,
in general, have a large unknown offset with respect to GPS
Time. This offset, however, will contribute the same timing
bias to all pseudorange measurements made at any particular
epoch and can be solved for along with the receiver
coordinates. Once determined, the bias can be used to
synchronize the receiver clock to GPS Time. GPS Time or
UTC can then be displayed by the receiver, used to time-tag
recorded data, or used to generate a one pulse per second
electrical signal for controlling other equipment. The
relationship among the satellite, receiver, and system time
scales and pseudorange measurements is shown schematically

in Figure 3.2. The raw measurement made by the receiver is
the time interval d.r which, multiplied by the speed of light, c,
gives the measured pseudorange, p. Correcting this
measurement for the offsets with respect to GPS Time of the
satellite clock, dt, and receiver clock, dT, gives the true
geometric range p (ignoring propagation delays and other
potential biases).
Figure 3.2 Relationships among GPS Time and satellite

and receiver time scales.
Several manufacturers offer GPS receivers specifically
designed to be used as sources of precise time information.
Generally operated from a fixed site, such a receiver once it has
accurately determined its location can provide synchronized
time signals even when only one satellite is in view.
Accuracies to within about 100 ns are readily achievable.
The USNO, BIPM, and others have developed sophisticated
techniques for using GPS to synchronize clocks to a precision
of I O ns or better even when the clocks are on different
continents.
3.3.6.1 Relativistic Effects. The frequency output of the

atomic clocks in the satellites is 10.23 MHz. This fundamental
frequency corresponds to the chipping rate of the
pseudorandom noise P-code and, when divided by IO, gives the
rate of the CYA-code. The fundamental frequency is multiplied
by 154 to produce the LI carrier frequency and by 120 to
produce the L2 carrier frequency. Actually, in order to account
for the effects of relativity, the fundamental frequency of the
satellite clocks is not set exactly at 10.23 MHz but very slightly
less.
Table 3.1 Required GPS Signal Properties

Work with one-way measurements (receive only)
Serve tens of thousands to millions of both military
and civilian users
Provide accurate, unambiguous, real-time range
measurements
Provide accurate Doppler shift measurements
Provide accurate carrier phase measurements
Provide a broadcast message
Provide ionospheric delay correction
Allow simultaneous measurements from many satellites
Einstein showed in his Special Theory of Relativity published

in 1905 that a clock moving with a constant speed relative to
another clock will appear to run more slowly. A clock in a
satellite travelling in a circular orbit about the earth would
accordingly lose time compared to one on the ground. But 11
years later, in 1916, Einstein in his General Theory of
Relativity deduced that clocks in different gravitational
potentials will also appear to run at different rates. Due to the
difference in gravitational potential, a clock in a satellite would
appear to run faster than one on the ground. The net effect on a
satellite clock is the combination of the two effects. A clock in
a GPS satellite i n a circular orbit with a nominal radius of
26,560 km, gains 38.4 ps per day compared to one on the
ground. This time difference corresponds to a relative
frequency offset of its oscillator of 4.45 x
In order to
compensate for this, the fundamental frequency of the satellite
clocks is reduced by 0.00455 Hz to 10.22999999545 MHz.
Have interference protection

Have multipath tolerance
If the GPS satellites were in circular orbits no further action

would be required to compensate for relativity to give ranging
I62
accuracies at the metre level. However, the orbital eccentricity

of a GPS satellite can be up to 0.02 which means that both its
speed and the gravitational potential it experiences change with
time. The result is an oscillating time offset which is
proportional to eccentricity and varies sinusoidally with the
position of the satellite in its orbit. The magnitude of this
effect can be as large as 45.8 ns which corresponds to a ranging
error of 13.7 metres. A GPS receiver must correct its measured
pseudoranges for this variation using the satellite orbit
description contained in the navigation message.
3.3.7 Selective Availability and Anti-spoofing. GPS was

initially developed to serve the needs of the DoD and the
military forces of U.S. allies. In keeping with its national
security mandate, the DoD plans to limit the real-time absolute
positioning capability of GPS afforded to most civil users. The
service to be provided to all GPS users on a continuous,
worldwide basis without direct charge is called the Standard
Positioning Service (SPS). SPS users will have access to the
coarse acquisition (C/A) code and navigation message on the
LI frequency. The accuracy afforded by SPS has been
intentionally degraded through a process known as Selective
Availability [Georgiadou and Doucet, 19901. The planned
predictable horizontal positioning accuracy to be provided by
SPS is 100 metres 2 drms at the 95% probability level with a
corresponding vertical accuracy of 156 metres (2 sigma). At
the 99.99% probability level, the accuracy will be 300 metres.
During the buildup of the GPS constellation, the precision (P)
code transmitted on both the LI and L2 frequencies will
generally be available to all users. However, when GPS is
declared fully operational, access to the P code by most civil
uses will be denied through the use of an encryption technique
known as Antispoofing.
A higher accuracy service, the Precise Positioning Service, will
be available to DoD-authorized users. The PPS provides
access to the P code on both the LI and L2 frequenies and the
undegraded navigation message. The predictable accuracy of
the SPS is 17.8 metres horizontal (2 drms) and 27.7 metres
vertical (2 sigma).
The PPS community includes civil and military users in the
U.S. federal and allied governments and a limited number of
other users who can demonstrate to the DoD that their use of
the PPS is in the U.S. national interest, that certain security
requirements can be met, and that a reasonable alternative to
the use of the PPS is not available.
Through the use of differential techniques, SPS users can
achieve positioning accuracies that are higher than those
otherwise available. In differential GPS (DGPS), a monitor
station operating at a known location provides range
corrections to users' receivers using some sort of
communication link. I t is also feasible to transmit the actual
data collected by the monitor station rather than range
corrections. The user's receiver combines the monitor station's
data with its own to obtain a high accuracy relative position
[DeLoach, 19911. A non-real-time implementation of this
technique was pioneered by surveyors using the GPS prototype
constellation in the early 1980s.
3.4 Control Segment
In order for a GPS receiver to compute its position in real time,
it must know the positions of the satellites. These positions are
established by the GPS Operational Control System (OCS) and

provided in the navigation messages broadcast by the satellites.
The OCS, which is operated by the Air Force Space Command,
includes five tracking stations, spaced in longitude around the
globe. Three of these stations are on small islands, primarily
used for military purposes: Ascension Island, Diego Garcia,
and Kwajalein. The other two stations are in Hawaii and at
Colorado Springs, CO. The site at Colorado Springs also acts
as the Master Control Station (MCS). (A back-up MCS is
located at the Onizuka Air Force Base in Sunnyvale, CA.) The
MCS collects the pseudorange and carrier phase data obtained
by the tracking stations and, with sophisticated sqftware
models, predicts the future orbits of the satellites. These
computed orbits, called an ephemeris, are then uploaded to the
corresponding satellite using facilities at the island tracking
stations, and are subsequently broadcast by the satellite.
The OCS computes its orbits by fitting Keplerian elements to
the tracking data together with some additional parameters to
account for the perturbations of the orbits. The sixteen
parameters of the broadcast ephemeris are listed in Table 3.2
and illustrated in Figure 3.3. The square root of the semimajor
axis is used rather than the semimajor axis itself to speed up the
GPS receiver's calculation of the satellite's position. Likewise,
the longitude of the ascending node rather than its right
ascension is used in the message. The three parameters An, idot, and R-dot account for the linear changes in the orbit with
time; the six C-values are the amplitudes of sinusoidallyvarying correction terms.
Table 1. Ephemeris Parameters in the Navigation

Message
da
Square root of the semimajor axis
Eccentricity
'0
Inclination angle at the reference time
RO
Longitude of the ascending node at the beginning of

the GPS week
Argument of perigee
MO
Mean anomaly at the reference time
An
Correction to the mean motion computed using da
i-dot
The rate of change of the inclination with time
R-dot
The rate of change of the right ascension of the

ascending node with time
Cut, Cus Amplitude of correction terms for the computed

argument of latitude
Crc, Crs
Amplitude of correction terms for the computed

orbit radius
Cic, Cis
Amplitude of correction terms for the computed

inclination angle
toe
Ephemeris reference time
163
Ib'
uemon of Greenwich meridim

at start of GPS week
Figure 3.3 The parameters of the GPS broadcast ephemeris.
The GPS receiver takes the ephemens parameters and

computes the coordinates of the satellite in an earth-cenmd,
earth-fixed (ECEF) coordinate system. The details of this
computation have been carefully spelled-out hy the designers
of the Global Positioning System and tan be found in a number
of reference publications including the Guide to GPS
Positioning published by Canadian GPS Associates ~ d l ets
al.. 19871. The particular ECEF system used hy GPS is the
World Geodetic System 1984 (WGS 84) of the Defense
Mapping Agency (DMA). For all intents and purposes. the
reference frame of WOS 84 and the new North American
Datum of 1983 WAD 83) are identical.
A new set of orbital parameters is cwnputed for each one hour
period using over-lapping data spans of four hours. A GPS
satellite broadcasts the appropriate set of parameters during a
particular one hour interval.
Fresh orbital data is uploaded to the satellites up to three times
daily. The initial Block I1 satellites can store the navigauon
messages for the following 14 days and therefore are afforded a
certain degree of autonomy should something happen to the
OCS. This autonomy capability has been extended to I80days
for the Block IlA satellites.
The broadcast ephemeris is computed with sufficient accuracy

to guarantee the design goal of horizontally positioning a OPS
receiver with an accuracy of 16 meters. This is the accuracy of
the Precise Positioning Service, the service afforded to
authorized (primarily military) users. When GPS is fully
operabond, the accuracy of the broadcm ephemerides will be
intentionally degraded as one of the meehanisms for
implementing the policy of Selective Availability @/A)
[Georgiadou and Doucet, 19901. Already, extensive tests of
S/A have been c a n i d out. PPS usen will have
undegraded orbits through a decryption process. Unauthorized
users will have to be eonmnt with the lower accuracy Sfandad
Positiomng Service unless they use the technique of differential
or relative positioning (see 3.6.5). Differential positioning is
far less sensihve to orbit errors than point positioning.
For some applications even the undegraded broadcast
ephemerides are not accurate enough. Very high accuracy
geodetic surveys and surveys carried out to study deformations
of natural and man-made structures require orblts with the
highest accuracy possible. Such surveys are carried out using
the differential mode of positioning with a network of
receivers. Each receiver records the phase of the GPS carrier
signals for subsequent analysis. Since the data does not have to
he processed in real time, the data analysts have the luxury of
using post-computed rather than predicted ephemerides. These
ephemeri
spanning
g data from a tracking network

as that of the survey.
The DMA operates a global network of five tracking stations

for the purpose of computing precise ephemerides. Data from
these stations is combined with that from the five OCS stations
to produce the orbits. Several civilian agencies also operate
tracking networks for orbit determination. The largest of these
networks is the Cooperative International GPS Network
(CIGNET). This global network currently consists of about 20
stations which are operated hy various national agencies. The
collected data is transmitted to the CIGNET Data Center in
Rockville. MD, where it is archived and made available for
orbit computation and research.
on to its ephemeris, each satellite transmits

ate orbit information for all of the other safellites in
the GPS constellation. Known as the almanac, this information
can be used by a GPS receiver to determine the location of
each satellite so that it CBH quickly acquire the signals from
satellites that are above the horizon hut are not yet being
tracked. The almanac consists of values for computing the
Keplerian motion of the satellites and only the most significant
of the correction terms is provided - the change in the right
ascension of the ascending node with time. The almanac can
alga be used with various software planning tools to predict
satellite avlulahility at a given location and time. The almanac
is updated by the OCS at least once every six days.
3.5 GPS Receivenr
3.5.1 The Antenna. The job of the antenna Is to convert the
energy in the electromagneticwaves arriving from the satellites
into an electric current which can be handled by the dectronia
in the receiver. The size and shape of the antenna are very
impartant as these characteristicsgovern in part the ability of
the antenna to pick up the very weak GPS signals. The antenna
may be q u i r e d to operste at just the L1 frequency or at both
the LI and L2 frequencies. Alga hecause the GPS signals are
circularly polarized, all GPS antennas must be circularly
polarized as well. Despite
different types of antennag
receivers. These include monopole or dipole configurations,
quadrifilar helices (also known as volutes). spiral helices. and
microstrips.
Perhaps the most Common antenna is the microstrip because of
its NggedneSs and relative ease of wnstruction. It can be
circular or rectangular in shape and is similar in appearance to
a small piece of copperclad prinEed circuit board. Made up of
one or more patches of mefal. microstrips are often referred to
as patch antennas. They may have either single or dual
164
frequency capability and their exceptionally low profile makes

them ideal for airborne and some hand-held applications.
permit tracking of more satellites or the 1

ionospheric delay correction or both.
Other important characteristics of a GPS antenna are its gain

pattern which describes its sensitivity over some range of
elevation and avmuth angles; its ability to discriminate against
multipath signals. that is, signals arriving at the antenna after
being reflected off nearby objects: and for antennas used In
very precise positioning applications, the stability of its phase
centre, the electrical centre of the antenna to which the position
given by a GPS receiver actually refers.
The other channelization concept uses one or more sequencing

channels. A sequencing channel "listens" to a panicular
satellite for a period of time, making measurements on that
satellite's signal and then switches to another satellite. A single
channel receiver must sequence through four satellites to obtain
a three-dimensional position '"fix". Before a first fix can he
obtained, however, the receiver has to dwell on each satellite's
signal for at least 30 seconds to acquire sufficient data from the
satellite's broadcast message. The time to first fix and the time
between position updates can be reduced by having a pair of
sequencing channels.
Some antennas. such as the microstrip, require a ground plane

to make them work prnperly. This is usually a flat or shaped
piece of metal on which the actual microstnp element sits. In
high precision surveying, the ground plane of the antenna is
often extended with a metal plate or plates to enhance its
performance in the presence of multipath.
The GPS signals when they arrive at the earth are very weak:
they have roughly the same strength as those from
geostationary TV satellites. The reason a GPS receiver does
not need an antenna the size of those in some people's
backyards has to do with the structure of the GPS signal and
the ability of the GPS receiver to de-spread it. The power to
extract a GPS signal out of the general background noise of the
ether is concentrated in the receiver rather than the antenna.
Nevertheless, a GPS antenna must generally be combined with
a low noise preamplifier that boosts the level of the signal
before it is fed to the receiver itself. In systems where the
antenna is a separate unit, the preamplifier is housed in the base
of the antenna and receives power from the same coaxial cable
along which the signal travels to the receiver.
3.5.2 The RF Section. The job of the RF section of a GPS

receiver is to translate the frequency of the signals arriving at
the antenna to a lower one, called an intermediate frequency or
IF which is more easily managed by the rest of the receiver.
This is done by combining the incoming signal with a pure
sinusoidal signal generated by a component in the receiver
known as a local oscillator. Most GPS receivers use precision
quartz crystal oscillators, enhanced versions of the regulators
commonly found in wristwatches. The IF signal contatns all of
the modulation that is present in the transmitted signal; only the
carrier has been shifted I n frequency. The frequency of the
shifted carrier is simply the difference between the original
received carrier frequency and that of the local oscillator. It is
often called a beat frequency in analogy to the beat note that is
heard when two musical tones very close together are played
simultaneously. Most receivers employ multiple IF stages,
reducing the carrier frequency in steps. The final IF signal
passes to the work horse of the receiver, the signal tracker
3S.3 The Signal Tmckers. The omnidirectional antenna of a

GPS receiver simultaneously intercepts signals from all
satellites above the antennas, horizon. The receiver must be
able lo Isolate the signals from each panicular satellite in order
to measure the code pseudorange and the phase of the carrier.
The isolation is achieved through the use of a number of signal
channels in the receiver. The signals from different satellites
may be easily discriminated by the unique UA-code or portion
of the P-code they transmit and are assigned to a particular
channel.
The channels in a GPS receiver may be implemented In one of
two basic ways. A receiver may have dedicated channels with
which particular satellites are continuously tracked. A
minimum of four such channels tracking the L1 signals of four
satellites would be required to determine three cwrdinates of
position and the receiver clock offset. Additional channels
signals for
A variation of the sequencing channel is the multiplexing

channel. With a multiplexing channel, a receiver sequences
through the satellites at a fast rate so that all of the broadcast
messages from the individual satellites are acquired essentially
simultaneously. For a multiplexing receiver. the time to first
fix is 30 seconds or less, the same as for a receiver with
dedicated multiple channels.
Receivers with single channels are cheaper but because of their
slowness are restricted to low speed applications. Receivers
with dedicated channels have greater sensitivity because they
can make measurements on the signals more often but they
have inter-channel biases which must be carefully calibrated.
This calibration is usually done by the receiver's
microprocessor.
The GPS receiver uses its tracking channels to make
pseudorange measurements and to extract the broadcast
message. This is done through the use of tracking loops. A
tracking loop is a mechanism which permits a receiver to 'tune
into' or track a signal which is changing either in frequency or
in time. It is a feedback device which basically compares an
incoming (external) signal against a locally-produced (internal)
signal, generates an error signal which is the difference
between the two, and uses this signal to adjust the internal
signal to match the external one in such a way that the error is
reduced to zero or minimized. A GPS receiver contains twn
kinds of tracking loops: the delay lock, or code tracking, loop
and the phase lock, or carrier tracking, lwp.
The delay lock loop is used to align a pseudorandom noise
(PRN) code sequence (from either the C/A or P code) that is
present in the signal coming from a satellite with an identical
one which is generated within the receiver using the same
algorithm that is employed in the satellite. Alignment is
achieved by appropriately shifting the receiver-generated code
chips in time so that a particular chip in the sequence is
generated at the same instant its twin arrives from the satellite.
A correlation comparator in the delay lock loop continuously

cross-correlates the twn code streams. This device essentially
performs a multiply and add process that produces a relatively
large output only when the code streams are aligned. If the
output is low, an error signal is generated and the code
generator adjusted. In this way the replicated code sequence is
locked to the sequence in the incoming signal. The signals
from other GPS satellites will have essentially no effect on the
tracking process because the PRN codes of all the satellites
were chosen to be orthogonal to each other. This orthogonality
property means that a very low output is always produced by
the cnrrelator whenever the code sequences used by two
different satellites are compared.
Because the P-code sequence is so long, a P-code tracking loop
needs some help in setting its code generator close to the right
spot for obtaining lock with the satellite signal. Its gets this
165
help from information included in the broadcast message which
is available to the receiver by first tracking the UAcode.

The time shift required to align the code sequences is. in
principle, the time required for a signal to propagate from the
satellite to the receiver. Multiplying this time interval by the
s p e d of light gives us the distance or range to the satellite.
But because the clocks in a receiver and in a satellite are, in
general, not synchronized and run at slightly different rates, the
range measurements are biased. These biased ranges are called
pseudornnges. Since the chips in the satellite code sequences
are generated at precisely known instants of time, the
alignment of the receiver and satellite code sequences also
gives us a reading of the satellite clock at the time of signal
generation.
Once the code tracking loop is locked, the PRN code can be
removed from the satellite signal by mixing it with the locally
generated one and filtering the resultant signal. This procedure
de-spreads the signal, shrinking its bandwidth down to about
100 Hz. It is through this process that the GPS receiver
achieves the necessary signal to noise ratio to offset the gain
limitation of a physically small antenna.
The de-spread IF signal then passes to the phase lock loop
which demodulates or extracts the satellite message by aligning
the phase of the receiver's local oscillator signal with the phase
of the IF or beat frequency signal. If the phase of the
oscillator signal is not correct, this is detected by the
demodulator in the phase lock loop .and a correction signal is
then applied to the oscillator. Once the oscillator is locked to
the satellite signal, it will continue to follow the variations in
the phase of the canier as the range to the satellite changes.
Most implementations of carrier tracking use the Costas Loop,
a variation of the phase lock loop designed for binary hiphase
modulated signals such as those transmitted by the GPS
satellites.
The carrier beat phase observable is obtained in principle
simply by counting the elapsed cycles and by measuring the
fractional phase of the locked local oscillator signal. The phase
measurement when converted to units of distance is then an
ambiguous measurement of the range to the satellite. It is
ambiguous because a GPS receiver cannot distinguish one
particular cycle of the carrier from another and hence assumes
an arbitrary number of full cycles of initial phase when it first
locks onto a signal. This initial ambiguity must be solved for
mathematically along with the coordinates of the receiver if
phase observations are used for positioning. Because this
ambiguity is constant as long as the receiver maintains lock on
the received signal, the time rate of change of the canier phase
is freed from this ambiguity. This quantity is related to the
Doppler shift of the satellite signal and is used, for example, to
determine the velocity of a moving GPS receiver such as that
in an aircraft.
After the carrier tracking loop locks onto a satellite signal, the
bits in the broadcast message are subsequently decoded using
standard techniques of bit synchronization and a data detection
filter.
There is another way to measure the carrier beat phase other
than the code tracking I Costas L m p combination, hut it comes
with a penalty. This is the so-called signal squaring technique.
The GPS signal is simply a constant carrier who's phase is
shifted by exactly 180" more than a million times each second
as a result of modulation by the PRN codes and the broadcast
message. These phase reversals can be considered as a change
in the amplitude of the signal from +I to -1 or from -I to +I
and the instantaneous amplitude is therefore either plus or
minus one. Electronically squaring the signal results in a

signal with a constant amplitude of unity, although with a
frequency equal to twice the original. However. the phase of
this signal is easily related to the phase of the original &er.
Of course, in the squaring process both the codes and the
broadcast message are lost so code-derived pseudorange
measurements are not possible and the information describing
the orbits of the satellites as well as their health and the other
details in the message must come from another source. There
is also some inherent signal to noise loss in the squaring
process compared to code tracking whch may result in slightly
noisier phase measurements.
One of the first commercially avalahle GPS receivers. the
Macrometer@, used the squaring technique and a number of
cumntly available dual frequency receivers use this approach
for measurements on the L2 frequency. A variation of this
technique has been used in receivers which measure the phase
of the code modulations without having to know the actual
code sequences.
If anti-spoofing is turned on in the satellites resulting in an
encryption of the P-code, then multiplying the signal by itself
will be the only way to make measurements on the L2
frequency. Pseudorange measurements and the broadcast
message would still be available from the code and camer
tracking on the LI frequency.
3.5.4 The Microprocessor. Although the hulk of a GPS

receiver could be built using analogue techniques, the trend in
receiver development has been to make as much of the receiver
digital as possible resulting in smaller, cheaper units. In fact. it
1s possible for the IF signal to be digitized and to perform the
code and carrier tracking with software inside the
micropmessor. So in some respects. a GPS receiver may have
more in common with a compact disc player than it does with
an AM radio. Because it has to perform many different
operations such as initially acquiring the satellites signals as
quickly as possible once the receiver is turned on, tracking the
codes and carriers of the signals, decoding the broadcast
message, determining the user's coordinates, and keeping tabs
on the other satellites in the constellation, a GPS receiver's
The
operation is controlled by a microprocessor.
microprocessor's software, that is the instructions for running
the receiver, is imbedded in memory chips withn the receiver.
The micropmessor works with digital samples of pseudorange
and camer phase. These are acquired as a result of analogue to
digital conversion at some point in the signal flow through the
receiver. It is these data samples that the receiver uses to
establish its position and which may be recorded for future
processing The microprocessor may run routines which do
some filtering of this raw data to reduce the effect of noise or
to get more more reliable positions and velocities when the
receiver is in motion.
The microprocessor may also be required to carry out the
computations for waypoint navigation or convert coordinates
from the standard WGS 84 geodetic datum to a regional one. It
also manages the input of commands from the user, the display
of information. and the flow of data through its communication
port if it has one.
3.5.5 7'he Command Entry and Display Unit. The majority of

self-contained GPS receivers have a keypad and display of
some sort to interface with the user. The keypad can be used to
enter commands for selecting different options for acquiring
data, for monitonng what the receiver is doing, or for
displaying the computed coordinates, time or other details.
Auxiliary information such as that required for waypoint
navigation or weather data and antenna height for geodetic
surveying may also be entered. Most receivers have well

integrated command and display capabilities with menus,
prompting instructions. and even "on line" help. It should he
mentioned that some receivers have a basic default mode of
operation which requires no user input and can he activated
simply by turning the receiver on.
Some GPS receivers are designed as sensors to be integrated
into navigation systems and therefore don't have their own
keypads and displays; input and output is only via data ports.
35.6 Data Storage and Ourput. In addition to a visual display,

many GPS receivers including even some hand-held units
provide a means of saving the carrier phase. and/or pseudorange
measurements and the broadcast messages. This feature is a
necessity for receivers used for surveying and for differential
navigation.
In surveying applications the pseudorange and phase

observations must be stored for combination with like
observations from other simultaneously observing receivers
and subsequent analysis. Usually the data is stored internally
in the receiver using semiconductor memory. Some receivers
store data on magnetic tape or directly on a floppy disk using
an external microcomputer.
Some receivers, including those which store their data
internally for subsequent analysis and those used for real-time
differential positioning, have an RS-232 or some other kind of
communications port for transferring data to and from a
computer, modem or data radio. Some receivers can he
remotely controlled through such a port.
3.5.7 The Power Supply. Most GPS receivers have internal

DC power supplies. usually in the form of rechargeahle nickelcadmium (NiCd) batteries. The latest receivers have been
designed to draw as little current as possible to extend the
operating time between battery charges. Most receivers also
make a provision for external power in the form of a battery
pack or AC to DC converter.
3.6 Observation Equations

3.6.1 Determining Posirionsfrom Pseudomnges
The basic measurement made hy a GPS receiver is the time
required for a signal to propagate from a GPS satellite to the
receiver. Since the signal travels at the speed of light. c. this
time interval can he converted to a distance simply hy
multiplying it by c. Let's assume that the clock in the receiver
is synchronized with the clock in the satellite, and that the
ionosphere and troposphere. which slightly delay the anival of
the signal, do not exist. Furthermore. let's assume there is no
measurement noise; that is, no random perturbation to the
measurement, something which invariably affects all
measurements to a greater or lesser degree. With a single such
measurement of the distance or range to the satellite we can
determine something about the position of the receiver: it must
lie somewhere on a sphere, centred on the satellite. with a
radius equal to the measured range. call it p i .
If we simultaneously make a range measurement to a second
satellite then our receiver must also lie on a sphere, of radius
p2. centred on this satellite. The two spheres will interseet.
with the loci of intersection points forming a circle. Our
receiver must lie somewhere on this circle which is therefore
called a line of position. A third simultaneous range
measurement, p3, gives us a third sphere which intersects the
other two at two points. One of these points can he
immediately dismissed as being the location of our receiver,
since it far out in space. So the simultaneous measurement of
the ranges to three satellites is sufficient to determine a position
fix in three dimensions - at least in principle.
'
,'
/
I
Figure 3.4 Determination of receiver clock offset (dT)

and true user position (intersection of shaded lines)
from the intersection of spheres centered on the
satellites; pseudoranges are shown by arcs of solid
lines.
When we started our analysis. we assumed that the clock in the
GPS receiver was synchronized with the clocks in the satellites.
This assumption, however, is fallacious. When a GPS receiver
is switched on, its clock will in general be mis-synchronized
with respect to the satellite clucks, by an unknown amount.
Furthermore. the atomic clocks in the satellites are
synchronized with each other and to a master time scale, called
GPS time, only to within ahont a millisecond. The range
measurements the receiver makes are biased by the receiver
and satellite clock errors and are therefore referred to as
pseudomnges.
A timing ermr of a millisecond would result in an error in
position of about 3M) km, clearly m intolerable amount. It
would be possible to better synchronize the satellite clocks by
frequently sending them adjustment commands from the
ground, but it has been found that clocks actually keep better
time if they are left alone and the readings of the clock
corrected. The United States Naval Observatory monitors the
GPS satellite clocks and determines the offsets and drifts with
respect to GPS time. These parameters are subsequently
uploaded to the satellites and transmitted as part of the
navigation message broadcast by the satellites. A GPS receiver
uses these satellite clock offset values to correct the measured
pseudoranges.
However, we still have the receiver clock error to deal with.
Because of this error, the three spheres with radii equal to the
measured pseudoranges corrected for the satellite clock offsets
will not intersect at a common point. However. if the receiver
clock error can be determined, then the pseudoranges can be
corrected and the position of the receiver determined. The
situation, compressed into two dimensions, is illustrated in
Figure 3.4.
So we actually have four unknown quantities or parameters

that we must determine: the three coordinates of our position
(say latitude. longitude. and height) and the receiver clock
offset. Now it is a mathematical impossibility to uniquely
determine the the values of four parameters, given only three
measurements. The way out of this conundrum is to
simultaneously measure an additional pseudorange to a fourth
satellite.
167
But just how does the GPS receiver actually extract the
position coordinates and the clock offset from the
measurements? In the software embedded within the GPS
receiver is an algebraic model that describes the geometrical
arrangement we've just looked at. For each pseudorange
measurement, an equation can be written which relates the
measurement to the unknown quantities:
The pseudorange measurement made by the receiver, in units

of distance, is on the left hand side of each of the equations.
The expression under the square root sign is the true range to
the satellite. It is actually a representation of the sphere centred
on coordinates x,y,z, the position of the satellite. The satellite
coordinates are obtained from the navigation message. The
coordinates X.Y,Zrepresent the position of the receiver. The
term c dT is the contribution to the pseudorange from the
receiver clock offset, dT. The set of four equations must be
solved simultaneously to obtain the values for X,Y,Ztogether
with the clock offset, dT. Although the equations are written in
terms of geocentric Cartesian coordinates, the resulting X,Y.Z
values can easily be converted to latitude. longitude. and height
in any gwdetic d a w n or into map grid coordinates.
Linearization of the pseudorange equations. Because of the
squares and square roots in the equations, the pseudorange
measurements are dependent on the receiver coordinates in a
nonlinear way. Consequently, the equations cannot be solved
in the usual fashion we all learned in high school. Instead a
procedure known as Newton-Raphson iteration is used. In this
procedure, each of the equations is expanded about a set of trial
values or guesses for X,YZ,and dT into an infinitely long
polynomial. Then each series is truncated after the first degree
term resulting in an equation which is linear in increments or
corrections to the trial values. The four linearized equations
can then be solved to determine the values of these increments,
and the trial values adjusted accordingly. Since the linearized
equations are an approximation of the nonlinear ones, this
process, in general, must be iterated. with subsequent iterations
yielding smaller and smaller increments. The final solution is
the one that satisfies the original nonlinear equations to within
an acceptable tolerance. Several iterations may be required to
converge to the final solution. However, if the initial position
estimate is close to the actual position, the GPS pseudorange
equations may be solved in just one iteration.
If one or more of the receiver coordinates is already accurately
known, then the remaining coordinates and the receiver clock
offset can be determined using fewer than four pseudoranges.
For example, say that the height of the GPS receiver is known.
Then pseudoranges to three satellites will suffice to determine
the two horizontal coordinates and the clock offset. In using
GPS for synchronizing a clock at a site with known
coordinates, only one pseudorange measurement to a single
satellite is actually required.
Overdetermined solutions. What if more than four satellites
are above the GPS user's horizon? If the user's receiver can
only track four satellites at a time. then the receiver will have
to choose which four satellites to track. We'll have something
to say a b u t a possible selection method a littie later on. But if
the receiver can track five or more satellites simultaneously,
then we have the situation where we have more measurements
than unknowns; that is. we have five or more equations, like
those above, but still with only four unknown parameters. It is
impossible to solve such a set of equations in the same way as
we did for the case of four observations. Why? So far we have
neglected the fact that there are other errors in our
measurements in addition to the satellite and receiver clock
offsets. The presence of these errors means that any subset of
four measurements taken from the full set will produce slightly
different solutions. We say that the solution is inconsisten:.
What do we do? We could discard the extra observations but,
although expedient, that seems wasteful of data. The best
approach is to use a method that was devised in the early 1800s
by the great German mathematician and father of modern
geodesy, Karl Friedrich Gauss: the method ofleast squares. In
this method. we obtain a unique solution for the unknown
parameters which best fits all of the measurements. This
solution is the one which when substituted into the
pseudorange equations gives the smallest discrepancies with
respect to the measurements in a sum-squared sense. That is,
the sum of the squares of the discrepancies is a minimum.
Without going into the mathematical reasons for adopting this
criterion, we can see qualitatively that it assumes that positive
and negative discrepancies are equally likely to occur and that
smaller discrepancies are more likely to occur than larger ones.
3.6.2 The Pseudorange. Before discussing the pseudorange,
let's quickly review the smcture of the signals transmitted by
the GPS satellites Each GPS satellite transmits two signals for
positioning purposes: the L1 signal, centred on a carrier
frequency of 1575.42 MHz. and the L2 signal, centred on
1227.60 MHz. Modulated onto the LI carrier are two
pseudorandom noise (PRN) ranging codes: the 1 millisecondlong CIA-code with a chipping rate of about 1 MHz and a
week-long segment of the P-code with a chipping rate of about
10 MHz. Also superimposed on the carrier is the navigation
message, which among other items. includes the ephemeris
data describing the position of the satellite and predicted
satellite clock correction terms. The L2 carrier is modulated by
the P-code and the navigation message -- the CIAcode is not
present.
The PRN codes used by each GPS satellite are unique and have
the property that the correlation between any pair of codes is
very low. This characteristic allows all of the satellites to share
the same carrier frequencies.
The PRN codes transmitted by a satellite are used to determine
the pseudorange - a measure of the range, or distance, between
the satellite and the antenna feeding a GPS receiver. The
receiver makes this measurement by replicating the code being
generated by the satellite and determining the time offset
between the arrival of a particular transition in the code and
that same transition in the code replica. The time offset is
simply the time the signal takes to propagate from the satellite
to the receiver. The pseudorange is this time offset multiplied
by the speed of light. The reason the ohservable is called a
pseudorange is that it is biased by the lack of time
synchronization between the clock in the GPS satellite
governing the generation of the satellite signal and the clock in
the GPS receiver governing the generation of the code replica.
This synchronization error is determined by the receiver along
with its position coordinates from the pseudorange
measurements. The pseudorange is also biased by several
other effects including ionospheric and tropospheric delay,
multipath. and receiver noise. We can write an equation for the
pseudorange observable that relates the measurement and the
various biases:
p = p + c . (dt-dT) t dion+ dq
+
EP
where pis the measured pseudorange, p is the geometric range

to the satellite, c is the speed of light, dt and dT are the offsets
of the satellite and receiver clocks from GPS time. dion and
dtrop are the delays imparted by the ionosphere and
troposphere respectively. and
represents the effect of
multipath and receiver noise. The receiver coordinates are
I68
nominally constant frequency whereas the received carrier is

changing in frequency due to the Doppler shift induced by the
relative mOhOn of the satellite and the receiver. The difference
between the received camer and the receiver-generated one is
sometimes referred to as the carrier beat phase since the phase
differencerelates to the difference or beat frequency of the two
carriers. Such a beat frequency is well known to musicians
who tune their instruments by listening for the beat note
generated when they play two notes slightly different in
frequency. The phase of the received carrier is related to the
phase of the carrier at the satellite through the time interval
required for the signal to propagate from the satellite to the
receiver.
NYO
Figure 3.5 Typical Variation in L1 pseudorange

measurements over approximately a one-hour period.
hidden m the geometric range along with the coordinates of the
satellite.
The objective in GPS positioning is to
mathematically describe all of the terms on the right-hand side
of the equation -- including the initially unknown receiver
coordinates in the geometric range term -- so that the sum of
the terms equals the measurement value on the left-hand side.
Any error in the description of the terms will result in errors in
the derived receiver coordinates. For example, both the
geometric range term and the satelhte clock term may include
the effects of Selective Availability (SA) which, if
uncompensated. introduce e m into the computed position of
the receiver.
Figure 3.5 illustrates the variation in the pseudorange of a
particular satellite as measured by a stationary GPS receiver.
The large variation is of course dominated by the change in the
geometric range due to the satellite's orbital motion and the
rotation of the earth.
Pseudoranges can he measured using either the CIA-code or
the P-code. Figure 3.6 shows typical CIA-code pseudorange
noise. This "noise record" was obtained by subtracting the
geometric range, clock, and atmospheric contributions from the
pseudomnge measurements illustrated in Figure 3.5. What
remains is chiefly pseudorange multipath and receiver
measurement noise. Because of its higher chipping rate, the Pcode generally provides higher precision observations.
However recent improvements in receiver technologies have
resulted in higher precision UA-code measurements than were
previously achievable.
3.6.3 Carrier Phase. Even with the advances in code

measurement technology, a far more precise observable than
the pseudorange is the phase of the received carrier with
respect to the phase of a carrier generated by an oscillator in
the GPS receiver. The carrier generated by the receiver has a
So, ideally, the carrier phase observable would be the total

number of full carrier cycles and fractional cycles between the
antennas of a satellite and a receiver at any instant. The
problem is that a GPS receiver has no way of distinguishing
one cycle of a carrier from another. The best it can do,
therefore, is to measure the fractional phase and then keep
track of changes to the phase; the initial phase is undetermined.
or ambiguous, by an integer number of cycles. In order to use
the carrier phase as an observable for positioning, this
unknown number of cycles or ambiguity, N. must be estimated
along with the other unknowns -- the coordinates of the
receiver.
If we convefl the measured camer phase in cycles to equivalent

distance units by multiplying by the wavelength, A, of the
carrier, we can express the carrier phase observation equation
as
Q=p
+ c . (dt - dT)t h . N - di,
+dq
t E*
which is very similar to the observation equation for the

pseudorange -- the major difference being the presence of the
ambiguity term. In fact, the camer phase can be thought of as
a biased range measurement just like the pseudorange. Note
also that the sign of lhe ionospheric term in the carrier phase
equation is negative whereas in the pseudorange equation it is
posihve. This comes ahout because the ionosphere, as a
dispersive medium, slows down the speed of propagation of
signal modulations (the PRN codes and the navigation
message) to helow the vacuum speed of light whereas the
speul of propagation of the carrier is actually increased beyond
the speed of light. Don't worry, Einstein's pronouncement on
the sanctity of the speed of light has not been contradicted.
The speed of light limit only applies to the transmission of
information and a pure continuous carrier contains no
information.
Although all GPS receivers must lock onto and track the carrier
of the signal in order to measure pseudoranges, they may not
measure or record carrier phase observations for use in
navigation or positioning. Some however, may internally use
carrier phase measurements to smooth -_ reduce the high
frequency noise - on the pseudorange measurements.
Incidentally, in comparison with the carrier phase,
pseudoranges when measured in units of the wavelengths of
the codes (300 meters for the UA-code and 30 meters for the
P-code) are sometimes referred to as code phase
measurements.
I
131
IS 4
1
I.
(I.
1.
W*rdTrn.h.L
I. 2
,I 4
I
(4
'
Figure 3.6 The ditference between the L1 pseudorange

measurements shown in Fig. 3.5 and the corresponding
phase measurements.
3.6.4 Point Positions. Most civilian receivers intended

primarily for navigation exclusively use CIA-code pseudorange
measurements to establish the position of a point or the
trajectory of a moving platform. The accuracy of point
positions afforded civilians is primarily limited by SA rather
than receiver measurement precision. In fact, before SA was
169
approximatelytu scale.
170
adjustment along with the coordinates of one of the receivers

(the coordinates of the other receiver are held fixed at a priori
known values). Usually these ambiguities are initially
estimated as real or floating point numbers rather than integer
numbers. If the data quality is good enough and the model
used in the adjustment accurately describes the OhSeNatiOnS,
the real number estimates from the float solution will turn out
to be very close to integers. The estimates can then he rounded
off to the nearest integer and then held fixed at these integer
values in a second adjustment of the data. The so-called fixed
solution will in general provide more accurate results than
those afforded by the float solution.
3.6.5.3 The Triple Difference. A potential problem with the

double difference observable is associated with the integer
ambiguity term. As long as the receivers do not lose carrier
lock on the signals, the integer ambiguities remain constant for
the whole data set. If, however, one or both receivers loses
lock due to a satellite passing behind an obstacle. low signal to
noise ratio, rapid motion of the receiver, or a severe
ionospheric disturbance, then one or more carrier phase cycles
will be lost or slipped. The receiver essentially loses track of
the continuous cycle count. This introduces a discontinuity or
cycle slip into the data. If any data gap accompanying the
cycle slip is short and any noise corrupting the data is minimal
then it is generally possible to determine the correct number of
slipped cycles and to correct the phase measurements to
produce a continuous phase record. If this cannot be done,
then a new set of ambiguities must be adopted for the
measurements following the cycle slip.
An observable which is insensitive to both the initial integer
ambiguities and cycle slips is the triple difference. Triple
differences are formed by sequentially differencing double
differences in time. If we have double differences at epochs
one. two, and three for example, then we can create two triple
differences: double difference two minus double difference one
and double difference three minus double difference two. The
triple difference observation equation is written as
SVAO=SVAp&'Mi,
+SVMtmp+Esv~e
where S is the time difference operator. The differencing in

time results in an observable which has less information
content and as a result receiver coordinates estimated from
triple differences tend not to be as accurate as those obtained
from double differences especially when the latter are used in
fixed ambiguity solutions. Nevertheless. triple difference
results may be accurate enough for certain applications. Also.
the triple difference observable is very useful for obtaining
initial estimates of receiver coordinates which can then be used
as a priori coordinates in a double difference solution. Triple
differencesare also very useful in spotting and correcting cycle
slips. In a triple difference data series. a cycle slip usually
appears as an easily-identified spike. If the data is particularly
noisy -- for example when it is corrupted by severe ionospheric
irregularities, identifying and repairing cycle slips can be quite
difficult.
3.6.6 Other Linear Combinations. The single, double. and

triple difference ohservables are known in mathematical
parlance as linear combinarionr of the measured carrier phases
(or pseudoranges). Several other linear Combinations of the
basic GPS observables have also found utility in GPS
navigation. positioning. and time transfer. Foremost among
these perhaps is the ionosphere-free linear combination of the
raw or undifferenced LI and L2 canier phases measured by a
single receiver. The ionospheric delay term in the carrier phase
observation equation is, with negligible error. inversely
proportional to the square of the carrier frequency. Therefore,
by combining the equation written for the L2 observation with

the equation for the LI observation. we can create an equation
and hence an observable which is essentially free of the
ionospheric effect. This observable is sometimes referred to as
Lc or L3 although the latter term should be deprecated because
of possible confusion with the L3 signal associated with the
nuclear burst detection package on the GPS satellites. We can
remove the ionospheric delay from pseudorange data with a
similar operation on the LI and L2 pseudoranges. Ionospherefree observations may be combined into single, double, or
triple differences and processed in almost the same manner as
single frequency data. We say "almost" because in the
ionosphere-free ObseNatiOnS, the ambiguities are no longer
integers and ambiguity resolution is more complex than in the
single frequency case. Another distinction is that for baselines
shorter than about 20 kilometers or so, ionosphere-free
observations are noisier than their single frequency
counterparts. This characteristic arises from the fact that the
dominant error source on short baselines is multipath and
receiver noise which in large part is uncorrelated between the
L1 and L2 signals. Therefore for short baselines it may be
preferable to use single frequency obsewations.
Instead of combining the LI and L2 phase measurements to
remove the ionospheric delay, we can simply difference the L1
and LZ phases measured in distance units and determine
changes in the ionospheric delay. Unfortunately, we cannot
determine the absolute ionospheric delay from phase
measurements because of the unknown integer ambiguities.
Nevertheless. since orbit, clock, and tropospheric effects are
removed, this linear combination of the L1 and LZ phases is
particularly useful for locating and potentially removing cycle
slips.
Other linear combinations of the L1 and L2 phase
measurements have been devised to help with the ambiguity
resolution problem. Differencing the LI and LZ phases
measured in cycles results in an observable with an effective
wavelength of about 86 cm whereas summing the L1 and LZ
phases in cycle units gives an observable with a wavelength of
about 10.7 cm. Algorithms have been developed to use the socalled wide-lane and narrow-lane ambiguities to help resolve
the L1 and L2 ambiguities. Still other linear combinations of
both phase and pseudorange observations are being
investigated for data cleaning, particularly under
ionospherically noisy conditions.
We can also combine together carrier phase and pseudorange
measurements. In fact you have seen such a combination in
Figure 3.6. Differencing the camer phase and CIA-code
pseudorange measurements on LI removes all of the common
effects: geometric range, clock terms. and tropospheric delay.
What remains is the effect of the ionosphere (doubled because
of the sign difference), multipath, receiver noise. and the
carrier phase ambiguity. An estimate of the ambiguity and the
constant part of the other effects can be removed by subtracting
the arithmetic mean of the camer phase - pseudorange
difference. We are left with an observable which is usually
dominated by the pseudorange multipath and the pseudorange
measurement noise. Also remaining is the variation in the
ionospheric delay. Over the one-hour period of the
measurements shown in Figure 3.6, the ionospheric delay
changed by only about 1 meter in a very smooth fashion. Of
course, if we had used data from a dual-frequency P-code
receiver, we could have removed the effect of the ionosphere
ab initio by using the Lc carrier phase linear combination and
the corresponding P-code pseudorange combination.
With the advent of receivers providing low-noise LZ
pseudorange observations, we can effectively use the
pseudoranges to help determine the integer ambiguities of the
171
carrier phase measurements. This synergistic combination of

pseudoranges are carrier phases has spawned the rapid static
surveying technique with which we can determine the
ambiguities with observation times of mere minutes.
3.7 Measures of GPS Position Accuracy.

As we have mentioned, the pseudorange measurements are
contaminated by the satellite and receiver clock offsets. Even
after solving for the receiver clock offset and correcting the
pseudoranges for the satellite clock offset using the parameters
in the navigation message, errors still remain i n the
measurements. These errors will, of course, affect the accuracy
of the position determination. Since these errors will, in
general, change with time, repeated determinations of the
position of a fixed location will give slightly different results.
The pseudorange errors come from several sources. The
parameters in the navigation message describing the behaviour
of the satellite clock account for almost all of the clock offset
with respect to GPS time. However because the model used to
describe the clock behaviour is quite simple and the parameters
of the model are predicted ahead of time, there are some small
residual clock errors remaining in the pseudoranges. The
positions of the satellites as computed from the predicted
ephemerides in the navigation messages are also slightly in
error. Other errors in the pseudoranges include unmodelled
effects of the ionosphere and troposphere, multipath, receiver
measurement errors, and (for the civilian user) the additional
clock and orbit errors due to Selective Availability (S/A) when
it is in effect.
3.7.1 User Equivalent Range Error. Each of these errors,

regardless of their origin, can be expressed as an error in the
range between the user and the satellite. When an error is
expressed in this way, it is known as a user equivalent range
error (UERE) or just user range error (URE). Over a
sufficiently long period of time, these errors can be considered
to be random in nature, with negative and positive errors being
about equally probable giving a mean value of zero. Also,
smaller errors are more probable than larger errors. A graph of
the frequency of occurrence of an error of a certain size,is
referred to as aprobability densiQ function. The probability
that an error will occur with a value between el and e2 say, is
just the area under the curve between these two values on the
horizontal axis.
The shape of the probability density curve depends on the
particular parameter being measured. However, it is often the
case in science and engineering that the probability density
curve is of a particular shape known as a Gaussian or normal
distribution. To quantify such a distribution or dispersion of
possible errors with a single number, we use the standard
deviation and usually represent it by the Greek letter sigma
(a). It is possible to determine a experimentally by making a
large number of observations. It is just the square root of the
sum of the squares of the errors in the observations divided by
one less than the number of observations made. It is from this
method of computation that gives a its alias of root-meansquare (r.m.s.) error. For the Gaussian distribution, there is a
68% chance that the magnitude of the error we actually get will
be smaller than the standard deviation. There is a 95% chance
that it will be smaller than twice the standard deviation and a
99.7% chance that it will be smaller than thrice the standard
deviation.
3.7.2 Dilution of Precision.

Since these errors originating from the different sources are
independent of each other, a combined error can be obtained by
taking the square root of the sum of the squares of the
individual errors. This value is the total user equivalent range
error.
The total UERE is clearly not the error in the position

determined by a GPS receiver. It is only a measure of the error
in the distance to one of the satellites. In order to determine the
three dimensional position error, we must also take into
account where the satellites are in the sky with respect to the
receiver. Because we can only receive signals from GPS
satellites that are above our horizon, the effect of the satellite
geometry is always to dilute the precision of the position
determination. I t is possible to quantify this dilution through
the position dilution of precision (PDOP) parameter. PDOP is
the number which when multiplied by the r.m.s. UERE gives
the r.m.s. position error, that is, the square root of the sum of
the squares of the standard deviations in latitude, longitude,
and height.
PDOP is a mathematical function involving the relative
coordinates of the receiver and the satellites and can easily be
computed for a particular satellite arrangement. PDOP can
also be visualized geometrically using four satellites by looking
at the tetrahedron formed by the end points of vectors of unit
length pointing from the receiver to each of the satellites.
PDOP is inversely proportional to the volume of this
tetrahedron. The more spread out the satellites are in the sky,
the larger the volume of the tetrahedron, the iower the PDOP,
and hence the lower the r.m.s. position error. If more than four
satellites are in view, the GPS receiver can select the four
which give the lowest PDOP. The lowest value of PDOP that
could be obtained is with one satellite at the user's zenith and
three satellites evenly spread out in azimuth on the user's
horizon. The maximum value PDOP could theoretically have
is infinity. This would occur if the four satellites were to lie in
the same plane. The final GPS constellation has been designed
to provide users anywhere in the world with a PDOP of less
than 6 (except for occasional very brief periods of time),
assuming four satellites are used with a minimum satellite
elevation angle of 5". Users can expect the PDOP to be less
than 3 most of the time.
Several other related DOP factors have been defined. HDOP is
the dilution of precision in the two horizontal coordinates;
VDOP is the dilution of precision in the vertical coordinate;
and TDOP is the dilution of precision in the range equivalent
of the receiver clock offset. A factor which combines the
effect of geometry on both position and clock offset is the
geometric dilution of precision, GDOP.
3.7.3 Other Accuracy Measures.

In general, the three coordinates of a three-dimensional
position fix will have different error probability distributions
and hence different standard deviations. Also the errors
between any two coordinates may be mutually correlated; that
is, an error in one coordinate will have an effect on the other.
If we trace out a contour of equal probability density in all
three coordinates, we get an ellipsoid centred on our position
fix. The shape of the ellipsoid is determined by the standard
deviations of the coordinates and their correlations. Note that
in general, due to the correlations, the ellipsoid axes are not
oriented in the same directions as the coordinate axes. There
is a certain probability that the true position lies within the
ellipsoid. If this probability is 20%, then this ellipsoid is
referred to as the standard error ellipsoid.
If the standard deviations in the three orthogonal directions are
identical, the ellipsoid degenerates into a sphere. The radius of
such a sphere inside of which there is a 50% probability of the
true position fix being located is called spherical error
probable (SEP). The term SEP is also used to indicate the
radius of a sphere inside of which there is a 50% probability of
being located even when the actual error figure is an ellipsoid.
If we make a large number of position fixes, we can say that
172
the SEP is the radius of the sphere containing 50% of the

individual fixes. The Department of Defense's accuracy goal
for GPS is to have a worldwide SEP of 15 metres.
If we forget about the height coordinate for the time being and
consider just the horizontal coordinates, we can construct the
two-dimensional analogue to the error ellipsoid: the error
ellipse. It is defined as the contour of equal probability density
in the two horizontal dimensions. There is a certain probability
that the true horizontal coordinates lie within the ellipse. For
the srundurd error ellipse, this probability is 39%. As with the
error ellipsoid, the semi-axes of the error ellipse are not in
general equal to the standard deviations. However, given the
standard deviations in the horizontal coordinates and their
correlation, the semimajor and semiminor axes of the ellipse
can be calculated.
The two-dimensional analogue of SEP is circular error
probable (CEP). CEP is the radius of a circle inside of which
the true horizontal coordinates of a position have a 50%
probability are being located.
Another accuracy measure frequently used in navigation is
twice the root-mean-square of the horizontal distance error or
2 dms for short. It is equal to twice the square root of the sum
of the squares of the semimajor and semiminor axes of the
error ellipse. A circle of radius 2 drms will contain the true
horizontal position with a certain probability. Unfortunately a
drawback of 2 drms as a measure of error is that it does not
correspond to a fixed value of probability for a given value of
error. The probability varies with the eccentricity of the error
ellipse, ranging from 95.4% (ellipse collapses to a line) to
98.2% (ellipse becomes a circle). Because of this variation in
probability, there is not a constant relationship between values
of 2 dms and CEP. The ratio of 2 drms to CEP varies with the
eccentricity of the error ellipse from 2.4 to 3.
Because of its wide use in navigation, 2 drms is used to specify
the designed level of horizontal positioning accuracy for the
GPS Standard and Precise Positioning Services. The latest
issue of the Federal Radionavigation Plan (FRP) states that
when GPS is declared operational, the horizontal accuracy for
SPS is planned to be 100 m 2 drms at 95% probability. This
means that 95% of all horizontal position fixes should be
within 100 m of the true position. But what about the other
5%? Theoretically, if the position errors due to the various
UEREs including SIA are from a Gaussian distribution, we
could occasionally get extremely large errors. However, the
Department of Defense will control SIA such that excursions
will not exceed 300 m 99.99% of the time. The corresponding
designed 2 dms (95%) horizontal accuracy for PPS is planned
to be 17.8 m.
The FRP describes the designed accuracy of the vertical
component of a GPS-derived position at the 20 level. As this
corresponds to a 95% probability level it is consistent with the
accuracy quoted for the horizontal position. For SPS, the
designed vertical 2a is 156111; for PPS it is 27.7m.
The FRP gives the designed accuracy of receiver clock
synchronization at the la level. For SPS, a t is planned to be
167 nanoseconds. For PPS, a t is given conservatively as 100
nanoseconds.
It should be pointed out that the stated PPS position and time
accuracies are designed estimates of GPS capabilities and
superior results have already been obtained in practice.
Significantly greater accuracies can be obtained for both PPS
and SPS users by operating in a differential mode with two or
more receivers used simultaneously. In fact, almost all of the

effects of SIA can be removed when operating in this mode.
3.8 Coordinate Systems

3.8.1 WGS84.
With the requirement to tie different regional datums together
for military and other purposes and with the advent of satellitebased positioning systems, there was a clear need for a global
geodetic reference system. One of the first such systems was
the U.S. Department of Defense World Geodetic System
(WGS) introduced in 1960. WGS 60 was created from a global
data base of conventional geodetic measurements as well as
satellite observations and Hiran (High Precision Short Range
Navigation), an airborne trilateration or range measuring
system developed during and after World War 11. In the years
following the introduction of WGS 60, the accuracy and
number of satellite observations greatly increased and led to
the development of WGS 66 and subsequently WGS 72.
WGS 72 was adopted as the CTS for describing the orbits of
the GPS satellites in their navigation messages. But like that of
WGS 60 and WGS 66, the accuracy of WGS 72 eventually was
found wanting. WGS 72 was superseded by WGS 84 and has
been used for the GPS navigation messages since January
1987.
The reference ellipsoid of WGS 84 is essentially that of the
International Union of Geodesy and Geophysics (IUGG)
Geodetic Reference System 1980 (GRS 80) with some minor
changes. This ellipsoid was adopted by the international
geodetic community at the 17th quadrennial meeting of the
IUGG in Canberra in 1979 as best representing the size and
shape of the earth. The WGS 84 ellipsoid is specified by the
value of 6,378.137 km for its semimajor axis, a form factor
describing the earth's equatorial bulge from which a flattening
of 11298.257 223 563 is derived as well as values for the mean
rotation rate of the earth and a value for the product of the mass
of the earth and the gravitational constant. How well does the
WGS 84 ellipsoid represent the actual earth? The equatorial
radius is probably in error by no more than 1 or 2 metres and
the value for the flattening is accurate to about 3 parts in a
million.
The coordinate system of WGS 84 is a realization of the CTS
as established by the Bureau International de I'Heure (BIH) on
the basis of coordinates adopted for BIH stations (the BIH was
a forerunner of the International Earth Rotation Service). The
Z axis of the coordinate frame is parallel to direction of the
CTP; the X axis lies at the intersection of the CTP's equatorial
plane and the zero meridian; and the Y axis completes the
system. The centre of the coordinate frame coincides with the
centre of the WGS 84 ellipsoid and the coordinate axes
coincide with the rotational axes of the ellipsoid.
The WGS 84 coordinate system was established using a set of
coordinates for a worldwide network of U.S. Navy Navigation
Satellite System (Transit or Doppler) stations. However, it had
been learned from co-siting Doppler stations with VLBI and
satellite laser ranging (SLR) stations that the Doppler
coordinate system had metre-level errors. For example, it was
discovered that the origin of the coordinate system was about
4.5 metres above the earth's centre of mass. So the Doppler
station coordinate set was modified in origin, scale, and
orientation to agree in the mean with the VLBl and SLR
results.
WGS 84 also includes a description of the earth's gravity field.
Knowledge of the gravity field is needed, for example, for
modelling the orbits of satellites. The field is described by a
series of coefficients or values which account for smaller and
smaller features with each additional term. These terms are
I73
called spherical harmonics. Spherical harmonics are identified

by a pair of indices called the degree , m, and order, n. The
harmonics with m = 0 are independent of longitude and are
called zonal harmonics. Those with m = n are independent of
latitude and are called sectorial harmonics. The others are
functions of both latitude and longitude and are called tesseral
harmonics. The WGS 84 gravity field is complete up to degree
and order 180. This means that variations i n the field
originating from an area of roughly 200 km by 200 km in
extent are described. Becuase of the military importance of
this model of the gravity field, only the coefficients up to
degree and order 18 are unclassified. The field coefficients
were obtained from the analysis of Doppler satellite tracking
data, SLR data, surface gravity data, oceanic geoid heights
from satellite altimetry, and GPS data as well as results from
the analyses of the orbits of a number of other satellites.
The coordinates directly computed by a GPS receiver refer to
the same system as the satellite coordinates that are used in the
computation. Since these normally refer to WGS 84, so do the
receiver coordinates. These coordinates may be expressed as
Cartesian coordinates (X, Y, Z) or geodetic ellipsoidal
coordinates (+, A, h). However, most GPS receivers also
provide an option to transform the coordinates to one of a
number of different regional datums such as those listed in
Table 3.3. In the table, the datum shifts give the mean offsets
of the reference ellipsoids from the centre of the WGS 84
ellipsoid, the reference frame rotations represent the
misalignment of the axes the coordinate systems with respect
to that of WGS 84, and the scale parameter accounts for
differences in the scale of lengths in each datum with respect to
the scale of WGS 84. The parameters in Table 3.1 are mean
values from a theoretical study and should not be used for
everyday mapping, charting, and survey applications. Such
mean values do not account for variations in scale and the
distortions in the datums which might exist.
For
transformations between datums, DMA provides a set of
multiterm polynomials for specific regions that it has derived
from extensive least squares analyses.
The receiver can also display orthometric height rather than
geodetic height if the geoidal undulation is known.
3.8.2 NAD 83.

GPS users in North America currently have a slight advantage
over users in other regions as a result of the recent introduction
of a new datum, the North American Datum of 1983 (NAD
83). NAD 83 replaces the North American Datum of 1927
(NAD 27). NAD 27 coordinates had become inadequate for
many purposes. Many of the published coordinates of survey
control stations were unreliable due to errors and distortions in
the system. I n fact, relative coordinates in NAD 27 were
sometimes in error by as much as 1 part in 15,000. A further
disadvantage of NAD 27 was that its reference ellipsoid was
non-geocentric and was not precisely oriented with respect to
the CTS as established by the BIH.
The need for a readjustment of North American networks was
realized in the late 1960s and work on the r.ew datum officially
began in 1975. NAD 83 was obtained by a least squares
adjustment of over one and three quarter million geodetic
observations at sites in the United States, Canada, Greenland,
Mexico, Central America and the Caribbean. In additional to a
large number of observations obtained using traditional
surveying techniques, Doppler and VLBI observations were
also used. Originally scheduled for completion in late 1982,
the new datum was christened NAD 83. But due to various
delays and extensions the first NAD 83 coordinates were not
published by the National Geodetic Survey until March 1987.
The reference ellipsoid and coordinate system of NAD 83 are

almost identical to those of WGS 84; the two systems agree at
about the 0.1 mm level. So, WGS 84 coordinates provided by
a GPS receiver can be taken to be NAD 83 coordinates.
However, the GPS surveyor should realize that if he occupies a
particular reference marker for which published NAD 83
coordinates exist, the coordinates he computes for the marker
from GPS observations may occasionally differ from the
published coordinates by a metre or more due to remaining
errors and distortions in the datum. Nevertheless, the vast
majority of published first- and second-order NAD 83
coordinates are accurate relative to one another at the 3 to 4
parts-per-million level.
To provide geodetic reference coordinates at a higher accuracy

than afforded by standard NAD 83 coordinates, many states
have established special GPS-derived "NAD 83 high precision"
networks.
3.8.3 UTM.
The GPS receiver may also be able to project the ellipsoidal
coordinates onto a mapping plane, that is, a flat map. It is
impossible to project an ellipsoidal surface onto a flat surface
without some distortion taking place. However, projections
have been developed which minimize these distortions. One
such projection is the Universal Transverse Mercator (UTM).
The UTM projection, which can trace its lineage back to Karl
Friedrich Gauss, has been adopted by the IUGG, NATO and
other military organizations and many civil administrations
worldwide for their various mapping needs.
The UTM projection divides the world between 80"s and 84"N
into 60 zones each with a width of nominally 6 degrees of
longitude onto which is superimposed a grid. Each of these
zones, which constitutes a segment of a reference ellipsoid, is
projected onto a cylinder whose axis is parallel to the earth's
equator and whose radius is chosen to keep the scale errors of
the projection within acceptable limits. Coordinates of points
on the ellipsoid within a particular zone can then be
transformed to coordinates on the UTM grid. The UTM
coordinates are generally referred to as eastings and northings
and are expressed in metres. Eastings are reckoned from the
central meridian of a zone and have 500,000 m added to them
so that all coordinates remain positive. Northings are reckoned
from the equator which has a coordinate value of 0 m for work
in the northern hemisphere and 10,000,000m for work in the
southern hemisphere. An important property of the UTM
projection is that is uses a conformal mapping which means
that the magnitude and sense of angles measured on the
ellipsoid are preserved when coordinates are transformed to the
mapping plane.
The U.S. State Plane Coordinate System uses a transverse
Mercator projection or another conformal projection, the
Lambert conic map projection, in one or more zones, to map
each state of the Union, Puerto Rico and the US. Virgin
Islands, onto a plane rectangular coordinate system. (The
panhandle of Alaska is a unique case with its own special
projection.) The transformations from NAD 83 geodetic
coordinates to grid coordinates yield errors less than about 1
cm for points within the boundaries of the appropriate zone so
that either geodetic coordinates or the corresponding grid
coordinates of a point may be used depending on the
application.
4. OTHER SATELLITE NAVIGATION SYSTEMS
4.1 INTRODUCTION
The following discussion briefly summarizes the
characteristics of several other satellite navigation systems.
4.2 GLONASS
174
The former Soviet Union has developed a system very similar

to GPS. Called Glonass, for Global Navigation Satellite
System, it uses satellite orbits, transmission frequencies, and
signal structure which bear a striking resemblance to the U.S.
system. However, the signals are different enough so that
presently-available commercial GPS receivers are unable to
access Glonass signals. The first three Glonass satellites were
launched on 12 October 1982 on a single rocket from the
Baikonur Cosmodrome near Tyuratam. Since then a further
nine trios have been launched, including one set that went into
undesirable elliptical orbits. As of early 1988, only five of
these prototype satellites were fully functional. The full
operational constellation scheduled for deployment by about
1995 will consist of 24 satellites, with 3 i n standby mode
[Anodina, 19881.
The Glonass satellites are placed into circular orbits with a
nominal semi-major axis of 25 507 km giving them an orbital
period of about 675.7 minutes. These satellites have ground
tracks which repeat every 17 orbits, i.e., 8 sidereal days.
However, not all of the prototype satellites are presently in
such synchronized orbits. Glonass orbital planes are separated
from one another by 120" and have target inclinations of 64.8".
The spacings of satellites within a plane are 30" or 45".
Like the GPS satellites, Glonass satellites transmit two pseudorandom noise signals with binary phase-shift keying. However
the clock rate of the Glonass signals is about one-half that of
GPS: 5.1 1 MHz for the equivalent of the P-code and
0.51 1 MHz for the equivalent of the C/A-code. A 50 bits per
second message lasting 2.5 minutes is also superimposed on
the signals. The signals are transmitted within two bands: LI,
I597 - 16 I7 MHz and L2, I240 - 1260 MHz, with different
satellites using different carrier frequencies. The channel
spacing for LI is 0.5625 MHz and for L2, 0.4375 MHz. As
with GPS, the P-code is transmitted on both LI and L2,
whereas the C/A-code is only present on the LI signal. The
C/A-code is 51 I chips long. The length of the P-code is not
presently known. Unlike GPS satellites, all Glonass satellites
appear to transmit the same codes. Signal timing and
frequencies are derived from one of two on-board atomic
clocks [Klass, 19881. The signals are right-hand circularly
polarized, like GPS signals, and have spectral power densities
of -44dbW/Hz in the L1 frequency band and -57 dbWlHz in
the L2 band. Some of the prototype satellites transmit signals
which include narrow spectral lines or tones. It is not clear if
these are intentional or spurious. If intentional, they could be
used for instantaneous velocity determination.
Accuracies of positions determined from the C/A-code of 100
m horizontal, and 150 m vertical are claimed, with velocity
component and time accuracies of 15 cm s-' and 1 ~s
respecti vely [Anodi na, 19883.
Some of the details about Glonass have been sleuthed through
monitoring of the satellite signals, primarily by Daly and his
co-workers at the University of Leeds [Dale and Daly, 1986;
1988; Klass, 1987al.
4.3 Transit Satellite Positioning

Experiments performed by scientists at the Johns Hopkins
University Applied Physics Laboratory (ARL) following the
launch of Sputnik 1 showed that it was possible to determine
the satellite's orbit by analysing the Doppler shift of the
satellite's radio transmissions. It was subsequently realized
that if the position of a satellite could be determined by
measuring the Doppler shift at a station of known position,
then it should be possible to determine the position of the
station if the position of the satellite is known. This realization
led to the development by ARL of the U.S. Navy Navigation
Satellite System, commonly known as Transit.
The first prototype Transit satellite was launched in 1961.

Following deployment of a number of test satellites, the system
was declared operational in 1964 and became classified.
However, in 1967 Transit was declassified and became
available for civilian use.
The Transit system consists of three components: the satellite
tracking and control facilities, the satellites themselves, and the
users. The Transit satellites are controlled by the U.S. Naval
Astronautics Group (NAG) at Point Mugu, California. Four
tracking stations in the U.S. record Doppler measurements on
each pass of every operational satellite. These data are sent to
Point Mugu where the orbit of each satellite is determined and
extrapolated into the future. This ephemeris is transmitted by
one of two injection stations to the satellite where it is stored in
memory for subsequent rebroadcast.
Six Transit satellites are presently fully operational, including
one launched in May 1967. Of these six, four are of the older
Oscar type and two are of the advanced NOVA class. Another
four Oscars and one NOVA have been stored in orbit for future
use. The satellites are in circular, polar orbits with altitudes of
roughly 1100 km and their nodes more or less evenly spaced
around the equator. Their corresponding orbital period is about
107 minutes. Each satellite transmits two harmonically-related
carrier frequencies, one at 400 MHz, the other at 150 MHz.
Both frequencies are actually offset from these nominal values
by -80 ppm to make receiver operation simpler. The use of two
frequencies permits correction for the dispersive effect of the
ionosphere (see Chapter 9). Superimposed on both carriers by
balanced phase modulation is a broadcast message containing
the orbit information previously injected into the satellite by
the NAG. The message is continuously transmitted as a series
of two-minute paragraphs, each paragraph consisting of 6103
binary bits. From the message, a new instantaneous orbit can
be computed every two minutes.
The received signals are compared with the local oscillator
frequency generated i n the user's receiver and the Doppler
shift, or frequency difference, is integrated to yield the
observable. By combining these Doppler counts with the
satellite orbit data, accurate coordinates of the receiver can be
determined.
The Soviet Union operates a system similar to Transit, called
Tsicada.
4.3.1 Principle Of Doppler Positioning.

The Doppler effect, discovered by Christian Doppler a
nineteenth century Austrian physicist, is familiar to anyone
who has waited patiently at a railway level crossing for a train
to pass. The pitch of the train's horn or whistle changes as the
train passes. It starts out high, changing imperceptibly as the
train approaches, then drops noticeably as the train goes
through the crossing, and maintains a lower pitch as the train
recedes in the distance. This same phenomenon which is so
readily apparent at audio frequencies also affects
electromagnetic waves. The frequencies of both radio and light
waves are shifted if the source (transmitter) and the observer
(receiver) are in relative motion.
The classical explanation of the effect is that the observer
receives more wave crests per second, i.e., the frequency is
increased if the source and the observer are moving closer
together, whereas fewer wave crests per second are received,
i.e., the frequency is decreased, if the source and the observer
are moving farther apart. If the relative speed of the source and
observer is much less than the speed of light, then the received
frequency is given approximately as
175
Suggested further readings are Bomford [ 19801 and Stansell

[ 19781.
where fs is the frequency at the source, c is the speed of light,
and S the distance or range between the source and the
observer; dS/dt is the range rate.
Returning to the train at the level crossing, you may have
noticed that the closer you are to the track, the faster the
change in pitch of the horn. And even if you could not see or
feel the train, you can tell when it passes the crossing (the point
of closest approach) by noting the instant when the pitch of the
horn is mid-way between the high and low extremes (fs).
Therefore by monitoring the frequency of the received sound
as the train passes and knowing its schedule (position and
speed), you can establish your position in a two-dimensional
coordinate system where the x-axis, say, runs along the track
and the y-axis runs perpendicular to it. The origin may be
assigned arbitrarily. This is the principle of Doppler
positioning.
In the case of a Transit satellite (or any other satellite for that
matter), the position of a receiver can be established by
continuously recording the Doppler shift of the received signals
(or the number of cycles Of the Doppler frequency which is a
more Precisely obtained observable). Subsequently these data
are combined with accurate coordinates of the satellite to
determine the position of the receiver. As with the passage of a
train, a single satellite pass can provide at most O n l y
coordinates of the receivers position. Whereas this may be
satisfactory for navigation at sea where the height above the
reference ellipsoid is approximately known, three-dimensional
positioning requires observing multiple satellite passes.
4.3.2 Transit Doppler Measurements.

The approximate frequency of a received satellite radio signal
(ignoring relativistic effects) is given by
fr=fs (I
-;
$)
where fs is the frequency of the signal measured at the satellite,

c is the speed of light, and dS/dt is the range rate. The Doppler
shift frequency, fr - fs , is approximately proportional to the
range rate, the component of the satellites velocity vector
along the line of sight from the receiver. The maximum range
rate of a Transit satellite is about 6.3 km s- implying a
maximum Doppler shift when the satellite rises or sets of 21
ppm of the transmitted frequency. This corresponds to 8.4 kHz
at a frequency of 400 MHz.
The Doppler shifts may be measured by differencing the
received frequencies from constant reference frequencies in the
receiver. For most Transit receivers, these frequencies are 400
MHz and 150 MHz precisely. The satellite transmitter
frequencies are approximately 80 ppm lower than the receiver
reference frequencies in order that the frequency difference
does not go through zero. If the transmitter frequencies were
not offset, the receiver would have difficulty distinguishing
between positive and negative Doppler shifts.
Most Transit Doppler receivers count the number of
accumulated cycles of the Doppler frequency (actually, fo - fr)
rather than measure the instantaneous Doppler frequency itself,
since counting cycles can be carried out more precisely than
measuring the instantaneous frequency. The counter is read
out at intervals and the data stored. The counter is reset either
after each two minute paragraph or at the end of the pass.
Sequential differences in counter readings actually constitute a
series of biased range differences.
4.4 ARGOS
Another satellite system which uses the Doppler effect for
positioning is Argos [Service Argos, 19841, a cooperative
project of the French Centre National dEtudes Spatiales
(CNES), NASA, and the U.S. National Oceanic and
Atmospheric Administration (NOAA). However, in contrast to
the Transit system, the transmitters are operated by the users
and the receivers are in the satellites. An Argos transmitter on
an instrumented platform of some sort (oceanographic or
navigation buoy, radiosonde balloon, remote weather station,
etc.), periodically emits a 401.65 MHz signal carrying
information from the platforms sensors. One of two passing
U.S. TIROS/N-class weather satellites picks up this signal and
records its Doppler shift along with the sensor data. These data
are subsequently played back when the satellite is in range of
one of three tracking stations: Wallops Island, Virginia;
Gilmore Creek, Alaska; or Lannion, France. The tracking
stations relay their data to NOAAs National Environmental
Satelliteand Data
Service in Suitland,Maryland,
where they are sorted and then passed on to the Argas Data
Processing Centre at cNES in Tou~ouse,France, CNES
computes the position of the platform from the recorded
~~~~l~~
This information along with the Sensor data
can be conveyed to the operator of the platform by a variety of
including packet switching data networks, telex, or
letter. The two-dimensional (latitude and longitude) positions
can be as accurate as f 1 5 0 m (Io). The actual accuracy
Of the
Obtained depends to a very large degree On the
users transmitter. A new Argos service is being planned that
would orovide even more accurate oositions. UD to 5000
platforms requiring location service can be handled by the
Argos system, assuming these are uniformly distributed over
the earths surface.
The Argos receiving system was first implemented on the
prototype TIROS/N spacecraft, orbitted in 1978. Subsequent
TIROSlN satellites, NOAA-6 through NOAA- IO, have also
carried the Argos Data Collection System.
Position determination from signals uplinked to a satellite is
also utilized i n the COSPAS-SARSAT search and rescue
system [McPherson, 1981; Elliot and Exter, 19871.
5. SUMMARY
The electronic components of a satellite navigation receiver

are very amenable to high volumn - low cost production.
The impact of adding satellite navigation to an aerospace
vehicle in terms of size, weight and space is minimal.
Hence the major technical challenge in application of
satellite navigation is in the design to functionally and
physically integratae its manifold capabilities in the most
efficient manner.
176
5. FEDERAL RADIONAVIGATION PLAN
Langley, R. B. (1990). Why is the GPS signal so complex?

GPS World, MayIJune, Vol. I , No. 3, pp. 56-59.
Acknowledgements
Much of the information contained in this chapter is based on
material the author has written for GPS World magazine and
for the Guide to GPS Positioning published by Canadian GPS
Associates, Fredericton, Canada.
References
Ackroyd, N. and R. Lorimer (1990). Global Navigation, A
GPS User's Guide. Lloyd's of London, 202 pp.
Advisory Group for Aerospace Research and Development
(1988). The NAVSTAR GPS System. AGARD Lecture
Series 161. Neuilly-sur-Seine, France, September.
anon. (1967). GPS: NAVSTAR GPS Space Segment Navigation User Interface, Report ICD-GPS-200,
November.
anon. (1987). System Specifications for Navstar GPS Space
SegmentfNavigation Users Interface. Interface Control
Document - Global Positioning System - 200 (ICD-GPS200), 10 November.
ARlNC Research Corporation (1991). GPS Interface Control
Document. ICD-GPS-200,3 July, 115 pp.
Canadian GPS Associates (1993). GPS Biliography - July 93
edition.
Langley, R. B. (1991). The GPS receiver: An introduction.

GPS World, January, Vol. 2, No. 1, pp. 50-53.
Langley, R. B. (1991). The mathematics of GPS. GPS World,
JulylAugust, Vol. 2, NO. 7, pp. 45-50.
Langley, R. B. (1991). The orbits of GPS satellites. G P S
World, March, Vol. 2, No. 3, pp. 50-53.
Langley, R. B. (1991). Time, clocks, and GPS. GPS World,
November/December, Vol. 2, No. IO, pp. 38-42.
Langley, R. B. (1992). Basic geodesy for GPS. GPS World,
February, Vol. 3, No. 2, pp. 44-49.
Leick, A. (1990). GPS Satellite Surveying. John Wiley and
Sons, Inc., New York, Toronto, 352 pp.
Logsdon, T. (1992). Navstar Global Positioning System. Van
Nostrand Reinhold, 249 pp.
May, M. B. (1992). Measuring velocity using GPS. G P S
World,.September, Vol. 3, No. 8,
McDonald, K. D. (1991). GPS in civil aviation. GPS World,
September, Vol. 2, No. 8, pp. 52-59.
Seeber, G. (1989). Satellitengeodasie. Verlag de Gruyter.
Georgiadou, Y., and K.D. Doucet (1990). The issue of

selective availability. GPS World, September/October,
Vol. 1, NO. 5, pp. 53-56.
Hofmann-Wellenhof, B., H. Lichtenegger, and J.Collins
(1992). Global Positioning System, Theory and Practice.
Springer-Verlag, 326 pp.
Institute of Navigation, The (1980). Global Positioning
System. Vol. I. Collected reprints of papers published in
Navigation. The Institute of Navigation (U.S.), 246 pp.
System. Vol. 11. Collected reprints of papers published in
Navigation. The Institute of Navigation (U.S.), 257 pp.
System. Vol. 111. Collected reprints of papers published
in Navigation. The Institute of Navigation (US.),293 pp.
Kleusberg, A. (1990). Comparing GPS and GLONASS. GPS
World, November/December, Vol. I , No. 6, pp. 52-54.
Kleusberg, A. (I 992). Precise differential positioning and
surveying. GPS World, July/August, Vol. 3, No. 7, pp.
50-52.
Kleusberg, A., and R.B. Langley (1990). The limitations of
GPS. GPS World, March/April, Vol. 1, No. 2, pp. 50-52.
Klobuchar, J. A. (1991). Ionospheric effects on GPS. G P S
World, April, Vol. 2, No. 4, pp. 48-51.
Langley R.B. (1992). The Federal Radionavigation Plan. GPS
World, March, Vol. 3, No. 3, pp. 50-53.
Seeber, G. (1992). Satellite Geodesy.

Wells, D., and A. Kleusberg (1990). GPS: A multipurpose
system. GPS World, JanuaryFebruary, Vol. 1 , No. 1, pp.
60-63.
Wells, D.E., N. Beck, D. Delikaraoglou, A. Kleusberg, E.J.
Krakiwsky, G. Lachapelle, R.B. Langley, M. Nakiboglu,
K.P. Schwarz, J.M. Tranquilla, and P. VanEek (1987).
Guide to GPS Positioning. 2nd printing with corrections.
Canadian GPS Associates, Fredericton, N.B., 600 pp.
Yunck, T. P. (1992). A new chapter in precise orbit
determination. GPS World, October, Vol. 3, No. 9, pp.
56-6 1 .
177
AN OVERVIEW OF O m G A
RADIO NAVIGATION SYSTEM
by
D.F. LIANG
Directorate Research and Development Communications and Space
National Defence Headquarters
Ottawa, Ontario,
Canada K1A OK2
1.
SYSTEM DESCRIPTION
Omega is a very low frequency (VLF)

navigation system using frequencies
allocated for radio navigation in
the band between 10 KHz and 14 KHz.
Such a low frequency band enables
Omega navigation system to achieve
the long operating ranges required
for global coverage, at the same
time provides a stable and
predictable propagation environment.
This also makes it the only radio
navigation system applicable to
completely submerged submarines.
Omega is supported by 8 transmitting
stations located in Norway, Liberia,
Hawaii, North Dakota, La Reunion,
Argentina, Japan and Australia.
These stations are, in general
5000-6000 nm (nautical miles) apart.
They are designated alphabetically
with the letters A to H. These
transmitters are independent of each
other in the sense that they do not
operate in a master-slave
configuration. Each station
transmits three frequencies, (10.2,
11.33 and 1 3 . 6 kilohertz (KHz), time
sequenced within a basic repetition
rate of 10 seconds. A fourth
frequency at 11.05 KHz as well as
one discrete frequency for each
station, have been added to
facilitate station identification,
conventional navigation and also to
support special applications such as
buoy-tracking, and search and rescue

operations.
Omega is a hyperbolic position
fixing system, the hyperbolae are
loci o f constant time difference
between the arrival of signals from
two transmitting stations. The time
difference is measured as a
difference in phase of the two
received signals. Each hyperbola is
known as a line of position (LOP).
Signals must be received from at
least three stations with one o f
them serving as a common station, to
obtain a position fix from the two
LOPs. The hyperbolic mode is
attractive, since it removes the
need to use a precise and expensive
local oscillator. It is also
possible to work with two stations
using a circular mode, which
requires a precision local
oscillator. In such a case, the
position fix is obtained from the
intersection of circular rather than
hyperbolic LOPs.
Figure 1 shows two transmitting
stations A and B radiating phasesynchronized signals simultaneously.
The concentric circles around each
transmitter are used to represent
wavelengths at the transmitted
frequency. For a receiver located
at point X on the baseline joining
the two transmitters, the signals
are received from the two
178
transmitters in phase. A full cycle

difference occurs.everyhalf
wavelength in distance, since in
moving from point X to Y, the
receiver has moved by exactly a half
wavelength towards B. The area
between two adjacent zero phase
difference LOPs is called a lane.
The phase measurements are repeated
through 360" cycle. Since the lane
is subdivided into 100 centilanes,
each centilane (cel) is then
equivalent to 3.6" of phase
difference. At 10.2 KHz, the lane
width, which is at half a wavelength
is approximately 8 nautical miles.
2.
LANE AMBIGUITY
Since Omega stations are far

apart, on the average there are
600-700 lanes between any of them.
To reduce ambiguity, each lane is
numbered for its identification.
The lane number must be initialized
to an accuracy of half a lane and
the lane count must be properly
maintained. The lane counting can
be manually maintained or performed
automatically in most modern Omega
receivers. When the Omega receiver
is unable to keep the correct lane
count, this is referred as a lane
slip or lane jump, which can be in
the multiples of its lane width.
Single frequency receivers use the
10.2 KHz signal which has a lane
width of about 8 nm on the baseline
between stations. Because of the
lane ambiguity, the receiver must be
preset to a known location at the
start of a voyage. The accuracy of
that position must be known to
sufficient accuracy to be within the
lane that the receiver is capable of
appropriate lane identification.
This has to be achieved to within a
mere 8 nm for a single frequency
receiver at 10.2 KHz. To
effectively use a single frequency
receiver, it is essential to
carefully maintain a DR plot, and
count the number of lanes it crosses
in the course of a voyage. It must
be periodically compared to its
Omega position s o that any lane
ambiguities can be detected and
corrected. Unless the Omega
position is occasionally compared to
the position fix of another
navigation system or to a carefully
maintained DR plot, the possibility
of lane count error increases with
time and distance. However, two or
three - frequency receivers can be
used to reduce the position fix
ambiguity. Navigational frequencies
of 13.6 KHz and 10.2 KHz have a
frequency difference of 3.4 KHz, the
phase of which can be measured by
some receivers. 3.4 KHz LOPs
coincide with every third 10.2 KHz
one, which has exactly a lane width
of 8 nm to give a lane width of
24 nm. With this, the operator only
need to know the platform location
to within an accuracy of 12 nm along
the baseline without any ambiguity.
Having established the lane count,
it should then be possible to
determine its centilanes without
ambiguity.
The lane resolution process can be
further extended by using even lower
frequency differences. The
frequency difference between
11.33 KHz and 10.2 KHz is 1.13 KHz,
which can provide a lane width of
72 nm. The frequency difference
between 11.33 KHz and 11.05 M z can
be used to generate LOPs with a lane
width of 288 nm, which is 36 times
the normal lanes width of 10.2 KHz.
The use of the latter frequency
difference is aimed primarily at air
navigation where the high speed of
the platform makes the use of wider
lane width imperative.
179
3.
PROPAGATIONAL PERFORMANCE
The Omega position fixing depends on

the differences in phase
measurements of computed LOPS. The
signals are propagated, in effect,
by a waveguide, with the earth's
surface and D layer of the
ionosphere as the boundaries.
Therefore, ionospheric conditions
greatly affect Omega signal
propagation. A key factor is the
variation of ionospheric height
which changes from seasons to
seasons and from day to night. The
major error sources then are due to
the inability to precisely determine
atmospheric phenomena and signal
conductivity. Meteorological
effects, however, are known to
introduce negligible effects to VLF
propagation.
3.1
Diurnal Variations
Fundamental to any VLF propagation

is the propagation velocity, which
can be used to relate a phase or
time difference to the geography of
the hyperbolic position lines. The
height of the ionosphere changes
from approximately 70 Km by day to
90 Km by night, such a variation is
a considerable change in the
dimension of the waveguide, which
affects the signal propagation. The
resulting variation can be of the
order of one whole lane.
Fortunately, such diurnal changes
are cyclic, and therefore
predictable over a 2 4 hour period
using skywave corrections. For an
all-daylight propagation path, the
correction can be accurate to five
centicycles, whereas for an allnight-time path, the accuracy can be
up to ten centicycles. The
necessary corrections are presented
in US Defence Mapping Agency Omega
Propagation Tables. They take into
account changes in propagation
velocity due to ionospheric height

variations, earth magnetic field
orientation, and to surface
conductivity. In general, the
navigational accuracy of the Omega
system depends on the quality of
these published corrections.
However, there are other
unpredictable disturbances that can
affect the propagation conditions.
3.2
Modal Interference
Omega transmissions are

launched from vertically polarized
antennas which generate transversemagnetic (TM) propagation modes
within the waveguide. At a distance
of 1,000 Km from the transmitters,
only the first and second modes,
TM 1 and TM 2 (see Figure 2 ) are
large enough to warrant
consideration. Close to the
transmitters, the TM 2 mode
predominates, but its attenuation
rate is greater than that of the
TM 1 mode. The TM 1 and TM 2 modes
propagate with different phase
velocities, and therefore in areas
where their amplitudes are
comparable, the two modes will
produce interferences that can
advance or retard the phase
propagation. It is generally unwise
to use Omega within 2 0 0 nm of any
Omega transmitters.
Modal interference from the Liberia
transmitter has also been severely
experienced as far away as the east
coast of the US during night time.
It can cause large errors,
especially when each mode
predominates separate day - night
time paths.
Some receivers have the software
algorithms to prevent near-field
modal interference for both day and
night-time operations and
automatically disable the selection
180
of stations whose ranges are less

than a preset minimum range.
3.3
Solar Flare Effects
Solar Flare Effects are known to

have major impacts on Omega
propagation, especially Sudden
Ionospheric Disturbances (SID) and
Polar-Cap Absorption (PAC) event.
3.3.1 Sudden Ionospheric
Disturbances (SID)
Sudden ionospheric disturbances are
primarily caused by an increase in
the energy of x-ray flux emancipated
by the sun due to solar eruptions or
flares .
'I
Increased energetic x-ray radiation

in the D-layer, can reduce the
waveguide height affecting the VLF
signal propagation. This causes a
phase advance on propagation paths
that are exposed to the sun. These
phase changes are known as sudden
phase anomalies (SPA).
Such a disturbance can happen
suddenly peaking within 5-6 minutes
followed by a slow decay of over a
period of 50 minutes. A typical
phase shift is about 10-20
centicycles and occurs mostly in
lower latitude regions. In the year
of sunspot activity, the SID may
occur as much as once per day.
3.3.2 Polar Cap Absorptions (PCA)
This is due to the precipitation of
solar particles, primarily protons,
in the vicinity of the earth's
magnetic poles shortly following
significant solar flares. It causes
an increase in ionization of the Dregion, resulting in an effective
lowering of the D-layer ionosphere.
The amount of propagation anomaly
depends on the extent of the
transmission path crossing the

magnetic pole region. Therefore, a
propagation path away from the polar
region will not be affected by PCA.
For users in high-latitude regions,
it may be difficult to completely
ignore signal paths over polar
routes, especially when some
transmitters are off the air. The
PCA effect does not occur as
frequent as the SID. The effect of
PCA tends to come on over several
hours and slowly decay over a period
of 1-2 weeks. The maximum effect
can vary from 10-50 centicycles.
4.
PHASE PROPAGATION CORRECTIONS
A most effective way of accounting

for Omega propagation errors is the
application of published Phase
Propagation Corrections (PPC). The
1980 PPC Model [l] is widely
employed through the Omega user
community. The application of
propagation corrections reduces the
positioning errors to the 2 - 4 nm
range. Figure 3 presents the error
of an Omega LOP, over a 4 day period
in November 1987, which shows
significant repeatability of the
phase error as a function of the
time of the day. The PPCs initially
was intended to be updated every
5 years, which unfortunately has not
been fulfilled.
Recent examination of the PPC [2]
has led to the conclusion that the
1980 Model contains a term which
grows exponentially with time. The
retention of the time term into the
1990s should be a cause for concern
For this reason, the US Coast Guard
has eliminated it from its
"Official" PPC Model.
5.
OEMGA ERROR MODEL
A significant part of Omega phase

errors can be compensated by using
181
Phase Propagation Correction (PPC)

tables. The residual error can be
represented by the following four
major random components:
I
a. A slowly varying component

representative of the
spatially correlated phase
error due to slowly
varying factors such as
ground conductivity;
b.
A 24-hour periodic diurnal
component used t o
represent errors in phase
propagation corrections;
c. A short time constant
Gauss-Markov process phase
error ;
d. A white noise representing
receiver dependent
measurement noise.
i
I
:
I
,1
1
!
I
1
I
I
The auto correlation function

of the Omega phase errors are
represented in Section 4 . 8 of [3].
A four state error equation t o model
each Omega phase error is also
represented in the same reference.
Since an Omega line of position
error is the difference between two
independent phase errors, the
statistical properties of the LOP
error will be basically the same as
that of a phase error. For the sake
of simplification, a two state
representation can be used as a
substitute for the four state model.
This was accomplished by omitting
the periodic states and adjusting
the parameters of the weakly
correlated Markov processes to
maintain the correct mean values and
correlation time. The phase error
of each Omega signal is therefore
modelled by the following two Markov
processes:
Where T1 and T3 are correlation time

constants.
6.
DIFFERENTIAL OMEGA
Differential Omega is a method of

providing increased relative
position fixing accuracy with
respect to a local reference
station. Normal Omega system is
subject to errors in diurnal
corrections and propagation effects,
therefore limiting the accuracy of
the system. Even though Omega
transmits VLF signals at long
distance, its propagation errors are
highly correlated over a small
geographic region. Therefore, a
reference receiver can be
established at a precisely known
location t o measure the LOP errors
for compensation to users in the
neighbourhood of about 200 nm.
Studies have shown that the position
accuracy can be reduced to
200 meters at a few miles of the
reference stations, and up to 400600 metres at 100 nm for a full
power operational system [4].
However, at a distance of 200300 nm, the accuracy deteriorates to
that of basic Omega performance.
The Differential Omega has not been
widely adopted because the
corrections can only be applied over
a relatively small area, and that
reference stations must be installed
and maintained together with the
associated communication links to
transmit the corrections. In spite
of this, there are currently 22
Differential Omega reference
stations installed around the world,
182
and some 2000 receivers have been in

service worldwide.
The policy statement of 1 9 8 8 Federal

Radionavigation Plan (FRP) had
generated considerable concerns
within Omega User community that the
Omega was headed for oblivion by the
end of the century. The FRP
reported that the normal transition
period of 10-15 years might be
shortened.
In the 1 9 9 0 FRP, even though the DOD
I
I
requirement did not change from the

1 9 8 8 version, it states that "Omega
aviation service requirements will
remain until there is a suitable
replacement". Some in the aviation
community can argue that there will
never be a suitable substitute for
Omega in low cost global coverage.
The recognition of aviation service
requirement is significant in light
of DOD's termination o f its
requirement beyond 1 9 9 4 . Without
such a substantiated aviation
requirement, the need for Omega
beyond the end of this century
becomes uncertain. With the
widespread use of Omega in aviation,
Omega can be expected to serve as a
global supplementary aid to GPS and
I83
References :
[l]
E.R. Swanson and P.B. Morris, "New Coefficients for the Swanson Propagation
Correction Model," Proceedings of the Fifth Annual Meeting, International Omega
Association, Bergen, Norway, August 1980, pp. 26-1 to 26-4.
[2]
R.J. Wenzel etc, "The 1980 Omega PPC Model Time Term Should Not Be used,"
Proceedings of the Fifteen Annual Meeting, International Omega Association, Bali,
Indonesia, September 1990, pp. 10-1 to 10-10.
[3]
D.F. Liang "An Overview of Generic Multi-sensor Integrated Navigation System

Design," This ACARDOGRAPH.
[4]
Nard, "Results of Recent Experiments With Differential Omega," Navigation Journal of

the Institute of Navigation, Vol. 19, No. 2, Summer 1972.
184
Figure 1.
Simplified Omega Hyperbolic Lattice.
185
Ionosphere D-layer
Figure 2.
TM Modes Propagation.
186
rn
14
14
c,
.rl
rn
k
0
k
k
aJ
..
m
aJ
k
(D
cv
++
187
AGARDograph on Advanced Astroinertial Navigation Systems

Seymour (Sy) Levine
Northrop Corporation
Electronics Systems Division
2301 West 120th Street
P.O. Box 5032
Hawthorne, CA 90251-5032
USA
ABSTRACT
This paper discusses modem Strapdown Astroinertial Navigation
( S A I N ) systems as autonomous navigators for manned aircraft,
ships, missiles, and remotely piloted vehicles. These systems,
which approach Global Positioning System (GPS) accuracy,do not
depend upon man-made electromagnetic radiating devices that
may be intentionally shut down,destroyed, or become unreliable
in ahostileenvironment. Thepaper analyzesthe gyroscopic accuracy, axtificial stellarimage stabilization, star density, sky visibility,
and sky background irradiance effects on system performance. It
concludes that a high-precision, reliable, low-cost stellar inertial
system can be achieved by eliminating gimbals and combining a
strapdown Inertial Navigation System ( I N S ) with an Optical WideAngle-Lens Staxtracker (OWLS).
1.0 INTRODUCTION
gimballed astminertial navigators were successfully used as the

navigation reference for reconnaissance, precision rendezvous,
and indirect. coordinated firing of munitions.
Figure 1shows the performance of pure, unaided inertial systems,
stellar inertial systems, and GPSs, starting in the year 1950 and
projected to the beginning of the 21st centuxy. As the curves
indicate, the e m r of a terrestrial inertial system operating for
10 hours is measured in miles whereas the stellar system error is
measured in hundreds of feet. Figure 1 further shows that the
strapdown system position accuracy is rapidly approaching the
precision of the gimballed system. In addition, the strapdown
systems attitude accuracy (e.g., roll, pitch, and azimuth) is now
superior to its gimballed predecessor. This strapdown attitude
superiority is primarily due to the attitude being a
computer-generated n w b e r as compared to an inaccurate gimbal
angular readout such as from a resolver or optical shaft encoder.
For several hundred years, man successfully used star positions to

reset his long-range navigation systems. During his earthly
These strapdown INSs not only provide position, and the Euler
journeys, the stellar augmentation subsystem-the sextant-provided
angles of roll, pitch, and heading, but they also supply the timely,
updates to minimize the accumulated errors in his continuous Dead
accurate angular rates required for the vehicles flight control and
Reckoning (DR) navigation. As he approached his destination or
autopilot. The angular rate data, flightcritical to dynamically
wartime target. he resorted to a Relative Navigation (RN)solution.
unstable vehicle control systems, is relatively simple to obtain in a
His eyes took on the function of a terminal seeker. They focused on
strapdown INS as compared to the gimballed INS process. which
the terminus of his mission (e.g., dock, etc.) or on the position of his
required the costly and noisy differentiation of the gimbal resolver
enemy. By the nineteenth century, man had an autonomous
readouts.
precision-navigation suite that utilized DR, stellar updating, and
terminal seeking to achieve a Circular Error Probable (W)
In 1990, w o k proceeded on the development of a Strapdown
measured in feet.
Astroinertial Navigation ( S A N ) system capable of producing the
same autonomous high-precision navigation that exists on a
Although present autonomous navigation suitesbear little physical
gimballed stellar inertial system.
resemblance totheirpredecessors, the system elements are the same
(see Table 1). They consist of a DR continuous navigator, i.e.. an
2.0 SYSTEM PERFORMANCE
Inertial Navigation System (INS),periodically updated or reset by
Figure 2 presents the canonical model of a stellar INS;Figure 3
a DR reference system such as a stellar tracker and altimeter. After
shows asimplexmodel. The strapdowninertialsystem,utilizingits
the navigation system guides the vehicle into an operational area or
gyroscopes and accelerometers,provides estimates of thevectorsof
basket of the terminal seeker, the seeker provides precise closure
position (e.g.. latitude, longitude, and altitude); velocity (e.g.,
for docking, landing, or interdiction of fixed and maneuvering
north, east, and vertical); and Euler angles (roll, pitch, and
targets. This autonomous suite is free from radio aids, which may
heading).
be unreliable, jammable, or unavailable during a twenty-first
century wartime encounter.
The raw inertial vectors are combined with a self-contained
ephemeris and chronometer (an error of 1 s of time in the chroSince the 1950s, inertial systems have been the dominant DR
nometer represents a 1,500 ft position error) to predict a star fix
navigators aboard ships and air vehicles. During the 1950s. these
angle. The astrotxackerthenuses thisestimatedcoordinate location
devices were made with gyroscopes and accelerometers mounted
to find the actual star and to provide an error vector representing the
on one gimbal and surrounded by two or three other gimbals to
difference between the predicted stars angular coordinates and the
isolate the inertial sensors from the vehicles angular rates. The
angular coordinates of the astrotrackers measured star. A single
inertial element assembly was usually called the stable member, or
star observation does not provide enough information to
yaw gimbal. To provide high precision, a Narrow Field of View
compensate three gyroscopic orthogonal axes of rotation. This is
(NFOV) startracker, embedded in train and elevation pointing
because the single stellar observation is unable to sense rotations
gimbals, was mounted on top of the yaw gimbal. In this
around its line of sight. Thus, to measure all three gyroscopic axes,
c o n f i p t i o n , there were usually five gimbals, multiple gimbal
a second noncoincident observation, preferably orthogonal to the
resolver angular readouts, a servo-mechanism, and associated
first, must be made. Additional measurements are used to further
electronics. A limited quantity of twmgimbal versions, the
characterize and refine the error vector. This error vector is then
minimumnumber of gimbals necessary to point a telescope at any
processed optimally in a Kalman filter to yield the best estimates
point in three-dimensional space, were also developed. These
188
TABLE 1. AUTONOMOUS NAVIGATION SUITE CHRONOLOGY

NAVIGATION COMPONENTS
1940
lAGNmC COMPASS-FLUX
1950
1960
1970
VALVE
GIMBALLEDINERTIAL
NAVIGATION
GYROCOMPASS-&b
DEAD RECKONING
1980
19pO
I
I DOWN
DOPPLER RADAR
4CREMENTAL CLOCK
b
b
HRONOMmA
RUBIDIUMTIME STANDARD
b
GIMBALLED
STRAPDOWN
NARROW FOV
WIDE F O V F
STAR TRACKER
STELLAR TRACKER
"r
1
I
VlSUAU
I
I
. M L
AIDS
LOW LIGHT LEVELN

IRTHERMAL
TERMINAL SEEKER
MlLLlMfER
ACOUSTICAL(AIR)
PROCESSOR
9uI.121.I 7
UNAIDED STRAPDOWN
lNERTlAL,NAVlGATlON
100.ooo
10.ooo
bl
N
I ERTA
I TA
I
NAVIGATION
bPlTOTTUBVEM LOG
00
2000
,UNAIDED INERTIAL
NAVIGATION SYSTEM
'\
STRAPDOWN
'
PROJECTION OF NAVIGATION
SYSTEM PERFORMANCE FOR
GPS, ASTROINERTIAL, AND
UNAIDED INERTlAL
NAVIGATION SYSTEMS
FOR l(1HOUR FLIGHT
'\
'\
STELLAW
'. ',
1,000
2%
POSITIONING SYSTEM (GPS)
- --- -
100
4 HOUR COVERAGE
,o
-
43D24HOUR
COVERAGE
10
1950
1960
1970
1900
1990
2000
I NOTE: THE STRAPDOWN SYSTEM
ERROR PROJECTION ASSUMES A

MILITARY VEHICLE WITH A MIX OF
BENIGN AND SERPENTINE
MOVEMENTS. FOR BENIGN ONLY
THE STRAPDOWN APPROACHES
THE GIMBAL SYSTEM ERROR.
2010
Figure 1. Startrackers Enhance Inertial Navigator Performance
92M-12141
I89
POSITION CORRECTIONS
VELOCllY
CORRECTIONS
2
A
2
GW
GYRO DRIFT AND

SCALE FACTOR
CORRECTIONS
~LWT
ROLL
f = SPECIFIC FORCES
W =RATES
LT = BODY TO LEVEL TRANSFORMATION
LT E = EARTH TO LEVEL TRANSFORMATION
WE = EARTH RATE
Wg = BODY RATE
Wc = CRAFT RATE
G<>= GYROSCOPEFRAME
V =VELOCITY
B<>= BODY FRAME
f g = GRAVITY
L<>= LEVEL FRAME
WT =TOTAL INERTIAL RATE
R =EARTHRADIUS
A< = ACCELEROMETER FRAME
OZM.12191
Figure 2. Strapdown Stellar Inertial System, Canonical Model
STRAPDOWN INERTIAL SENSORS

INERTIAL SENSOR PROCESSOR
GYROSCOPES
* DIRECTION COSINES
ACCELEROMETERS
*CONING COMP. ETC.
ALTIMETER
CHRONOMETER
-4
INTEGRATION PROCESSOR
II
KALMAN FILTERS
EPHEMERIS, ETC.
STRAPDOWN IMAGE SENSORS
OPTIMUM
ESTIMATED
NAVIGATION
SYSTEM OUTPUT
POSITION
VELOCITY
EULER ANGLES
ANGULAR RATES
--------
IMAGE SENSOR PROCESSOR
STELLAR
TERRESTRIAL
HORIZON
CORRELATOR
9
ENHANCERS. ETC.
Figure 3. Simplex Strapdown Stellar Inertial System Model with Terrestrial and Horizon Image Options
of the vectors of position, velocity, and Euler angles, as well as
gyroscopic errors, such as bias, scale factor, and misalignment, and
observable accelerometer errors.
After the inertial stellar observations are made, every additional
measurement continues to minimize the error vectors.
It should be pointed out that the astrotracker is similar to a marine

sextant in that it looks at the star angle with respect to the horizontal
level, but, unlike the sextants observed horizon, the astroinertial
system uses the inertial systemsestimatedvertical. This deviation
leads t o m inability to observe certain emrs. Figure 4 is a heuristic
error model of an astminertial system. As can be seen. the
I90
Note: Laplaclan Operator 1R la umdfor an lntegrntor.
Where:
I = Acceleratlon of Gravity
6 A = Acceleratlon Error
6 v = Velocity Error
6 = Posltlon Error
6 9 =Tilt Error
6 w = Angular Rate Error
6 (0 E = Error In Earth Rate Compensation

6 g A = Error In Gravity Horizontal Component
6 s = Accelerometer
~ ~
Scale Factor
6 s =~
Gyroscope
~
Scale Factor Error
6 EL = Telescope Angular Error
9.?M.l21&
Figure 4. Heuristic Astroinertial Error Model (the astroinertial system cannot compensate for some acceleration errors)
startracker cannot distinguish between a tilt in the inertial system

andanerrorinindicatedpositionwhenbothareequalandopposite.
Thus, the tracker is unable to observe, and therefore cannot aid in
damping, the INSS84 min Schuler oscillations. Uncompensated
gravity horizontal deflections have correlated tilt and position
errors. Since this correlation produces a position error opposite to
the tilt errors, the astrotracker is unable to sense (and therefore
compensate for) these differences. Also for uncompensated
gravity horizontal deflections, the astroinertial systems exhibit the
identical error response as do the pure inertial navigators. The
errors can be minimized by accurate gravity compensation that
takes into account spherical harmonics, terrain data, and geological
information. Horizon sensors and velocity augmentation reference
systems can be. used to minimize these errors, as shown in Figure 5.
This figure shows a simulation of a strapdown astroinertial
navigator aided by a velocity reference system after an 18 h flight.
The velocity reference augmentation is used to damp the
navigators Schuler oscillations and thereby minimize the vertical
error. When the vertical oscillations are damped, the stellar
sightingsareabletoobserveandminimizepositionerrors.Theless
than 200 foot simulated errors shown in Figure 5 were ma& using
a 1.5 =sec astrotracker. When the simulations are run with a
0.5 arcsec stellar sensor, the Circular Error Probability (CEP),is
only 60 ft.
Gyroscopic drift terms, on the other hand, are easily distinguished,
since the one integration response to tilt is much faster than the
three integration processes required for the error to go into position.
Figure 6 is an example of how stellar augmentation reduces the
effect of gyroscopic drift on navigation errors. It illustrates the
performances of four different Ring Laser Gyroscopes (RLGs)in a

Strategic Penetrator scenario. Freeinertial system navigation errors
are compared with a stellar augmentation. As can be seen, the
stellar augmentation reducesthe errors significantly. By isolating a
40 cm RLG (i.e., the most accurate gyroscope at 0.00025 d e d h ) ,
a graphical high-resolution comparison can be made both with and
without stellar augmentation. Figure 7 shows that the error after
8 h of flight with the 40 cm RLGs is 9,OOO ft unaided and only
600 ft with stellar augmentation (a 15 to 1 improvement factor).
As the flight extends past 8 h, the improvement factor increases

further. Figure 8 compares 10, 20.30, and 40 cm RLG stellar
inertial performance. The stellar navigation performance just
begins to degrade with 10cm RLGs. It should be noted that the 40
cm gyroscopic random walk angular drift rate is 16 times lower
than that of a 10 cm RLG or Resonant Fiber Optic
Gyroscope/TnterferometricFiber Optic Gyroscope (RFOG/IFOG);
yet even with this significant disparity in gyroscopic drift
performance, astroinertial system accuracy isessentially the same.
The astrotracker compensates for gyroscopic drift by monitoring
the navigators tilt and azimuth angles and by providing tilt,
azimuth, and gyroscopic driftcorrections. These curves illustrate
one of the values of astroinertial systems: they minimize the
systems performance dependence on ultra-precise, large, costly
gyroscopes. It should be pointed out that the strapdown
astroinertial simulation responses have higher navigation errors
that the present class of high-precision gimballed astroinertial
instruments. This difference is expected to disappear over the next
three years as the strapdown acceleration inertial sensors (i.e.,
accelerometers) improve.
191
2500
2000
1500
E
AUGMENT WITH VELOCITY
REFERENCE AT T = 18 H
1000
500
0
16
20
18
22
24
26
HOURS
92M12l.CX
Figure 5. Simulated Results: Strategic Penetrator Position CEP of Inertial Navigator with Low-Rate Stellar
and Velocity Reference System Augmentation
STELLAR AND FREE INERTIAL WITH ALTIMETER AUGMENTATION

STRATEGIC PENETRATOR POSITION CEP SUMMARY (FEET)
+13.2-80000Fl'
FREE INERTIAL
9.9
10 CM RLGs
OR
RFOWIFOG
- 60000 FT
6.6 40000 FT
20 CM RLGs
30 CM RLGs
CM RLGs-
-40
3.3 20000 FT
STELLAR INERTIAL
0 - 0 FT
OSEC
20000SEC
40000SEC
60000SEC
HRS
5.5 HRS
11.1 HRS
16.6 HRS
-<1000FT
80000SEC
22.1 HRS
82M-121-06
Figure 6. Simulation Results: Astrotracker Inertial Augmentation Essentially Eliminated the RLG as a Dominant Error
Source (CEP vs Mission Time)
1.92
14000
INERTIALWITH ALTIMETER
AUGMENTATION
12000
10000
--
6000
z
P
8
nl
6000
4000
INERTIALWITH ALTIMETER
AND STELLARAUGMENTATION
2000
1000
500
0
0
10
12
14
16
MISSIONTIME IN HOURS
02M.12147
Figure 7. Strapdown Stellar Inertial System Performance--Strategic Penetrator Mission with 40 cm RLGs
(CEP vs Mission Time)
n n
30 AND 40 CM
12
16
MISSION TIME (HOURS)
Figure 8. Strategic Penetrator Strapdown Stellar Inertial System Performance: 10,20,30, and 40 cm RLGS or
RFOGAFOG (CEP vs Mission Time)
Another major attribute of astroinertial systems is the ability to

observe north directly. In ineltial systems, the initial azimuth
determination processis cormpted by the effective east gyroscopic
drift. Furthermore, the azimuth error vector can be changed by
effective azimuth gposcopic drift. Startracken can observe and
correct these e m r vectors. This endows ashinertial systems with
excellent response times for accurate performance, since they can
correct these quantities in the sky. Stellar inertial systems also

provide a high level of self-monitoring and self-healing. For
example, apureinertial system cannot observe an initial or updated
longitude insertion error during flight, but a stellar inertial system
can detect this potentially catastrophic problem. This is because
the pure inertial navigators dynamic response is transparent to
longitude data,while the stellarinertial star ephemeris depends on
193
longitude. When the star detection probability is high, the stars

should be in the Field of View (FOV). If the star is not in the FOV,
the flight computer could trigger either a selfcorrective action
response or a mission safety abort mechanism. The choice of
action can be programmed into the flight computer and depends on
subsequent self-tests.
Thus, an astrotracker provides an excellent augmentation device
for the inertialnavigator. Furthermore, by the addition of avelocity
reference (see Figure 5) or horizon sensor, S A I N can approach
GPShertial performance (60 vs 45 ft) without the use of or
dependence on man-made radiation devices. If GPS were
available, it wouldbe integrated into the navigation suite as another
This augmented
inertial navigator augmentation device.
navigation suite provides a level of navigation position and
velocity redundancy. Nevertheless, the missions performance
goals need not be jeopardized by a dependence on satellite data
availability. Therefore, the astroinertial system yields the
high-performance autonomous mission capability necessary for
the 21st century.
3.0 STAR IMAGE CONSIDERATIONS

Image stabilization is a major problem with solid-state
astrotrackers. Figure 9 shows how a vehicles motion causes a
stars image to traverse a fixed-to-vehicle F o c a l - P l a n e h y P A ) .
Because of the vehicles motion, ahigh-speed snapshot of the stars
image must be taken to prevent blurring.
Aftertheimage snapshotistaken,thedataintheimagepmcessoris
transformed into stabilized coordinates with the aid of the inertial
navigators direction cosine matrices. The image data integration
for signal-to-noise enhancement, as shown in Figure 10,is a
complex process in which the shutter opening is operated at 100to
1,OOOps to prevent star image bluning. Thishigh-speed shuttering
of data is necessary even for vehicle attitude rates as low as 1 de&.
4.0 STELLAR TRACKER PERFORMANCE

PARAMETERS
The driving factor in the design of shipboard and airborne stellar
trackers that augment inertial systems is their capability to
accurately measure star positions in daylight at sea level. For
simulated Hyper Velocity Vehicle (HVV) missions, the
Signal-to-Noise Ratio (SNR)improves by afactor of 2.5 (one star
magnitude) as the vehicle ascends from sea level to a 20,000 ft
altitude. This is primarily due to the effects of the atmosphere. As
the air density decreases, the sky W e n s because of a decrease in
light scattering. Figure 11 shows the sky background as afunction
of wavelength for sealevel and at a 100,OOOftaltitude. The spectral
radiance is a hundred times brighter at sea level than at the 100,OOO
foot altitude. In addition, atmospheric molecular absorption
degradation factors decrease at higher altitudes. Figure 12 shows
the optical atmospheric transmittance as a function of wavelength
and zenith angle. The atmospheric transmission is attenuated at
wavelengths where there is 02.03. and H20 molecular absorption.
For an astrotracker that utilizes an FPA as its photon detector, the
S N R can be approximated by the following expression:
VEHICLE:
ROLL R1
PITCH P1
STAR PATH ACROSS THE FPA
Figure 9. Image Trajectory Across an Unstabilized FPA Star Detector
194
VEHICLE PIANE,
/*
INERTIALLY STABILIZED
FPA IMAGE
TRANSFORM
TO INERTIAL
/*
I
I
I
I
I
I
I
-------1
Figure 10. Strapdown FPA Charge Enhancement

n
x 1.6-
-1.6
- 1.4
5.
T
L
- 1.2 4
U)
- 1.0 E
v)
0.2-
0
J
Wavelength (pm)
0
r
9261-121.11
Figure 11. A Typical Sky Background at Sea Level and 100,000 ft (Sun Zenith Angle 53 deg, Observer Angle 58
Sun-to-Observer Azimuth 90 deg)SEYRAY Electro-OpticalSystem Analysis, page 139, Pub.
"EO RESEARCH CO."(AFAL Report of NEMD)
195
100
90
80
70
YE r n
6o
540
30
20
10
0
0.3
0.5
0.4
0.6
0.7
0.8
1.1
1.0
0.9
1.2
1.3
Wavelength (pm)
9W-121-12
Figure 12. Atmospheric Tranmission for Various Air Masses or, Equivalently, From
the Ground to Space at Various Angles from the Normal; One Air Mass
Corresponds to the Amount of Air Through Which a Beam Must
Pass When Going Straight Up to Space (Courtesy RCA)
Signal-to-Noise Ratio (SNR)
QW = J
where:
signalcm cmes verpixel

Electronic charge
Stellar spectral radiance
Wavelength bandwidth of telescope system
Atmosphere transmission
Aperture area
Telescope transmission
Quantum efficiency
Integration time of a frame (stellar snapshot)
Number of frame snapshots utilized
Number of pixels containing the star image
(star blur factor)
QSP
4%
.AA
AT
A,
TT
rl
4
Nf
NS
q
QBP
q5B
* A A * A , TT * FOV. t l * IT * Nf
N * 0,
where:
QBP
4B
FOV
N
0,
where:
=
=
=
=
=
QDP
ni
=
=
VE
Backeround uhoto charees uer pixel

Sky background spectral radiance
Fieldofview
Total number of pixels in sensor array
Sky background attenuation of optical fdter
tector da& current charcres

- ~e
Detector intrinsic carrier concentration
Detector da& current charge generation time
Detector charge generation bucket volume
(QSP
+ QSP + QDP)
QES
where:
QNP
Star signal chaxges -per uixel
&p
QDp
=
=
=
Qas
Star photo charges per pixel

Background photo charges per pixel
Dsuk detector charges per pixel
ElectroNc system bandwidth noise coefficient
&p
SNR =
QSP
/,
(Qsp
+ Qsp + QDP)
QES
For sea level daytime tracking in a non-nuclear

event environment, the S N R equation simplifies to:
QBP
> > QSP + QOP
- Je,.e,
QSP
sNR
196
Unfortunately, the trackers SMpmvidesonly apart of the insight

necessary for the design of a stellar inertial system. Knowledge of
the detected stellar density is also of major importance. For
example, the number of stars brighter than a given magnitude, M,
is a function of the spectral contents of the stars. Stars radiate in a
similar fashion to a blackbody at different maximum temperatures
based on their age andcomposition. Table 2 is a comparison of the
stellar densities for the R band (red) stars and that of B band (blue)
space or satellite startracker with only a 0.72 in? aperture and a 1

ms exposure time has the same capability as the much larger
sea-level tracker with a 36 in.* a p e ~utilizing
e
a l00ms exposure
time. Figure 15 shows a system size comparison for both
gimballed and strapdown astroinertial navigators that have the
same daytime sea-level tracking capability. The dramatic disparity
shown in Figures 14 and 15 illustrates how the tracker technology
and application dictate the physical startracker design.
Stars.
A few simple terms dominate the astrotracker design tliou&&t

process. The S M is proportional to the square root of the trackers
aperture and the star photon integration time (stellar exposure
time). The reason for only a square-mot improvement in the S N R
as a function of aperture is that the expansion of the apenot
only increases the star signal photon collection but also has an
associated increase in the collection of sky background noise flux
Therefore, a6.25 timesincreaseinapertureareais~quiredinorder
to observe a single magnitude increase (i.e., 2.5) in dimmer stars.
Similarly, the 6.25 increase factor holds for the stellar exposure
time, since extension of the integration increases the collection of
both the star photons and the sky background noise flux. In the
opposite direction, a wide FOV detracts from the stellar
observation by allowing more sky brightness background noise
flux into the optical system. For example, when the FOV is very
small, as when looking at the sky from the bottom of a deep mine
shaft (illustrated in Figure 13), the unaided eye can observe stars in
the daytime. The sky brightness noise increases as the inverse of
the square root of the FOV. In order to achieve satisfactory
performance, abalanceisstruckbetween thebenefit of awideFOV
increasing the number of bright stars observed in the tracker at a
given time vs. the deteriorative effects of the sky brightness noise.
Figure 14 summarizes some of these considerations for a silicon
FPA and a 3 deg FOV Optical Wide-Angle Lens Startracker
(OWLS). Itis aplot of starmagnitudecapability versusaltitude. A
Figure 16 shows the probability of seeing the sky from sea level.
Above 13.7 lun (45,000 ft) in altitude, the probability of having an
unobstructed view of the starsis essentially 100percent. Although
at an altitude of 13.7 km. an unaided observers view of the sky is
unobstructed by cloudcover, the sky background light will prevent
observing the starsin daylight. Yet at night, the unaided observers
view of the stars from that altitude is spectacular. Above 13.7 km,
the daytime sky background, not in the direct vicinity of the sun,
grows pmpssively darker with increases in altitude until it
essentially turns black and is indistinguishable from the nighttime
sky. Thus, at altitudes above 60miles, a startracker the size of an
eyeball provides sufficient signal to reset an inertial guidance
platform.
The quality orpureinertialperformanceof the gyroscopes needs to
be high when operating for sustained periods under cloud cover.
Under these conditions the gyroscopes may only get stellarupdates
a few times an hour. On the other hand, when operatingabove 13.7
km with a high-speed startracker, the system navigation
performance will be outstanding even with low cost, modest
performance gyroscopes. Furthermore. when operating at zero
velocity on the ground, while tracking stars, the stellar inertial
system can calibrate many of the inertial instrument errors
including the accelerometerbias. This featwe.reduces some at the
accelerometer long-term emor sources.
TABLE 2. COMPARISON OF STELLAR DENSITIES FOR M E

R & B BAND STELLAR WAVELENGTHS
B Band
0.4 f 0.049~
Mr
R Band
0.70 f 0.11~
Densitv per Deg2
1
2
3
4
5
6
~~
0.0002
0.0006
0.0022
0.0075
0.0240
0.0710
0.0006
0.0022
0.0077
0.0250
0.0750
0.2300
197
Rgure 13. Miner Observing StarllgM in the Daytime
A IAPERTURE AREA
12
TERRESTRIAL OWLS
A = 36 aq In.
E l1
3 lo
E
s 76
SPACE
MinKlWLS
A = 0.72 rq In.
0 3
z
U
2
1
bi
-p
-2
0.06
0.6
I-IER:brAL
60
600
6,000
60,000
ALTITUDE IN MILES
- & 3 P A C E MiniOW
IN DAYUGHT
9ucIII-14
Rgure 14. Stattracker Sensed Magnitude va AHitwJe for a

Silicon Detector and a 3 Deg FOV
198
1992 - Gimballed
1996 - Strapdown Including Power

Supply and Processor
4:l
Reduction
7
; in. w x 7
; in. H x 20 in. D
45 Ib
Satellite and Spacecrat System
120:1
Reduction
14in. W x 15 in. D x 2 0 in. H
Including Power Supply 8 Processor (165 Ib)
2 in. H x 1 in. W x 4 in. H

250 Grams
92M-121.15
ngure 15. Comparison of Glmballed and Strapdown Stellar lnertlal Navigators for
Alrcrafi Along with a Spacecraft Stellar Inertial System
40
30
ALTITUDE
(FrXlO))
20
10
6
26
Rgure 16. Probabllity of Clear Llnes of Sight Over the Northern Hemisphere
for All Seasons Combined (72,000 Observatlons)
I99
5.0 CONCLUSIONS
Stellarinertialcapabilityis aprecisionautonomousnavigationstrategic asset. The technology encompasses gravity compensation
techniques, stellar catalog, telescope design, and inertial navigation. It provides the user with a unique surgical strike advantage in
an area where, for a host of reasons, radio navigation has been
eliminated or temporarily denied. The SAIN enables reducing the
cost, size,andreliabilitypenalty for autonomousprecision strike to
an affordable option in futurevehicles. When sensor fusion is taken into account, the low-cost stellar option provides in-& alignment regardless of the vehicles flight path or speed, and a level of
navigation redundancyto enhancemission success. Thus, as in the
past, modem man should depend on the stars for navigation.
3.
Kenton L. Bachman, Ronald Dennis, and Seymour Levine,

StrapdownAstro-Inertial Navigation (SAIN) An Idea That
Has Come of Age, Central Inertial Guidance Test Facility
(CIGTF)-Fouteenth BiennialGuidanceTestSymposium,p.
181. October 1989.
4.
Kenton L. Bachman, Ronald Dennis, and Seymour Levine,

Strapdown Astro-Inertial Navigation Utilizing the Optical
Wide-angle Lens Staxtracker, Navigation, Joumal of the
Institute of Navigation, p. 347, Winter. 1990-91.
5.
Fredrick R. Nadeau and Seymour Levine, Strapdown

AstmInertialNavigation as aSecure Methodofsatisfying the
lbenty-first Century Precision Mission Requirements,
Central Inertial Guidance Test Facility (CIGTF), Fifteenth
Biennial Guidance Test Symposium,p. 19, September 1991.
6.
Major Rick Miller, Fredrick R. Nadeau, and Seymour Levine,

Mini-OWLSin Autonomous Stellar Angular Reference
Systems (STARS)for Hypervelocity Strategic Systems and
SpaceBased Interceptors, IEEE!Aerospace and Electronics
Systems Society, Plans Symposium, M m h 1992.
BIBLIOGRAPHY
1.
2.
Joseph F.Caligiuri. The Navigation System for the Polaris

Submarine,Navigation JournaloftheInstituteofNavigation,
p. 3, Spring 1960.
Vem A. Blumhagen, Stellar Inertial Navigation Applied to
Cruise Vehicles, IEEE Transactions of Aerospace and
Navigation Electronics, p. 235, September 1963.
200
Magnetic Heading References

by
Donald L. Moore
Avionics Systems Engineering
Smiths Industries
Aerospace & Defense Systems Inc.
Grand Rapids Division
4141 Eastern Avenue, MS 240
Grand Rapids, Mi 49518
USA
Summary
This paper addresses the most m m o n system for pro
vidinganaircraftheadingreference: theme~~~ticheedng
r e f e m system. It begins brietly explaining the importanceof a magnetic heeding reference(section1). tl then
addresses the fundementel characteristics of the enrth's
megnetic field and exqlains the co(IoBp1of 'magnetic miatiorr'(section 2). Itdisarssestwo of the more common
slyles of magnetic heading 88nsors along with their inherent e m (section 3). After explaining how these b(terent compasses operab, the paper explains the dillerent
types of heading enom that occur. secoion 4 addresses
the in-flieht m,
Wle section 6 addressesthe magnetic
disturbances caused by the aircraft W .Section 6 discussescalibration techniques that providefor the curection against these ma~neticdisturbance induced mors.
Finally, this peper cardudes nrcth a Mef disarssion of fucuretmds(sectian7).
1 Introduction
Many pieces of information must be available to a navigator (either a person or an OlFbOard computer) to reliably
fly an aircraft from point A to point 8. Obviously, one very
useful piece of infomationis the correctdirection of niM
to eventually arrive at point B. But, before turning the aircraft in the desired direction, the navigator must know the
ament headingof the aircraft. The magnetic heading reference system providesthii uitical piece of information.
Figure 1 : Earth's Magnetic F i
at the poles. All points on the globe betwen the equator

and the poi8s have some megnihde of e horizontal and
verticalcomponent.
Ab previoustyetated,the eerth'sfIugneticfield is only

'similar' to a bar magnet. The earth'sQu81does contain
materials that affect the earth's magnetic field. There is
a diurnel change and a slow drift over the years 111. The
reasonsfor these anomalies are not wfthin the scope of
thii text but it is important to note that the crust's elTed is
fairly stable and well know. The diurnal change is only
M.1"
but in certain regions, such as Alaska, variations
of 3 to 4 degrees per mirude have been observed. Also,
duing a magnetic storm,variations up to 12.8"have been
0bseNedover a 19hour period. The slowdrift, which in on
the order of fO.1 per year, is mulefive over a number
of years but eventually reverses and drifts in the other
direction. Reference provides eddib;onal information
pertaining to the 6atth's magnetic field.
[a
The poles of the earth's megneaic lield are referred to

as the north and swth 'magnetic' poles. As previously
stated,to determine a heading, an eppropciate heading
reference must be chosen. The north magnetic pole provides this reference in a magnetic heeding reference system.That is, all derived headings are basedon magnccic
north, whii is definedas 0".
MAGNETIC
NORTH POLE
2 The Earth's Magnetic Field

A W i n g is based on a coordinate system that we8 a
predefined direction as a reference. The angle created
betweenthe vectors pointingin the referencedirection and
the desired directionis known as the heading. To fadlite
the job of air lmfiicconl~)Ilera,
all aircraft navigatorsshould
use the same reference direction. The eerth's magnetic
field can provide that reference.
The properties of the earth's magnetic field are similar

to that of a powrful bar magnetic located at the eerth's
center (see@urn 1). The.fieldconsists of horizontal and
vertical components of the total intensity of the magnetic
field. H e n the held is parallel to the earth (near the
equator), the horitootel intensity is the same as the total
intensity of the held. There is no vertical component to the
field at the equator. Likewise, at the magnetic poles of the
earth,the verticalintensity is the same as the total intensity
of the field. There is IK) horizontal component to the field
Fgure 2: Magnetic and Geographic Poles

The eerth'sposition-locating'grid' system (latitudeand
longitude) is based on the earth's rotational or polar axis.
The poles of the &s
rotational axis are referredto as
the north and south 'geographic" poles. Unfortunately,the
magnetic poles and the geographic poles do not coincide
(seemure 2). In fact, they are over lo00 miles apart.
20 I
W?mnpbuing a coufs~from pointA to point B,the navigator (which may be an on-board computer) determines
thedesiredheadingbasedon the earth's@ system. This
heading references the no& geographic pole, and is r e
fend to as the ?rue' heading (asopposed to 7nagnetic"
headq). Course plotting is usually performed using bue
heading, and course navigating is performed using magnetic heeding.
The dmencebetweenthe true heeding (which reterernes the geographic north pole) and the magnetic heading (vuhii references the magnetic nath pole) is know
as magnetk variation. Aeronautical cherts are aveileble
that povidetheeast~orwesterfymegneticvariationfor
any point on the eerth. These chertsareknowas 'Mag
neticVariation'charts and are p u M i i by the Defense
MepQine Asency H Y d m J m center PI. It
closes the gap between course plotting to true h e a d i
and navigating to magnetic heading. For erample, if the
plotted ooum is 210" tma heading, and the chart indicatesawesterlyvariationd 50, the rnlqnetic course tony
wDu(d be 215".
For bystems without magnetic senaors, an

exists nrhich creates a MAG VAR look-uptaMe.The magnetic dation algorithm (GEOMAG) is based on a 12th
order spherical harmonic expansion of the Earth's mag
netic fie#, the coefficients of H l h i comprise the world
Magnetic Model (WMM). These codfhnb areproduced
jointly by the U.S. Naval Oceanographic Office and the
British Geological Sunrey and are disbibuted by the U.S.
Defense Mepping Agency end the British Hydrographic
office. The World Magnetic Models are usually produced
at Syear intervalsand are composed of bo parts: a main
field model, which describes the Earth's magnetic field at
some base epoch, anda SeaJervariationmodel,M i i
accounts for the slow temporal variations in the main geomagnetic field from the base epoch to a maximum d 5
years beyond the base epoch. For example, the bese
epoch d the WMM-90 magnetic field model is 1990.0.
This Model is therefore considered valid betvveen 1990.0
and 1995.0 and i l l subsequently be replaced at 1995.0
by the WMM-95 magnetic field model.
3 Magnetic Sensors
The mainfUnc(i0nof a magnetic heading referencesensor
is to provide the actual magnetic heading of the aircraft.
Sensors cepeble of providing this data are often refened
to as %agnetometers". Mqnetometersmust be capable
of detecting the direction of the horizontal component of
the earth's magnetic field. From this, they can derhre the
actual aircraftheeding. T w of the most m m o n methods
for determining this are the 'simple magnetic compasss
andtherluxgetestylempass'.
The M e V compass consists of a disk (also
as acard) &ached to a buoyant material, with hrvo brig

magndSWnted Underneath. The CWdmtaiMmarkingscormpoding to all pointsofthecompass. Betwen
the magnets is a pivot point allowing the disk to rotate
upon a bearing thus allowing the two magnets to freety
aliinthemsehreswtththemegneticnorlhpde. This '(kating'assembly is then encased in a chamber that is filled
with aGquid such as alcohol or white keroaem(hence the
name W e y ' compass). The liquid inhibits the walk
m h g ofu-le compassdue to hrrbulenceor abrupt W S e
141.
Someadventagesin using the-ey
cunpassare:
l.Theyarerdalhdyi-.
2. Theydonotrequireelecbicalpowertooperate.
3. They do not require i n p m from other orrboard 8 8 ~

m.
4.They are very durable and rdiable.

Some of the disedventegesof avvhiskey compess are:
1. The aircraft heeding display b reliable only when
the aircraft is in less than a 18" bank [4] due to the
physical limitationof most d.

2. Theliquidcausesanoticee#edampinge~dubrg
turn.
3. The heading data from t h e c o m p a s s ~ b e e l e e

biCally transferred to other aircraftcomputers that
may need magnetic heading information.
3.2 The Fiux-gate Compass

Most, if not all, of the technically advanced aimaft te
day use a fluxgate style sensor as the primaty 8 w c e of
magnetic heading detection. There are several vafiations
of thii type of sensor, but all are based on the elecbical
principleofnux+ating'.
flux-gating is a very complex electrical concept. L

detailedexplanationis outsidethe scupe of this document.
For a detailed description of fluxgatiw, refer to [SI.A top
level desuiption, howver, is providedto giw the reeder
a basic understandingof how the nuxgete magnetometer
detects the aircraft'sament heading.
The flragate style megnetometer consists of a highly

permeable material (the core)such that the earth's magnetic t
b lines are draw into one end of the core and
released from the other end of the axe (seefigure 3). A
wire is wapped aroundthis core material, and an altemating c u m t is then applied to the wire. This Hlire is called
the 'primary' coil. A second Are is also wrapped around
t h e m and is used to sense the voltage peaksgenerated
3.1 The Simple Magnetic Compass
from the AC current. This wire is often refened to as the
'
coil.
Thesimplemagneticcompess,betterkno~asthe~ey' 'sense' coil or compass, is fourd in most aircraft today. It is used as the
sole heeding reference in m e small aircraft and as a
backup headng reference in larger aircraft with an advanced cockpil design.
202
I\ -.m7
PRIMARY
COIL
CORE
F ~ u r 3:
e flux Lbres UnsaturatedCore
0
When the current is applied to the primary wire, there

is a point at which the core becomes saturated. When
aturated, the core is no longer permeable to the earth's
magneticflux l i(the earth'smagnetic fmld is no longer
drawn into one end of the core and outtheotherend4m
figure 4). The key to the entirenuxgate' dedg is that the
amount of arrent needed to saturate the core is ploportional to the horizontalcomponent of the earth'smagnetic
field. In other words, m e current is requiredto saturate
the core when the detected horizontal component of the
earth'smagnetic Reld is stmg (heading north or south:
parallelto the horizontel component). LikeAae, less currentinneededwhenthedetectedhorizontalcomponentof
the earth's magnetic field is wak (heading east or west:
perpendicular to the horizontal component).
FLUX UNES
CORE
I
I
PRIMARY
VSl
ERST
ew.1
Figwe 5: Peak Vottages in Secondary Coil

tometer becomes divided into types.
Recall that the sensed heading from a flux-geUe me^netometer is a function of the horizontalcomponent ofthe
earth'smagnetic field. If the aircraft is banked, the sensed
horizontalcomponentof the field isnot of the same magnitude as whenthe einraA i
s
min a leml atbihrde. miis
due to the couplingofthe vertical field component and r
e
sub in an error in the calculationof the aircraft'sheading.
To correclthis wor,the compass keeps the 'axessin the
horizontalposition h unecceleratedflight by either placing
the unit in a 'gimbal", or by setting it in a fluid similar to
that used in the "lhiskey' compass. The VuW approach
poses the same damping problems as the 'vuhiskey'com
pass. Magnetometersthat are 'gimbaled" or set in fluid
are referredto 8s pendulous magnetometers.
4 In-Flight Errors
COIL
Figure 4: flux Lines Saturated Core

The amount of current needed to saturate the core
en be detected by the output of the secondary coil (the
881158 coil). The angle af the m e ( M iis the same as
the airaaKs) relative to the earth's magnetic field can be
derived by reading the peak vokgeson this SeCOndery
cdl (seeWre 5 as taken from I6D.
Inthe fluxgatingprinciplestyle justdescribed, two posSiMe headings exist for every current needed to saturate
We core. In the example given in figure 5, the current of
0.707 amps implies that the angle of the core relative to
the earth's magnetic field is either 45" or 315". To r e s o h
thii conflict, another core (with a primary and wumdary
coil) is offset such that the ambiguity can be resolved. The
number of additional cores,the physical placement of the
additional core(s), the type of materiel used as the core,
and the shape of the core is where the fluxgate magne-
Two of the most common in-flight errors are the 'northerly

turning' and 'accelerationMeceletatiation'enors,and these
are am'buted to what's known 8s 'magnetic dip'. As explained earlier, the earth's magnetic field has both a horC
zontal and vertical component. The earth's'magnetic dip
angle' (or inclinetion) is the angle betweenthe earth'stotal
magnetic field vector and the horizontal. At the magnetic
equator this angle is Oo, and at the north and south magnetic poles it is 90". When the dip angle is large, the ratio
of the earth'svertical to horizontal field strength is large.
Tipping the magnetic heading sensing element causes a
component of the srertical vector to be coupled into the
plane of the sensing element. This component adds to
the sensed horizontal vector, resulting in a heading e m .
The lage dip angle encountered near the magnetic
poles renders the typical magnetic heading system use
less. For this reason, the use of the magnetic heading
system is usually limited from 70"Nto 60"s.
203
4.1
Northerly Turning Error
When in a coordinated turn, the centrifugal force causes

the gimbaled or pendulous magnetometer to swing in a
direction parallel with the Roor of the aircraft as opposed
to remainlng parallel with the horizontalplaneof the earth.
This deviation from the horizontal introduces the vertical
component of the magnetic field which, as stated earlier,
resultsin an error in the calculation of the aircraft's heading
(see figure 6 as taken from
m).
As seen in figure 6, the error is at its greatest when

the perpendicular to the horizontalfield component (
is greatest. Since a turn direction from a northerly or
southerly direction causes this "perpendicular tip", the er-
wv)
=??
Most magnetometers are designed to correct for this

error by utilizingthe sensed acceleration/deceleratbnfrom
an accelerometer. Usingthis information,the magnetometer knows and compensates for the vertlcal component of
the magnetic field being introduced.
5 Aircraft Magnetic Disturbances

Anything that affects the direction of the horizontalcomponent of the earth's magnetic field at the magnetic heading
sensor is a disturbance. The magnetometer's housing
itself may cause a disturbance to the field. There are
four major sources of magnetic interferences associated
with an aircraft. The magnetometer is therefore usually
located in an area which suffers the least influence from
these sources of interference(e.g., the wing tips or the tail
of the aircraft).
5.1
Hard Iron Disturbances
The first major source of magnetic interference is the

permanent magnetism of various ferromagnetic structural
parts from the aircraft, or the cargo it contains. This type of
field turns with the vehicle, thereby changing its relationto
the earth's field vector and causing a change in the magnetic field surrounding the sensor. These disturbances
are usually referred to as nard iron" disturbances. For
one revolution of heading through a 360"turn, the hard
iron generates a heading error with one cyde of variatbn
(688 figure 7). Hard kon disturbances are therefore saM
to induce a monecycle'error.
C
I-
W - M
Figure 6: Generationof North Turning Error

ror is often called a northerly turning error. While in the
turn, the error causes the heading indicator to either show
that the aircraft is turning in an opposite directionor that it
is turning at a greater rate than the actual turn rate. The
effect on the heading indicator is a function of the hemisphere in which the aircraft Is being flown, and whether
the turn is away from a northerly or southerly direction.
Most magnetometers are designed to correct for this
error by utilizingthe sensed roll angle from a vertical gyroscope. Usingthis roll Information. the compass knows and
compensates for the vertical component of the magnetic
field being introduced.
4.2
Acceleration/Deceleration Errors
When an aircraft accelerates or decelerates, the gimbaled

or pendulous compass tins from the horizontal. This deviation from the horizontal plane introduces the vertical
component of the magnetic field, which causes an error in
the calculation of the aircraft's heading.
The geometry behind this error is the same as explained in the 'northerlyturning error" section. That is, the
error is at its greatest when the sensor tips in a direction
such that the perpendicular to the horizontal component
of the field is greatest. An acceleration/decelerationwhile
flying in an easterly or westerly direction causes this "perpendicular tip". This causes the heading indicator to show
that the aircraft is in a turn, even though it is in a straight
and level attitude.
.W
IW
zm
-v
Figure 7: HardSoft Iron Errors
5.2 Soft Iron Disturbances

A second major source of magnetic interference. produced
by the aircraft, is the induced magnetic fields created in
the vehicle's ferromagnetic structures by the earth's magnetic field. These fields are generally associated with soft
iron parts, although they also occur, to a limited extent, in
hard iron parts. Unlike the permanent field, the induced
field polarity and magnitude are determined by the relative direction and magnitude of the earth's fmld. These
disturbances are generally referred to as 'soft iron" disturbances and are more prominent in smaller aircraft [I].
For one revolutionof heading through a 360"turn, the soft
iron generates a heading error with two cycles of variation
(seefgure 7). Soft iron disturbances are therefore said to
induce a Two-cycle" error.
5.3 Eddy Current Disturbances

The third major source of interference is that produced
by eddy-current magnetic fields. These occur in all air-
204
waft skin, ribs, frames, and other stmctud units without

resped to the magnetic materials from tnhich they are
combucted. The eddy currentsrequireody the axistence
of electrical conductivity and aircraft maneuy~~s,
end are
generatedin the same fashion as the cwents produced
in a coil of wire rotating in a magnetic field. An eledrical
conducting path is formed around the outside edgeofthe
airufbws conducting sheets or dosed-loopsbuchwes. As
the aircraft maneuvers in the magnetic field of the earth,
electrical currents are produced in d i m t metallic fmrts
proportional to time changes of the fluxes of the eerth's
magnetic Reld through these parts. The fkwofament
mates a magnetic fidd in a direction perpendiilartothe
plane of the sheet or conductingbop.
5.4
DC Conductor Disturbance
The fourth major eourceof interferenceisthat prodwedby

~ ~ o r ~ c a n yDCarrrenta.
i n g Hlhen ament
is passed througha condudor,ltnesoftluxaregenarated
in drdes concentric with the CQnduCtol end in a plane

perpendiculartoit. Thesetypesofdaretwocyde in nature. They can be minimizedby using a twisted
pair of wires to counteraa the distwbame,as opposed
to using one wire with the aircreft sbucture as a ground
return.
5.5 Index Errors

Index enor is an error which is dueto misalignment of the
flux valve or compass system synchroa
mor has
the same magnitude at all headings.
In the pest, index enor has been comcbdby mtating
the flux valve in its mount until the ayBIBB8 8 c r o ~ o nthe
cardinalheadingswas zero. This techniqueusuallyleft the
flux valve misalignedwith respectto the air& centdine.
Present day techniquescall for accurately a l i i n g the flux
valve with the aircraft centerline. The redduel system index error is then removed by providing an elecbical differential between the flux valve and its control bansforer,
adjustment of which permits compensetionof index error
to less than 0.10.
6 Calibration Techniques
The aircraft magnetic disturbances, 88 Mned in the p f e
vious section,anect the detected direction of the horirontal component of the earth's magnetic field. To account
for these disturbances and prewnt an inbced am,the
compass must be 'calibrated'. T\ruo of the more common
methodsusedto calibrate the magneticheedingreference
sensor are 'manual swinging" and 'elecbicel swinging'.
Thesetwmdhodsprovidelheactuelenocceusedby
thealrcraftitaelf. Theactualmpensabjonoftheemus
canbeperformedby:
1. Installingdegaussingcoils or permanentmagnetsat
the location of the sensor to counterad the aircrefl
field.
2. Installing soft iron pieces in a position that will compensate the soft iron (2-cyde) ~ 0 1 s .
3. Compensating for the soft iron (2cyde) 8mwg by

the
adding appropriate impedances in series
m n e t i c sensor fransnission &es.

4. Digitally compensating for the 8mx in an
computer.
aircraft
6.1 Manual swinging

The 'manual swinging- method of d h U o n is nonnally
performedon the grouui by physicab swinging or turning
the airmft360", and,a! differen! points on the mpess,
measuring the enor between the a c W W i n g and the
heedingas detected by the smsof.
When performedonthegound,thesvvingmustbeacmplished on Rat terrain in an area free from tn4pdc

field distubnces such as nearbybddhgs or aircraft. Aircraft swinging goundaitesareavaileblethat meet these
aiterhandpro~asurveyedcompessroseforanactual
heeding reference.
lf an on-bowd gyro-bamd inertial heading reference
is available (e.g., an Inertial Navigat&m System), ltwn it

canbe used as the heading referencein placeofthe surveyed points on the compass me. This epproachalkws
the manual d n g to be performed either on the grwnd
(still in a magnetic distuhme free emn'ronment) or i+
Mht. A also faciliies automation of the data cdlection
and processing functions, thereby SimpNfying the calibrationprocedureandgreattyshorteningthetimerequiredto
accomplish it. An airborne swing reqrrires the a b a f t to
fly several straight and level, UnBCcBJBIBled flkjht legs in
dment directions. The more lliiht
wed, the better
the accuracy in the error calculation.
6.2 Electrical Swinglng

The'elecbicelswinging'methodofcalbrationisperfonned
onfluxgatestylecompessesbyap@yinga oontrolled DC
current into the Rux valve to simulatea manual360O swing
of the air& 191. In essence,this sianulates the eerth's
magnetic field rotating around the aircraft. As in manual swingins', this method of calibration requires that the
measurementsbe taken on a llet twain and in en area
free from external magnetic field distuhnces.
Beforethe elwrical swing canbeperformed,thecharacbrMcsof the fluxgate senm must be knourn. That is,
the amount of DC current required in each coil to sim
ulate any heading must be known for a magnetic dis
turbance free environment (indding the aircraff disturbances). This informationmay be provided by h e VBndOl
of the fluxgate sensor, or the information am be determined independently. In the latter method, the fluxgate
88nso1is physically removed from the aircrafi and placed
in a magnetic disturbancefree envirorment that povides
a m y e d compass rose. The drarecteristics of the sen801 are then recorded as the rnit is manually rotated to
severaldiffetentheadings.
Vvhen performing the 'electrical swing' on-boerd the
aircraft, M= currents are applied to the mils to simulate a
headingin a distufree envirment. The dilferern
between the displayed heading end the desired heading
are then recorded,svld another data sample is taken at a
differentheading. Again, the more data points taken,the
more accurate the error analysis will be.
205
Notethat the 'electrical swiqing' methodrecordsonly

the aircraft's hard-iron (1-cyde)errors. This method d
cafibration does not account for soft-iron (2-qde) errors
since they are a functionofthe relativedirectionandmeg
nihde of the earth's field. Soft-iron error analysis still
requires that the aircraft be menually rotated. However,
since soft iron enorstend to be a function dthe aircraft

design, they typically are steble. Thus, once calbrated
u a i q the manual swing technique, soff iron wom usually
do not need to be recalibrated,and subsequent elecbical
avdngp therefore provide edequate resub.
[e] W. A. Shapiro and C. E. Roemer, 'A Strapdow Magnetic Azimuth Detector", Navigatim: Jorrmal of The
Imtitute of Navigation, Vol. 17, No. 4, Winter 19701971.
[9] D.Baker and M. Skaar, Maans Fur Compass Swim
ing wilhout Rotabng the Aimfl, Proceedings of
the IntemationalAirTrfuwportAssociationsSixt~
Technical conference,Miami, FL, April 1965.
7 Future Trends
The magnetometersdiscus& in this papware very r e
IlaMe, relatively inexpensive, and suffidentty accuramfor
most uses today. Some military a p p l i i , however, re
quire more acaracy, especielly during accelerated light
endaggressivetums. Developerahavethereforecmted
a magnetometer that is strapped down 88 opposed to
baled. To detect the rdl a@eofIhe eircraft, it u s e s a w
andehxqpte coil placed at an angle that sensesthe YBIticalcomponent of the enrth's magnetic field. Couplingthii
rollinfonnation with accelerationdatafrom anaccakometer, the magnetometer can then provide very eccurate
heerfing information under all n i t c o n d i i . W s 3
cuds sIrapdowmmagnetometer' [8] is already in use today
in m e aircraft, and shouldbecome more widely used in
the W e .
The most dramatic change vvith magnetometerswill
pobably be in the way they are calibrated. cunently,
oegmund manual swinging is the most common melhod
usedtoday and it takesat least 3 housto perform. The
timeusedforcalaxationusingthemaMlalswi~~technique
may 80011 be reduced to no mmOre than 10 minutes by including amparisonsagainst m inertialheading reference
such as an INS or Attitude Heading Reference System
(AHRS). Ether an on-board computer or a canyon canputer wdl cdlect and process the information duing the
swing and digiily relay the appropriate calibration coeffidents backto the unit for storage. A developmentalmodel
of such a system (an improved version d the U.S. Army's
ANIASN-43) calibrated a compass to 0.5" RMS witkin 6
minutes using 1975 tedmdogy.
References
[l] M. Kayton and W. R. Fried, Avionics NabigfdimSystern, John Wiley 8 Sons, Inc., New Y& 1969.
R. 1.Menill and M. W. McElhimy, The Mh's Meg

ns(ic Fieki, Academic Ress, New York,
1983.
[3] Nathaniel Bowditch, Amencen Prectical Nawtor,

DefenseMappingAgency Hydrographic/Topogr@ic
Center, P a . No. 9, WeisMgton, D.C., 1984.
[SI plivate Pilot Manual, Jeppesen sanderson, Inc.,
1983.
[5] H. D. Qardner, I m p r o d k - G a t e Magnetometer
NASA Langley Research Center,UR-13560.
[e] J. 0. Fmar, US. Patent4,918,824,
Compass, April 24,1990.
m Technical Newsletter,Spmy FliiM Systems Division,

TA-595TN-1,March 1963, Revision 2, Oct 31,1969.
207
SECTION IV
SYSTEM ANALYSIS, DESIGN
AND SYNTHESIS METHODOLOGIES
THE INTRODUCTION
by
Dr. D.F Liang
Directorate Research and Development
Communications and Space
Ottawa, Ontario
Canada
K1A OK2
INTRODUCTION
For aerospace navigation

systems, the rapid advances of
sensor technology, estimation
techniques, high speed processors as
well as the necessity to satisfy
mission requirements in a cost/
weight/reliability/performance
conscious environment have led to
increasing applications of
integrated multi-sensor navigation
systems. In general, stand alone
dead-reckoning navigation subsystems
require periodic updates, which may
be time consuming and can affect the
operational effectiveness of the
aerospace platforms. Radio
navigation systems are inherently
noisy over the short term, but the
errors are normally bounded. Even
though the GPS is an accurate radio
navigation system, which can act as
a "stand alone" navigation system,
it cannot on its own provide
attitude information of sufficient
interest to most aerospace
navigation users. In contrast, the
synergism and fault tolerance
realized by automated complementary
sensor data blending provides
performance accuracy and reliability
far beyond the capabilities of
individual system components.
This section deals with

system analysis, design and
synthesis methodologies of
multi-sensor system integration. As
an introduction to this Section, a
general overview of multi-sensor
integrated navigation system design
is presented. In most aerospace
navigation system applications, the
INS and GPS are considered the two
most important navigation subsystems, therefore the emphasis of
our discussions on the system
analysis, design and synthesis
methodology is on GPS/INS
integration looking at deep
integration, general integration
overview and federated faulttolerant integration. For real-time
suboptimal filter design, a
practical implementation of an
airborne transfer-of-alignment
system aiming at a challenging
Synthetic Aperture Radar Motion
Compensation application is
presented. This is followed by a
detailed discussion on the vertical
channel design of an inertial
system.
To be more specific, in
Dr. D. Liang's Chapter on "An
Overview of a Generic Multi-Sensor
Generic Integrated Navigation System
Design", he presented a general
overview of the design of a
208
Multi-Sensor Generic Integrated

Navigation System. It provides a
tutorial of navigation system error
models of interest in forming
robust, complementary integrated
systems. Considerable attention is
placed on the Kalman filter design
and on the selection of error states
to reduce computational requirements
without compromising performance
accuracy and stability. Lessons
learned from his development of a
number of integrated navigation
systems have been summarized to
assist the integrated system
designer s o that practical
development can be completed with
minimum risk.
In Col Lewantowicz and

Major Paschall's Chapter on "Deep
Integration of INS, GPS and Other
Sensor Information", the authors
presented the methodology to perform
optimal deep integration of
information provided by the GPS, INS
and other sensors. Their chapter
discusses the need to perform
vigorous error modelling and
validation in support of system
level analysis and filter synthesis.
High fidelity simulation analysis
provides a sound basis for
understanding parameter
sensitivities in Kalman filter
integration schemes. The simulation
analysis is essential for sound
engineering trade-offs in Kalman
filter implementations.
Introductory material is provided at
the outset to discuss various sensor
operational concepts, error
modellings, linear system tools and
estimation theory. Integration
issues are illustrated by way of
analysis case studies looking at
single versus multiple filter
integration and stability. The
analysis of differential GPS
integration with INS and other
sensors is the final example
presented. The final section
presents a broader discussion from a

mission context of the role of
INS/GPS in future avionics and
weapon systems.
In Dr. Carlson's Chapter
on "Federated Filter for FaultTolerant Integrate'd Navigation", he
presented an efficient, federated
Kalman filtering methodology, based
on rigorous information-sharing
principles. The method applies to
decentralized navigation systems in
which one or more sensor-dedicated
local filters feed a larger master
filter. The local filters operate
in parallel, processing unique data
from their local sensors, and common
data from a shared inertial
navigation system. The master
filter combines local filter outputs
at a selectable reduced rate, and
yields estimates that are globally
optimal or subset optimal. The
method provides major improvements
in throughput (speed) and fault
tolerance, and is well suited to
real-time implementation. Practical
federated filter examples are
presented, and discussed in terms of
structure, accuracy, fault tolerance
throughput, data compression, and
other real-time issues.
In Dr. Greenspan's Chapter

on "GPS/Inertial Integration
Overview", he presented a rather
thorough overview of GPS/INS
Integration. It presents an
overview of the benefits and tradeoff issues of integration; an
overview of integration
architectures and algorithmic
concerns: and sample results from
performance evaluations of
integration case studies.
In Mr. DiFilippo's Chapter
on "Design Considerations for a
Suboptimal Kalman Filter", he dealt
with the practical implementation of
a transfer alignment filter for a
209
Synthetic Aperture Radar Motion

Compensation System.
He started with the
optimal filter design and proceeded
at length to show how a practical
suboptimal filter can be designed to
accommodate state deletion and
decoupling. The criteria used in
making the filter model
simplification provide useful
guidelines for general suboptimal
filter design. Simulation results
were presented to confirm that the
suboptimal filter provided similar
performance to the optimal filter.
In Dr. Ausman's Chapter on
"Vertical Channel Design
Considerations", he first discussed
barometric and inertial errors,
followed by a detailed look at the
LN93/94 conventional vertical
channel mechanization, and finally
described the design and performance
of a 5-state Kalman filter using
barometric altitude updating. It is
shown to have achieved superior
performance by modelling the bar0
scale factor error as a linear
function of altitude, a relationship
which the real atmosphere tends to
follow.
210
AN OVERVIEW OF A GENERIC MULTI-SENSOR INTEGRATED

NAVIGATION SYSTEM DESIGN
Dr. D.F. Liang
Directorate Research and Development
Communications and Space
Ottawa, Ontario
Canada KIA OK2
1.
INTRODUCTION
Modern avionics systems are becoming
increasingly sophisticated as the
demands for better mission
performance and higher reliability
continue to escalate. Many of the
missions must be carried out at
ultra-low altitude under all weather
and visibility conditions. The
increased range, speed and accuracy
of modern weapon systems, impose
stringent accuracy and reliability
requirements upon the aircraft
navigation system. To enhance
mission success in a hostile
environment, the pilot amongst other
things needs to operate weapon
systems, target acquisition and
designation systems, radar
detection, night vision systems and
perhaps engage in air-to-air combat.
This paper describes the application
of Kalman filtering technology to
the design and development of a
multi-sensor Generic Integrated
Navigation System (GINS).
2.
SYSTEM OVERVIEW
A typical structure of the GINS can
be illustrated by Figure 1. The
primary system component is the
integrated sensor unit which
consists of all the selected
navigation sensors (subsystems).
They in general, can be classified
into two main categories, dead
reckoning and radio navigation
systems. The dead reckoning (DR)
systems are the inertial navigation
system (INS), doppler radar and air
data computer. These systems are
self-contained and provide velocity

information which can be integrated
to obtain the position of the
aircraft. Position estimates s o
obtained have a tendency to drift,
slowly accumulating a large error
over an extended flight. DR systems
(except doppler) are generally nonradiating and non-jammable and hence
attractive for operation in a
hostile environment. On the other
hand, radio navigation systems such
as Omega, TACAN and GPS (Global
Positioning System) rely upon radio
signals transmitted from several
external sources which make them
vulnerable to enemy actions in any
serious confrontation. These
systems provide absolute position
information with error
characteristics which do not grow
with the passage of time, because
errors are not accumulated through
an integration routine. In the short
term however, radio aids excepting
the GPS generally do not produce as
stable and accurate a track as the
DR systems.
In order to take advantage of the
short-term stability of the DR
system and the long term stability
of the radio navigation system, one
can apply the theory of optimal
estimation in the form of a Kalman
filter (K.F.) to combine all the
available measurements and provide a
statistically "optimal" estimate of
the aircraft position, attitude and
velocity, etc. The workload of
optimal integration, navigation and
21 1
guidance functions can be handled by

the data processing computer. This
means that large amounts of data can
be handled at high rates, and the
computer can be entrusted to carry
out very sophisticated computations
to provide much more reliable
navigation and control output data.
To achieve the desired performance,
the GINS processor must perform
signal conditioning and sensor
compensation functions (Figure 2 ) ,
and it must contain efficient K.F.
integration algorithms, error
control and performance monitoring
routines. The signal conditioning
function controls the sampling rate
of incoming data and works as a
prefilter to average the higher
frequency data input. The sensor
compensator compensates for
predictable deterministic errors
intrinsic to each sensor, and also
provides compensation to remove the
effects of unstable positive
feedback in the GPS rate aiding
loop. The performance monitoring
routines serve as a watchdog to keep
track of the overall system
performance, and also provide CEP
estimates to indicate the expected
performance accuracy of the
automated system.
The GINS integration filter will
optimally weigh and combine all
available sensor data to estimate
aircraft position, velocity and
attitude with greater accuracy than
available from any of the sensors
individually. This means that
mission requirements can be met by
using sensors which are individually
less accurate and less expensive
than would otherwise be required.
In the event of subsystem failures,
the system performance degradation
will be graceful and the integrated
system will automatically configure
itself to the next optimum
operational mode.
The use of a digital Data Bus

(MIL-STD-15538) facilitates
configuration selection and data
interface of different brands and
types of equipment, and provides a
higher data flow capacity, self
check on each data transmission and
reduces susceptibility to electromagnetic interference. It also
reduces weight and improves
reliability because less wiring and
fewer connectors are needed.
3.
SYSTEM DEVELOPMENT APPROACH
For the system design of a generic
multi-sensor integrated navigation
system, a large number of 'equipment
configurations are possible. The
typical approach is to use previous
experience in selecting a candidate
configuration in an ad hoc manner.
This has the potential danger of
eliminating good alternatives early
in the project and could result in a
suboptimal configuration. It is
therefore more prudent to dedicate
some resources to simulate and study
two or three potential configurations with the aim of identifying,
developing and testing an integrated
navigation system which best
satisfies the mission requirements.
The prototype of the selected system
configuration can then be developed
and tested with the aim of
subsequent engineering development
for use on board the chosen mission
platform. The development effort
can be divided into at least the
following two phases:
Phase I - System Definition

and Design.
To properly analyse the performance
of various candidate configurations,
a versatile simulation package must
be developed. Since the fidelity
and performance prediction of these
configurations are of prime
importance, a substantial portion of
the development effort should be
dedicated to generating complete and
3.1
212
accurate error models. These

activities are described as follows:
a.
Survey existing or soon to be

available navigation sensors
to determine their
suitability.
b.
Identify several costeffective candidate system

configurations that can
potentially satisfy the
performance, weight, size and
reliability requirements.
C.
Generate the navigation sensor

error models.
d.
Generate a set of mission

profiles to be used for
trajectory generation.
e.
Develop the integration

algorithms to blend the sensor
outputs.
f.
Develop control and display

software .
g.
Develop diagnostic software to

detect sensor failure.
In the detailed simulation analysis

both covariance analysis and Monte
Carlo simulation software should be
used. Covariance analysis software
is effective for the design of
integration filters because it
provides ensemble statistical data.
Because ensemble statistics are the
outputs, covariance analysis can (in
many circumstances) reduce the need
for computationally intensive Monte
Carlo simulations. This statistical
information can be used to assess
candidate Kalman filter designs and
to project the performance of a
particular navigation system
configuration. Covariance analysis
is very useful for assessing the
effects of mismatch between the
filter design model and the "real
world" or truth model. In addition,

it can be readily used to establish
error contribution tables and error
budgets which let the filter
designer focus in on the major error
contributors.
However, Monte Carlo simulation also
has its place in the design of
suboptimal Kalman filters.
Simulation can be particularly
effective in assessing the effect of
nonlinearities which are difficult
to address in the covariance
analysis framework. Also, certain
types of mismatches between the
Kalman filter model and the truth
model are more conveniently
addressed with the simulation
program than with the covariance
program. An example of this is the
sensitivity assessment of the effect
of sea current correlation time
mismatch.
Another area in which the Monte
Carlo simulation can be of more use
than the covariance analysis program
has been in the investigation of the
effects of unmodelled manoeuvredependent sensor errors. As well,
covariance analysis software will
never be able to replace the
function of simulation software for
final checkout of Kalman filter
code.
3.2
Phase I1 Development and

Testing
Acquire the sensor hardware
for the preferred configuration and complete the
development of the design
produced in Phase I.
Develop the data base and
interface electronics to
connect the sensors.
Refine and convert the
preliminary forms of the
Phase I Kalman filter and
213
integration software into

flyable programs. Develop the
control and display software
and transfer these programs
into the airborne processor.
Inertial Navigation System

(INS)
Attitude and Heading Reference
System (AHRS),
Doppler Radar,
TACAN ,
Omega,
Air Data System,
Strapdown Magnetometer,
Radar Altimeter.
9
Conduct static tests of the

completed system in an
integration laboratory and low
dynamics tests in a mobile
van.
Conduct flight tests in a test
aircraft to evaluate system
performance in a realistic
vibration environment.
Assess the navigation system's
performance by conducting
extensive flight tests in a
test aircraft on an
instrumented range.
The Phase I1 development can be
aimed at realizing in hardware the
Phase I system design by
constructing and developing through
ground and flight testing the
Advanced Development Model (ADM) to
produce a fully developed and flight
validated integrated navigation
system prototype.
4.
SYSTEM ERROR MODELLING
Detailed simulation and error models
for all the relevant navigation
equipment and environmental
disturbances must be developed for
the performance evaluation and
sensitivity analysis.
A wide variety of types and brands

of navigation sensors can contribute
to meeting a specific set of
requirements. The following list of
generic navigation subsystems can be
considered having merit in any GINS
applications:
However, many of the generic

equipment categories can be further
subdivided. For example, in the INS
category there are high and medium
accuracy inertial equipments
available. The inertial sensors can
be mechanized either as a gimballed
platform or a strapdown configuration. Strapdown configurations may
use conventional or ring laser
gyroscopes. Similarly, Doppler
velocity sensors and Doppler
navigation systems are available in
the Doppler category.
Now that the GPS system is fully
deployed, the utility of Omega has
become rather limited, however for
the sake of completeness of
treatment, Omega has been included
in this study as an alternative.
TACAN can supply absolute navigation
data from ground stations, and its
transmitter can be used for relative
navigation and as a homing signal.
The air data system and magnetometer
are also considered as part of the
standard airborne equipment.
Furthermore, due to the relative
inaccuracy of the barometric
altitude information, a radar
altimeter is considered necessary
for low altitude flight. Therefore,
a radar altimeter is also included
in the evaluation of all candidate
configurations.
4.1
(GPS) ,
Inertial Measurement Unit

( I N ) Models.
214
The models described in the

following subsections are basically
developed for an inertial sensor
assembly. The same IMU error model
is utilized for AHRS systems with
appropriate changes in the
magnitudes of the sensor error
parameters. This approach is
reasonable, should an AHRS be
selected as the inertial reference,
it is expected that strapdown
computations would be implemented in
the processor. Thus the inertial
measurements made by the AHRS would
be processed just as if the AHRS
were an inertial sensor assembly for
an inertial navigation system.
8 V - p x 6R,
attitude error vector,

I
velocity error vector,

position error vector,
local vertical reference
frame relative to an
earth-fixed frame,
4.1.1 IMU Error

Several mechanizations can be
employed to represent errors of
Inertial Measurements Units (IMU).
One such error model is the 'IT angle" formulation, which is a
classic set of navigation error
equations applicable to either
strapdown or platform systems. The
error model is described by the
following set of vector differential
equations:
gyro output error vector,

accelerometer output error
vector,
gravity error vector.
6 V - 0 2 8 R + 3 o2
S
(8R
The total gyro and accelerometer

error vectors are expressed in terms
of individual sensor errors as
R R + 8s.
-)-
- r r
r9
Ta
This has the advantage of simpler

attitude error propagation provided
that the position error is
propagated via
CHX,,
CGX,,
vector of gyro error (bias

plus random noise),
vector of accelerometer
error (bias plus random
noise),
where
xg
xa
where
215
transformation matrix
relating inertial sensor
assembly reference frame
to the local-vertical
frame ,
coefficient matrix which
transforms the collection
of gyro errors into a net
angular rate error in the
sensor assembly frame,
coefficient matrix which
transforms the collection
of accelerometer error in
the sensor assembly frame.
In the event the IMU is strapdown in
nature, the transformation matric C
is defined to be the direction
cosine matrix relating the vehicle
body axes to the local-vertical
frame. When the IMU is of the
gimballed platform type, the matrix
C is an identity matrix, since the
inertial sensor assembly is at all
times coincide with the local
vertical reference frame.
4.1.2 IMU Simulation Model.
The IMU simulation model implements
the classic set of navigation
equations
transformation matrix from

body frame to the localvertical frame,
transformation matrix from

the local-vertical frame
to the earth reference
frame,
angular velocity vector of

the local vertical frame
relative to the earth
reference frame,
velocity vector of the

vehicle relative to the
earth reference frame,

the earth relative to an
inertial frame,

the vehicle relative to an
inertial frame,
gravity vector,
AL
vehicle nongravitational
acceleration,
[.I
skew-symmetric realization
of the enclosed vector.
4.1.3 IMU Sensor Errors.

Most of the gyro and accelerometer
sensor errors can be treated as
random constants or slowly varying
Gauss Markov processes. They are:
Gyro
G2 - drift coefficient,
Bias drift,
Scale factor error,
Input axis misalignment,
G - sensitive drift coefficient.
Accelerometer
Where,
Bias error,
Scale factor error,
IA misalignment,
Vibration induced Bias.
216
The remaining errors can be treated

in the following manner:
Gyro random drift (random walk),
Gyro attitude random walk (white
noise drift),
Gyro and accelerometer turn-on
transients (exponential decay).
The typical RMS values for gimballed
and strapdown INS errors are shown
in Table 1.
This error is intended to model a

bias in the Doppler frequency
tracker. The RMS value for this
error is in the order of 0.05 to
0.1 kt.
4.2.4 Ocean Current Error.
The effects of ocean currents can be
accounted with north and east ocean
current states represented as first
order Gauss-Markov processes.
Typical RMS values of the current
states range from 0 . 8 to 2 . 5 kts.
4.2
Doppler Radar.
Doppler radar measures the aircraft
velocity with respect to the fixedantenna coordinate system. The
errors include:
a.
Doppler fluctuation noise;
b.
Scale factor error in the

along heading velocity
component;
c.
Cross coupling scale factor

errors; and
d.
Speed offset errors.
4.2.1 Doppler Fluctuation Error.

This error represents the noiselike
character of the Doppler spectrum.
It is due to variations in the
backscatter of the terrain over
which the Doppler measurements are
being made. Typical correlation
times for this error are on the
order of 0.25 to 1.0 second, with
power spectral density of the order
of 0.002 - 0.005 kt2/rad/sec/kt.
4.2.2 Scale Factor Error.
Typical values of scale factor error
due to antenna calibration and
transmitter frequency deviation are
in the range of 0.1 to 0.25% of
total velocity.
4.2.3 Speed Offset Error.
' -- - 1
vcn
T,
rlcn
and
1
where
Tc
rlcn
correlation time constant

of 1 to 2 hours for ocean
current,
north ocean current white

noise,
east ocean current white

noise.
and
rl ce
4.2.5 Wind-Induced Error

The wind-induced water motion
velocities are commonly represented
by
217
v,
= 1.3
(v,)
Bij
-1
vwe
3 - I
cross coupling scale

factor error,
V"
for which i
where
x , y, z .
The typical (1RMS) Doppler error

parameters are presented in Table 2.
VW
vm
magnitude of wind,
north surface wind
velocity,
and
VW,
east surface wind

velocity .
The doppler velocities in the

aircraft coordinates are given by
4.3
GPS
GPS three dimensional position and
velocity errors can be modeled as

independent first order Gauss-Markov
processes with relatively long
correlation times of one or more
hours.
and
i-axis position
correlated error,
i-axis velocity
correlated error,
position error
correlation time,
where
true velocity along
i-axis,
velocity error
correlation time,
north-east-down to x-y-z
frame transformation
matrix,
i-axis position
correlated noise,
speed offset along i-axis,

i-axis velocity
correlated noise,
fluctuation noise along

i-axis,
for which i
scale factor along i-axis,
N, E, D.
218
Typical 1 U values for qigp and qigv

are 15 metres and 0.1 m/s,
respectively. It should be noted
that GPS velocity errors generally
have correlation times short enough
to be represented as uncorrelated
white noise processes.
altitude contains four potential

sources of error. They are:
a.
Error due to altitude

variation in a constant
pressure surface;
b.
Temperature dependent scale

factor error;
C.
Static pressure measurement

error; and
d.
Instrument errors.
4.4
Magnetometer Model
The magnetometer provides a true
heading measurement referenced to
true north. The heading error can
be modeled as
where
true heading,
llrt
slowly varying
instrument bias error
with typical value of
0 . 5 - 2 degrees
*b
Ill
I3
I5
I2
I4
single cycle Gauss Markov heading errors

with typical RMS of
0.5 - 1.0 degree,
Depending on flight profiles, the

dominant source of altitude errors
could vary. For a maritime aircraft
with low altitude flight profile the
dominant source of error can be
reasonably modeled as a first order
Gauss-Markov process
where
hb
correlated barometric
altitude error,
Tb
barometric altitude
correlation time which
equals to the correlation
distance divided by the
vehicle velocity.
Ib
barometric altitude white

noise error.
two cycle
Gauss-Markov heading
errors with typical
RMS of 0 . 5 - 1.0
degree,
uncorrelated white
noise in the order of
0.1 degree.
4.5
A i r Data Model
The air data subsystem provides air
speed and barometric altitude
measurements. The barometric
The air speed can be modeled as the

groundspeed corrupted by winds.
They are given by
219
vns
northerly surface wind

velocity ,
Vna
northerly aloft wind

velocity,
where
vna
northerly airspeed,
vea
easterly airspeed,
vha
airspeed along the

heading ,
v,,
lateral airspeed,
The surface and aloft wind

velocities can be represented by the
combinations of slowly and rapidly
varying first order Gauss-Markov
processes with corresponding
correlation time.
4.6
TACAN Error
The TACAN measurement is corrupted
by range and bearing errors. The
errors sources can be modelled by:
a.
The northerly and easterly airspeeds
are given by the equations.
vea=
Vet
(1
- 3 , v,, ha
ht
ha
0 . 2 nm;
b.
First order Markov bias

ranging error to account: for
time-delay between signal
reception and transmission
with typical correlation time
of 1 to 2 hours and RMS value
of 0.1 to 0.2 nm;
c.
Random white noise bearing

error of the order of 0.1 to
0.5 degree;
d.
First order Markov bearing

calibration error; and
e.
First order Markov slowly

varying bearing correlation
error.
Vea
and
where
h,
true altitude,
ha
wind aloft altitude,
v*s
easterly surface wind

velocity ,
Vea
easterly aloft wind

velocity,
Random white noise ranging

error o f the order of 0.1 and
4.7
Radar Altimeter Error

The radar altimeter measures
altitude above ground level. Thus
for overland operation the largest
error in altitude will be due to
variations in the terrain itself.
Radar altimeter errors are modeled
as a random scale factor error plus
220
generally agreed that when PPC

tables have been used, the remaining
phase error at the Omega receiver
has an autocorrelation function of
the form.
an uncorrelated (white noise)

altitude error.
The radar altimeter simulation model
uses the error model to produce
simulated radar outputs. The radar

altimeter error in altitude is added
to true value of altitude obtained
from the trajectory generator data
to produce simulated radar altimeter
outputs.
4.8
OMEGA Error
Omega is a hyperbolic system that
utilizes lines of position based on
phase difference measurements from
at least three transmitters. It
operates in the 10 to 14 KHz band
and, since each station has a range
of 13,000Km, only eight stations
are sufficient to provide world-wide
coverage.
where auto-correlation times T1 and

T2 are respectively 1 8 0 , 0 0 0 and 4800
seconds, o is the earth rate and A,
B, C are respectively 1520, 500 and
850 metres. This phase error can
therefore be adequately described by
the sum of three stochastic
processes: two first order Markov
processes and a periodic process
with a period of 24 hours. Since
the periodic process is second
order, a total of four states are
required to properly model each
Omega phase error. Determining
three independent Omega lines of
position (LOPS) requires reception
of Omega signals from 4 distinct
stations.
Diurnal changes in the ionospheric

propagation characteristics and the
inhomogeneity of the earth's
magnetic field and surface
conductivity cause anomalous
variations in the Omega phase
measurements. These deterministic
errors can be partially compensated
for by using phase propagation
correction (PPC) tables. However,
there remain four major random
components of residual error. The
statistical properties of this
residual error have been studied in
detail by various groups and it is
dt
4 states are required to model

each Omega phase error which
propagates according to
MARKOV
PERIOD
0
0
0
pz
03
L- J L
1/T2
22 1
where W, is a zero mean Gaussian

white (ZMGW) noise process, and P,
is 01 the extra state needed to model
the periodic error. T1 and T2 are
the Markov process correlation times
and 01 is a constant. Here the
strongly correlated Markov process
is referred to as a bias to
distinguish it from the more weakly
correlated one (i.e. Tl>>T2).
where H(t) is the measurement

matrix,
and v(t) is the measurement noise
with spectral density R.
The discrete form Kalman filter
equations may be written as
5.
THE KALMAN FILTER DESIGN
The dynamics of the error state
vector
and
where F(t) is the fundamental matrix

and w(t) is a zero mean Gaussian
white noise process with spectral
density matrix Q. The dynamic
system model defines the manner in
which the system errors propagate in
time and the geometry relating
estimated error states to error
observations constructed from the
primary and auxiliary sensors.
There,are many possible suboptimal
filter models that could be derived
from simplified truth models. The
major problem in Kalman filter
design is to determine which error
states can be ignored or grouped
together to reduce the size of the
filter (and hence the computational
burden), without significant
performance degradation.
The measurement model describes the
relationship between the inputs to
the Kalman filter, z , called
measurements, and the state vector
that is to be determined from these
inputs. For linear systems this is
generally of the form
where
The error covariance propagation is

given by
The filtering algorithm most

commonly implemented is Biermans
U-D factorised Kalman filter. This
algorithm avoids the explicit and
computation of the estimation error
covariance matrix Pk by propagating
in terms of its factors U and D:
Pk
UDUT
where U is a unit upper triangular

matrix and D is a diagonal matrix.
The UD factors are calculated by the
modified weighted Gram-Schmitt
(MWGS) algorithm.
The U-D algorithm is efficient and
provides significant advantages in
222
numerical stability and precision.

Specifically, the factorization of
Pk provides an effective doubling in
computer word length in covariancerelated calculations, and avoids
filter divergence problems which can
arise in more conventional filter
mechanizations due to loss of the
positive (semi-) definite property
of P, through the accumulation of
round-off and truncation errors.
For the practical Kalman filter
design of any inertially based
integrated navigation systems, two
categories of modelling simplifications are possible. The first one
deals with inertial error dynamics.
In Section 4.1 we have commented
that the 'P formulation has become
very popular because of simpler
attitude error propagation, in
addition, some of the following
simplifications should be
considered.
Elimination of Vertical Axis
Modelling
For most aircraft navigation system,
it is possible to eliminate vertical
axis error modelling. Conventional
fixed-gain error control mechanizations using altitude measurements
might provide adequate vertical axis
performance. A s well, there is
normally little cross coupling from
the vertical to horizontal axes to
demand the modelling of vertical
axis error state. However the
vertical velocity error can affect
horizontal error propagation through
the horizontal Coriolis acceleration
components.
filter designer to eliminate

vertical axis dependencies from his
dynamic error model. In most cases,
the dynamical coupling of errors
associated with the system computed
Coriolis acceleration errors can be
eliminated.
5.3
Kalman Filter Error States
In view of the above, a Kalman
filter design for a generic multisensor integrated navigation system
might adequately be represented by
the following error states:
5.3.1 INS
-
INS horizontal position

and velocity errors
( 4 states)
INS attitude errors

( 3 states)
5.1
5.3.2 Doppler
Doppler boresight and

along heading scale factor
errors ( 2 states)
Sea surface currents for

doppler over water
application ( 2 states).
5.3.3 GPS
-
5.2
Elimination of Horizontal
Coriolis Acceleration
In most aircraft application, the
Coriolis acceleration error due to
the vertical velocity is
substantially less than the
uncertainty in gravity model. Such
an observation allows the Kalman
3 gyro drift rates and 2

acceleration bias errors
(5 states)
No GPS'error states are

necessary except when GPS
pseudo range and range
rate data are used instead
of position and velocity
measurements.
5.3.4 Omega
8 Omega states (1 long
term and 1 short term
223
error state for each of

four stations)
5 . 3 . 5 For most applications, it is
not necessary to dedicate special
error states to estimate errors
associated with the air data
strapdown magnetometer, and radar
altimeter, etc. The vertical
channel design can normally take
care of errors in the vertical
channel, and this is fully addressed
in S . Ausman's Chapter on Vertical
Channel Design Considerations.
5.4
GPS Rate Aiding
The INS data can be used to rate aid
the GPS receiver. The aiding
permits narrower bandwidth code
tracking loops in the GPS receiver
to improve its anti-jamming
performance. It also reduces the
satellite acquisition time by using
a priori antenna velocity data to
account for Doppler shift of the
carrier frequency.
The closed loop system design of

using GPS measurement based Kalman
filter estimates to reset INS was
found to be unstable at high Kalman
gains in the noncoherent mode.
Widnall [ 3 ] has suggested an
alternate approach of using reduced
control gains for correcting INS
errors. Such an approach was also
found to be insensitive to tracking
loop bandwidth.
6.0 Covariance Analysis
Covariance analysis is an efficient
and powerful tool for sensitivity
performance analysis to determine
the contributions of distinct error
sources. It is essential for
developing a robust filter design of
minimum state size. Its simulation
analysis consists of three major
components [ 41 .
a.
an aircraft trajectory
generator which provides
nominal flight data,
b.
a reference sensor error truth

model which characterizes all
the sensor errors, and
C.
the Kalman filter reducedorder design to be evaluated.
To conduct sensitivity analysis, a

covariance analysis is performed
with the reduced order Kalman
filter, and the gain history is
recorded. Then another covariance
analysis is performed using the
truth model for all sensor errors,
with the Kalman filter gain computed
from the earlier step. The
performance obtained in the second
covariance analysis represents the
predicted performance of the reduced
order filter design.
Figure 3 illustrates the real-world/
filter-world analysis flow diagram
for integrated navigation system
performance assessment. The upper
half of the diagram represents the
iterative solution of the filter
covariance equations. These are
solved in order to generate the
sequence of filter gains, Kk*, which
is a necessary input to the lower
half of the diagram, representing
the iterative solution of the truth
model covariance equations. Thus
two passes through the covariance
program are required to assess the
performance of a suboptimal filter.
7.
DESIGNER BEWARE
For any integrated system design, it
might be useful to consider some of
the following design tips.
a.
Be generous on size of truth

model,
b.
Be precise on noise level of

truth model,
224
C.
Minimize filter states,
d.
Be pessimistic on noise level

of filter model,
e.
Implement easily modifiable

computational algorithms
(Don't take short cuts too
early), and
f.
Potential causes of
divergency:
Numerical error (negative
diag. P can be eliminated with
appropriate choice of P
propagation equation),
Programming error,
Nonlinearity,
Unmodelled error,
Optimistic noise level,
Incorrect model.
DESIGNER'S ROAD MAP

8.
For the designer, a simple and easy
to understand road map is offered
for reference:
system output is compared to

the input, or/ and significantly increase system errors,
verify that the output errors
are reasonable and behave as
expected,
i.
Implement filter algorithm

(could waste lots of time if
this is done before #h),
j.
Review filter results,
k.
Tuning: good agreement

between state estimate error
and covariance, and between
measurement residual and
residual variance (through
Monte Carlo and covariance
analysis),
1.
Determine achievable optimal

performance,
m.
Robustness test (filter should

perform well with increased
noise level of truth model),
n.
Filter sub-optimization
(reduction of filter states to
improve efficiency while
increasing other noise levels
to absorb unmodelled errors),
a.
Determine system truth error

model,
b.
Determine true measurement

model,
C.
Determine filter error model,
0.
Substitute simulated data by

real data whenever possible,
d.
Determine noise model and

statistics,
P.
Conduct controlled test of

partial operation, and
Determine initial conditions,

noise and error variance,
q.
Redesign or retune the filter

if necessary.
e.
f.
Simulate truth model and

physical environments,
g.
Verify truth model simulation

(How realistic?),
h.
Verify simulation
all error sources
verify that there
unintended system
tool: set
to zero to
are no
errors when
9.
CONCLUSION
This paper has presented a brief
overview of the.design of a multisensor Generic Integrated Navigation
System (GINS). Lessons learned from
the development of a number of
integrated navigation systems have
been summarized to assist the
integrated system designer s o that
225
practical development work can be

completed with minimum risk.
REFERENCES
Benson, D.O., "A comparison of
Two Approaches, to Pure
Inertial and Doppler-Inertial
Error Analysis," IEEE Trans on
Aerospace and Electronics
Systems, Vol. AES-11, No.4,
July 1 9 7 5 .
Bierman, G.J., "Factorization
Methods for Discrete
Sequential Estimation,"
Academic Press, N.Y., 1977
Windall, W.S. , "Alternate
Approaches for Stable Rate
Aiding of Jamming - Reistant
GPS Receivers, "National
Aerospace and Electronics
Conference, May 15-17 1 9 7 9 .
Dayton, Ohio.
Gelb, A., Applied Optimal
Estimation, MIT Press,
Cambridge, 1 9 7 4 .
226
.-
...................................................
- 1I
$5
a
...........................................
....................................................................................................
w
U
4-
.
f
ra
a
228
"i
* x
7,s
n
3
+-a
2
U
w
0
3U
0
0
229
2
r
3
m
d
d
c
\
m
0
U
CO
0
0
0
v)
>
0
Y
L
'
m
d
d
U)
v)
0
X
5
d
E
Q
N
0
v)
L
0
0
\t
VI
c
r
v)
0
m,
X
c
\
zl
VI
c
r
0
n
0
U)
L
0
0
c
VI
0
0
L
N
L
v0
c
\
m
- 4
M
0
VI
0
M
0
c
VI
0
0
d
o
N
o
em
o
c
3m
E
.-
0
0
F
0
0
Y
L
U)
.C
.-E
v)
i i
. x
v)
L
0
.C
230
.-U
m
al
c
.-
U)
L
0
2L
c
\
01
m
al
U
In
In
In
r
0
m
r
0
m
'
al
c'!
E
3
w
0
m
rn
r
v)
'c
E
3
m
m
m
m
C
0
L
0
\
v)
L
0
v)
al
m
\
m
v)
0.
c
In
v)
0)
v)
Pal
.-U
'
al
C
0
0
0
rr
In
v)
r.l
'
c c c
'
'
'
'c
w
0
0
s
L
'c
m
aI
m
Q
m
0
v)
v)
.-nm
E
L
a,
v)
I-
>.
.-r
C
m
.-
.-r
al
>
al
r
r
x
.-
.-
Pal
.-
U)
U)
m
m
m
m
C
0
C
0
Pal
Pal
U)
v)
.r
U)
v)
e
0
m
m
C
0
d
v)
23 I
Deep Integration of GPS, INS, SAR, and Other Sensor Information

Zdzislaw H. Lewantowicz, Colonel, USAF
Deputy Director, Avionics Directorate, Wright Laboratory
WUAA Bldg. 22
2690 C St., Ste 2
Wright-Patterson AFB. OH 45433-7409
USA
Randall N. Paschall, Ph.D., Major, USAF

Chief, Advanced System Integration Research Team
System Avionics Division, Avionics Directorate, Wright Laboratory
1.
INTRODUCTION
This discussion is presented in the context of avionics sensor

integration methodology. However the concepts developed and
illustrations presented are in no way limited in application to
military avionics systems. There is a growing number of civil
applications, where information from multiple sensors is combined to improve performance, provide redundancy management, increase robustness, or achieve graceful degradation
when sensor failures (or outages) occur. We are rapidly moving
beyond the classical examples of sensor information integration
such as in aircraft navigation, or in control of chemical processes, nuclear plants, and jet engines. For example, the automotive industry is using several sensors for engine emissions
and fuel control, vehicle active suspension, and yes, vehicle
electronics (vetronics). Communication systems, position,
velocity, and attitude (rotation) sensors, and other information
are rapidly appearing in the consumer automobiles. These sensors provide, at affordable cost, functional capabilities, which
until recently were reserved for the more expensive commercial
air and ground systems. Other applications of sensor integration are either being implemented, prototyped, or considered in
toys, electronics, consumer products, such as the household
washers and dryers, communications, and home environment
monitoring and control systems. Although the sensor integration possibilities are expanding into these other domains, this
discussion focuses on deep integration' of Global Positioning
System (GPS), inertial navigation systems (INS), synthetic
aperture radar (SAR), and other sensors which are a subset of
modem aerospace systems.
Historically, the aircraft avionics have evolved from dedicated,
single function, mechanical sensor assemblies to, more recently,
sensor systems which have become sophisticated in their functionality and accuracy. These earlier sensors were developed,
refined, and added to the aircraft as stand-alone (federated)
devices which provided crew members with displayed information which increased their mission performance. Fusion of this
information provided by a multitude of these avionics subsystems has historically been one of the required pilot (or crew
member) skills. For example, instrument flying requires the
pilot to methodically, rapidly, and effectively cross-check several instruments that make up the performance and control
instrument groups. He must absorb this dynamically changing
information and continually make real-time decisions about the
type and quantity of control input(s) to achieve the desired/
required flight path precision. In this case the pilot mentally
performs the fusion of the various elements of information
required for safe instrument flight. Incidentally the autopilot,
when engaged, performs similar fusion and control decisions
for a subset of the flying tasks. The fusion of displayed information by the human operator is an increasing challenge due to
the increases in quantity and complexity of information provided in the cockpit.
1. Deep integration. for the purpose of this paper. is defined as a process
which applies the system theory at integraled system level to the ensemble of
individual sensor raw measurements.
In recent past either adequate computational and data bus technologies did not exist or the mission did not require the deep
integration of information that the various sensors provided.
The sensor performance and the computational and data bus
capabilities have increased, and the mission requirements warranted some exchange of information outputs among the avionics black boxes. For example, INS velocity information was
provided to the synthetic aperture radar (SAR) to enhance its
motion compensation performance. These were the earlier
stages of integration and could be called limited integration or
output inregration. Numerous constraints limited the quality,
quantity, type, and other attributes of information in that limited
integration information exchange.
In addition to the various computational and data bus constraints, standardization constraints also affect avionics sensor
subsystem. For example, the USAF Standard INS has stringent
specifications for not only the required physical fit and function
but also for the digital output format, rate, and content, including the data word length, precision, structures, time tagging,
and other related parameters. The output specifications, which
were developed primarily to drive cockpit displays, are often
not suitable for deep integration. Some constraints are imposed
by the availability of technology, such as the speed limitation
(capacity) of the MILSTD 1553 data bus while others are driven
by the near-term focused specification process.
An early example of limited sensor integration is the central air
data computer (CADC) for aircraft applications. The CADC
combines several pitostatic and dynamic measurements of
atmospheric conditions with other information, and provides
altitude, airspeed, and rate-of-climb information corrected for
several nonlinear distorting effects. The integration exemplified
by the CADC is in combining of the dynamic and static atmospheric pressure-sensor information and enhancing it with relatively sophisticated, but open loop, compensation algorithms to
provide higher accuracy outputs than those possible from the
basic sensors. In other instruments the combining of information is nonexistent or limited. The GPS/INS integration,
described in a later section is an example of limited integration.
Of significance here is the fact that each sensor evolved primarily from the early necessity for stand-alone operation, that is
without the necessity for communication of information to or
from other sensors. Where information was needed to be
shared with other sensors, the format of that information was
typically constrained to that available at the post-processed output format specified by the primary sensor function. The more
recently developed sensors have imbedded in them powerful
computing capabilities which enhance sensors' output performance. In these cases, functional outputs from other sensors
may be combined to enhance this performance.
Continuing rapid advances in processors, processing, detectors,
networking, and other technologies are motivating the reexamination of the sensor information integration architectures and
the integration of navigation information with that from other
sensors. These advances in technology and rapid reduction in
component cost provide opportunities for integration at sensor
232
signal levels deeper (closer to the raw signals) than just recently
possible. This deeper integration is an unexplored area, especially when considered from a broader system level perspective.
where the system might contain a variety of sensors, including
those classically dedicated to navigation. While this assessment
is based on information available in the professional literature,
it is important to recognize that substantial work in this area has
likely been accomplished and not yet reported.
In many avionics sensors the output signal is a result of substantial processing of raw signals. An example of a raw signal in
this discussion might be the digitally coded radio frequency
(RF) in a communication system such as the phase shift keying
(PSK). Examples of raw measurements are: GPS receivers,
pseudo- and delta-range measurements derived from the binary
phase shift keying (BPSK) L, and/or L2RF signals, outputs of
gyros and accelerometers in inertial measurement, or pixel level
image elements in imaging devices. The acceptance of the GPS
as a prominent system for navigation and the explosion of the
receiver technologies, even before the full (24 space vehicle)
constellation is deployed in 1994, are first motivators for examining deep integration. Additionally, the rapid processing capability advances, with the simultaneous reduction in cost, are no
longer bamers to implementation of complex algorithms requiring substantial processing capacity.
Deep integration, for the purpose of this papes is defined as a
process which applies the system theory at integrated system
level to the ensemble of individual sensor raw measurements.
Essential ingredients of this deep integration process are proper
(adequate) modeling, system synthesis and simulation, error
analysis, and engineering trade-offs. A variety of approaches to
modeling and analysis are available. Most frequently useful, in
these types of analysis, are the error model approaches. One
critical step in this deep integration process is the access to (or
development of) validated truth error models. This truth error
model becomes the foundation of system level analysis. [ 1, 2, 3,
45.6.7, 8.91.
Linear algebra is a rich linear system toolbox well suited for
system level modeling, analysis, synthesis, and engineering
trade-off analysis [7, 10, 11, 12, 13, 141. In many system applications, linear system theory is widely used in the modelling,
analysis, and synthesis purposes. For nonlinear systems, where
linearization is practical, the same linear system tools are
applied. One of the profound impacts on system level integration is the development of the recursive linear estimation theory.
The most significant recent contribution to estimation theory is
the publication by R. E. Kalman, in early 1960s, of his work in
various articles, such as: A New Approach to Linear Filtering
and Prediction Problems [15]. This work in discrete-time
recursive estimation theory, combined with the advent of wider
use of digital computers, is known as the Kalman filter theory
and application. It has provided a powerful tool useful in system level integration. Although the Kalman filter is most
widely used, other methods of integration, or sensor fusion, are
used. This discussion is limited to linear estimation theory
applications.
Estimation theory is an essential ingredient for optimally
extracting the desired information given the random nature of
the measurement information received by the sensors and which
is inherent in processing of the information. The Kalman filter,
in its various forms, is the most useful estimation algorithm,
which is well suited for operation with linear systems tools [7,
16, 17, 181. In this endeavor, it is important to properly (adequately) characterize the dynamic and stochastic error behavior
of the sensor signals. Wherever possible, the least processed
(least correlated) information should be operated on by the estimation (integration) algorithm, because sound error modeling
becomes difficult for signals which have been modified with
various filtering and decision algorithms. Theoretically it is
possible to use processed (correlated) information, provided
that the processing, which had modified the basic signal, is
accounted for analytically. However this is not practical for a

variety of reasons, including the difficulty of obtaining the
actual processing algorithms which operated on (correlated) the
particular signals. Frequently these algorithms include nonlinear processing, which can be troublesome, especially where
processing jump discontinuities are present.
There exist various factors which cause barriers to deep integration in the aerospace vehicles. Federated integration architectures are barriers to deep integration since these architectures do
not provide sufficient communication protocols and bandwidth
and sufficiently precise timing required for deep integration.
The federated architectures and sensor outputs were not
designed to provide signals (measurements) with deeper integration in mind and constrain progress in deeper integration.
The sensor outputs, of the current inventory sensors, are
designed to drive cockpit displays and are inadequate for deep
integration. The sensor output information is typically processed, (functionally correlated) through a variety of filtering,
compensation, and transformation algorithms. The unprocessed (uncorrelated) signals are often available in the sensor
box. They are often created as intermediate variables in the
implemented algorithms and stored in the sensor digital memory, but not made available on the outpuddata bus ports. These
barriers to deep integration prevent achievement of maximum
performance, robustness, and cost benefits possible with deep
sensor benefits. Later sections illustrate these benefits.
This section discusses the historical evolution of sensor and
computational technology. The analytic process and the tools
required for deep integration are identified. This process consists of necessary modeling, simulation, analysis, and synthesis,
and the linear systems and estimation theory domain tools. In
later sections, several examples of modeling, simulation, and
analysis of deep integration are presented. These illustrate the
methodology with comments and interpretations. Simulation
results, with empirical data comparisons (where available) are
presented in later sections.
2. ANALYSIS, ESTIMATION, AND SIMULATION
In analyzing various engineering systems, it is normally necessary to represent the physical quantities involved with mathematical functions. There are many different types of functions
one can use, and choosing an appropriate function usually is
determined by the type of problem under study, and the dominant characteristics of the quantities involved. For example, a
process may be either deterministic or random. A deterministic
process is one whose future realization can be exactly predicted
based on knowledge of past behavior, whereas a random process is one in which there exists a degree of uncertainty with
regard to future activity. Typically, the more one knows about
the governing factors of a phenomena, the more likely they are
to use a deterministic description, and conversely, the less one
knows about a process, the more likely they are to use a random
description. The advantage of using deterministic models is the
ability to use straightforward explicit mathematical formulas.
Another distinction in representing a physical quantity is the

dimension of the problem. Depending on the complexity of the
problem, and the level of detail required, the model may
describe a single entity or a group of related quantities. This
leads to the use of scalar versus vector representations. However, since all scalar problems are a subset of vector representations, we will only consider vector representations.
Many excellent texts are available which present linear system
fundamentals [IO, 11, 131, and the material found in this section
can be found in any one of these references. This section will
present only those concepts used in later sections to design integration algorithms for GPS, INS, and other sensors.
2.1 Linear Systems

A general dynamical system mathematical model is written
233
3 (2) = f ( X ,
U ,t )
(1)
with associated general measurement equation
y ( t ) = h (x, U , t )
(2)
The time-rate of change of state vector x(t). 3, is a vector valued functionf (not necessarily linear) of x, input U , and time t.
The system states may not all be physically meaningful quantities, but any physical quantity of interest can be obtained from a
linear combination of the states.
For any system, an infinite number of linear models exists, but

the models are constructed so that the equations describing the
system dynamics are combined into a unifying structure
expressed by a matrix differential equation of the form
I ( t ) = A ( t ) x ( t )+ E ( t ) u ( t )
(3)
In many problems, the elements of the A and B matrices are
constants, implying that the system is time invariant in addition
to linear. Thii is not the case for the navigation systems considered in this paper. In particular, the elements of the A matrix are
a function of the position, velocity, and acceleration of the systems.
The biggest reason for using linear models to approximate real
world systems is the ability to solve the system of dynamic
equations using relatively simple linear algebra techniques, and
to achieve a closed form solution whose form does not depend
on the nature of the inputs or initial conditions (as is often the
case with nonlinear system models). For linear systems, the
output equation can be formulated as the matrix equation
are also easy to transform into discrete algorithms for microprocessor implementation. Computer programs to perform analysis for nonlinear systems are not as easy to use nor as simple to
implement. Performing parametric studies becomes extremely
challenging when nonlinear models are used.
Example of Dynamic Analysis Utility

Once a model is in the form of Equation (3). the stability of the
system can be easily analyzed by computing the complex eigenvalues of the A matrix. If the eigenvalues all have negative real
parts, the linear system is asymptotically stable. The stability of
a linear system is not a function of the model chosen, but rather
the eigenvalues remain the same between any models related by
a similarity transformation. This provides additional design
freedom to the integration engineer. Stability may be time varying as in the case when the A matrix (and thus the eigenvalues)
is a function of time. In this case one may need to study the
eigenvalue migration as was done by Cunningham [ 191 where
he explored the eigenvalue migration characteristics of the INS
during various maneuvers. The resulting eigenvalue migration
is depicted in the next three figures.
2.2
0.00.
....................
."
xxI
- .......-........
-... ......................
i.
I
j
0.00)
.....................
x x x
.*.. .......".. .......i
'
xr*
% X U
,.. ....................../ ...............'.......X.
..................I
...........................
.-c
0.w1
.-E
o.m,: .................................................................
x.
. . . . . . .
0 ;
Y ( t ) = C ( t ) x ( t )+ D ( t ) u ( t )
....................
.................................................
=+'.&z
. . .
(4)
"::
........................
.
In many problems of interest, D(t) is often zero as is the case for

the problems considered in this paper.
In the case where the system model is nonlinear, a commonly
used approach is to linearize the system of equations about a
nominal using Taylor series expansion techniques, and to keep
perturbations from this nominal condition small so as to maintain the validity of the linear approximation [ 131. When the
nominal Eqoation (1) and the higher order terms (HOT) are subtracted, the linearized error perturbation equation provides
63 = A ( t ) S x + B ( t ) S u + w
0.001
I;
xix
0.001-
'
........x.x. ........../. ...........................................
I ;
....... ...................j . ...........................................
......................
: x
!... ...........................................
...i
.........E." ..........j
x
;..................5 . . ...;
:
........................
(5)
where A ( f ) = a f ( x , u , r ) / d x and E ( ? ) = af(x,u,t)/du

are Jacobian matrices which model the linearized error
dynamics A and the control distribution E matrices. This perturbation differential equation in &, for notational convenience
is often expressed in terms of x, that is x is used in lieu of 6r.
When this substitution is made, then Equation (5) is in the form
of the linear vector differential Equation (1). When both the
whole value and perturbation quantities are discussed, the symbology ambiguity exists, and caution (or variable redefinition) is
required.
Similarly, the linearized measurement Equation (2) is
Sy = H ( t ) S x + D ( t ) S u
1 :
(6)
where H ( t ) = ah (x, U ,f ) /ax and D ( t ) = ah (x, U , r ) / a u

evaluated at the nominal conditions provide the linearized measurement function. The H matrix contains the geometric, and
other relationships, between the sensor measurements and the
state vector variables. Thus Equation (6) is in the form of linear
measurement Equation (2). In many applications, the control
matrix D is zero, stating that the control has no direct effect on
the measurement.
Because all linear models can be put into a common form. and
thus the form of the solution is the same in every case, one can
use standard computer aided design packages to perform system analysis regardless of the system under study. This allows
integrators to use a variety of commercially available products
to conduct design and analysis with. The form of the equations
Figure 2. East Heading Loop Eigenvalue Migration

000.
0001-
4.001
.i......................
.o.m
.&.c ..................... ;.............................
I
rea?
0.002
1
t
0.00.
Figure 3. North Heading Loop Eigenvalue Migration
In Figure 1 the important fact is the general location of the

eigenvalues during the maneuvers. Note that at least one complex pair of eigenvalues is always in the right half plane. The
eigenvalue on the positive real axis is due to the vertical channel
instability. The eigenvalues for the vertical maneuvers have different characteristics. In the maneuver which has an initial east
heading, there is a complex pair of eigenvalues which breaks
away from the imaginary axis, migrates to the real axis, and
returns to the imaginary axis. For the maneuver with an initial
north heading, only eigenvalues associated with the vertical
channel appears in the right half plane. This indicates that east/
west maneuvers result in an additional pair of unstable eigenvalues. This type of analysis lends useful insight to an engineer as
to which type of maneuvers will cause. the greatest problems to
234
the systems under test. This information allows one to fully test
conditions that might otherwise be overlooked until too far into
the development cycle. A nonlinear system's stability is a much
more complicated issue and cannot be analyzed as easily. One
is often forced to talk about stability in the sense of Lyapunov
or bounded input bounded output stability. This is another reason to use a linear model whenever possible.
In addition to stability, one can easily compute the eigenvectors
of a linear system to gain geometric insights into the system
behavior. Eigenvectors form a basis vector set which spans the
n-dimensional state space. There is an eigenvector ( (t) associated with each eigenvalue hi, for i = l ,..., n. This, when coupled with Equations (3) and (4), yields insights into the
controllability and observability directions of a given system.
Knowledge of the eigenvectors coupled with the system matrices gives one insight into the limitations imposed by a given line a r model. One must realize that observability and
controllability are not physical properties of a system as are
eigenvalues and stability. Controllability and observability are
determined by the form of linear model chosen to represent the
physical system [13]. When using a Kalman filter, it is necessary to use a model where the states to be estimated are observable. States which are not observable cannot be estimated
accurately by the Kalman filter.
For time varying systems, the observability and controllability

of a system is determined by the rank of the Grammians defined
The rank may change as a function of time which is useful to

determine. For example, one may wish to address the question
of what type of geometry or dynamics may enhance the observability of states of interest. For large dimensional problems
which are time varying, the evaluation of these Grammian functions is computationally intensive. For this reason one often
B , and H
looks at specific instances in time, evaluates the 0,
matrices of the system and for those times the invariant Grammian matrix. This is an easier computation to perform, given
the available software products. By evaluating the rank of this
matrix, one can determine the observability (or controllability)
of the system at a given time. To get some idea of how the
observability changes with time, several of these evaluations
must be made at different time instances.
Another property of linear systems that aids in the analysis is
the property of linear superposition which says that the
response of linear system to multiple inputs is simply the sum
of the responses to individual inputs. This allows one to study
the response of a system to one input at a time for parametric
evaluations, and then to make performance evaluations to combinations of inputs without performing additional analysis.
This is not the case for nonlinear systems where the response
depends on the exact form of the input and changes unpredictably depending on what input combinations are applied.
One can also analyze linear systems using frequency domain
techniques as well as time domain techniques. Using Laplace
transforms, a system described by Equations (3) and (4) can
also be described in the Laplace domain
X(S)
= H[sl-A]-'Bu
(s)
(9)
The matrix H[sl-A]-'B is the transfer function matrix of the system. Using Bode, Nyquist, or Nichols charts one can graphically evaluate items such as stability margins, bandwidth,
settling time, peak overshoot, and crossover frequency. These
issues are important when one starts to question how well a system performs in the presence of external disturbances, in various dynamic environments, and over frequency ranges of
interest. As an example, in a GPS receiver one must decide on
the tracking loop bandwidth which affects the receiver's ability

to reject jamming signals. If one chooses a low tracking bandwidth to give maximum jam resistance, then the dynamic range
over which one can maintain track is reduced. Similar questions for nonlinear systems are very difficult to analyze.
If a linear model cannot be obtained, which accurately represents the behavior of the system being modelled, then one is
forced to use nonlinear techniques, and the problem becomes
much more difficult to manage. On the other hand, even if a
good linear model can be achieved, it is critically important to
perform system level evaluations using simulations which
incorporate nonlinear effects and thus more accurately reflects
the real system's behavior. This is important to recall when one
is developing a truth model as opposed to a design model or an
operational filter model. This is the real reason behind system
testing using real hardware. The actual system contains all of
the nonlinear affects that were not modelled in a linear system
design and analysis. It is common for engineers to not use nonlinear truth models since they are difficult to understand and
develop. This leads to a linear truth model which ignores the
nonlinear affects, often leading to unexpected outcomes in system testing. One must always keep in mind the limitations of
the linear models used.
2.3 Stochastic Models

Section 2.1 discusses mathematical models useful in the analysis of deterministic dynamical systems. In the physical world
there are no known deterministic phenomena. Thus the mathematical model developed in that section must be modified to
account for the stochastic variable behavior and the corruption
of the measurements by random processes.
In general, the geometric information from each (or all) sensor
outputs can be modeled by a set of stochastic diflerential equations of the form of
i = f ( x , U ,t ) + w ( t )
(10)
where w is an additive random driving noise vector which is a
zero-mean, white Gaussian noise process with covariance kernel [7, 16, 17, 181
E{w(r)wT(t+z)}= Q(t)6(z)
(1 1)
The measurement vector y, here in the discrete-time sampled
measurements form, follows the form of Equation (2)
y (ti) =
[x ( t i ) , til
+y (ti)
'
(12)
These output measurements are corrupted by additive white

Gaussian noise vector Y of mean zero and covariance kernel
E { Y ( t i )V T ( f j ) } =
R ( t i ),ti = tj
0, ti f ti
The pair of Equations (10) and (12) is sufficient to describe the

dynamic and stochastic behaviors of the sensors. The noise statistics of w,and v, which represent the actual sensor e m r behavior must be provided a-priori. This set of stochastic dynamic
and measurement equations provide the framework for modeling and integration of sensors from a system level perspective.
Estimation Concepts
Whenever one attempts to integrate two or more navigation sensors, a choice must be made as to the type of integration algorithm to use. A Kalman filter is an optimal recursive algorithm
(estimator) that processes measurements from sensors in order
to compute a minimum error estimate (in a well defined statistical sense) of the state of a system. It does so by taking advantage of knowledge of systems dynamics, the statistics of the
system error sources and measurement errors, and available initial condition information. A Kalman filter can also be operated
in a batch mode, but recursive operation is the mode presented
2.4
235
in this section. This section briefly presents the fundamentals of

the Kalman filter in order to enhance the readers understanding
of the integration results presented in later sections. For a more
complete presentation of the theory, the reader should refer to
any of several excellent texts on the subject [7, 161.
A Kalman filter may be used to perform one of three different
types of estimation operations. The operation of filtering refers
to the situation where the time at which an estimate is computed
corresponds to the time when the last measurement is provided.
The operation of smoothing is when the filter estimate corresponds to a previous time when measurements were provided,
and prior to the present measurement. The operation of prediction involves performing an estimate corresponding to a time
after the last available measurement. This section limits its discussion to the filtering operation.
A Kalman filter computes estimates of states which cannot be
accurately measured (perhaps not measured at all), but which is
observable (as defined in Section 2.1). The Kalman filter has
several underlying assumptions. First it is assumed that the system and measurement equations are linear. When the system
andlor measurement equations are not linear one uses a linearized or extended Kalman filter. Second, it is assumed that the
noises that act as forcing functions to the system are white.
This implies that the noises are uncorrelated. and have equal
power at all frequencies. Third, it is assumed that the noises
have Gaussian probability distributions. None of these assumptions are completely satisfied for actual systems, but in many
cases they are very good approximations. Because of concepts
such as the central limit theorem, shaping filters, systems bandwidths, and because one often knows the mean and variance of
the noise processes, the assumptions underlying a Kalman filter
are not very restrictive in application [7].
The general problem for a Kalman filter to solve can be cast as
follows. Suppose there are certain states associated with a system which one cannot determine directly and exactly, but for
which there are measuring sensors that provide measurements
corrupted by white Gaussian noise and which are functionally
related to the states of interest. The Kalman filter computes an
estimate of the states based on the measurements available, but
to do so requires basically three elements. First one must have a
mathematical model which describes the relationship between
the states and the measurements. Measurement errors are modeled by a noise vector Y. When measurement uncertainties are
non-Gaussian, shaping filters driven by white noise are used to
augment the state vector. Second, a dynamics model is required
that describes the time varying nature of the states. This model
uses a noise term Gw to account for model uncertainty and random behavior of the states.
These measurement and dynamic models describe the a-priori
knowledge of the states as well as the statistics of the noises. In
developing these models it is assumed that the dynamics noise
w and the measurement noise Y are uncorrelated. The final element associated with the Kalman filter is the performance index
to be minimized. The Kalman filter is based on a Bayesian
viewpoint. The estimates of the states are based on all measurements which is reflected in the conditional probability density
function written as
There are three possible measures of optimality: median, mode,

and mean. If the density function described by Equation (14) is
Gaussian, then all three measures of optimality are satisfied by
the Kalman filter. When the measurement noises are statistically independent (i.e. uncorrelated) then the matrix R defined
by
E{Y(t),Y'(t-T)}
= R6(t)
(15)
is diagonal. If however the noises are correlated, then R is not
diagonal. This degree of freedom is usually avoided in order to
simplify the computations. However, by accounting for correla-
tions which are present, the estimation performance of the filter

is enhanced. The same is true when one considers the system
dynamic noises.
The Kalman filter does not have knowledge of the actual noises
except for the statistics provided. This is especially important
to recall when filter models are being tested against truth models in simulation analysis. The truth model will include all
known noises, and should be of a higher order than the Kalman
filter. Along with the state propagation equation, there is a
propagation equation for the covariance matrix. It is this equation one evaluates when performing covariance analysis.
2.5
Linearized and Extended Filters
In many cases, especially those involving navigation sensors, a

linear model does not provide an adequate model of the system
under study. In this case the system is described by a set of nonlinear differential equations as discussed in Section 4. Given a
nonlinear model it is still desired to produce an optimal estimate
of the states. To do this, one can assume that a nominal state
trajectory can be generated. Associated with the nominal trajectory is a nominal measurement
The measurement may be a nonlinear function of the states.

Using Taylor series perturbation techniques (see Section 4.1)
the perturbation equations are derived
6 i = A[x,u,t]6x+E[x,u,t]6u+Gw
(17)
6z = H [ x , t ] 6 x + D [ x , u , t ]( 6 u ) + Y
(18)
Using these linearized equations one can now apply linear filter
theory. The input measurement for such a filter is the difference
between the actual measurement and the nominal measurement.
The output of such a filter is an optimal estimate of 6x. To
establish an estimate of the state, one simply forms the total
state estimate as
E(r) = x , ( t ) + 6 E ( t )
(19)
This form of the Kalman filter is called a linearized Kalman filter [7]. It is computationally efficient compared to a nonlinear
filter, but it can experience large erron if the nominal state trajectory differs greatly from the true trajectory. This leads to the
use of an extended Kalman filter.
The basic idea of the extended Kalman filter is to evaluate the
Jacobian matrices about the last estimate once it is computed.
As soon as a new estimate is produced, a new,state trajectory is
computed and included in the estimation process. In this way
the validity of the small perturbation assumptions inherent to
the Taylor series approximation is maintained. The extended
Kalman filter is defined by the update equations given as follows [7]
= f ( r ; ) + K ( t , ) [zi-h[3(t;),tl]
(20)
P (ti' ) = P ( t ; ) - K ( t i ) H ( P , t j ) P ( t ; )
(21)
i ( f i + )
and the propagation equations are
f = F [ f , t ] P + P F ' [ P , t ] +GQGT
(23)
Notice that in the extended Kalman filter, the propagation is

done using the nonlinear equations evaluated about the latest
estimate. AlSo, notice that the update equation for the state
involves the nonlinear relation described by h[x,r]. These are
important differences between the extended filter and the linearized filter. The extended filter is used in this paper to integrate
GPS with other sensors. This is due to the fact that the measurement is nonlinear for the navigation sensors.
236
2.6
Simulation Tools
Software which explicitly calculates time-varying satellite

position for a full GPS constellation was written, and analyses were conducted using the Multimode Simulation for Optimal Filter Evaluation (MSOFE) [20]. It is referred to as
multimode in the sense that it is capable of simulating a variety
of system problems. Its two most widely used modes are Monte
Carlo and Covariance simulations. MSOFE implements the upper-diagonal (U-D) form of the extended Kalman filter algorithm
[20, 71. MSOFE is used to establish a performance baseline
against which other full-order and reduced-order filters may be
compared. Time histories of variables, such as aircraft position
in R3 space, attitude, and acceleration, are created and serve as
the true aircraft trajectory in all simulations conducted and presented in this research.
Profile generator (PROFGEN) is a trajectory generation program developed by the Avionics Directorate [21]. It computes
position, velocity, acceleration, attitude, and attitude rates for a
maneuvering air vehicle profile based on user provided inputs.
This flight trajectory is directly interfaceable to MSOFE simulation. These data are used in the MSOFE simulation as nominal quantities about which the truth model is relinearized after
each integration step [7]. The extended Kalman filter is relinearized about its best estimate of the trajectory.
the earth center, attitude with respect to the earth surface, and
the Greenwich Meridian Time (GMT). When two systems,
each represented by a geometric vector G,and G2respectively,
operate relative to each other, we can define a relative geometric vector AG
AG = G , - G ,
3.2
3. GEOMETRIC SENSORS AND ERROR MODELS
Among the avionics sensors aboard a typical aircraft, most contain common information in the geometric domain, that is these
sensors are geometrically related. Their measurements contain
some, or all, the elements of geometric information which is
defined as a set consisting of position vector ( r ) . orientation
(attitude) parameter array (g), and time scalar (t). The orientation of the vehicle can be represented in any of several orientation parameter sets; the direction cosine matrix (DCM),
quaternions, Euler angles, and rotation vecto2. Two most commonly used are the DCM and the quaternion because these offer
singularity-free representations [22]. The orientation array 9,
can represent any chosen orientation parameter set. These are
related and can be transformed between the sets. The rotation
vector is a useful geometric concept and is the choice for this
discussion. Thus the symbol (0)
- represents the rotation vector
in the following discussion. Temporal and/or spatial derivatives
of these information elements may also be directly or indirectly
available. For many applications it is convenient to use the
derivative of position, that is the velocity vector Y explicitly.
3.1
Geometric Vector Concept
(25)
Most every sensor aboard an aerospace system, provides an output containing some or all elements of G or AG. For example,
an INS provides r. v, and 9. The GPS provides r, Y , and t, while
a crude approximation of 9 can be extracted for a receiver
which is not maneuvering rapidly. In another example, a modern aircraft radar system has the capability of providing various
forms of information accumulated along the line of sight, however the basic infomation available from most radar systems
consists of the LOS azimuth and elevation (@) and/or range R.
This information is derived from elements Ar and A$ of the
basic AG,where G, is that of the radar carrying platform and G2
is the target. The presence of G and/or AG components in these
sensors is the foundation for deep integration. Because these
sensors refer to the same physical geometric elements, the measurements that these sensors provide are thus spatially correlated.
Error Model Format
The spatial correlation, between the various sensor outputs, discussed in Section 3.1 is mathematically modelable in the context of linear systems theory and a stochastic process. This
model can be processed in a context of a Kalman filter, where
the off-diagonal elements of the filter covariance matrix contain
the cross-correlations representing the spatial correlation. This
cross-correlation between the sensors is the cornerstone of deep
integration benefits. By proper modelling of all relationships
and processing in a single (joint3) Kalman filter, the optimal
(best possible) estimate of the cross-correlated errors is possible. This can be thought of as a mutual (cross-) calibration process of the various sensor errors which share the same physical
geometric domain (or have a known relationship with the same
geometric vector). Again, this cross-calibration is limited by
the quality of the dynamic and stochastic error models, the relative magnitudes of the uncorrelated noises among the sensors,
and the noise (both measurement and driving noises) magnitudes relative to the signal strengths. The truth model analysis
establishes the best possible cross-calibration performance.
The performance of the subsequently synthesized reduced-order
filter can be compared to that of the truth model. The next discussion presents key linear algebra tools and concepts in modeling of the sensors and their error behavior.
A ten-element array G of geometric parameters is defined:
Although this array does not always satisfy all vector properties
as discussed above, it is referred to as the geometric vector G
and denotes the represented physical variables. To be useful in
mathematical and algorithmic calculations, this vector must be
coordinatized in some reference frame, such as body b-frame,
navigation n-frame, earth e-frame, inertial i-frame, etc. For
definitions and discussion of reference frames see [ 1.2, 3.4, 5,
6, 7, 8, 9, 231. Of significance, for this discussion, is the fact
that this geometric vector can be used as an absolute or a relative reference. An example of an absolute geometric vector
might be the position and velocity of a vehicle with respect to
2.
Caution is required in interpreting the Orientation parameter array ES a

vector. as it does not satisfy the vector property of commutation. although.
for small rotations. the mfafionvecfor does satisfy vector properties. to firstorder approximation 1221.
3.3 Sensor Classes

The quantities position, velocity, attitude, range, line-of-sight,
and time form the basic set of variables measured by sensors.
These quantities are either elements of the geometric vector G,
or the relative geometric vector AG. The sensors aboard an aircraft can be basically classified into the following categories:
inertial, direct position/velocity/attitude,and/or relative position/velocity/attitude.
A particular sensor may provide one or more of the G or AG
elements. The differences between sensors providing these
variable outputs are in the quality of the measurements and their
environmental capabilities. For example an RF radar penetrates
atmospheric fluids, however a laser radar, with more restricted
propagation due to moisture and other attenuating matter, provides a much more accurate measure of both the LOS and
range. The key point is that for deep integration, the structure
of the eeometric and error models for these LOS sensors are
3. Joint in lhejoint probabilily density function sense 17,16. 17,
181.
231
similar. The following discussion presents the basic principal

of operation for each sensor class, followed by a structure for
the error model. The emphasis is on those aspects of error models that relate to the elements of absolute or relative geometric
vector. This is not limiting nor exclusive. If other, non-geometric but related, quantities are of interest in a particular system
design, then the models need to be expanded to those domains
and appropriate analytic tools applied.
The goal in integration of a set of sensors is to take advantage of
the fact that all of the sensors share a common geometric reference. They all experience the same reference position, velocity,
and attitude at any instant of time. Even when the sensors may
be looking in different directions, and their lines-of-sight
might be varying, their origins are either fixed with respect to
each other, or are varying at some rates. These relative sensor
orientations are known, or measured, except for relative errors,
such as installation errors, boresight errors, randomly varying
errors, and others.
3.3.1
I
Sensor Truth Models
Sensor errors can be characterized as random processes, representable in the context of system level models. If linearization
and Gaussian uncertainty model assumptions are valid, then linear estimation theory is applicable. When all of these sensors
are completely (adequately) modeled in their dynamic and stochastic domains, then these models can be assembled in a single, joint error model. This joint model then becomes a truth
model. Estimation theory is then applied at the truth model
level where Kalman filter is the tool of choice for linearhinearized error models. Various desirable outcomes are immediately
available.
First, the Kalman filter computes the joint estimate of the mean
2 and covariance matrix P which contains the cross-covariances
in the off-diagonal elements [7, 16, 17, 181. These off-diagonal
elements represent the statistically (and dynamically) derived
error dependencies among the various sensors. These crosscorrelations, through the sharing of the common geometric vector, produce in the filter the error estimates, or sensor cross-calibrations.
Second, the truth model system-level analysis provides the
baseline for the best possible system level performance. This is
useful for various trade-off studies, engineering designs, sensor
level parameter trade-offs, baselining of contributions of each
sensor to the system level solution. and other analyses.
Third, the truth model becomes the reference system infrastructure for suboptimal filter design. Various system level tradeoffs in the filter design, filter tuning, and other sensitivity studies are well performed at this level.
3.3.2 Sensor Error Truth Models

The most important aspects of this process is the availability of
sufficiently correct sensor error models represented in the
dynamic and stochastic domains. The error model development
must apply sound scientific method. The necessity for sufficient correctness of these models can not be understated.
3.4
Inertial Navigation System (INS)
A key sensor available in many airborne systems is the inertial

navigation system (INS) [ 11. It is an amazing example of
advances in various technologies from improvements in gyros
and accelerometers, and gimbal platform performance, to computational capability explosion, strapdown systems, and system
level integration. These improvements were driven by two key
historical events requiring accurate navigation. The ballistic
missile and the space programs of the 1960s era played a key
role. The endeavor to the moon by the Apollo program was a
noteworthy milestone in that evolution. Scientific process
applied to error model development produced essentially what
is known today about inertial system error behavior structure [ 1.

2,3,4,5,6,71.
Navigation, in general, is the determination of a physical bodys
position and velocity relative to some reference coordinate
frame or coordinate grid. The case of terrestrial navigation
involves the determination of a vehicles position and velocity
relative to the earth. The grid coordinates usually used for this
application consist of the spherical coordinates, latitude, longitude, and altitude. An INS utilizes the inertial properties of sensors mounted aboard the vehicle to execute the navigation
function. The system accomplishes this task through appropriate processing of the data obtained from specific force and inertial angular velocity measurements.
All INSs must perform the following functions:
1) Instrument a referenceframe, e.g. n-frame
2 ) Measure specijic force, f
3) Have knowledge of the gravitational field, model g
4 ) lime integrate specijic force to obtain velocity and position
The first function is accomplished by the use of gyroscopic
instruments. Typically three gyros are mounted, with their sensitive axes mutually orthogonal, on a rigid platform whose orientation is measured and tracked by either mechanical or digital
computer closed loop integration and control. The gimbaled
inertial platform can be mechanized to instrument, for example,
the local geographic coordinates, such as the commonly used
north, east, and down navigation n-frame. If three structurally
mounted gyros are used, the relative orientation between the initial and present vehicle coordinate frame can be determined.
Systems which analytically instrument a reference frame are
popularly referred to as strapdown (strapped-down) systems [ 1,
2, 3.4, 5, 6, 8.91. The current inertial navigation systems utilize the strapdown configuration, where typically ring laser
gyros (RLG), hemispheric resonating gyros (HRG), fiber optic
gyros (FOG), or other gyros, provide the angular orientation
information.
The second function, the specific force measurement, is accomplished with devices commonly called accelerometers.
Although there are many ways of measuring specific force,
most of the devices in common use are sophisticated variations
of the simple pendulum. The motion of the pendulous element
is related to the motion of the platform, or structural element
upon which the accelerometer is mounted, via Newtons second
law of motion. Three accelerometers with sensitive axes typically in mutually orthogonal orientation are required to measure
the three components of specific force.
3.4.1 INS Sensor Model

The inertial sensors provide measurements of inertially referenced Newtonian motion. The specific force f and angular
velocity 0 vectors are measured by accelerometers and gyroscopes, respectively. The specific force vector f,consists of
gravitational force vector g and the reactive force vector due to
inertial acceleration pfr
f =dr-g
= piv-g
where p is the time derivative operator, pfr is the acceleration
(second time derivative of position) with respect to inertial
frame i. Gravity g must be computed and subtracted from f in
order to extract the inertial acceleration. This satisfies the third
required function of an INS. With proper computations. information from the gyros provides the platform orientation, where
the platform can now be mechanized to represent a desired reference frame. Equation (26) is coordinatized in a selected reference frame, e.g. the navigation frame n, and solved for earth
referenced acceleration (rate of change of earth referenced
velocity) v
238
The generalized INS error state equation is of the form [ 1.71

(with symbols defined in Section 9.)
The product terms in Equation (27) represent the accounting for

the nonlinear Coriolis and centrifugal force terms. The processing includes double integration of the acceleration information,
starting with appropriate initial conditions, to provide the
desired position and velocity. The gyro outputs are given in
terms of an attitude variable differential equation as illustrated
by the direction cosine matrix (DCM) representation of a strapdown system
where
represents the relative orientation of the platform p-
frame with respect to the computation c-frame, flyp is the skewsymmetric form of the angular velocity vector between the inertial i-frame and the platform p-frame coordinatized in the platform p-frame, and the C17c represents the angular velocity
between the computation and inertial frames [I]. Similar differential equations exist for the other orientation parameter sets.
For detailed discussion see any of the following: [I, 2, 3.4, 5 ,
6, 8, 91, with [I] providing the best analytic discussion with
complete error analysis methodology. Equations (27) and (28)
are in the nonlinear stochastic differential equation form of
Equation (10). These two vector differential equations form a
set of nonlinear, coupled, second and first order differential
equations, respectively, which are easily convertible to a set of
nine coupled, first order differential equations and form the
foundation of the INS relationships. The initialization of an
INS includes the initialization of the orientation parameter sets
through a self alignment process using the information provided
by the measurements of the gravity vector and the earth rotation
vector [I]. These equations, when integrated from a set of initial conditions, provide continuous navigation solution in the
desired reference frame. However due to errors in the initial
alignment, gyros and accelerometer, and other sources, the INS
errors are nonzero and grow in time.
3.4.2
INS Stochastic Error Model
Britting [ 11, and others [2, 3,4,5, 61, develop error models for
various INS mechanizations. The basic INS error model is represented by nine fundamental variables, three position, three
velocity, and three orientation (attitude) errors. The errors form
an intricately coupled set of nine, linear differential equations
which are characterized in their dynamic behavior be nine
eigenvalues, corresponding to the dynamic modes of the system. These error dynamics are the linearized representation
(the A(t) matrix of Equation ( 5 ) ) of the nonlinear, coupled differential Equations (27) and (28). The gyro, accelerometer,
instrument misalignments, component transients, and other
errors are added to the basic nine error equations.
When the full error truth model is assembled it can exceed 90
states. There exist various versions of error models, with primary differences resulting from the choices of the coordinate
frame. For additional discussion on the error model development see references such as [ I , 3, 231. The Litton Guidance
and Control Systems LN-93 INS error model is used in each
illustrative analysis described in later sections.
3.4.2.1 Litton LN-93 INS Error Model
In the class of navigation sensors, the INS error model is likely

the most complex in structure and in the extensive modelling
work that has been required. Significant physical error sources
were identified and modeled and the components were extensively tested to validate the error models. This process is the
most significant step in any sensor error model development.
& ( t ) = A(t)6x(t) +Gw(t)

(29)
The nine general error states describe the fundamental INS
behavior [24]. Britting [I], (and others) analytically demonstrates that an invariant property of the error dynamics is represented by three fundamental oscillation modes. Errors of a
stationary INS consist of the classical oscillations called
Schuler (84.4 minute period), Foucault (24 hour period), Foucault-modulation of Schuler (projection of earth rate onto
Schuler), and the vertical channel instability. Elegant discussion of these topics is found in the referenced texts and literature. Other error states coupling into these general states,
grouped by categories, are added.
The Litton Guidance and Control Systems LN-93 INS error

model is used in the analyses presented. It defines 93 error
states (see [24] and Appendix). The Litton LN-93 documentation defines the earth e-frame, the true t-frame, the computer cframe, the platform p-frame, the sensor s-frame, the gyro gframe, the accelerometer a-frame, and the body b-frame [24].
For discussion of INS errors, the reader is referred to any good
text on inertial navigation systems, such as [ 1,2,3,4,5,6. 8,9,
221.
The LN-93 is a strapdown. local-level, wander azimuth INS.
Position is specified in terms of latitude, longitude, altitude, and
wander angle. Generally, wander angle is the angle about the
navigation frame z-axis which results when the INS platform is
not torqued to maintain precise (ENU) orientation. In this
mechanization, the platform is torqued only in the level axes in
order to maintain the local level orientation. Since the LN-93 is
a strapdown mechanization the torquing is performed in the
INS computer software. The WGS-844 model is used as the
ECEF reference for this effort:
Random constants (biases) represent variables which may take
on values within some specified probability distribution and
thereafter retain this value [7]. Approximately 64 bias error
states are classically lumped into the four subcategories: gyro
bias, accelerometer bias, trending, and gyro compliance states.
First-order Markov processes describes the gyro and accelerometer drifts and thermal transient errors characterized by random,
but time correlated, wandering. This process is modeled by a
first order lag driven by zero-mean white Gaussian noise [7].
Litton partitions the error states into the six subvectors [24]
6x1 represents the general error vector containing position,

velocity, attitude, and vertical channel errors. 6x2consists of
gyro, accelerometer, and baro-altimeter correlated errors, and
trend (first-order Markov) states. 8 x j represents gyro bias
errors modeled as random constants in the truth model and are
(later) modeled in the Kalman filter as random walks with small
magnitude pseudo-noises. &,+is the accelerometer bias error
state vector. 6x5 depicts accelerometer and gyro initial thermal
transient (first-order Markov) states. &, models the gyro compliance errors as biases in the system model and as random
walks in the Kalman filter.
The Litton LN-93 error model is inadequate in its depiction of
the baro-altimeter error sources in that it contains only a single
state for the barometric altimeter. Other authors discuss several
error sources normally linked with baro-altitude [25]. A revised
baro-altitude error model, developed and embedded in the INS
error model, includes states for correlated noise effects, bias
error, and scale-factor error [25. 261. These error sources are
4. GPS positions are based on an earth-centered. earth-fixed (ECEF) Cartesian coordinate frame designated as the WGS84. This reference frame designates the origin at the center of earth mass having the z-axis passing through
the instantaneouspole at the year 1984.
239
combined in a more complete error model. For the analyses

performed and reviewed in the following sections the 98 states
are reduced to 68 states, with the added baro states retained in
tact. These states are listed in the Appendix.
The vertical INS channel instability requires aiding by an additional altitude (or height) sensor. Classically, the baro-altimeter
has been used successfully in bounding this instability.
Although the baro-altimeter will remain aboard the aircraft for a
long period of time, today the GPS can also superbly perform
the vertical channel aiding function. For completeness, the baro
altimeter model is discussed in Section 3.5.
3.4.3
Other Inertial Sensor Applications
Barometric Sensor Error Model
Because the altimeter error affects many of the INS states, the
fidelity of the baro error model is important. The single baro
state included in the Litton error model is adequate to represent
only one of the major sources of baro altitude error. Other baro
altimeter error sources are statistically independent and should
be included in the truth model. The baro output error state, 6hB,
is directly coupled to four states in the INS dynamics equations,
and is indirectly coupled to several others. An adequate baro
altimeter error model, includes the timelposition varying bias
(function of atmospheric pressure-altitude spatial gradient and
vehicle speed) 6hpco.instrument dynamic lag 6h,, scale factor
error 6h,,, and an instrument bias 6h,.
Inertial components, particularly accelerometer and gyros, are

used in a variety of applications other than the inertial navigation systems. They may be mounted at various locations
throughout the aircraft to provide the vehicle acceleration or
rotation with respect to the earth frame. In these applications
the gravitational force components must be subtracted from the
accelerometer outputs, while the centripetal, coriolis, and tangential acceleration forces may be ignored due to their small
relative magnitudes. Often a flat, non-rotating earth model is
assumed as in an aircraft flight control system. This paper does
not elaborate on such applications, only mentions them for
completeness. The current trend is to include more of such
information for purposes such as boresight error reduction and
flexure control. When information from such sensors is used in
system-level integration appropriate sensor models are required.
The reader is referred to various texts, e.g. [23], and articles,
e.g. [27,28], for these sensor models.
3.5
3.5.2
The instrument dynamic lag error is typically modeled by first

order lag with a time constant on the order of I second. The
instrument bias is modeled as a random constant whose initial
1-0 value is chosen to be 15 feet. The choice is based on a baro
altimeter model contained in [26]. The scale factor error is a
multiplicative combination of aircraft altitude and the random
constant scale factor state. The initial 1-0 value for the scale
factor state is 0.01 [26]. Although the error contribution from
this state is negligible at low altitudes, it can be a significant
source of error at higher altitudes. The last error state in the
revised baro altimeter model is a first order lag which sums and
filters the previously described errors.
An example of a CADC (barometric) altitude error model, in
the form of a first order stochastic differential equation, is
Barometric Altimeter
Barometric altimeter provides an indication of altitude by measuring the static air pressure. Altimeter output is a robust
bounding signal to the INS vertical channel. Like most pressure
sensors, the altimeter transducer typically consists of an aneroid
bellows with mechanical linkages to a dial or drum display.
This sensor is rather inaccurate due to several factors. The relationship between the static air pressure and altitude depends on
current weather conditions. The altimeter must be compensated
with a bias input (altimeter setting). Transducer time lag,
instrument bias, and scale factor are significant error sources.
The pilot and aircraft system need for accurate air data (speed,
Mach number, altitude, rate of climb) drives the trend to use air
data computers.
3.5.1
Barometric Sensor Model
Typical central air data computer (CADC) provides a more

accurate static error corrected pressure altitude encoded for
automatic altitude reporting and in analog form for pilot display. Even the CADC corrected altitude is of limited accuracy.
The operational tolerance for the altimeter output error on the
ground is f 7 5 feet, and 53300 feet at altitude. Figure 4 illustrates
a typical CADC altitude and error model. The variables in Figure 4 are defined in the next section and Section 9.
atmospheric variation
inst
T~~ = SOOsec
T~~ =
I.Osec
true altitude h
. .
Figure 4. CADC Stochastic Model
The output error equation of this filter represents the total,

lagged baro-altimeter error 6hBis
6hB = 6h,+6hb+h(6h,,) + v
(32)
This error model represents a CADC corrected barometric

altimeter measurement error reflected in Figure 4. This model
framework is adequate for most air vehicle applications. For
those cases where the CADC internal compensation computations can affect either the dynamic or stochastic error, this
model can be enhanced appropriately to account for such processing.
3.6
Global Positioning System (GPS)
The Global Positioning System (GPS) is a navigation system

which transmits signals from which a user can compute navigation information. The three elements of GPS are the space,
user, and the control segments. The space segment, when completed in 1994, will consist of a group of 24 satellites in semisynchronous orbits around the Earth. Each satellite, known as a
GPS space vehicle (SV), transmits on two microwave frequencies L, and L,. Each frequency has binary phase-shift keyed
(BPSK) coded information as well as a 50 bps navigation message.
The user segment consists of all users of GPS information with
suitable equipment to receive the electromagnetic energy transmitted from a subset of visible SVs. Using a combination of
phase lock loop (PLL) and delay lock loop (DLL) techniques,
the receiver measures the apparent propagation time of the signal from each SV. The receiver determines the position of each
SV based on ephemeris data in the navigation message. Know-
240
ing the positions of the tracked SVs, the apparent propagation

times, and user time the receiver solves for its own position.
The control segment functions as the system administrator. It
consists of a master control station at Falcon AFB, Colorado,
and a few remote sites throughout the world. It is responsible
for maintaining the satellites and the data they transmit. This
includes updating the SV clock error coefficients, ephemeris
data, health status, etc. A thorough discussion of the GPS segments can be found in the literature [29].
Although the GPS can act as a stand-alone navigation system,
for dynamic applications it is most powerful when integrated
with an INS. Specifically, GPS pseudo-range measurements are
used to periodically improve the navigation system's accuracy.
A free-running INS is subject to various errors which generally
worsen with time. GPS range measurements, when processed
by an appropriate Kalman filter, estimate the INS position error
and can be used to improve position estimates. The Kalman filter combines satellite range measurements, with its own internal
models of the system to derive an estimate of the user position
error. This is then algebraically summed with the indicated
position to yield the best estimate of user position [30].
3.6.1 GPS Measurements
The GPS error model consists of three model domains. First, is
the deterministic dynamic behavior of the GPS errors. Second,
the measurement model functionally relates the measured quantities, the pseudo-range and delta-range, to the quantities of
interest, including the error variables. Third, the stochastic
attributes of the model describe that behavior which is caused
by the random nature of the measurement noise as well as the
non-deterministic character of the error which drive the
dynamic error variables.
3.6.1.1 True Range

The GPS receiver measures a range which is based on the time
offset between the SV generated BPSK code and the user generated code. When this time offset is scaled by the speed of light
c, a measure of range results. The true range, RI,is simply the
shortest distance between the two points in the three dimensional position space:
RI = lrrv-rul
(33)
When the SV (rSJand user (r,,) coordinates are expressed in the

same reference frame in R3 space, this range is
RI = &'xu)
'+ (ySv- Y , )
+ (zSv- z,)
(34)
Note that the range is a nonlinear function of the whole-valued

user ru and SV r,, position coordinate vector variables. These
vectors are frequently calculated in an ECEF coordinate frame,
e.g. r:", and r:. The whole-valued position of the user is generally expressed as a latitude, longitude, altitude triplet, which can
be converted to the Litton ECEF (or other desired reference)
frame by a transformation such as:
respect to the GPS time. The signal from the SV traverses the
ionosphere which distorts the path and causes an apparent time
delay. When the signal traverses the troposphere, the fluid
retards the propagation. The accuracy with which the user generated code can be correlated to the received code is a function
of noise, code loop bandwidth, and code chip length. The
uncalibrated propagation delay through the receiver also adds to
this error. The antenna-to-receiver cable propagation calibration error causes additional measurement error. However, this
error is often indistinguishable from the user clock delay and is
lumped therein. The raw pseudo-range model consists of the
true range and various time delays (scaled by c )
RRP
= R~+ Rclk, + Rcodc + R ~ r o p+ Riono
'elk,"
(36)
The range domain variables are defined in Section 9 and the following paragraphs. The white Gaussian additive measurement
noise v statistical properties are in Equation (13).
3.6.1.3 Compensated Pseudo-range
Three components on the right-hand side of Equation (36),

namely RlmP.Rio,, and R,,k,v,can be estimated open-loop. The
total tropospheric delay is a function of the GPS receiver altitude, elevation angle to the SV. and other environmental factors.
The ionospheric delay can be open-loop compensated using
dual-frequency reception or by a model involving total electron
count (TEC) estimates, elevation angle to the SV,and other
environmental factors. The SV clock advance can be open-loop
estimated by using the SV clock correction coefficients which
are transmitted in the navigation message. Even after these
three open-loop corrections are subtracted from both sides of
Equation (36), the left hand side can be called the compensated
pseudo-range, or simply pseudo-range. The corresponding
terms on the right-hand side are replaced by the residual error
terms after open-loop compensation. This yields the pseudorange error equation (with error variables defined in Section 9.)
Rp = RI
6Rc/km
&Rcode+ 6RIrop 6Riono+ 6R,lk + V (37)
3.6.1.4 Coordinate Transformations
In a typical modelling. simulation, and analysis task, several

different coordinate frames are necessary. Although the details
of coordinate transformations are not presented in this paper,
caution is necessary to correctly relate the required vector and
matrix quantities in proper reference frames. The reader is
referred to texts and other publications for explanations and
examples of such applications [ 1,22,28,24].
3.6.2 GPS Error Dynamics

The dynamic and statistical nature of the error variables is
developed in a form suitable for Kalman filter implementation.
First, we develop the range error dynamic relations as a system
of first order linear differential equations in variable x driven by
white Gaussian noises w. These equations follow the format
discussed in Section 5.2.
3.6.2.1 User Clock Error 6R,,,
U
( R , + h ) coscpsinh
=
= [ [ R , ( l - e ; ) +h]sin.]'
(35)
( R , + h ) coscpcosh
Generally, user equipment sets use a quartz crystal oscillator

which is less precise than the SV clocks. The user clock error is
modeled as an offset and drift from GPS system time. Over a
typical mission duration, the user clock error 6RCIkcan be
U
The symbols are defined in Section 9. Equations (33) and (34)

are truth model representations of the range. In a simulation
environment the values of x,, ym z,, xu,yu. z,,. (or 4, h, h) can
be provided as true, whole-valued trajectory data.
modeled with two-skltes
3.6.1.2 Range Measurement Errors, 6R
where x , is the range equivalent of user clock bias, and x2 is the

velocity equivalent of user clock drift. The numerical subscripts on these and subsequent GPS states are specific to the
development of the GPS truth model; the same states are num-
The measurement of true range R,, called raw pseudo-range, is

corrupted by several errors, the most significant being the user
clock bias [31, 321. The SV clock itself may be in error with
24 1
bered differently in the combined GPSlINS model (Appendix)

The stochastic properties for these random variables (states) are
[331
(to) = 0
(39)
sd,, =
:
i 9
40
where oIz= 9x1O14ftZ, 022= 9X10'0fr/secz for a typical user

clock bias and drift. Note that the Equation (38) is a stochastic
differential equation describing the code loop emor dynamics of
the form: i= A x + B u + G w ,with u=w=O. These random
variables are assumed to be zero-mean Gaussian processes,
where Equations (39) and (40) describe the mean P c l k y ( t o ) and
'rr
-1
'11
--I
0 - 0
'IIr
-I
TI!
(47)
0 o;, 0 0
0 0 o;, 0
covariance P,,, ( t o ) initial conditions.
0 0 0 o;,
3.6.2.2 Uncompensated Code Loop Error Mcde
--I
where d,,
= 1.M. The driving noise q, = 0.004 f$/sec?.
E{w,,)
The GPS phase lock loop can be modeled as a first order lag
driven by white Gaussian noise, with a time constant of approximately I second [34]. Assuming that the user set has a separate channel for monitoring each of four SVs, and assuming no
correlation between channels, the dynamic pations for user
set code loop am (with z,= I sec)
=o
(49)
3.6.2.4 Uncompensated Ionospheric Range Error Mi
The ionospheric-induced range error can exceed 150 feet [31].

This error varies as a function of solar activity, time of day, etc.
and is frequency dependent. Single L-band receivers can correct for some ionospheric error. GPS receivers with dual Lband capabilities can compensate for most of this error. However even after open loop compensation, the residual error can
be about 3 meters [31]. This can be further reduced with multiple measurement time averaging subject to temporal and spatial
ionospheric delay gradients. Whatever the uncompensated ionospheric error is, it is expected to behave dynamically in a manner similar to the tropospheric error, but with mutually
uncomlated statistics. The form of the ionospheric error model
is again a stochastic differential equation driven.by a white
Gaussian noise with appropriate initial conditions and statistics
where :
o = I.Oj8, for i=3,4,5,6. The driving noise is
E{wcode} =
where qc = 0.5f?/seC?. The code-loop error stochastic differential equation is in the form of Equation (38). but with
w,,,, ( 1 ) f 0. Equations (44) and (45) describe the first and
(51)
with T;= 1500 sec. It should be noted that this ionospheric

delay error time constant is based on the work of [31,32, 351
and may require additional empirical validation and revision as
necessary. However the basic structure and the chosen parameters should suffice for many applications.
Pi(to) = 0
second moments of the white Gaussian driving noise probability density function.
(52)
[o; 0 0 01
3.6.2.3 Uncompensated Tropospheric Range Error 6Rlr
0 o ; o 0
The range error introduced by tropospheric delay can be as

large as 80 feet, and is a function of altitude, environmental
conditions, and SV LOS elevation angle [31]. Much of this
error can be open-loop compensated, with the residual error
modeled as first order Gauss Markov with z~,=500 sec [34]
0 O o f O
10 0 0 oi
8 i
= 1.w.The driving noise w with qi = 0.004f8/seC?
(53)
242
3.6.2.5 Uncompensated SV Clock Range Error 6Rclks.
The GPS navigation message contains correction coefficients

for SV clock bias, drift, and drift rate. Open-loop corrections
are accurate to about 1.0 meters [31]. This error is assumed to
be a relatively constant over time, albeit random, thus it is modeled as a random bias. The dynamic equations for SV clock
advance range error are:
velocity error data, including the error statistics and dynamics,

should be available from the GPS management. Each satellite
position error is in three dimensions, thus 12 first-order Markov
process states are required. In the actual Kalman filter, this
three dimensional error does not entirely affect the user measurement. Only the scalar projection of the three-dimensional
true error vector unto the line-of-sight (LOS)affects the measurement. Thus the filter requires only a single state for each
satellite line-of-sight to account for this effect. With time, as
the error vector changes its orientation with respect to the LOS,
the projection magnitude also varies. The magnitude of this
slow variation is a small contribution to the overall pseudorange measurement error. Thus a useful truth model for GPS
satellite position errors is
;r
with=,z
...
01
1500 sec
[o: 0 0 01
where &,= 25f8. The driving noise statistics are

E {WCIk,1 = 0
where bzSy=25f?. The driving noise statistics are described
completely by the first and second Gaussian moments
E{w,,1 = 0
where qs = O.Ofl/sec? for the truth model, and qs = I.OxlQ'o
f8/sec? for the filter model. The non-zero driving noise in the
Kalman filter is used to keep the filter gain, associated with this
set of variables, from approaching zero. The non-zero, but
small, value keeps the filter gain open to permit parameter variation, or other model inaccuracies from detracting from filter
performance. This is one of various filter tuning artifices discussed in various texts, such as [7,161.
3.6.2.6 Satellite Position Errors 6rs,
In addition to the above described 18 error states the satellite

position and velocity errors also affect the navigation solution
accuracy. The satellite position and velocity are known with an
accuracy limited by the ground segment of GPS. The ground
measurements are accumulated from the monitor stations, processed in a joint Kalman filter, and uplinked to the satellites.
Even at the time of highest achievable accuracy, the satellite
ephemeris errors are non-zero. More significantly, these errors
grow in time due to various random orbital perturbations. The
position and velocity errors increase with time, from the last
update, at some rate. The time-tagged position and velocity
vectors are transmitted to the user, however the transmitted
quantities differ from the true ones. For most applications, the
SV position errors alone are sufficient. Should an application
require velocity error models, similar methodology can be used.
The following model develops only the SV position errors.
The magnitude of these errors can be up to 2-4 meters, as
reported in early literature on GPS. Actual satellite position and
E{ws,(r)W,T,(t+T:)l
(64)
=
0
... 9,"
The GPS error truth model state vector consists of 30 elements

(see Appendix for a complete listing of these states). In the
actual Kalman filter only one state per LOS is required. In fact,
attempting to estimate the other eight components of this error
may lead to filter instabilities. This is due to the fundamental
unobservability of the SV position error components which are
orthogonal to the LOS.
3.6.3 Differential GPS

Differential GPS concepts are described in Section 6. The error
models for differential GPS applications are derived from those
of the absolute GPS. As is explained in that later section, the
differential GPS error model, basically consists of variations of
the model derived in this section.
3.7 'kansponder Error Model
The Range/Range-Rate System (RRS) is a navigation aiding
system which comprises a critical part of the Completely Integrated Reference Instrumentation System (CIRIS). The RRS is
used in the Navigation Reference System (NRS). It is an example of a system utilizing measurements from ground-based
ranging transponders to obtain relative navigation information.
243
The RRS is quite similar to the orbiting transponder system,

that is the GPS. Each provides relative range and range-rate
measurements. In fact, during early stages of GPS testing, an
inverted GPS system was used. The GPS transmitters that were
placed on the Yuma Proving Grounds surface, performed a
ranging function similar to that of the RRS. The few differences in the resulting range and range-rate measurements are in
the character of the measurement errors. Six to ten transponders are typically used for Kalman filter updating.
modeled by zero mean random bias states, and the atmospheric

error states are represented by first order Markov processes with
a time constant of 300 seconds. For each transponder, four
states are defined [36]
0
pa,, ( t o )
25ft2 0
0 25fr2
Navigation information is obtained by interrogating ground

transponders and subsequently processing the signals with
which the transponders reply. The received information allows
high quality range and range-rate measurements to be calculated by the RRS interrogating hardware [36]. Using these
range and range-rate measurements, refinements to the NRS
position and velocity estimates are then possihle.
The RRS transponder measurement error model is in a form
similar to that of the GPS [36. 371. The lever-arm effect discussed in [36] is not included in this work. The lever-arm effect
must be considered in hardware applications. Its omission in
these studies does not change the validity of the conclusions.
The RRS range measurement is computed from the time delay
detected between the transmission of an interrogation request
and the reception of a reply from the transponder. This temporal difference is scaled by the speed of light, c/2, to obtain an
uncorrected range measurement. Correction factors are applied
for tropospheric propagation delays and calibration residual
[36]. The uncorrected range measurement RRRSis (with variables defined in Section 9.)
R,,,
= R , + 6Rafm+ 6Rbr+ v
(66)
A difference of two measurements is useful. The first is the
RRS range measurement. The second is the range computed
from the INS-indicated and surveyed-transponder positions.
The INS and transponder positions are represented by vectors
expressed in the Litton ECEF frame. The calculated range from
the INS to the transponder is given by
This equation is linearized about the error variables 6x, and 6x,
in a Taylor series expansion truncated to first order [ 1,7]
Note the unit-line-of-sight vector, uLoselements in the brackets.

The difference measurement is formed
6~ = RINS-RRRS
(69)
The true range present in each range measurement is cancelled
[ 1.73. Also note that the bracketed coefficients in the equation
above determine the Kalman filter's H matrix.
The RRS error state vector consists of 26 elements. The first
two states are zero mean random biases which model the airborne equipment range and range-rate calibration errors. The
initial covariances for these states are
These two states apply to all RRS measurements. There are two
error sources unique to each transponder. First is the error due
to R' transponder surveyed position uncertainty (x, y. z components in ECEF frame), and second is the error due to atmospheric propagation delays between the user and each
individual transponder. The three position error sources are
with o2=1Q'". Equations (71) and (72) are for a single transponder. Six RRS transponders are used in this model.
The Kalman filter combines range measurements with its predicted measurement to calculate an optimal estimate of the state
vector. The indicated positions of the transponder and user are
modeled by
= x, + 6x,
(73)
i" = xu+ 6x,
(74)
XJ
The truth model measurement equation is

z = uLOs( 6 ~-,&xu)-6Ra,,-6Rbr + v
(75)
Caution is required in performing the computations. It is necessary that all of the vector operations be performed in the same
coordinatized vector space. Typically the ECEF is chosen for
these computations.
3.8 SAR Model.

Synthetic Aperture Radar (SAR) is an airborne radar capable of
generating high resolution images of surface target areas and
terrain. A fundamental limit in angular resolution ares of an
imaging system is due to the diffraction limit and is defined by
a,,,v = 1.22 ( D / h ) , where li is the electromagnetic wavelength, and D the system aperture (lens, mirror, or RF antenna).
Thus the larger the aperture, the smaller minimum resolvable
angle (higher resolution).
The term synthetic aperture is used because SAR radar utilizes
the motion of the aircraft, or other platform, to synthesize a
large aperture antenna from a smaller physical antenna aperture.
It accumulates motion-induced, spatially-distributed information over a period of time and phased properly to within a fraction of the wavelength h. As a result, high resolution images
are obtainable using an lower instantaneous resolution antenna.
High resolution SAR radar images are employed in various
applications. As with any physical device there exist measurement uncertainties (errors). Combining the SAR measurements with other sensors provides an opportunity to reduce the
measurement errors. This discussion, presented in [7, 39, 40,
41,42,43,44] provides the SAR measurement model. The
goal is to develop key ideas sufficient for implementation in a
centralized Kalman filter.
3.8.1 SAR Targeting Background
The basic SAR geometry is shown in Figure 5. The SAR is
characterized by a wide beam antenna which illuminates a large
area on the ground. For an assumed instant in time, the SAR
operates by transmitting a radar pulse and samples the magnitude and phase of the return signal. These samples are stored
into a vector of data generally referred to as a range bin. As the
aircraft travels along a given ground track, the process of illumi-
244
nating the ground with a radar signal and sampling the magnitude and phase of the return is repeated every T seconds for
consecutively overlapping areas on the ground. The result is a
two dimensional array of data which contains magnitude and
phase information of radar returns in both range and azimuth
directions.
3.8.2 Coordinate Frames and Transformations

The coordinate frames and coordinate transformations matrices
needed for radar measurements are the navigation n-, body b-,
and path p-frames defined as a right handed orthogonal set. The
path frame defines the attitude and direction of flight relative to
the navigation frame via roll, pitch, and yaw angles U,,(+and
4. Its origin is at the aircraft center of mass with 9pointing in
the line-of flight (LOF) direction, >p points out the right wing,
and f pointing through the bottom of the aircraft'. Coordinate
transformations between frames is required to perform vector
additions and multiplications.
3.8.3 Transformation Matrix Perturbations
Error analysis utilizes perturbation methods which linearize the
nonlinear differential equations [ 13. For coordinate transformations matrices, this involves approximating the indicated transformation obtained from the INS as the sum of the true and
indicated transformation matrix error
The SAR image alone cannot provide all the information

needed to determine the target location, nor is it used for aircraft
navigation information. The primary function of a SAR radar is
to generate high resolution radar images of ground terrain. As a
result, SAR images are employed to achieve the first task for
target location, namely target designation. The operator designates the target by manually placing a cursor over the desired
target image. SAR systems must employ other modes of operation to actually locate the target. These modes generally consist
of conventional radar techniques which measure parameters
associated with the target location. For instance, the range (R)
and range-rate ( d )for each sampled cell in the image may be
measured using conventional range-Doppler radar. These quantities are computed
R = -cT
2
.
R =
fd
(77)
where c is the speed of light, Tis the round-trip transit time of

the transmitted pulse, fd is the Doppler frequency shift, and h is
the wavelength of the transmitted signal.
SAR target location measurements are often classified by two
basic types: range-Doppler measurements and range-monopulse
angle measurements. Both of these methods provide just
enough information to compute the target location with only
one set of measurements. When classical observer or Kalman
filter is used with multiple-look measurements, it is possible to
locate a target with a smaller set of measurements. Also, the
Kalman filter provides a means to continually improve the estimate of the target location by utilizing the complete measurement history. In this report, we derive the Kalman filter
measurement equations for both the range-Doppler and the
range-monopulse measurements. Presented next is a brief
explanation of both types of radar targeting methods.
q=q++q
(78)
The indicated transformation matrix errors are 6Ci and
<,
6q.
Orthogonalizing
about a small angle perturbation matrix
askprovides the required linearized error model.
(<),=
(1-a)q
(79)
where askis given by a skew-symmetric matrix representation
qY 44 1451
of the rotation error vector & =[b,
@*k
Ly;J
$z
-+,
This attitude error matrix a s k (or rotation vector

usable in a Kalman filter [45].
(80)
is directly
3.8.4 Range and Range-Rate Measurement Model

In this section the measurement equations for the radar range
and range-rate measurements are presented. The basic vector
geometry of an aircraft and a target in an earth centered earth
fixed (ECEF) coordinatized frame is shown in Figure 6.
aircraft
Figure 6. Target Location Geometry
In the range-Doppler method, range and range-rates are measured and stored for each cell in the SAR image. In addition to
range and range-rate measurements, this method requires a
radar altitude measurement for computing the location of a
ground based target.
The vector r = Ru , where R is the vector magnitude, and U is

the unit vector indicating the vector direction, such that tu1 = 1,
and Irl = R = uTr. The time-rate of change of vectorr is:
The range-monopulse method utilizes the range and precise

angle measurements to compute the target position. In a
monopulse radar, two spatially separated antenna form two similar beams. An RF signal receiver/combiner. obtains a sum and
difference outputs. The antenna pattern for the difference port
exhibits a null directly between the beams. Likewise, the pattern for the sum port exhibits a peak directly between the
beams. This forms a precise pencil-beam tracking and/or angle
measurement radar.
where d is the range-rate, and U is the line-of-sight (LOS) rotation rate. Range-rate can be expressed (using Equation (81),
uTu = 0,and uTu = 1)as
d
d
i = -r = - ( R u ) = R u + R u
dt
dt
R =
UTi
(82)
5. This problem is simplified by the assumption (when valid) that the longitudinal axis of the aircraft is aligned with the velocity vector.
245
The ECEF coordinate frame is shown for clarity, however, we

assume that all equations are solved in the navigation frame.
Note from Figure 6 that the vector raris computed from r, and r,
by the relationship r,, = r, - r a , where vectors raf,r,, and r,
are expressed in ECEF coordinates. Thus
RaPat = RPt-Rauo
(83)
= U:, (&, + RfUf- Raua- ROUO)

(84)
In special cases Equation (84) is simplified. For example when
the target is stationary RI = 0 and li, = 0 resulting in
Rat
RaI = -uzIia
t
3.8.4.I
range and azimuth, is assumed equal to infinity. The SCNR is

computed and compared to a threshold (of IOdB) to establish that
the individual targets are recognizable by a human operator.
SCNR =
SNR
CNR+0.1
The SNR is given by

r0
SNR = p,AC2LaL,Tl(-)r
where p, is the target cross-sectional area. For a side-looking

SAR, range and azimuth resolutions are given by
(85)
Radar Measurement
As previously stated, the radar is capable of obtaining an indirect measurement of R and d . However, these measurement are
not perfect and therefore must be modeled as the sum of true
value and the measurement noise. R and R denote the radar
measurement, and may be modeled by
Ay =
D
-
2
respectively,where c is the speed of light, B is the bandwidth of the
transmitted pulse, z is the unmodulated pulse length, and D is the
antenna diameter. For a spotlight mode SAR, the range resolution
is the same as above and the azimuth resolution is expressed as
R = R+nR+vR
(93)
R = i i + nR. + vR.
(87)
where nR and n R. denote colored measurement noises, vR and

v . denote white Gaussian measurement noises.
R
In general, the correlated measurement noises nR and n . are

R
composed of many error sources. With an appropriate Kalman

filter design it is possible to estimate the specific errors which
make up the colored measurement noises nR and n R.. provided
that these errors are individually observable. This requires that
the errors comprising nR and ni be modeled as Gauss-Markov
processes driven by white Gaussian noise. Likewise, the error
sources comprising vR and v . must be modeled in a truth
R
model. The truth model simulates with higher fidelity the true
radar measurements.
The primary measurement noises observed in the range measurement include the target designation error and the clock
error. Other errors which may also be included are multi-path
and range glint errors. These are not modeled here because they
are not always present in the measurements.
The target designation error is often modeled as a function of four
parameters, namely clutter-to-noise ratio (CNR), signal-to-noise
ratio (SNR), signal-to-clutter-plus-noiseratio (SCNR),
and the resolution cell size. The CNR is computed to establish that the image
quality is sufficient for individual fixed targets to be designated.
with
=
a
10(2.16x10~')r
atmospheric loss
L, = 10 (5.4xIO')qr
rain attenuation
p, =: 0.1 ( A R ) ( A a z ) clutter cross-section
11 is the rain rate in mm/hr, R is the range to the target, AR and

Aaz are the range and azimuth resolutions, respectively, and ro
is the range at which the SNR is unity for a l m 2 target, I sec
array time, and undegraded main-lobe antenna gain. A typical
value of ro is 1.5x105m. If the CNR ratio is less than (3 dB), the
resulting image is clutter. Consequently, when the CNR is less
than 3dB, the variance of the target designation error, in both the
where h is the wavelength of the transmitted pulse and @ is the

angle between the aircraft and the target created by aircraft
motion during SAR image data collection. The designation
, azimuth direcerror in the range direction, denoted ~ R Dand
tion, denoted 6vD,
have standard deviations given by
(T
RD
(T
VD
= KAR
(94)
= KAv
(95)
respectively, where typical values of K are on the interval [0.5,

2.51. In [40] K is modeled by the following expression
K =
1
2 (SCNR)
This choice of K is given without discussion.

Two possible scenarios exist for SAR measurements each
resulting in entirely different spectral characteristics for designation error. The first is a situation where SAR measurements
are made only after target designation. In this scenario, the target designation errors in both range and azimuth directions are
modeled as white Gaussian noises
6RD = wR,
(97)
6VD =
(98)
w VD
where wR, and wyr, denote white Gaussian noises. The second
scenario consists of initially designating the target with a SAR
then tracking it with monopulse radar. The errors present in the
radar tracking mode are a direct result of the initial designation
errors. Consequently, in this scenario a random bias model is
appropriate for the designation error
6RD = 0
(99)
8jlD = 0
(100)
In the Kalman filter design, a weak white Gaussian pseudo-noise
is added to Equations (99) and (100) to compensate for possible
hardware failures and model uncertainties. The clock error 6T
contribution to the range measurement 6Rc,, is expressed by
where c =: 3.0 x IO*( n d s e c ) is the speed of light. The clock

error 6Tcan also be modeled as a random bias
246
6T= 0
(102)
Again a weak pseudo-noise is added in the Kalman filter to
compensate for model inaccuracies.
The primary error in the range-rate measurement is the rangerate ambiguity error. Other errors which may also be included
are the errors associated with surface wave motion when targeting is performed over bodies of water.
To formulate the framework for incorporating the angular radar

measurements, other-source derived LOS angles are again
required. These are again generated from the INS provided
information. The two error angles, azimuth 6~ and elevation
60 are given by [45]
The range-rate ambiguity Aka,,,b represents the smallest difference between two range-rates that could be detected by the
radar and is characterized by
60 n - R V 0 ( u : ~ ) $,I ~+ v 0 6 4 - V06<
where V 0 is the gradient of
where z is the radar pulse length and fo is the frequency. As an
approximation, we assume that the error in range-rate 6Rambr
resulting from the range-rate ambiguity, is white Gaussian noise
with a standard deviation of Akamb
v0 =
(109)
e, and defined as
-- (U;) nT [ I - u : ,
Rsin (0)
(U:,)
3.8.5.1 Radar Measuremenr

A monopulse radar is capable of obtaining direct measurements
of azimuth and elevation angles. However, these measurement
are not perfect and therefore are modeled as the sum of the true
where
is the standard deviation of
(I .
6Ramb.
Use of this
Ram6
approximation results in an error that is slightly larger than the

true errors that are normally expected from this source.
3.8.4.2 Error Measurement Model

The inputs to the Kalman filter are the differences between the
indicated (e.g. INS) and the radar measured values. The range
measurement model, for the general case where target and aircraft may both be moving with respect to navigation frame, differences the INS computed range El and the radar measured
range Rs, and is expressed in terms of Kalman filter error states

6r,,bo,
and nR [451
zR = R I - R s + n R + v R
T
= (U:,) (6r:
- 6ra") + nR+ vR
(105)
The range-rate measurement is similarly formed from the INS

computed, RI, and the radar measured, Rs range-rates. An
example of such measurement for the case where the target is
stationary with respect to earth's surface is [45]
,nT
-ra
= -[ I - U : , U ; , ~ ] [ 6r: - 6r:]
value and the measurement noise. If $ and 0 denote the radar

measurement, then they may be modeled by the following equations
+ nd + vi
(106)
3.8.5 Azimuth and Elevation Angle Measurement Model

Next we present the measurement equations for the monopulse
radar azimuth and elevation angle measurements. The basic
vector geometry defining the azimuth angle y~ and elevation 0
angle are shown below in Figure 7.
= w-q,-v,
0 = O-qe-ve
where q, and qe denote colored measurement noises and where
v,,, and ve denote white Gaussian measurement noise. In reality,
measurement noises q,,,, qe, v,,,,and V e consist of a number of
physical error sources. Next follows a brief description of the
primary error sources for use in the Kalman filter truth model
design.
The primary error source for azimuth and elevation measurements include: target designation error, antenna boresight error,
radar-INS harmonization error, antenna servolpointing error,
and radome correction error. First consider the effects of the
designation error. In particular, consider the effects of an azimuth designation error on the azimuth angle. A similar development applies for the effects of range designation errors on the
azimuth angle as well as the effects of both an azimuth and
range designation errors on the elevation angle.
An azimuth designation error may be modeled as a vector
which points in the direction of the aircraft velocity vector at the
time that the image was formed. For detailed discussion the
reader is referred to [45]. The radar azimuth designation errors
are presented
6$,
= vVu:sVD
(1 13)
60, = veu:6ylD
(1 14)
The two radar line of sight (angular) errors are
I/
Clov"d PI-
Figure 7. SAR Angular Geometry
The boresight, radar-INS harmonization, antenna servo/pointing. and radome refraction errors have both an azimuth and an
elevation component which directly affect the azimuth and elevation angles, respectively. The azimuth and elevation errors
consist of several physical components. With the exception of
the radome errors, each of the error sources is characterized as
random biases. In general, the radome errors are spatially
241
related to the azimuth angle y and elevation angle 8. However,

often targeting is performed at large stand off distances with the
aircraft in essentially a straight and level flight. Given this scenario, the azimuth and elevation angles change very little. In
those cases they may also be modeled as a random biases.
3.8.5.2 Error Measurement Model

The inputs to the Kalman filter are the differences between the
INS indicated, and the radar measured values
z, =
? - $ = - R V v u k- $ + V y 6 < - V v 6 r : + q v + v v
(117)
- RVBu:,.,$ + V96r: - V96r: + qle+ V ,
(I 18)
ze =
6 -9
3.8.6 SAR Summary

I
The radar and SAR error model presented is neither complete,

nor unique. It does present a framework for development of the
required models. It is important to emphasize that required is
relative to the task at hand. The truth model must always be of
sufficient quality, such that the unmodelled errors would provide in the physical world a contribution which in effect is
below the noise level of the system.
3.9 Other Sensors

As presented in the Section 1.. the benefits of deep integration,
go far beyond INS aiding. When a group of sensors, which
share the same geometric vector, are processed jointly the natural result is that the stochastic cross-correlations are optimally
computed. These cross-correlations become the agent for benefitting each of the sensor. The geometric errors of these geometrically related sensors are cross-calibrated, to the maximum
degree possible (optimally).
A large variety of additional sensors, and other navigation
update inputs have been demonstrated as useful in updating the
INS. Some examples are: doppler radar, radar altimeter, map
correlation, starlsolar sensors, pilotage inputs, and others. The
integration of those sensors for navigation updating has been
studied, and reported on, by various organizations, corporations, and individuals. For an example, see Models f o r Aided
Inertial Navigation System Sensor Errors by The Analytic Sciences Corporation (TASC) [38].
3.9.1 Angle (Azimuth, Elevation) and Range Sensors

One application of deep integration is the combining of the navigation sensors with other mission sensors, such as a laser radar
(LADAR) or a synthetic aperture radar (SAR) which are capable of providing navigation update information as well as targeting information for a weapon system. The key advantage
possible with deep integration is the calibration of the line-ofsight (LOS) and range bias and scale factor errors. Multiple
looks by these sensors at calibrated and surveyed targets build
up over time strong cross-correlations. These cross-correlations
develop when the same common geometric vector reference is
varied in time. This is analogous to the concept of inducing
observability in the estimation theory [7]. The proper modeling, analysis, and simulation provides the baseline of performance and a foundation for implementation of trade-offs. The
models for a LADAR or a SAR follow the framework developed earlier. Adequate models for the geometric vector component errors must be expressed in terms of their dynamic,
stochastic, and measurement characteristics.
3.10 Geometric Sensor Model Summary
This section presented several models, of varying fidelities,

which are deemed of sufficient quality for the analyses that had
been performed in various studies and summarized in the following sections. The models presented in Section 3 were not all
developed fully. Reader is referred to the referenced papers and
reports for sufficient level of detail. The intent in this paper is to
present the concepts of geometric vector commonality among
all of the sensors modeled. This commonality, when represented in appropriate mathematical models suitable for Kalman
filter implementation, offers opportunities to jointly estimate
the errors that are geometrically related. This joint estimation
is the key to substantial sensor error calibration and registration
against some reference points, such as the illuminated target.
When combined with the GPS. for referencing within the GPSdefined framework, tremendous opportunities are opened for
relative registration on a global scale. This registration can be
in relative GPS, differential GPS, or absolute GPS domains.
Again the key to this precise, yet globally available, sensor calibration and registration is joint modeling and estimation with a
single Kalman filter.
4. CIRIS: EXAMPLE OF GPS/INS INTEGRATION

CIRIS is a transponder aided INS test reference currently used
by the Air Force for the development and testing of aircraft navigation systems [37,46]. It is discussed in the transponder error
model (Section 3.7). The office of primary responsibility
(OPR) for CIRIS is the Central Inertial Guidance Test Facility
(CIGTF), 6585th Test Group, Air Force Materiel Command
(AFMC), Holloman AFB, NM. The operation of CIRIS
involves flying the INS (or other systemj to be tested, referred
to as the test article, and the CIRIS through a pre-planned trajectory across a transponder range while data from both is
recorded in-flight. The test article performance is compared
(post-flight) to CIRIS.
CIRIS has served as the navigation system test standard since
becoming operational in 1975. Until recently, CIRIS has been
considered an accurate test baseline for determining the performance of aircraft navigation systems. CIRIS obtains an
accurate navigation solution by combining information from
three major subsystems: INS, barometric altitude (from
CADC), and rangehange-rate (RRS) transponder aiding [47,
481. CIRIS determines the aircraft latitude and longitude
with an accuracy of 14 ft horizontal and 40 ft vertical; the
north and west velocity to 0.1 fps, and the vertical velocity to
0.4 fps (1-0) [37, 461. Improved navigation systems already
exceed the CIRIS accuracy. CIRIS is being enhanced to provide a navigation solution an order of magnitude more accurate.
By augmenting with DGPS measurements, it is possible to
increase the CIRS navigation accuracy by an order of magnitude. The system presented in this section is called the Navigation Reference System (NRS). The higher CIRIS accuracy
is possible by combining the strengths of the three systems.
An extended Kalman filter which incorporates an LN-93 INS
model, transponder aiding (as implemented in CIRIS), and GPS
aiding is designed and analyzed. Solomons research specifically addresses GPS-CIRIS integration [37] while research performed by Snodgrass [36], and Cunningham [ 191 address the
problem offilter-driving-filterinstability which can occur when
the integration is inadequately engineered. Solomons GPS
error model [37] and extensions developed at the Air Force
institute of Technology (AFlT) form the GPS error model basis
(Section 3.6.)
The following assumptions are made in this study. The truth
models are of sufficient quality for these analyses and the conclusions drawn. The INS LN-93 error model, with the barometric altimeter error modifications [24], the RRS transponder error
model by Snodgrass [36], and the GPS error model based on
articles by Cox [30]. Martin [31], and Milliken and Zoller [32]
form the truth error model basis.
MSOFE is used to establish a performance baseline against
which other full-order and reduced-order filters are compared.
Time histories of variables such as aircraft position in R3 space,
attitude, and acceleration are created, and serve as the true aircraft trajectory in MSOFE simulations.
248
Extended Kalman Filter

A navigation solution is computed from a set of non-linear, coupled, deterministic, differential equations, such as illustrated in
Equation (27). The accuracy of the solution depends on the INS
error dynamics which are driven by several random sources
whose statistics are known with varying degrees of accuracy.
Given a-prioristochastic and dynamic models and initial conditions, the Kalman filter combines all measurements to produce
optimal estimates. A system model must be linear in order to
satisfy the optimality assumptions of a Kalman filter. Extended
Kalman filter allows for relinearization about the best estimates
at each computation cycle, thus enhancing the validity of the
linearization process and improving filter performance [7].
4.1
4.2 INS and Altimeter Models

The Litton LN-93 strapdown INS.is chosen for use in the proposed Navigation Reference System (NRS). The INS error
model with the revisions made to the Litton baro-altimeter
model, are presented in Sections 3.4 and 3.5. The INS subsystem, aided by a baro-altimeter, is combined with the RRS
and GPS models to form the NRS. See Sections 3.7 and 3.6 for
the respective error model descriptions.
order and reduced-order Kalman filters may be tested against a

common, high-fidelity standard.
Several system-level configurations are tested and analyzed
using Monte Carlo simulations. Generally, analysis of a configuration consists of performing a series of 10 alignment runs,
followed by a series of 10 flight simulations. The stochastic
time history of error-state variables of interest is recorded. In
the case of large dimensioned models with many measurement
updates, fewer runs may be used, somewhat reducing confidence in the data sample statistics [7]. At the beginning of each
alignment run, the truth model state vector is initialized. Each
truth state is set to a random value based on the states initial
covariance and a random number to represent actual INS, RRS.
and GPS error condition. The intent is to determine the effect
of the randomization of truth model states on the Kalman filters
ability to perform its estimation task. Having initialized the
truth model error states, the system is allowed to propagate its
states for the period of the eight-minute alignment. During the
alignment, the Kalman filter is provided with alignment measurements. At the end of the alignment runs, terminal conditions for the truth and filter state vectors and the final
covariances are used as the initial conditions for flight simulations. Thus, a simulation is continuous for the entire analysis
sequence.
4.6
Transponder ID I
Latitude
005
33 01 36.14
Longitude
-10608 20.74
I
1
Altitude
4339 ft
102
32 55 58.59
-106 08 50.33
4074
181
33 44 58.03
-106 22 14.63
7932
21 1
33 17 55.99
-106 31 44.31
8842
3247 16.41
I -10549 15.47 I
9202
32 42 12.23
-106 07 38.90
448 I
212
216
Results
In the next step the RRS transponder aiding is added. The 26

RRS states are combined with the 72 INS states to produce a
98-state model whose performance establishes the baseline
INSRRS performance. The impact of providing RRS measurements to the Kalman filter is dramatic indeed. An improvement
for the alignment simulations is evident in the horizontal position estimates. Latitude and longitude mean errors are closer to
the zero-mean as predicted by estimation theory. Comparison
of the vertical scales for horizontal channel states shows that the
addition of RRS transponder measurements during alignment
reduces the true- and filter-computed 1-0 values by more than
50 percent. A more substantial effect is evident in the flight
runs. With RRS transponder measurements, horizontal position
errors are reduced to less than 40 feet during all phases of the
Bight. Variations in the tme and filter one sigma values are due
to two sources. First, significant transitions occur in the altitude, velocity, and tilt states during periods of high dynamic
maneuvering. Second, the aircraft range from the transponders
has an obvious effect in horizontal position errors. Flight
regimes in which the aircraft is at low altitude or is a long distance from the transponders result in increasing position uncertainty. When the aircraft is high overhead the transponder
locations, much better estimates are possible. In all cases, the
filter appears to be reasonably tuned in that the filter 1-0 values
bound the mean error fb,,, traces the majority of the time,
without being overly conservative (Figure 8).
I
4.5 Simulation
The system truth model is constructed from the LN-93 INS, the
RRS, and the GPS subsystem models embedded in the MSOFE
simulation routines. The truth model, along with PROFGEN
generated trajectory and the SV orbit calculation software, generates measurements as well as reference variables which are
used to test the performance of a full-order Kalman filter. This
constitutes a simulation environment in which a variety of full-
In Figure 8, latitude error is indicated by the solid trace and longitude error is depicted by the dashed trace. Addition of the
RRS measurements improved performance by roughly three
orders of magnitude when compared to the INS performance
aided by baro-altitude only. This is essentially the performance
currently experienced by CIRIS. In the final phase, the GPS
model is added to the 98-state INSlRRS model, bringing the
249
total number of NRS states to 128. Two modes of testing are

performed with GPS addition. First the alignment and flight
runs are performed with GPS measurements; no RRS measurements are included. Second, the alignment and Right series are
re-run using both GPS and RRS measurements. Ten one-hour
flight runs are performed for the 128-state NRS model. To
facilitate a direct comparison of the horizontal positions states
between this, previous, and subsequent configurations, Figure 9
is included.
1
E
'0
o
-100
-200
2000
4000
(sec)
kigure 1 1 . Latitude, Longitude, and Altitude trrors
1ooa
ooa
1
-2ooo.
-2ooo.
3000
3000
(sec)
(sec)
Figure 9. 128-state Model Filler Error (ft) 1-a
In Figure 9, latitude error is indicated by the solid trace and longitude error by the dashed trace. Note the significantly reduced
1-0 magnitudes in this configuration compared to Figure 8. In
some flight regimes (notably those far from the transponder
sites, or those with depressed elevation angles between the user
and transponders) performance appears to have improved by an
order of magnitude compared to the INS performance achieved
with RRS aiding alone.
In final configuration, both RRS and GPS measurements are
used. For the alignment simulations, 10 Monte Carlo runs are
performed. Alignment results are similar to those obtained with
either GPS or RRS alone. The general trend is a slight reduction in the filter estimates of the overall error magnitudes during
the alignment phase. In order to facilitate a direct comparison
of the Kalman filter horizontal positions states between this,
previous, and subsequent configurations, Figure 10 is
included. In Figure 10, latitude error is indicated by the solid
trace and longitude error is depicted by the dashed trace. The 1a magnitudes in this configuration are further reduced compared to Figures 8 and 9.
I Figure 10. 128-stateModel GPS & RRS Error

1.5
2000
1000
3000
.8
o
-.s
-'
-1.8
2000
-00
6000 (Sec)
=igure 12. North, East, and Vertical (Up) Velocity Errors
(sec)
(ft) 1-0
The results achieved from the complete system simulation and

analyses are presented in Figures 11-13. Key variables such as
position, velocity, and platform tilt errors are plotted in the following figures. The true standard deviations of the indicated
mean error variable is a,,,. The traces shown by dotted lines
which bound and hack the mean error time history represent the
mean error plus and minus a,,,,,.The pair of traces shown by
solid lines represents the filter-computed krj,,,r for the error
variables indicated. They are symmetrical about zero because
the Kalman filter ussumes that its errors are zero-mean [7].
I
"
Fi
3
-rm
moo
4400
eo00
(sec)
Figure 13. North, East, and Azimuth Tilt Error
-120
'
ZOO0
4000
(sec)
After reviewing the different configurations, GPS aiding alone

appears to offer a noticeable improvement over simple transponder-aided INS. In several flight regimes, the GPS-aided
solution is significantly better than that of the RRS-aided INS.
This fact is a direct result of two features which are inherent to
250
this simulation. The flight path extends far beyond the optimal
coverage areas for the fixed transponders, and only six transponders are used; many actual flights use twenty or more transponders. The other important conclusion is that combining
GPS and RRS measurements, in a joint Kalman filter, offers the
best overall performance.
5. GRACEFUL DEGRADATION OF GPSnNS WITH
FEWER THAN FOUR SATELLITES
This study examines the graceful degradation of a deep integration of GPS with INS and compared it to that of a federated filter architecture integration [50]. Precision first pass weapon
delivery with exceptionally accurate navigation are vital to the
operational survivability of attack aircraft. Systems such as the
GPS are currently being fielded and integrated with INS to support such a capability. It offers the potential for less costly
weapon targeting systems if target coordinates can be provided
relative to the GPS reference. The current standard military
GPS receivers requires simultaneous signals from 4 satellites
(with good geometry) to calculate and output useful position
and velocity information. However, signals from 4 satellites
may not always be available, especially in and around heavily
defended target areas protected by high power jammers. These
high power jammers, even when recognized and appropriately
nulled out by adaptive array GPS antenna, can result in large
sectors of the sky surrounding these jammers no longer being
monitored for any GPS satellite signals. This study examines
Kalman filter algorithms which provide graceful degradation of
aircraft navigation performance when only one, two, three, or
intermittent satellite signals are available, and quantifies the
level of relative performance expected for these conditions.
This study also analyzes the relative performance of various
Kalman filter mechanizations as a function of filter size.
The Standard GPS Receiver IIIA does not provide the pseudorange and delta-range data from the GPS satellites on the MILSTD 1553 data bus. The receiver only provides position and
velocity data as an output of its own internal Kalman filter.
This filter is based on a generic INS error model that is not optimized to any particular INS type or technology (it is crude in
accuracy). If an aircraft requires a navigation solution optimized to a particular INS or if other sensors are required in the
optimal navigation solution this GPS receiver output standard
limits the aircraft Kalman filter designer to a cascaded filter approach. Use of GPS filtered position and velocity outputs to
drive a separate aircraft Kalman filter can lead to filter instability, in part because the GPS and the INS position and velocity
data are time correlated. This potential problem is minimized
by processing GPS measurements in the aircraft filter at a
much slower rate than available from the GPS receiver. This
helps in reducing the GPS and INS measurement correlation.
A spacing of 10- 12 seconds is frequently considered sufficient
when 4 satellites with good geometry are available.
When fewer than 4 satellites are available, the GPS receiver filter outputs degrade and closely track the rapidly growing INS
errors. To avoid filter instability under these conditions the aircraft filter is designed to disregard inputs from the GPS receiver when fewer than 4 acceptable satellites are available. This
cascaded filter approach therefore results in the binary ordoff
incorporation of GPS information. However, if pseudo-range
and delta-range data were directly output, a single non-cascaded (joint) Kalman filter could be implemented. Such a filter
would facilitate graceful position and velocity performance
degradation when fewer than 4 satellites are available. One
concern for GPS signal availability is during, or near, the
weapon delivery time when jamming is likely to be present.
Use of even momentary measurements from single random satellites by such a filter during a high jamming period will provide significantly improved performance over the cascaded
filter approach. This discussion expands on the work presented
in [50]. It describes the sensitivity analysis of the filter perfor-
mance as a function of the number of significantly contributing

states. Subsequently, this performance is quantified against
various GPS availability conditions.
5.1
Recommended Filter Mechanization
The GPS receiver should output pseudo-range and delta range,

satellite position, and GPS time tag data. A joint Kalman filter
that uses these GPS pseudo-range and delta-range measurements (instead of the filtered GPS position and velocity outputs) is recommended. The complexity of such a filter is not
necessarily increased. For example a Kalman filter incorporating range and range-rate measurements from ground transponders has been flying since early 1970s as part of the CIRIS
at the Central Inertial Guidance Test Facility (CIGTF), Holloman AFB, NM [51]. Modeling of GPS measurement error
states in the Kalman filter results in higher estimation accuracy.
When raw GPS data is available from 4 satellites with good geometry, three-dimensional position errors and user clock errors
are directly computed by the aircraft filter. Furthermore, since
raw GPS measurement errors are not time correlated with
those of the INU. all GPS data can be used at the maximum
GPS output rate. This enables INU errors, user clock bias and
drift, and other errors to be rapidly and accurately estimated in
the aircraft integrated (joint in the probability density function
sense) Kalman filter. If, at a subsequent time, fewer than 4 satellites are available the now-calibrated user clock and other
measurement errors (both GPS and barometric altimeter) will
facilitate use of pseudo-range and delta-range data from the
available satellites for up to 30 minutes and keep the navigation
solution bounded [52]. Accurately estimated GPS and barometric measurement errors, when modelled in the filter, permit
substantially improved navigation performance.
5.2
Filter Development
A 98-error state truth model is developed for the fully integrated

GPS/INS navigation system. It combines 68-state INU truth model based on the LN-93, the Litton Inc. version of the Air Force
Standard RLG inertial system [24] and a 30-state GPS error truth
model developed by Solomon, in support of the advanced CIRIS
for CIGTF [51]. The Littons 93-state INU error model is expanded to include all of the inertial component misalignment states and
the barometric altimeter measurement errors then reduced to 68
states by eliminating the trending, thermal transient, and compliance states (see the Appendix). From this truth model various reduced-order filters are created and evaluated using MSOFE
against a PROKEN-generated tactical mission trajectory.
The first reduced-order (51-state) filter analyzed is created by
identifying and deleting those states whose error contributions
are orders of magnitude less than the dominant error states.
Additionally other error states, with similar coupling paths, are
linearly combined. 15 INS states are an order of magnitude
smaller in error contribution than other significant error states
affecting either gyro platform tilt or velocity accuracy (e.g., the
gyro scale factor asymmetry errors). Also 12 other minor
INS error states are sufficiently similar in effect to other existing states that they are combined with respective dominant
states (e.g., gyro misalignment errors with gyro scale factor error). Similarly the twelve GPS satellite position error states are
not fully observable, except for the position error vector projection onto the respective line-of-sight, and can be deleted or
combined. Additionally, the ionospheric and satellite clock errors can easily be combined with the tropospheric errors to reduce the GPS model by another eight states. Except for the
user clock errors, the GPS measurement errors to each satellite
are assumed to be statistically independent. (In fact some correlation exists, for example the low altitude atmospheric delay
has some correlation as a function of angular proximity of the
lines-of-sight and elevation angles.) Using this approach the
98-state truth model is reduced to a 51-state reduced-order fil-
25 1
ter. These 51 states are listed in the Appendix. The resulting

51-state reduced-order filter is tuned and its performance compared to that of the truth model. As expected, little or no tuning noise is added to selected states to insure filter stability
(filter tuning). The 51-state rzduced-order model filter performance is nearly identical to that of the 98-state truth filter.
The next, 32-state, reduced-order model combines the following 19 states into the remaining error terms (see the Appendix).
a. 3 gravity anomaly states
b. 3 accelerometer correlated bias states
c. 3 accelerometer scale factor asymmetry states
d. 6 accelerometer misalignment states
e. 4 CPS code loop errors
These 19 error states are primarily the least significant error
contributors of the remaining 51 states. The resulting 32-state
reduced-order filter is tuned and its performance compared to
that of the truth model. The amount of tuning noise added to
selected states to insure filter stability is relatively small. The
performance of the 32-state reduced-order filter is discussed in
the Filter Evaluation Section. Its performance is slightly degraded from that of the 5 l-state filter.
The subsequent, 23-state, reduced-order filter is generated by
combining effects of the following 9 states into the remaining
error terms (see the Appendix).
a.
3 accelerometer scale factor states
3 gyro scale factor states
b.
C.
3 vertical channel states
These 9 error states are the most significant error contributors
of the remaining 32 states. The performance of the resultant
23-state filter is tuned. A moderate amount of noise strength is
added to insure filter stability. The performance comparison of
the 23-state filter is discussed in the Filter Evaluation Section.
Performance of the tuned 23-state is somewhat degraded from
that of the 32- or 51-state filters.
vals. During the first 2000 second flight segment these errors
are reset at 30 second intervals
A representative weapon delivery mission for a modem attack
aircraft is generated using PROFGEN. This mission flight trajectory is shown in Figure 14. The significant segments are the
take-off, cruise, and descent during the 0 - 2000 second segment into the simulation. The second significant segment,
from 2000 to 2800 seconds, is the ingress, pop-up maneuver,
weapon delivery, evasive maneuvers, and escape, shown as an
enlargement in the bottom of Figure 14 This segment is examined with various GPS satellite availabilities. The final segment, 2800 to 7295 seconds, is the return to home base.
Single-run Monte Carlo simulation is performed for each case
analyzed for relative filter performance comparisons
--
1
..os0
.......................................................................................................
.,-- ........................................................................................................
...........:.. ...........;.............;...........;...............;. ........................... i......
.1070 .............................................................................
:.. ................
.............i ......................... ;.. ............;.............;................:...............i......
-il
..........................................................................................................
,-
.....................................................................................................
..D7..
..o-o
I 0 0 0
s-00
1
-
e-
............. ...........p..f
1-
m-
-0-
-00
..............
....,......
-000
0-
-0-
-00
The final, 18-state, reduced-order filter error model is generated by deleting the following 5 states (see the Appendix).
a.
4 code loop error states
b.
1 user clock drvt
These 5 e m r states m the least signifcant of the 23 primary emor
contributors. Deletion of these states should only be considered under severe computational limitations. A significant amount of tuning noise must be added to insure filter stability, especially for the
conditions of reduced numbers of available satellite measurements.
The performance comparison of the 18-state. reduced-order filter is provided in the Section 5.3. The performance of the 18state filter is significantly lower than that of the other three filters. The projected performance of the joint, 18-state filter
promises to be better than the F-16 15-state cascaded filter performance demonstrated during recent F-16 GPSANS flight test
[53]. With four satellites available, the projected performance
of the joint 18-state filter is 10 feet compared to the demonstrated performance of 27 feet (9 meters) of the cascaded filter.
5.3
Filter Evaluation
Before the simulated flights each filter is aligned using the

flight filter. During alignment zero velocity, barometric altimeter, and the four GPS measurements are incorporated. This
provides the filter with the opportunity to immediately begin
estimating those error states which are observable. The significant advantage during the alignment is that the flight filter begins building up the cross-correlation terms in the covariance
matrix between those observable states. Additionally only a
single filter is required, rather than separate alignment and
flight filters. During the alignment the estimated INU errors
and GPS receiver user clock bias are reset at 10 second inter-
Evasive
Low Altitude
....
Figure 14. 2-Hour Fighter Flight Profile

These reduced-order filters are exercised using MSOFE simulation against this mission profile. During the takeoff, climb
out, and cruise enroute to the target area four GPS satellites are
simulated as continuously available until the 2000 second
(33.33 minute) point. A single-run Monte Carlo simulation is
performed from takeoff until the 2000 second point for each reduced-order filter and the truth model. This occurs immediately after the completion of descent to a low altitude. In this
analysis only pseudo-range measurements are simulated.
Position and velocity error pedormance plots, for this initial
2000 second segment, are shown in Figures 15 and 16 The appropriately labelled performance plots provide the Kalman filter confidences expressed in 1-0 values for the tuned 18, 23,
32, and 51-state reduced-order models (gray line) as well as the
estimation error (black line). These are presented as root sum
squared (RSS) three-dimensional position and velocity errors
(magnitudes of the position and velocity error vectors) versus
time. It is important to point out that each filter was tuned until
reasonably acceptable performance is obtained. No attempt is
made to locally optimize the performance of any of the filters.
252
"he four-satelliteperformance is the baseline for comparison of the

following three, two, one, nxo, and occasional satellite measurement availabilitycases. The GPS satellite geometry assumed in thii
analysis is with a GDOP of approximately 3 with one satellite di23-state
40.
rectly overhead. Note the relative performance of the signifcantly
30reduced-order filters against the 51-state filter, whose performance
is very close to that of the truth model (98-state filter) performance
(not shown). Also note the superior performance of the 51-state filter, especially in velocity. When the four satellite measurementsare
available the position performance differences are not significant,
approximately6-7 ft for the 5 1-statecase compared to approximate51-State
ly 8-10 ft for the 18-state case, but more noisy and with less filter
confidence for the 18-state case. The velocity performance is sig10 ;
nificantly better for 51-state versus the 18-statecase.
18-State
40
30
500
1000
1500
200
Figure 15. 3-0 Position Error (ft), Take-off to 2000 Sec.
I
I
When the overhead satellite is assumed not available (the

three-satellite case), the 18-state filter position-performance in
degrades by approximately a factor of two. The other filter
performance remains essentially the same as the full satellite
coverage case. Note that the 18-state filter confidence (1-0
value) is significantly larger than that of the other filters. Initially a higher confidence (smaller 1-0 value) was chosen.
That worked well with four satellites. However when satellite
availability was reduced, this particular filter became unstable.
At that point the filter was detuned resulting in the larger 1-0
value as shown in the following performance plots.
Figure 16. 3-D Velocity Error (fps), Take-off to 2000 Sec.

At approximately 10 minutes prior to time-on-target 0,
(2000
seconds) varying qualities of GPS coverageare assumed. A separate
Monte Carlo simulation run is performed for an additional 800 seconds for each reduced-order model against each assumed GPS coverage case. Three-dimensional RSS position and velocity error
plots, for the reduced-ordermodel filter cases, Figures 17-28, provide
the Kalman filter confidence (1-0 value) and actual estimation emrs.
51-State
18-State
40
30
--
Figure 19. 3-0 Position Error (ft), Three Satellites
lo
18-State
40
30
0.8
0.6
23-State
3020-
10-
wv
2000
2200
2400
2600
28Q
0.4
0.2
51-State
Figure 20. 3D Velocity Error (fps), Three Satellites

23-State
32-State
0.2
0.84
51-State
Figure 18. 3-D Velocity Error (fps), Four Satellites
In the next case, only two satellite measurements are incorporated during this mission segment. Note that in all cases analyzed, the vertical channel model includes at least one state for
the barometric altimeter. and the associated atmospheric pressure variation errors modeled as first-order Markov processes.
The presence of this state appears to significantly enhance the
performance of each filter, including the 19-state filter. In the
two-satellite case the position error performance of the various
filters varies substantially. The 32- and 51-state filters perform
better than the 23- and 18-state filters. This is especially true
during the first 400 seconds of this flight segment.
253
3-State
175
32-state
25
175..
175
23-State
150
12 5
75
50
50
25
25
32-State
175
2000
2200
2400
2600
kigure23. 3-U Position trror (tt), m e satellite
280
0.4..
0.2-
kigure 21. 3-U Position trror (ti),
IWO
satellites
0.8T
+
51-State
0.8
l!:il
Figure 24. 3-D Velocity Error (fps), One Satellite

23-State
0.80.6.
:::: - - = = = = =
0.80.60.4.
0.2.
0.8.
0.6.
t
32-State
51-State
0.4-
Figure 25. 3-D Position Error (ft), No Satellites
0.84
32-State
254
.._
II14l
2 4'0 0
2600
Figures 25 and 26 show the performance of each filter when

only barometric altimeter measurements are available. In this
case each filter relies on the quality of estimates of INS and
barometric measurement errors. The estimation quality of
these errors depends on the estimation quality of the GPS measurement errors during the time when GPS measurements are
available. The position error performance is surprisingly good
for the 51-state filter compared to that of other filters. The velocity error performance is also quite good, especially when
compared to the velocity performance of cascaded filters [53].
ments and modelling significant GPS and barometric altimeter

measurement errors can perform significantly better in actual
flight than the current cascaded approach.
The results of the zero satellite case provide a best case approximation (upper bound) for a cascaded filter performance,
with fewer than 4 satellites under these Same conditions. This
is because the cascaded filter does not use the filtered GPS position and velocity data when fewer than 4 satellites are available. Note that for all the reduced-order filters the RSS
performance at the 10 minute point for the 3 satellite case is essentially the same as for the 4 satellite case, with the exception
of the 18-state filter. The one, two and three satellite case 23state RSS position error values at the 10 minute point are 130,
100, and 10 feet respectively. These values are significantly
better than the best case value that could be expected for the
cascaded filter approach using fewer than 4 satellites. Similarly the 23-state velocity errors at the ten minute point for the
one, two and three satellite cases are 1.0. 0.8, and 0.6 feet per
second. Note the position error growth rate for all Kalman filters is substantially less for the l-satellite case versus the zero
satellite case. Another interesting case is when only occasional
single GPS measurements are simulated by a single measurement from each of four satellites, in turn, at 30 second intervals. It is clear for all the reduced-order Kalman filters that
even occasional pseudo-range measurements provide sufficient
information to significantly improve filter performance over
that of the cascaded filter.
5.5
Centralized Filter Advantages
Precision first pass weapon delivery with exceptionally accurate

navigation are vital to the operational survivability of attack aircraft. Systems such as the Global Positioning System (GPS) are
2200
24'00
2600
currently being fielded and integrated with inertial navigation
Figure 27. 3-D Position Error (ft), Occasional Satellites
systems (INS) to support such a capability. It offers the potential for less costly weapon targeting systems if target coordinates relative to the GPS reference can be provided. The
current standard GPS receiver requires simultaneous signals
from 4 satellites (with good geometry) to calculate and output
useful position and velocity information. However, signals
from 4 satellites may not always be available, especially in and
0.2
around heavily defended target areas protected by high power
jammers. These high power jammers, even when recognized
0.84
and appropriately nulled out by adaptive array GPS antenna,
can result in large sectors of the sky surrounding these jammers
no longer being monitored for any GPS satellite signals. It has
been recommended [SO] that changes be made to the existing
GPS equipment and new aircraft Kalman filter algorithms to
provide graceful degradation of aircraft navigation performance
0.2 {
when only one, two, three, or intermittent satellite signals are
2000
2200
PA00
2600
2
available. In a paper by Lewantowicz and Keen [50], they
Figure 28. 3-D Velocity Error (fps), Occasional Satellites
quantify the level of relative performance expected for these
This final case is the most interesting. It is a simulation of a
conditions. They also analyzed the relative performance of varsituation where only occasional GPS measurements from variious Kalman filter mechanizations as a function of filter size.
ous satellites are available. Although in reality the time distriThe analysis performed in [50]demonstrates that a significant
bution of the sporadic availability of GPS measurements is
level of navigational performance improvement can be realized
expected to be random, here it is approximated with a single
for the case when fewer than four satellites are available by prosatellite measurement available each 30 seconds on a rotating
viding GPS pseudo-range and delta-range data on the MILSTD
basis, that is each satellite is available only once every 120 sec1553 data bus and by implementing a single integrated aircraft
onds. The results are interesting, especially for the 32 and 51Kalman filter. This same filter should be used for ground alignstate filters. The position error performance is near that of the
ment as well. In fact, this filter configuration continues to estifour-satellite case, within a factor of two or so. The velocity
mate all of the observable INU and measurement errors
performance of each of these filters is nearly unchanged from
including the misalignment errors. The quality of these error
the four satellite case!
estimates improves significantly during any maneuver segments, be it horizontal or vertical. In fact the flight trajectory is
5.4 Simulation Results
rather sterile in the sense that all maneuvers are perfect in the
sense that a straight and level flight is precisely straight and
The RSS position and velocity error results for each assumed
level. In reality pilots, and to a lesser degree autopilots, are not
GPS coverage case are shown i n the previous Figures 17
physically capable to fly these segments as perfectly. Thus any
through 28. These simulation results predict that a single Kalrolling, pitching, yawing, or linear accelerations, even relatively
man filter receiving pseudo-range and delta-range measure-
255
small in magnitude should help the filter by making the less

observable errors become more observable.
Additional significant performance improvement can be realized by incorporating as many significant error states into this
single integrated Kalman filter as possible subject to the computational memory and thruput constraints. For the applications
where the position, velocity, and attitude accuracy are the primary system contributors to weapon-on-target miss distances,
this GPSlINS integration approach promises to significantly
reduce those error components.
6.
DIFFERENTIAL GPS CONCEPTS
This section discusses the application of differential GPS

(DGPS) by way of an example to demonstrate how DGPS can
be used to enhance performance of CIRIS. The enhanced system is the ENRS, which takes advantage of newer ring laser
gyro strapdown INS (LN-93) technology, ground transponders
from the current CIRIS, and DGPS measurements. Analysis
conducted using MSOFE is presented. An enhanced navigation
reference system (ENRS) truth model is developed. This truth
model serves as a baseline for full- and reduced-order Kalman
filter designs. Results suggest that the proposed ENRS with
DGPS aiding can provide a navigation position solution at least
one order of magnitude better than the current CIRIS.
This section presents the development of a 48 state post-processing Extended Kalman Filter (EKF) to augment the CIRIS
navigation solution with DGPS pseudo-range measurements.
An EKF is developed instead of a smoother due to the limited
computer storage capacity available. The EKF is limited to 48
states to ensure 24 hour turn-around time for post-processing
real measurements during INS testing at CIGTF. An in depth
description of the design of this EKF can be found in [ 5 5 ] .
Sections 3.4.3.5, 3.6.3.7 describe the truth models used in this
research. The error-states for the test reference (INS and RRS)
are described and then a short explanation of the DGPS implementation and error-states is given. Section 6.5 presents the
DGPS measurement equation. Full- and reduced- order filter
models are described in the Filter Models Section 6.6. Results
of a 25-run Monte-Carlo simulation analysis are presented in
Section 6.7. This section ends with a discussion of conclusions
based on the performance of the ENRS.
6.1
Differential GPS Concepts
A GPS receiver determines its position by computing the range

from the receiver to the satellite vehicles (SVs). This range is
measured by determining the difference between the signal
transmit time and the signal received time. It is in error due to
several sources [32]. The DGPS concept improves the performance of the GPS receiver by eliminating errors that are common in a local vicinity between a user set and a GPS receiver
that is at a known location. This latter receiver is referred to as
the reference receiver. The placement of a receiver at a precisely surveyed location is the key to estimating DGPS corrections. After application of DGPS corrections, it is possible to
achieve accuracies of 1 to 3 meters instead of the normal 10
meter accuracy one can achieve using P code without selective
availability (SA). Use of DGPS allows one to mitigate much of
the SA affect.
A good discussion of DGPS methods and concepts is found in a

paper by Blackwell [56]. There are several methods of employing DGPS. Basically there are two methods to apply DGPS
corrections; corrections can be applied in the pseudo-range
measurement domain, or they can be applied in the solution
domain. The key difference between the two approaches is in
the specific data that must be transmitted between the two
receivers involved, and whether an uplink versus downlink
communication link is required.
Some DGPS implementations include:
I ) Uplink of pseudo-range corrections for user vehicle processing

2 ) Uplink of position corrections for user vehicle processing
3) Downlink user receiver raw data for ground station DGPS
processing
4 ) Downlink of uncorrected position data from user vehicle
for ground station DGPS processing.
Methods 1 and 2 will provide corrected position on board the

user vehicle. Method 1 is normally preferred when multiple
users are to be accommodated. Methods 3 and 4 are used in
applications where precise position information is required at a
central ground facility. Method 3 is the most demanding to
implement of these two.
There is also a concept called relative GPS where one achieves
differential accuracies of the relative position between two
users. This concept is not the same as DGPS. In relative GPS
there is no reference receiver, but one simply takes the difference between two receivers in this same area. When this is
done, the common biases seen by both receivers are essentially
removed. In this technique, absolute position accuracy is not
improved at all. However, relative position is often of greater
interest in applications such as targeting or formation control.
6.2 CIRIS
CIRIS is described in Section 4. Up to this point in time, CIRIS
has been considered more accurate than the test articles and has
formed the baseline for determining the performance of aircraft
INSs. Recently, state-of-the-art aircraft INSs have been developed (and many more in the design stage) approaching the
accuracy of CIRIS. Interestingly, many of these new INSs use
the Global Positioning System (GPS) to increase their accuracy.
In order to use CIRIS as a baseline against these new INSs,
CIRIS must be enhanced to provide an order of magnitude more
accurate navigation solution. By using DGPS measurements to
augment the navigation solution of CIRIS, it is possible to
increase the accuracy of CIRIS to produce an order of magnitude better estimate of the navigation solution.
6.3 Truth Models
The 89-state truth model used in this research is divided into 3
sub-models based on the 3 subsystems forming the ENRS. The
first sub-model contains a 41-state INS model consisting of 40
Litton LN-93 INS error-states and a single baro-altimeter errorstate. The second 26-state sub-model defines the error-states
associated with 6 transponders modeled in the RRS. The last
sub-model contains the 22 error-states associated with DGPS
measurements from 4 space vehicles (SVs).
The 41 error-state INS truth model is derived from the complete

93-state Litton LN-93 INS truth model [24]. The 41 error-states
utilized in this research are the dominant error sources present
in the 93 error-state truth model. This error-model is composed
of 13 general error-states as well as 28 dominant gyro, accelerometer, and baro-altimeter error-states. The 52 gyro and accelerometer error-states not present in this truth model are not
large in magnitude and have been combined with other states or
eliminated according to the recommendations of Lewantowicz
and Keen [50]. The first 13 states in the 41-state INS truth
model represent the general error-states, states which are combinations of several other states in the error model. Three position (latitude, longitude, and altitude), three velocity (east,
north, and vertical), and three platform tilt (east, north, and azimuth) errors along with four vertical channel errors comprise
this group of error-states. Following the first 13 general errorstates are 4 error- states modeled as first-order Markov processes. These states are the X, Y, and Z accelerometer noise
states and the baro-altimeter state. The final 24 error-states are
gyro and accelerometer error-states modeled as random biases.
The gyro error-states include gravity, drift, and scale factor
errors. The accelerometer error-states include bias, scale factor,
256
and misalignment errors. The 41-state truth model performance

has been compared to the 93-state truth model performance
with excellent correlation. A discussion of this comparison and
a complete description of the 41-state INS truth model is found
in [ S I .
The transponder based RRS provides the external measurements necessary to update the EKF for CIRIS as it is currently
implemented. The RRS uses an airborne transceiver. referred to
as the RRS interrogator, along with a computer to calculate
position and velocity based on EM signals received from
ground transponders. The transponders have been accurately
surveyed, so that once the interrogator sends a signal to a particular transponder and after receiving the transponders reply, the
computer can accurately calculate position and velocity information. Although only six transponders are modeled in this
research, many more (or less) can actually be in range of the
RRS interrogator depending on the location of the flight test.
There are dozens of transponders located on the CIRIS flight
test range and many more located across the country. The 26
error-state RRS truth model contains two types of error-states,
two states common to all six transponders and four states which
are transponder dependent. The first two states in the truth
model are random bias states modeling the effects of user hardware (RRS interrogator) on range and range-rate calibration
errors. These two states are modeled with initial variance values of 1 f? and 10-4fps2,respectively. The final 24 states, 4
states for each of six transponders, model transponder x, y. and z
position errors and an atmospheric propagation error. The position errors, stemming from the fact that the transponders positions are not perfectly known, are modeled as random biases
with initial variance values of 25f?. The atmospheric error,
induced by the propagation delay, is modeled as a first-order
Markov process with a time constant of 300 seconds and a
white dynamics driving noise of 6 . 6 6 ~
IQ'f?. Description of
RRS states is found in [ S I .
6.4
DGPS Truth Model
GPS is designed to be an accurate, stand-alone navigation system. However, for this research, GPS is used as a subsystem to
improve the navigation solution of the ENRS. GPS navigation
information is obtained from EM signal propagation through
the media (space and atmosphere) between the user (ENRS)
and each of 4 SVs which the user locks into a reception channel
of the GPS receiver. In a stand-alone GPS receiver, navigation
information is obtained by receiving GPS SV ephemeris data
broadcast continuously from each active (locked-on)SV,correlating the phase of the signal with a matching signal in the GPS
receiver, and correcting for known error sources to produce an
accurate range estimate between the user and each SV which is
monitored.
In this research, uncorrected range measurements (known as
pseudo-range measurements) are channeled to a Kalman filter
which provides estimates of the error sources. Common GPS
error sources which are considered dominant in this research
include receiver clock bias and drift, ionospheric and tropospheric (atmospheric) propagation errors, SV clock, and SV
position errors. Other non-dominant errors are also present in a
true GPS signal, but are compensated for in this research in the
GPS pseudo-range measurement noise. As in the RRS, GPS
range measurements make refinements to the ENRS navigation
solution possible.
Intermetrics, Inc. is the government sponsored contractor
responsible for the DGPS reference station at CIGTF [57,
58].The following discussion of DGPS comes from interviews
with Mr. Darwin Abbey and Mr. Scott Dance of Intermetrics,
and the DGPS error model described is a combination of Intermetrics description and a course given by Navtech Seminars on
DGPS error models [57,58,59].
In order to apply differential corrections to GPS measurements,

a ground based reference receiver (GBR) is needed as well as an
airborne GPS receiver (ABR). Figure 29 shows the basic DGPS
system as it is being implemented at CIGTF.
PR,DR
INS Output
Errors
Figure 29. GPS Reference Station

The ground based receiver must be capable of tracking all SVs
in view, possibly as many as eleven when the full GPS constellation is placed in orbit. The GBRs antenna position is permanently fixed and surveyed to within centimeter accuracy. A
high accuracy rubidium clock is used in place of the GBRs normal clock, greatly decreasing the large clock errors common to
GPS receivers. Using the transmitted SV ephemeris data, the
GBR computes its position (different from the surveyed position) known as the ground truth. The data (pseudo-range measurements, S V transmitted corrections, GBR applied
corrections, ground truth, clock errors, etc...) is fed to a 30386
personal computer which then estimates the SV position errors,
SV clock errors, and atmospheric delays with great precision.
Because the GBR's true position is accurately known and its
clock errors are much smaller than the ABR's, the SV position
and clock errors and atmospheric delays are highly observable.
This is unlike a normal GPS model where the large ABRs clock
errors and vehicle dynamics cause these states to be largely
unobservable. These errors, now called differential corrections,
are time tagged and stored on magnetic tape or disc. Note that
the differential corrections could be immediately transmitted
via a data link for real-time differential corrections if the need
ever arises [57,58].
Remembering that the ENRS is a post-processing filter, the raw
airborne pseudo-range measurements (which are also stored
magnetically and time tagged) now have the differential comctions applied before they are analyzed in the post-processing filter. Of course, this assumes the GBR is tracking the same four
SVs the ABR was tracking (a good assumption if the ABR is
within the CIRIS test range). Using differential corrections in
this manner, the SV clock error is eliminated from each pseudorange measurement and the SV position errors are nearly eliminated. Depending on the distance between the ABR and the
GBR during the flight profile, the atmospheric propagation
errors (ionospheric and tropospheric) can be almost totally
eliminated for close trajectories or greatly reduced for flights
within 200 miles of the GBR. Even with long-range DGPS
(flights which extend more than 200 miles from the GBR), the
post-processing in the 30386 personal computer eliminates
some of the atmospheric propagation errors. The largest
remaining errors in the pseudo-range measurements are the
ABR clock errors. With this basic knowledge of DGPS, a
DGPS error model is now developed assuming that differential
corrections have previously been applied to the raw pseudorange measurements from the ABR [57,58,59].
The DGPS error-model is composed of 22 states. The DGPS
error-states are similar to any GPS error-state model except for
elimination of the SV clock error and reduction of the atmospheric and SV position errors due to differential corrections.
The first two error-states in the DGPS error model are the ABR
clock error-states modeled by
257
symbols are defined in Section 9., and equivalencies applied:

xUclkb= Rclk. = range equivalent of ABR clock bias
Xuclk,,
= R c i k . = velocity equivalent.of ABR clock drift
The initial state estimates and covariances are [ S I :

1
Until the ABR clock error is determined, it is the single largest

source of error in DGPS range measurements. While the two
states discussed above apply to all DGPS measurements, the
remaining 20 DGPS error-states are unique to each SV measurement. Each of the 4 EM signals originates from a different
SV and travels through different amounts of atmosphere, hence
the need for individual atmospheric and SV position errors for
each pseudo-range measurement. Note that these errors have a
much smaller contribution after differential corrections than
they would in a GPS model [59]. The tropospheric and ionospheric errors are modeled as first-order Markov processes such
that:
P (to) =
i"'
0
E{w(t)}
01
Iff2
=o
( 124)
E{w(t)w(t+r)} = oBO.
L o
0 . d
Cft2/sec2)G(r) (125)
The SV position errors are modeled as random bias errors:
The sub-models described above for the INS, RRS, and DGPS
subsystems compose the state dynamics equations (with initial
conditions). These are entered into MSOFE simulation so that
true error-state values can be computed and analyzed against
the extended Kalman filter's estimated error-states. Now that all
the error-states used in the models have been described, the 3
types of measurement equations can be overviewed.
6.5 Measurement Equation

In order for any Kalman filter to properly estimate state variables, external measurements encompassing some of the state
variables must be provided to it. In this research, 3 types of
measurements are provided to the post-processing EKE The
first type of measurement is the baro-altimeter measurement
used to stabilize the vertical channel in the INS. This measurement occurs at I-second intervals. The second measurement
type is the RRS range measurement from each of six transponders, with measurements from all six occurring every six seconds. These measurements are well documented in [37] and
[55], respectively, if the reader wishes to further research these
areas. The final type of measurement, and the measurement this
research focuses on, is the DGPS pseudo-range measurement.
The DGPS pseudo-range measurement is best described as a
GPS pseudo-range measurement with differential corrections
applied to it, as mentioned in the previous section. These measurements are obtained from 4 geometrically optimal SVs (out
of 24 total and 11 possible, when the full GPS constellation is in
orbit) at 10 second intervals. The full satellite constellation has
been programmed into this simulation and geometric dilution of
precision (GDOP) calculations are performed to obtain the 4
SVs used for measurement purposes, with satellite changes
occurring when necessary for optimal GDOP considerations
[46]. After applying differential corrections, the measurement
equation is modeled as:
where the symbols are defined in Section 9.

The DGPS pseudo-range equation above includes the true range
(which can never be known exactly) along with terms which
reflect sources of error and uncertainty inherent to DGPS range
measurements. With this equation in hand, it is desirable to formulate a difference measurement in the DGPS model. However, to accomplish this, two sources of range information must
be obtained. The first source is the range measurement coming
from the DGPS reference station and modeled by Equation
(130). The second range estimate is constructed by differencing
INS-indicated position and SV (broadcast) positions to calculate the range. This derivation is explained in detail in [55], and
its result yields:
6z =
0.35ff2
P ( t ~=
)
0
O.35ff2
0
E{w(t)}
E{w(t)w'(t+r)} =
=o
- RDGPS
0.35ff2
Ixr ix 1
Lo
R,NS
(128)
C f t / s e c 2 ) G ( z )(129)
XXXJ
The set of equations above apply to a single SV. The initial

covariance values in Equations (123) and (127) and atmospheric
error dynamics noise variances in Equations (125) and (129) are
obtained from [59] and are modeled slightly larger than Abbey
and Dance [57, 581 recommend since a conservative model is
developed. There are four such sets of matrix equations for
DGPS SV errors modeled in this study. The error-state vector is
completely specified in [ S I .
- 6R,, - 6Rion- 6RUcrt- v
(131)
The terms ax,, 6y,, and 6zudirectly relate to the INS position
error terms (latitude, longitude, and altitude) while 6xs, Sy,, and
6z, are the SV position errors. The pseudo-range measurement
noise variance is 9ft' when DGPS pseudo-range measurements
occur every 10 seconds. The true whole-valued range (R,)formerly present is cancelled in the differencing operation. The
bracketed coefficients in the equation above appear in the EKF
update equations. The full derivation of this equation along
with the EKF update and propagation equations is found in
[ 5 5 ] . The DGPS error-state truth model and measurement
258
equations have now been shown, so now the two filter models
used in this research are described.
6.6
Filter Models
Two filters are developed to implement DGPS pseudo-range

measurements into the ENRS, a full-order filter of 89 errorstates and a 48 error-state reduced order filter. The first filter is
described as full-order since the 89 filter states are modeled
exactly like the 89 truth model states. For the reduced-order filter, 41 states are eliminatedafter analysis determined their magnitudes are small when compared to other error-states being
modeled.
The 89 error-state full-order filter model is used to establish a
baseline for comparison to any reduced-order filters developed. As a reminder, this filter is composed of 41 INS errorstates, 26 RRS error-states, and 22 DGPS error-states. Baroaltimeter, RRS range, and DGPS pseudo-range measurements
are used in this EKF to provide accurate position estimates. At
this time, no velocity aiding measurements are incorporated
into this filter, although research is continuing in this area and
eventually RRS range-rate and DGPS delta-range measurements will also be used to provide accurate velocity aiding as
well [%I.
In order to ensure 24-hour post-processing time of a typical
flight profile used to test INSs, CIGTF requested that the final
EKF be less than 70 error-states. This means that at least 19
error-states have to be eliminated. More could be eliminated if
position (and eventually velocity) accuracy can be maintained.
With these objectives in mind, filter order reduction was performed and a new 48 error-state EKF was developed. The first
step in state reduction was taken when 20 DGPS error states
were eliminated because their magnitudes were very small compared to the magnitudes of the ABR clock bias and drift states.
A typical ABR clock bias error is on the order of lo* ft, while
the differentially corrected atmospheric and SV position errors
are 1 or 2 ft. The EKF often had trouble accurately estimating
the small error-states because of this magnitude difference, so
their elimination did not affect filter performance. Of course,
the white Gaussian measurement noise variance was increased
slightly to compensate for the eliminated states [SI.
As a second step to eliminate more states from the EKF, the recommendations of Lewantowicz and Keen are again followed
when 21 more INS error-states are eliminated. The eliminated
states are small magnitude gyro and accelerometer errors, and
small increases in the dynamics noise variance in some of the
remaining states is necessary to compensate for their removal.
Note that the removal of these 21 states significantly increased
the attitude errors, and slightly increased the velocity errors, but
the position errors remain relatively close to the values of the
full-order filter. Again, for a complete description of the INS
error-state filter reduction, consult [ S I . At this point, enough
states have been eliminated to ensure quick analysis while producing accurate estimates of the navigation solution. The EKF
has been reduced to 48 error-states; 20 INS, 26 RRS, and 2
DGPS. It is important to note at this point that the RRS errorstates could also be reduced. However, since this research is
focused on the DGPS error model, no RRS filter state reduction
is performed. The results of filter performance are now presented.
6.7
Results
This sub-section presents and discusses the results of the filter

performance of the full-order and reduced-order filters. The
current CIRIS 1-0 position and velocity accuracies are used as a
baseline to judge the filters performance. Also, a 46 error-state
filter called CIRIS-46 is included in this comparison. This filter
is composed of the 20 error-state reduced-order INS model and
the 26 error-state RRS model and uses baro-altimeter and RRS
range measurements. This filter gives an indication of the per-
formance gained by using a higher-order post-processing filter,

and will lend itself for a better comparison in the performance
increase of DGPS alone. Each filter's performance is tested
with MSOFE utilizing 25-run Monte-Carlo analysis of a 2-hour
fighter flight profile., Figure 14 shows the latitude, longitude,
and altitude information as well as a 3 dimensional representation for this fighter flight profile. As is readily seen, it incorporates several turns and dives to simulate a realistic flight profile.
Figure 30 shows typical latitude, longitude, and altitude plots
utilizing the data provided by MSOFE. These particular plots
show the relationship between the true l o errors and the filter's
estimate of the errors. The single dashed line plots the mean
error, or the difference between the true latitude error and the
filter's estimate of that error. The two dotted lines are plots of
the mean error f the true 1-0 value. Finally, the solid outermost
lines are the filter's estimate of the errors. Again these plots are
an average over 25 Monte-Carlo runs and represents an adequately tuned filter.
5 -.-
-.
~
~
~
' rooo,.rrlT
~
~
-- _ _ _ ' ' . '~_ . '
igure 30. 89-State ENRS Filter, Latitude, Longitude,and
Altitude Errors
,000
I
-
-00
-000
-000
The 89-state full-order ENRS filter's position and velocity estimation error performance is shown in Table 2 along with the
current CIRIS and the CIRIS-46 filter. The north velocity error
results are not presented due to space considerations, but their
values are comparable to the east velocity error results. The
temporal average of the ensemble average simply means that
the true lo values for all 25 simulation runs are averaged
together at ten second intervals and then this average is averaged to give the number shown. As seen, the full-order filter
provides order of magnitude better position and velocity estimates than the current CIRIS filter and much better performance than CIRIS-46. The increase in velocity error estimation
is attributed to the fact that if the filter's position error estimation is increased, it can also estimate the velocity errors with
greater precision. It is easy to see the benefit that DGPS
pseudo-range measurements have on increasing the accuracy of
CIRIS.
The results of the performance analysis on the 48-state reducedorder ENRS EKF are also shown in Table 2. As a reminder, this
filter is composed of 20 INS, 26 RRS, and 2 DGPS error-states.
There is a slight decrease in position and velocity performance;
however, this filter performs much better than either CIRIS or
259
CIRIS-46. Notice that the reduction in filter states (gyro and

accelerometer bias error-states) did affect the filters ability to
estimate the velocity error (and attitude error, though not
shown). It is again safe to assume that DGPS measurements
dramatically increase the accuracy of the navigation solution. It
is obvious from the results that DGPS pseudo-range measurements do indeed increase the accuracy of CIRIS and the postprocessing 89- and 48-state ENRS filters increase the navigation solution position accuracy by one order of magnitude
Table 2. Temporal Average of True Error Ensemble
CIRIS
14.00
I
I
14.00
I
I
40.00
0.100
18.09
0.046
I
I
0.400
I CIRIS-46 I
3.12
ENRS-89
0.84
1.04
3.80
0.013
0.042
ENRS-48
0.90
1.32
3.05
0.027
0.044
6.84
0.100
in this tremendous capability is most clearly evident in the

growth of attendance at the Institute of Navigation ION/GPS
conferences and in the number and variety of commercial and
civil GPS application products. Of significance to this discussion is the fact that the free enterprise spirit has produced a multitude of ingenious solutions to such a vast number and variety
of commercial and civil problems.
There are various factors affecting the difference in utilization
of the GPS by the military services, especially when compared
to that of the commercial and civil sectors. Some of these factors are clearly identifiable and discussed in this paper while
others are not known or not clearly understood. The more
demanding military operating environment and mission requirements are among the more obvious differences. However there
are other factors, some of which are systemic. Of the systemic
factors, perhaps the most significant one is that of avionics integration architectures. Although GPS/INS integration is used as
an illustration of a systemic factor, the discussion is presented
from a broader perspective of system level avionics integration.
Avionics architecture options are examined from the standpoint
of the cost of integration, cost of ownership, and the yet unexploited mission capability.
7.2
Federated Architectures
The evolution and limitations presented by the federated architectures is described in the INTRODUCTION Section l. This
evolution, and federated architecture subsystem outputs resulted
in unintended constraints on deep integration. However, despite
some challenges in providing for multiple requirements, this
limited integration concept was successful in that performance
of some sensors was improved, to a lesser or greater degree, by
providing information from other sensors. Most of the information provided from other sensors was at the post-processed output level or in the domain of output integration. This postprocessing modifies the raw signal, available from the detector, in bandwidth, noise statistics, and other electronic compensation characteristics. Although theoretically it possible to
undo the output processing if the internal algorithms are
known, this is not practical due signal-to-noise losses, distortions, and typically high cost. The impact of this post-processing is that it limits the degree to which the information from
other Sensors can be utilized, where appropriate, to improve the
performance of other sensors. This federated architecture, with
its limitations, is the prevalent environment that the military
GPS user equipment is required to interface to.
7. GPS/INS- CORNERSTONE OF SENSOR FUSION

The GPS contribution to navigational accuracy has been clearly
established in military weapon systems. However the GPS signal provides enormously broader potential mission capability
which has not yet been exploited. The GPS signal, when properly combined with information from INS and other sensors,
provides position, velocity, attitude, and time of unprecedented
accuracy and robustness. These ten elements, defined in a geometric vector in Section 3.1, are common, in various combinations, to most of the avionics functions. When viewed from a
system perspective, this high precision information, can be
thought of as the integration basis, or a reference set, which
offers opportunities for reconfiguration of the offensive, defensive, communication, navigation, and other sensors. Various
integration architectures for fusion of these sensors can inherently enhance, enable, or severely limit these potential mission
capabilities. The choice of integration architecture, can directly
and profoundly affect performance, cost of integration, cost of
ownership, and exploitation of much greater mission capability.
7.1
Introduction
A multitude and variety of commercial and civil GPS applications are exploding, which in turn are fueling the explosion of
commercial GPS receiver hardware development. The interest
7.3
Military GPS Equipment
When the military user equipment segment of the GPS was

fielded, the navigation function also was provided in a black
box (really gray colored), Standard Rockwell Collins Receiver
IIIA. This box was required to perform as a stand-alone instrument, while at the same time it output information for several
other devices, such as the INS, the fire control computer, or fire
control sensors. These outputs were defined under various standards, such as the MILSTD 1553, but other restrictions were
also applied. For example, the GPS receiver did not include the
pseudo-range and delta-range information on the MILSTD
1553 bus output. Security considerations prevent the output of
the Selective Availability (SA) corrected precision pseudo- and
delta-range information outside some boundaries of an avionics
sensor container. Instead the Receiver IIIA outputs navigation
position, velocity, and other data after significant processing by
its own internal Kalman filter. This filter is based on a generic
INS model that is not optimized to any particular INS type such
as gimbaled or strapped-down system.
If an aircraft requires a navigation solution optimized to a particular INS or if other sensor information is required for an optimal navigation (or other variable) solution, this GPS receiver
constraint limits the aircraft Kalman filter designer to a cas-
260
caded, and suboptimal, filter implementation. This results in

significantly degraded performance and jam resistance, and
requires a strongly disadvantageous GPS availability or visibility definition. Also use of GPS filtered position and velocity
outputs to drive a separate aircraft Kalman filter (e.g. in a fire
control computer) can lead to filter stability problems, since the
GPS and the INS data are time correlated [60]. This potential
problem is minimized by typically processing GPS measurements in the aircraft filter at a rate much slower than they are
output from the GPS receiver. This helps in reducing the correlation between GPS and INS measurements, but not without
penalties.
of forms, from the raw to the fully processed for a variety of

integration approaches. Signal level sensor fusion ranging from
combining of two or three sensors to large scale fusion of a multitude of sensors is possible. In any of these cases, the availability of the raw pseudo- and delta-range GPS measurements
provides unprecedented opportunities. When the GPS is properly combined with INS, this two-sensor subset produces a significantly more accurate, robust, jam-resistant, less costly
navigation, attitude, and time reference set. This reference set
is very important because not only does it provide superior navigation solution, but it also becomes an (geometric) integration
reference for combining (fusing) of other sensor information.
Addition of the GPS receiver to the air weapon systems

increased the mission capability tremendously from that of INS
alone, or non-GPS aided INS. The GPS Receiver IIIA was
designed and installed in the aircraft (and other vehicles) essentially as a navigator. It provides an unprecedented improvement
in navigation accuracy over that of the previously available
from the INS, which was aided by barometric altimeters, doppler velocity sensors, and other navigation instrument outputs.
Fundamentally this increase in navigation accuracy provided
substantial increase in mission capability. Global scale navigation accuracies of 16m position and O.lm velocity are two
orders of magnitude better than possible before introduction of
GPS. However this performance increase is limited when compared to the potential performance, robustness, jam-resistance,
etc. of a properly integrated GPSDNS. Performance increase in
the navigation function as well as other improvements and benefits remain largely unexploited.
When other sensor information, which contains either absolute

or relative navigation, attitude, or time information, is processed
jointly with that of GPS and INS, then additional benefits are
available. For example the reference biases, and other errors, of
these sensors are often observable and estimable, thus providing
a cross-calibration function. For those sensors where this crosscalibration is significant, several opportunities immediately
appear. For example, the absolute performance tolerances on
those sensors perhaps could be relaxed, thus reducing the complexity, cost, and reliability of those sensors. In some cases
more than one sensor performs a similar or an overlapping function, or measures information also measured by some other sensor. When the information from this ensemble of sensors is
properly fused, some of those sensors are either redundant, or
their omission could result in minor and acceptable mission performance degradation but at a significant reduction in cost.
Another possibility may be the realization of a common and
highly accurate time reference that could be derived from this
joint fusion of signals which have the GPS time as a common
reference. Also, when information from other sensors is jointly
fused with that of GPS and INS, the reference set accuracy and
robustness is further improved.
7.4 Advanced Avionics Architectures

In contrast to the federated avionics architectures, the integrated
avionics architectures offer numerous advantages. These
advantages can be understood from a broader perspective of
avionics system flexibility, redundancy, fault tolerance, realtime reconfigurability, decreased life cycle cost, adaptability,
and other attributes. These architectures consist of multi-function, software programmable modules in common avionics
enclosures. Many of the mission functions can be performed by
individual or groups of modules. This concept replaces the discrete functions of the federated avionics architectures. The
Integrated Communications, Navigation, Identification Avionics (ICNIA) and the INtegrated Electronic Warfare System
(INEWS), are examples of such architectures, and are the basis
for the U.S. Air Force Advanced Tactical Fighter (ATF),
recently designated as the F-22, and the U.S. Amys next generation helicopter the LHX. The U.S. Navy is considering this
avionics architecture for its AX system. Selection of modular
avionics standard is a topic of high interest to the North Atlantic
Treaty Organization (NATO), in particular the Four Power
Group consisting of United States, United Kingdom, France,
and Germany. The essence of the question before this group is
not whether integrated avionics should be considered, but what
form should the standards take for the avionics architectures
and for one of its key elements, the modules.
Many advantages result from the integrated architecture concept. The essentially raw signals are brought into a single
framework where real-time management of the available
resources can be exercised. This immediately offers opportunities for real time reconfiguration for a spectrum of requirements, be it to account for failures, or reconfigure to adapt to a
changing mission phase, or to respond to changing mission
requirement, such as just identified enemy threat requiring a different and immediate response.
Perhaps the most significant advantage of integrated avionics is
that most of the signals are available in various forms and stages
of processing within the physical confines of this architecture.
The communication, navigation, and identification signals in
ICNIA, the defensive avionics sensors signals in INEWS, and
the offensive avionics sensors signals are available in a variety
These are only few examples of what may be achievable by

integrated architectures. It is essential to emphasize, that to
realize this vision, system and component level modeling, analysis, and simulation are required to identify the system level
sensitivities and perform various trade-offs.
7.5
GPSlINS Integration
The GPS/INS integration is a special subset of the larger set of

advanced avionics architectures and sensor fusion discussion. It
is treated separately because of the very synergistic properties
that a properly integrated GPS and INS functions can provide.
In many vehicles that do not, and may not have integrated avionics architectures, an opportunity exists for installing a properly integrated GPSDNS in a single box. The miniaturization of
the GPS receiver processing is more than ready for imbedding
into the INS. For those who may feel attachments to the GPS as
the primary function, a possibility exists of imbedding the ring
laser gyro (RLG), fiber optic gyro (FOG), or the hemispheric
resonating gyro (HRG) INS into the GPS enclosure. The benefits derived from such an integrated system are increases in
accuracy, jam resistance, reliability, fault tolerance, and
decreases in cost of the required INS and certainly in life cycle
cost.
Signal level fusion, as discussed earlier, produces better results.
For an eloquent discussion of the synergistic benefits of proper
integration of these two elegant and proven system, the GPS
and INS, see the now classic, in this authors view, paper by
Cox [30]. The vehicle dynamic maneuvering and the potential
jamming of the GPS signals require the proper integration of the
GPS with an INS. The significantly more demanding performance requirements of the military mission environment drive
the implementation solutions in somewhat different directions
than those of the civil and commercial applications.
Several analyses and experiments support the discussion by Dr.
Cox and have pointed out significant mission capability advan-
26 1
tages of the proper integration. For example in the analysis in

[54] and [50] the integration is performed at the raw pseudorange signal level. That analysis demonstrates that a significant
level of navigational performance improvement can be realized
for the case when fewer than four satellites are available and
pseudo-range and delta-range measurements are processed in a
single integrated aircraft Kalman filter. Incidentally, this same
filter should also be used for ground alignment as well. In fact,
this filter configuration continues to estimate all of the observable INS, GPS. and barometric altimeter measurement errors
including the gyro and accelerometer misalignment errors
throughout all flight phases from power-up.
Thus a GPS/INS, properly integrated in a single box offers several advantages which are not possible with the federated GPS/
INS integration. Those advantages range from significant
reduction in life cycle cost to significantly better overall system
performance. Placing the Kalman filter inside this box and on a
Kalman filter chip should unconstrain this GPS/INS integration and unburden the fire control computer. The GPS jamming
performance must be redefined, because even momentary lockon to the GPS signals at relatively large (several minute) intervals provides performance of an essentially unjammed GPS!
Of course the best solution is to integrate the GPS and INS with
other avionics sensors in a modular integrated architecture.
7.6 Conclusions
The challenge for the military research and development community is to vigorously exploit the simultaneous arrival of the
GPS, the explosion in computational capability, and availability
of the integrated and modular avionics architectures for weapon
systems. These factors offer unprecedented opportunities for
much greater exploitation of avionics sensor fusion. With
proper fusion of the multitude of information available from the
variety of sensors aboard a weapon system, much greater benetits can be derived from the information contained in the GPS
signal. This sensor fusion is a strong function of the avionics
architecture and the variety of forms of information readily
available to any of the fusion algorithms. In integrated and
well-fused sensor avionics architectures, in addition to significant increases in performance potential, the concepts of standardization, system failure, and mission capability require
redefinition. Entire sensor suites can then be revisited with the
question: are all of the current sensors required for military mission performance?
8. ACKNOWLEDGMENTS
The direct and indirect contributions to this paper are acknowledged. Dr. Jesse Ryles, Chief Scientist of Avionics Directorate
and Mr. John Anselmo, Delco Systems Operations have earned
our gratitude for their critical reviews. Particular tribute is to
the numerous AFIT graduate students, who through their thesis
research, contributed several of the modeling and simulation
results presented. Mr. Jeff Laynes technical report on SAR
forms the basis for SAR models.
9. LIST OF SYMBOLS
earths equatorial radius
aE
angular resolution
an*
A
system dynamics matrix
ABR
airborne receiver
AFIT Air Force Institute of Technology
B
system control distribution matrix
bits per second
bPS
BPSK binary phase shifr keying
CADC Central Air Data Computer
CIGTF Central Inertial Guidance and Test Facility
CIRIS Completely Integrated Reference Instrumentation
System
c: ( t ) direction cosine matrix relating e- to n-jirames
direct control input to nieasurement

direction cosine matrix, e.g.
direrential GPS
earth-centered earth-fired referenceframe
extended Kalman filter
eccentricity of earth s ellipsoid
expectation operator
specifcforce
vector: ~f,
f y f21T
f
FOG fiber optic gyroscope
geometric vector: [c v, 9,tlT
G
driving noise w distribution matrix
G
GBR ground-based receiver
GMT Greenwich Meridian lime
GPS
true altitude, altitude above the reference ellipsoid
h
altimeter
output
h
measurement distribution nmtrix
H
HOT higher order terms
HRG hemispherical resonator gyroscope
I
identity matrix
Institute of Navigation
ION
INS
inertial navigation system
barn altimeter model gain
Kl
1
longitude
L
latitude
1575.42 MHz L-band signal
Ll
1227.6 MHz L-band signal
L2
LOS
line-of-siaht
.,MSOFE Multinwde Simulation for Optimal Filter Evaluation
NRS
Navigation Reference System
pdf
probability density function
time derivative operator (with respect to i-frame)
pi
Pmffimprofilegenerator
PSK
phase shift keying
r
position vector: [rx ry r21T= R uLos
[x, yw zJT true SV position vector coordinates
rsv
D
DCM
DGPS
ECEF
EKF
ee
E(.)
[x,, y,, zJT transponder position vector coordinates

[xu,yu, zUlTtrue user position vector coordinates
ru
R
range; measurement noise covariance matrix
RCltY user clock delay
r,
Rclk,, SV clock advance

code loop delay
R D ~ p sDGPS pseudo-range measurement
Riono ionospheric delay
R,
RP
RRp
RI,
R,
RF
RLG
RRS
R
SA
SAR
SV
S
(.)sk
t
to
pjT
=aE/./compensated pseudo-range
raw pseudo-range
tropospheric delay
truerange
radiofrequency
ring laser gyroscope
rangekange-rate system
range-rate
selective availability
Synthetic Aperture Radar
space vehicle (GPS)
Laplace transform operator
skew-symmetric matrix
time
initial time
matrix transpose operator
262
TEC
U
uLos
V
V
total electron count

controlinput vector: [U, uy uzlT
unit-line-ofsight (ULOS) vector
zero-mean white Gaussian measurement noise
velocity vector: [v, v,, vzlT
white, Gaussian driving noise vector: [w, wy wZlT

WGS-84World Geodetic Survey I984 datum
X
state vector: [x, xz ... x,lT
6h
bar0 pressure altitude variation error
81,
baro instrument output dynamic lag error
6hhb baro instrument random bias error
6hB
total ban, altimeter error
6hsJ baro instrument scale factor error
6RclkU user clock bias (range equivalent) error
w
6Rclku user clock drifl (range-rate equivalent) error

8RcIksv SV clock advance bias range equivalent error
tropospheric delay residual
range designation error
ionospheric delay residual
tropospheric delay residual
[6x, 6yl, 6zJT transponder survey error
[tix,,, 6y,, 6z,,lT user position error
elevation designation error
azimuth designation error
dgerential geometric vector: [dr, AV, A$, AtlT
eigenvector
elevation angle (LOS)
eigenvalue, terrestrial longitude
vector of eigenvalues
range designation error standard deviation
range-rate designation error standard deviation
azimuth designation error standard deviation
noise standard deviation
noise variance
integration time variable
instrument dynamic time constant
pressure altitude variation correlation time constant
state transition matrix
attitude vectoc rotation vector: [$, $y $JT
geodetic latitude
azimuth angle (LOS)
earth angular velocity vector with respect to i-frame,
coordinatized in n-frame: [a, any0,7
skew symmetric matrix form of gYc
partial derivative operator
gradient operator: defied as a row vector
References
[ 11 Britting, K.R., Inertial Navigation Systems Analysis, New
York, Wiley-Interscience, 1971.
[2] Broxmeyer, C., Inertial Navigation Systems, New York,

McGraw-Hill Book Company, 1964
[3] Farrell, J.L., Integrated Aircraft Navigation, New York,

Academic Press, 1976.
[4] Fernandez, M. and Macomber, G.R., Inertial Guidance

Engineering, Englewood Cliffs, NJ, Prentice-Hall Inc., 1962.
[5] Kayton, M., and Fried, W., Avionics Navigation Systems,
New York, Wiley, 1969.
[6] Markey, W.R.. and Hovorka, J., The Mechanics of Inertial
Position and Heading Indication London, Spottiswoode, Ballantyne &CO, 1961.
[7] Maybeck, P.S.,Stochastic Models, Estimation, and Control,Volumes 1-3, San Diego, Ca, Academic Press, 1982.
[8] ODonnell, C.F., Inertial Navigation Analysis and Design

New York, McGraw-Hill, 1964.
[9] Pitman, G.R., Inertial Guidance, New York, Wiley, 1962.
[lo] DeCarlo, R.A., Linear Systems - A State Variable
Approach with Numerical Implementation, Englewood Cliffs,
New Jersey: Prentice Hall, 1989.
[l 11 Gaskill, J.D., Linear Systems, Fourier Transforms, and
Optics, New York, New York, John Wiley and Sons, 1978.
[ 121 Nering, E.D.. Linear Algebra and Matrix Theory, New
York, Wiley, 1970.
[ 131 Reid, J.G., Linear Systems Fundamentals. New York,
New York, McGraw-Hill, 1983.
[ 141 Strang, G.,Linear Algebra and its Applications, New
York, Academic Press, Second Edition, 1980.
[15] Kalman, R.E., A New Approach to Linear Filtering and

Prediction Problems, Transactions ASME, Series D: Journal of
Basic Engineering, 1960
[16] Gelb, A., and others, Applied Optimal Estimation, Cambridge, MA: MIT Press, 1984.
[ 171 Jazwinski, A.H, Stochastic Precesses and Filtering Theory. New York, Academic Press, 1970.
[ 181 Meditch, J.S., Stochastic Optimal Linear Estimation and

Control, New York, McGraw-Hill, 1969.
[ 191 Cunningham, J.R., Performance of GPS-aided INS During High-Dynamic Maneuvers, MS Thesis, AFITIGEIENGI
87D-12, School of Engineering, Air Force Institute of Technology, W-P AFB, OH, December 1987.
[20] Carlson, N., and Musick, S., Users Manual for a Multimode Simulation for Optimal Filter Evaluation (MSOFE).
AFWAL-TR-88-1138, W-P AFB OH, AFWAUAAAN-2, Apr
1990.
[21] PROFGEN - A Computer Program for Generating Flight
Profiles, Avionics Laboratory, AFWAL/AARN-2, WrightPatterson AFB, OH, November 1976.
[22] Vander Velde, W.E., Strapped-Down Inertial Systems,
Course 16.43 Notes, Department of Aeronautics and Astronautics, MlT, 1983.
[23] Wrigley, W., and others, Gyroscopic Theory, Design and
Instrumentation, Cambridge, MA, MIT Press, 1969.
[24] Litton Guidance and Control Systems, Performance
Accuracy (Truth ModeYError Budget) Analysis for the LN-93
Inertial Navigation Unit. DID No. DI-S-2 1433 B/T:CDRL
No. 1002. Woodland Hills, CA, Jan 1985.
[25] Pearson, R.A., Advanced Reference System. Proceedings of the Twelfth Biennial Guidance Test Symposium, October 1985.
[26] Windall, W.S., and Grundy, P.A., Inertial Navigation System Error Models, Technical Report, AF-SWC TR 73-26,
Intermetrics Inc., Cambridge, MA, 11 May 1973.
263
[27] Savage, P.G., Strapdown Sensors, AGARD Lecture

Series No. 95, Strapdown Inertial Systems - Theory and Applications, June, 1978.
[28] Savage, P.G., Laser Gyros in Strapdown Inertial Navigation Systems, Proceedings: IEEE Position Location and Navigation Symposium, San Diego, CA, November, 1976.
[29] Spillaker, J.J.,GPS Signal Structure and Performance
Characteristics, Navigation, Journal of the Institute of Navigation, Summer 1978.
[30] Cox Jr., D.B., Integration of GPS with Inertial Navigation
Systems, Navigation: Journal of the Institute of Navigation, pp
144-153, 1980.
[31] Martin, E.H., GPS User Equipment Error Models, Global Positioning System, Janiczek P.M., editor, Institute of Navigation, Washington, D.C., 1980.
[32] Milliken, R.J., and Zoller, C.J., Principles of Operation of
NAVSTAR and System Characteristics, Navigation: Journal of
The Institute of Navigation, 25:3-14, Summer 1978.
[33] Bate, R.R., et al, Fundamentals of Astrodynamics, New
York,John Wiley and Sons, Inc., 1971.
[34] Solomon, J.K., CIRIS Special Study, EENG699 Final
Report, School of Engineering, Air Force Institute of Technology, W-P AFB OH, M m h 1989.
[35] GPS Aided LN-93 INS, EENG735 Class Project Report,

Department of Electrical and Computer Engineering, Air Force
Institute of Technology (AU), W-P AFB, OH, Spring 1990.
[36] Snodgrass, F.B., Continued Development and Analysis of
a New Extended Kalman Filter for the Completely Integrated
Reference Instrumentation System, MS Thesis, AFITlGEl
89M-8, School of Engineering, Air Force Institute of.Techno1Ogy, W-P AFB, OH, March 1990.
[37] Solomon, J.K., Development of the Extended Kalman Filter for the Advanced Completely Integrated Reference Instrumentation System (CIRIS). MS Thesis, AFIT/GE/89M-8.
School of Engineering, Air Force Institute of Technology, W-P
AFB OH, March 1989.
[38] The Analytic Sciences Corporation (TASC), Models for
Aided Inertial Navigation System Sensor Errors. TR-3 12-3,
Prepared under Contract No. DMA700-74-C-0075, Reading,
MA, Feb 1975.
[39]Musick, S.H., Radar measurements and measurement
matrices, Tech. Report AFAL-TM-76-48, Wright Laboratory
(AFMC), W-P AFB, OH, May 1976.
[40]Grettenberg, T.L., and others, SAR Targeting Accuracy
Analysis Program (STAAP) - Hughes Aircraft Company, Final
Report, Tech. Rep. AFWAL-TR-88-1074, Wright Laboratory
(AFMC), W-P AFB, OH, October 1988.
[41]Jeffers, R.H.Biererman, and others, SAR Targeting Accuracy Analysis Program (STAAP) - Emerson Electric Company,
final report, Tech. Rep. AFWAL-TR-88-1030, Wright Laboratory, W-P AFB, OH, June 1988.
[42]Schmidt, D.J. Error Analysis of an Air-to-Surface Missile
with a Synthetic Aperture Radar Seeker, NAECON, pp. 363-370, Dayton, OH, May 1981.
[43]Nelander, J.C., and others, Performance Sensitivities of
Position and Velocity Determination Using SAR Vision Guidance Updates, Proceedings of the Southeast Conference on
Energy and Information Technologies in the Southeast, pp. 9094, Columbia, South Carolina, April 1989.
[44]Kenton, A.C., and others, Vision Guidance Update. Synthetic Aperture Radar (SAR) Multiple Image Exploitation for
Position and Velocity Determination, SPIE Millimeter Wave

and Synthetic Aperture Radar, vol. 1101, pp. 158--169, 1989.
[45]Layne, J.R., Integrated Synthetic Aperture Radar and navigation Systems for Targeting Applications, WUAAAS Draft
Report, W-P AFB, OH, Aug 1992
[46] Stacey, R.D., A Navigation Reference System (NRS)
Using Global Positioning System (GPS) and Transponder Aiding. MS Thesis, AFIT/GE/91M-04, School of Engineering,
Air Force Institute of Technology, W-P AFB OH, March 1991.
[47] Guidance Test Division. CIRIS II Test Summary Report,
In-House Report, CIGTF, 6585th Test Group, Holloman AFB,
NM, May 1988.
[48] Guidance Test Division. The Completely Integrated Reference Instrumentation System (CIRIS), Unpublished Contractor Informational Pamphlet, CIGTF, 6585th Test Group,
Holloman AFB NM.
[49] Green, G.B., and others, The GPS 21 Primary Satellite
Constellation, Navigation, Journal of the Institute of Navigation, 25:9-24, Fall 1989.
[50] Lewantowicz, Z.H., and Keen, D.W., Graceful Degradation of GPSlINS Performance With Fewer Than Four Satellites, The Institute of Navigation, National Technical Meeting,
Jan 1991.
[51] Solomon, J.K., CIRIS Special Study, EENG699 Final
Report, Department of Electrical and Computer Engineering,
Air Force Institute of Technology, W-P AFB, OH, March 1989.
[52] Hirt, E.H., and Yakos, M.D., Time Dissemination Using
NAVSTAR Global Positioning System (GPS) Phase IIB User
Equipment, 35th Annual Frequency Control Symposium,
USAERADCOM, Fort Monmouth, New Jersey, May 1981.
[53] Flight Test Report for Performance Validation of the Global Positioning System on the F-16 Block 40 Aircraft, General
Dynamics Test Report Number 16PR8816.19 June 1990.
[54] Lewantowicz, Z.H., and Keen, D.W., GPS Aided Navigation Performance for Weapon Release with Fewer Than Four
Satellites, Joint Services Data Exchange, October 1990.
[55] Negast, W.J., Incorporation of Differential Global Positioning Measurements for Improved Reference System Performance. MS Thesis, AFIT/GE/91D-41, School of Engineering,
Air Force Institute of Technology, W-P AFB OH, Dec 1991.
[56] Blackwell, E.G., Overview of Differential GPS Methods,

Global Positioning System Papers, Institute of Navigation,
Washington, D.C., Vol3, pp 89-100, 1986.
[57] Abbey, D., and Dance, S.D., Intermetrics, Inc. GPS Data
Analysis Reference Station, Briefing to AFIT Students, Holloman AFB, NM, June 1991.
[58] Dance, S.D., Personal Interviews., Intermetrics, Inc., Holloman AFB, NM, June 1991.
[59] Navtech Seminars, Inc., Dynamic Differential GPS,
Notebook from Course 31 1, Alexandria, VA, December, 1989.
[60] Johnson, G.B., and Lewantowicz Z.H., Closed Loop
Operation of GPS Aided INS, Proceedings, ION-GPS 90 Conference, Colorado Springs, CO, September 1990.
264
Appendix
State
Name
1 ~ ~ ~1
I I I
F51 F32 F23 F18
fvx
fvz
I
I
160A
61A
fZX
Accel. Misalignment
63A
6 R c l ~ User Clock Bias

6DC, User Clock Drift
6Rcdl
code h
p Errors (4)
SEZ
"
x,
I Gyro Misalignment (6)
v2
V?
32G
26
,
I33GI
20
S Q A ~ Accel S.F.Asymmetry
51A
33
SOAv
52A
34
19
71
44
41
15
48
49
Tropospheric Delay (4)
16
I Ionospheric Residual (4)

SV Clock Bias (4)
I9
18
1 I I I
83
6Rclkz
84
6Rclk3
85
6x,2
11
11
87
88
6x,3
89
k V 4
90
6Ysvl
91
6Ysv2
92
6Y-3
93
94
k v l
95
7
23
18
28
14
38G
29
21
43
86
I
31G
41A
42
IO
45
SV (x, y. z) Position Errors
I 1 1
69
46
6Rclkl
19
38
13
6RjOnl
25
37
65A
6Rcd3
w , 2
31G
36
12
6R,,l
I
I
64A
6Rcd2
&d4
SgY
I
I
62A
fZY
F1
I
I
I59AI
a. The LN-93 INS error state model actually contains 93 error states. 5 states
were added to complete the accelerometer misalignments and 3 vertical
channel states to account for the baro altimeter and pressure altitude errors.
From this total of 98 states, 30 least significant states were deleted to form
the error model in this table.
h. The GPS error model contains 30 error states, assuming five independent
receivers, ionospheric correction having been made using dual L, and L2 frequencies
265
FEDERATED FILTER FOR FAULT-TOLERANT INTEGRATED NAVIGATION

Neal A. Carlson
Integrity Systems, Inc.
31 Middlecot Street
Belmont, MA 02178 USA
SUMMARY
This paper describes federated filter applications to integrated, fault-tolerant navigation systems. The federated
filter is an optimal or near-optimal estimator for decentralized, multisensor data fusion. Its decentralized estimation
architecture is based on theoretically sound informationsharing principles. Federated filters consist of one or more
sensor-dedicated local filters, generally operating in parallel,
plus a master combining filter. The master filter periodically combines (fuses) the local filter solutions to form the
best total solution. Fusion generally occurs at a reduced
rate, relative to the local measurement rates. The method is
well suited to real-time system implementation, and can provide significant improvements in data throughput, fault tolerance, and system modularity. This paper discusses federated filter applications to integrated navigation systems in
terms of operating modes, accuracy, fault tolerance, computational efficiency (throughput), and real-time system features. Numerical simulation results are presented to demonstrate federated filter performance characteristics.
1. INTRODUCTION
Integrated multisensor navigation systems have the potential

to provide high levels of accuracy and fault tolerance. The
presence of multiple data sources provides functional redundancy as well as greater observability of the desired navigation states. However, that potential has not always been
fully realizable, via the application of standard (centralized)
Kalman filtering techniques. Applied to multi-sensor systems, centralized filters can result in severe computation
loads when implemented in strictly optimal fashion. Worse,
when used as the components of two-stage (cascaded) filter
architectures, standard Kalman filters can exhibit poor accuracy and unpredictable behavior under certain conditions.
During the past fifteen years, the development of decentralized (or distributed) Kalman filtering methods has received
increasing attention. Parallel processing technology, emphasis on fault tolerant system design, and availability of multiple specialized sensors strongly motivate the development of
such methods. Potential applications include multisensor
navigation systems, multisensor tracking systems, and other
data fusion systems.
Early contributions to optimal decentralized filter theory
were made by Speyer 111, Chang [2], Willsky et al [3],
Levy et al [4], and Castanon et al [5]. While providing
useful insights, these early methods did not appear practical
for real-time navigation system applications, due to their
computational burdens, inter-filter databus loads, and/or
model restrictions.
Kerr proposed an attractive decentralized filtering structure
[6], in which several parallel filters process data from separate navigation subsystems, and a master filter combines
their outputs. Fault-tolerant aspects of this structure were
This work was supported by the Defense Small Business Innovation Research (SBIR) Program under Contract F33615-87C-I520, administered by the Avionics Directorate, WLIAAAI,
Wright Laboratory, Wright-Patterson AFB, OH.
emphasized. Previous work (e.g., [l-41) was cited as a

general theoretical foundation, and no implementation equations for mechanizing the decentralized filter structure were
presented.
A theoretically sound yet practical decentralized filtering
method was developed by Bierman at JPL [6], for orbit estimation purposes. While this method was optimal and efficient on a per-cycle basis, it required the master filter to
operate at the maximum local measurement rate, and the
local filters to have infinite process noise (hence no memory). Bierman and Porter [7] extended that approach to
decentralized parameter estimation. Belzer and Cho [8] also
provided an extension of that approach.
Hashemipour et al [9] developed another unique decentralized filtering method during this same period. This method
required inter-filter data communication at each measurement update cycle, which can lead to heavy databus loads.
More recently, Carlson [ 10-121 developed the "federated"
filtering method based on rigorous information-sharing principles. This method provides globally optimal or near-optimal estimation accuracy, with a high degree of fault tolerance, and is practical for real-time, distributed navigation
system applications. It was adopted as the basis of the Air
Force's Common Kalman Filter for highly fault-tolerant,
next-generation navigation systems [131. The federated filter method is quite general, with several different information-sharing modes. Some modes require simultaneous updates from the LFs to the MF, while others permit independent updates.
Covino and Griffiths [14-151 subsequently developed the
related "net information approach" to decentralized estimation. This globally optimalhear-optimal method is based on
information-sharing principles similar to those of the federated method. It employs two local filters (one processes
measurements, the other does not) and a differencing algorithm to determine the new information gained over each
period. Local elements pass their new information to the
master filter independently for data fusion.
Decentralized filters have also been examined in connection
with multiple-model adaptive estimation. Watanabe [ 161
developed a decentralized multiple-model filter based on the
Speyer-Chang approach [ l , 21. Martin and Carlson [17] extended the federated filter method to obtain a computationally efficient distributed model adaptive estimator. Both the
number (multiplicity) and state-size of the adaptive filter
banks are reduced, by implementing multiple sensor models
at the local filter level, and combining the most likely local
filter solutions in the master filter.
This paper focuses on practical applications of the federated
filtering technique to integrated, fault-tolerant navigation
systems. The remaining sections of this paper describe: (2)
the distributed filtering problem, (3) limitations of standard
Kalman filters, (4) the new federated filtering technique,
(5) system architecture options, (6) real-time implementation
features, (7) numerical simulation results, and (8) conclusions. Appendix A presents implementation equations, and
Appendix B presents a mathematical derivation of the federated filter as a partitioned optimal estimator.
266
2. ESTIMATION PROBLEM STATEMENT

In concept, the federated filter is a partitioned estimation
method. It employs a two-stage (cascaded) data processing
architecture, in which the outputs of local, sensor-related
filters are subsequently combined by a larger master filter,
as illustrated in Figure 1. (This figure shows the major
flow of information, but not all possible data exchanges.)
As indicated, each LF is dedicated to a separate sensor
subsystem, and also uses data from a common reference
system, generally an inertial navigation system (INS).
...........
Reference
.................................
............. .. .. .. .. .
...........
......................
......................
PM,m
SENSOR 2
vij
are uncorrelated, with these statistics:
This last condition is crucial: measurement errors from dvferent sensors i and m are statistically independent. Disjoint sensor data sets permit the total estimation problem to
be divided into sensor-related partitions with independent
measurement processes, as described further in Section 4.
3. LIMITATIONS OF STANDARD KALMAN FILTERS
The primary limitations of standard Kalman filtering methods when applied to multisensor navigation systems and/or
systems with embedded local filters are these:
a) heavy computation loads, and potential inability
to keep up with high-rate sensor measurements;
.......
b) poor fault-tolerance in terms of detecting gradual

sensor faults, and of quickly regenerating a good
solution after a failure; and
-7.
Figure 1: General Federated Filter Structure

The federated filter technique comprises a weighted leastsquares solution to the following linear (or linearized)
estimation problem. First, consider a system state vector x
that propagates from time point t ' to t according to the
following dynamic model:
Here, e is the state transition matrix between time points t

and t , G is the process noise distribution matrix, and is
the additive uncertainty vector due to white process noise
acting ovzr the timestep. The error
in the initial state
estimate 5 and the sequential values g j are uncorrelated,
per the following error statistics (subscripts j and k refer to
different time points t j and t k , and 6jk is the Kronecker
delta):
Our system also has access to external measurements &

from i = l..n separate local sensor subsystems. Measurements from different local sensors are independent, and
comprise disjoint data sets. The discrete measurements
from sensor #i at time t j are linearly related to the true
state zj :
aj = HIj'
ues
zj +
vij
(51
Here, HI is the sensor #i measurement observation matrix

(often d e h e d without the transpose), and v i j is the additive, random measurement error. The sequential error val-
c) inability to correctly process pre-filtered data in

a cascaded (two-stage) filter structure.
The first limitation of a standard, centralized Kalman filter
(CF) is fairly obvious. In Figure 1, the presence of several
sensors generally implies a relatively large number of filter
states n, since each sensor typically introduces one to five
measurement bias states. For a single Kalman filter, the
per-cycle computation load grows roughly in proportion to
n 3 + a - n 2 where
,
cm is the total number of measurements
across all the sensors. The problem is especially severe
when the multisensor suite requires a large, high-rate filter.
The second limitation of a C F relates to fault tolerance.
Like any optimal filter, the C F attempts to make its data
inputs agree in a weighted least-squares sense, thereby suppressing their differences. Hence an undetected failure in
one sensor gets distributed into all of the navigation state
and sensor bias estimates, so that they all tend toward
agreement. Thus, while measurement residual tests can
readily detect sudden "hard" failures, they may completely
miss gradual "soft" failures. If the C F does incorporate
faulty data from any sensor, its full solution becomes corrupted, and must be reinitialized.
The third limitation of a C F relates to cascaded filter processing, and can best be illustrated by means of an example.
Figure 2 shows the major components of a cascaded filter
designed for an integrated navigation system composed of
an inertial navigation system (INS), Global Positioning System (GPS) receiver/navigator. and a radar subsystem. GPS
receiver measurements and INS outputs are processed locally by an embedded GPSlinertial Kalman filter. Periodically, position and velocity outputs from the GPS local filter
are incorporated as discrete "measuremergs" by a Kalman
master filter (MF) in the central computer. The M F also
uses the same INS reference solution as does the GPS local
filter (LF).
Two aspects of this ad hoc cascaded filter design comprise
potential causes of accuracy and/or stability problems, given
a Kalman MF. First, errors in the position and velocity
outputs from the local GPS filter are not sequentially random, but are time-correlated. Second, errors in the LF outputs are not independent of those in the M F state, but are
cross-correlated due to common INS errors.
261
implemented by the federated filter is this:
.........
I
FaU
............
R-hW
......
.:.:.:mLTeR:q............
I:.:.,:_.:.: .''I
............. .. .. .. .. .. .
..... .. .. .. .. .. .. .. .. .. .
.. .. .. .. .. .. .. .. .. .. .. ..
.:_:.:.:.:.:.:.
... . . . . . . . ,...
..
..
..
iq 1;:
MSMT
RADAR
22
M s m t Rate
..
..
......
........... .. .. .. .. .......
........... .. .. .. .. .. .. .
........................
Figure 2: Ad Hoc GPS/Radar Integration Filter

Sequential correlations in the GPS filter outputs mean that
each output contains some new information and some old
information. Treating each GPS output as entirely new information causes the MF to become over-optimistic regarding its own accuracy (i.e, its covariance gets too small).
The resulting problem is especially obvious if we suppose
that GPS measurements become unavailable for several minutes. The MF will continue to incorporate successive GPS
outputs as fresh "measurements", and to reduce its covariance accordingly, even though those outputs contain absolutely no new information. (During the outage, the GPS
filter output covariance will grow, but not fast enough to
prevent the MF from reusing the old GPS information.)
A common, ad hoc "fix" for these correlation problems is
to limit the GPS incorporation rate in the Kalman MF to
once every 10 to 20 seconds. Over these intervals, the GPS
navigator output errors typically become decorrelated (particularly the velocity errors), so that they appear sequentially random to the MF. (Rapid decorrelation is caused by
large process-noise terms in the GPS local filter; these
suppress the time-correlated INS propagation errors, and
emphasize the more random GPS measurement errors.)
This ad hoc cascaded filter design approach can yield satisfactory results in terms of estimation accuracy and stability,
for many applications. However, questions remain as to
whether the same approach will work with different sensors
and local filters. For example, would performance remain
satisfactory if the GPS local filter were tuned with less
process noise, so that its outputs had longer correlation
times? Or, if a less accurate sensor than GPS were used?
Or, if other sensors were added to the navigation suite?
Such questions provide further motivation to seek a theoretically correct foundation for designing and implementing
cascaded navigation filters.
4. FEDERATED FILTER METHOD

The new federated filtering technique presented here avoids
the theoretical and practical difficulties described in Section
3 by means of a simple yet effective information-sharing
methodology. The advantages of information sharing, as
implemented by the new federated filtering technique, are:
o increased measurement data throughput by parallel operation of local filters, and by data compression within local filters;
o enhanced system fault-tolerance by maintaining
multiple component solutions to improve fault
detection and recovery capabilities; and
o improved accuracy and stability of cascaded filter operations, via use of theoretically correct
estimation algorithms.
The basic concept of the information-sharing approach
o divide the total system information among several component (local) filters;
o perform local time propagation and measurement
processing, adding local sensor information;
o recombine the updated local information into a
new total sum.
The remainder of this section will illustrate how the federated filter applies information-sharing principles in its use of
the n local filters (LFs) and one master filter (MF) in Figure
1. The presentation here will be somewhat heuristic, to
emphasize the general concept. Appendices A and B provide implementation details and a more rigorous mathematical derivation of the federated method, respectively.
First, let the full, centralized filter solution be, represented
by the covariance matrix PFAand state vector xf; the local
filter #i solutionAby P I and x i ; and the master filter solution by PM and G . We will use index i = 1. .n for the
LFs alone, and k = 1. .n,m for the MF plus LFs, where
k = m represents the MF.
Now, if the LF and MF solutions are statistically independent, they can be optimally combined by the following additive information algorithm, where the inverse covariance
matrix P - I is known as the "information matrix":
The key to the new federated filtering method is to construct individual LF and MF solutions so they can be combined or recombined at any time by the above simple algorithm. In particular, the construction avoids the need to
maintain LF/LF or LF/MF cross-covariances. The procedure for doing so is the essence of the information-sharing
approach. (See Appendix A for equivalent algorithms that
simplify the P-' operations indicated above.)
Suppose we start with a full solution P F , &. Now, divide
that solution so that the k = 1. .n,m LFs plus MF each
receive fractions pk of the total information:
PP-'
= PM
PK-'
= PF
-1
-1
pk
P1
-1
or
..
PN
-1
P K = PF
(loa)
pk- 1
floc)
>x
fx
(111
k = 1. . n , m
To maintain constant total information across the sum in Eq.

(IO) (the conservation of information principle), the sharefraction values & must sum to unity:
n,m
E
k=l
pk=
pm+
n
E
i =1
pi=
(12)
It is clear that the LF and MF solutions given by Eqs. (10)

and (11) can be rec2mbined per Eqs.(9) to yield the correct
total solution P F , x f .
Next, consider the discrete time propagation process. This
process can also be performed via independent, parallel
operations of the LFs and MF, provided the common process noise information is divided in the same fashion as
268
were the fused solutions. The covariance propagation

equations from time t' to t are:
PK = @K PK'M
GK QK GK
(131
where O K represents the filter #k matrix values. For the

moment, let us assume that the LFs and M F are all fullsized, such that the transition matrices Q?Kequal Q?F,and the
noise distribution matrices GK equal GF. The process noise
covariance matrices OK are governed by the informationsharing principle, however:
Thus, if the PK' (k = 1. .n,m) values have been initially

obtained by Eqs.(lO), and the QK values are obtained by
(14), then the post-step values PK can again be summed by
the simple fusion algorithm (9) to yield the correct total
solution PF:
Third, consider the measurement update pocess. Each LF

#i incorporates discrete measurements
from its own
unique sensor #i. Measurement information is added to LF
#i as follows, where R I - ~is the #i measurement information matrix:
= P I -1
PI;'
PI-'&
HIRI
-1 Zi
where the subscript + refers to post-measurement values.

Again, combining the above results by the fusion algorithm
(9) yields the correct total solution, i.e., the solution that
would be achieved by a single centralized filter processing
all of the i = 1. .n sensor measurement sets:
(17)
=
PF-I
n
E HIRI-'
i =1
It can be shown (Appendix B) that this minimum-LF structure still produces a globally optimal solution in some cases.
One case is when the M F retains all the fused information,
and the LFs are reset to zero information after every fusion
update (i.e., when pm = 1 and pi = 0 ) . However, for general values of s,,, and pi, the minimum-LF structure introduces a slight loss of information with a conservatively suboptimal result. In practice, the resultant estimation accuracy
is almost indistinguishable from that of the globally optimal
filter.
Second, the federated filter can perform fusion updates at a
reduced rate relative to the LF measurement rates, implying
data compression in some or all of the LFs (multiple LF
measurement sets are "compressed" into the latest LF statevector estimate). LF data compression does introduce a
small loss of global information, equivalent to neglecting a
vector measurement of common process noise dimension at
each interior step. This information loss is negligible when
Q << P over the fusion interval, as is the usual situation.
Last, the federated filter method has been described here in
conventional covariance (P) and information ( P - l ) terms.
However, as shown in [18], the federated filter can be implemented in square root (factored U-D) form to maximize
computational efficiency, numerical stability, and effective
precision. One can choose covariance square root form,
information square root form [7,8], or even a mix to best
suit any particular application.
5. FEDERATED CONFIGURATION OPTIONS
H 1 R I - l HIT
(16)
e+=
-1 *
PI+
sensor4 bias states. Hence the matrices PK, #K, GK, and
QK contain only the common and bias-k partitions of the full
matrices. The & fraction values apply only to the common
INS states, since only those states are shared among the LFs
and MF. Appendix A provides representative implementation equations. (The M F can also be implemented in a global or large form containing all of the sensor bias states;
then, each sensor bias-i partition is shared by the M F and
one LF, with corresponding & values of 1/2.)
HI^
PF-'
a+.
A similar relation yields the correct total state vector

Now, Eqs. (10) to (17) demonstrate that the federated filter
solution is the same as that of a single, centralized Kalman
filter, and hence is itself globally optimal, when certain implicit assumptions are satisfied: i) each filter employs a
single gk value for all of the full-system states and process
noises, and ii) the information fusion and reset (dividing)
operations are performed after every measurement cycle.
However, much less restrictive conditions can be accommodated, with only a modest loss of optimality.
First, the federated filter can and should be implemented

such that the LFs are of minimum size, each LF #i containing only the common INS states and its own unique
The new federated filtering technique can be implemented in

a variety of design configurations. Each design derives
from different performance criteria that suggest a different
information-sharing strategy, or "mode". This section describes four federated filter configurations. All four are
suitable for "ideal" applications, in which the local and
master filters can be freely designed to support whatever
federated filter operations are required. Two of them are
also suitable for "constrained" applications, in which the
master filter may be freely designed, but the local filters
are assumed to have been developed elsewhere for standalone filter operations. (We assume that pre-existing LFs
can accommodate few if any software modifications.)
To make the discussion more concrete, we will consider as
an example a hypothetical multisensor navigation system
containing sensors typical of an advanced tactical aircraft.
These sensors and their data output rates (Hz) or intervals
(sec) are:
o Strapdown INS,baro-aided (50 Hz);
o GPS receiver ( I - 2 sec);

o Terrain-aided navigator (TAN) (1/4 - 1/2 sec);
o Synthetic aperture radar (SAR) (100 - 300 sec).
The strapdown INS is the common reference system, normally with a baro-stabilized vertical channel; it puts out
indicated position, velocity, body attitude, and angular velocity. The GPS receiver typically puts out four pseudorange and four pseudorange-rate (or delta-range) measurements per cycle. The TAN radar altimeter puts out one
269
height measurement per cycle. The SAR typically puts out

two landmark imaging or "map" measurements (range and
range-rate) and a set of 6-8 precision velocity measurements (multi-directional range-rate sets) per cycle. This
example system serves to illustrate how various federated
design options can be applied to advanced, integrated navigation systems.
INS
. . ..................
...........
PF,Xf
Figures 3 to 6 illustrate four federated filter modes embodying different information-sharing strategies for this
system. In each case, the federated filter structure consists
of the M F and three LFs. Each LF processes measurements from one external sensor (GPS, TAN or SAR),
while all of the filters use common INS data. Each LF
provides a navigation solution of varying accuracy, depending on the inherent capabilities of its assigned sensor,
and on environmental factors such as visibility, relative
geometry, aircraft dynamics, and electronic interference.
The M F performs generic fusion operations, and generates
the best total solution from the three LF solutions.
Figure 4: Federated Fusion-Reset (FR) Mode
..... . . . . . . . .
INS
!::::Wj&rfEhj
..
.:....:.
......FlkTki:
.. .. .. .. .. .. .. I:..
...........
PF.xf
4
p;-.j
...........
MF provides temporary short-term memory, in propagating

the full fused solution between fusion updates. The FR
mode involves feedback of fusion-reset solutions from the
MF to the LFs. Hence, the less accurate LFs operate at
higher levels of accuracy than they can alone, since they
gain information from the more accurate sensors via the
MF resets. Conversely, the more accurate LFs operate at
somewhat lower levels of accuracy, since they periodically
give up more information than they receive back from the
MF. The FR mode requires all the LFs to send solutions
to the MF for fusion at the same time. Each LF then waits
for a reset solution from the MF before proceeding.
..........
Figure 3: Federated No-Reset (NR) Mode
The primary difference among these four modes relates to

information storage. In the first two modes (Figures 3 and
4), the LFs collectively store the long-term system information, while the M F acts as a short-term information
combiner. Conversely, in the second two modes (Figures
5 and 6 ) , the M F stores the long-term system information,
while the LFs act as short-term data collectors, or data
compression filters. Other important differences exist
among these modes as well.
Figure 3 shows the federated No-Reset (NR) mode. Here,
the LFs collectively maintain the system long-term memory, while the MF provides short-term propagation of the
fused solution after combining (fusing) the LF outputs.
There is no information feedback from the MF to the LFs;
each LF retains its own, unique portion of the total system
information. The NR mode comprises the least optimal
but most fault-tolerant federated filter mode. This mode
permits the LFs to operate independently, as standalone
filters, with estimation accuracies essentially at their
normal levels (except for the minor effects of process noise
multipliers). The LFs all send solutions to the MF for
fusion at the same time. The MF propagates the fused
solution to intermediate time points, but does not use it in
the next fusion update, since the next set of LF solutions
contain all the accumulated system information.
Figure 4 shows the federated filter Fusion-Reset (FR)
mode. Here, the LFs again collectively maintain the
system long-term memory, each retaining a share of the
total fused information while the MF retains none. The
p p
Receiver
Figure 5: Federated Zero-Reset (ZR) Mode
Figure 5 shows the Zero-Reset (ZR) mode. Here, the MF

retains all of the fused information (p, = 1) or system
long-term memory. The three LFs retain none of the fused
information (pi = 0), but act as data compression filters
with short-term memory only. Each LF provides the MF
with the new information it has gained since the last update. The MF adds each new LF solution input to its total
information, in much the same way a standard Kalman filter adds new sensor measurement information. After the
update, the LF resets itself to zero information. An attractive feature of the ZR mode is that the MF can process LF
solutions at different times, according to how rapidly each
LF gains information.
Figure 6 shows the federated filter Rescale (RS) mode.
This mode can be considered a variation of the Zero-Reset
270
With regard to data processing rates, the four federated

filter designs of Section 5 all have the capability to adjust
the MF fusion rate relative to the LF measurement update
rates. When maximum accuracy is required, the M F can
perform a fusion update after every measurement cycle of
the most accurate LF (the GPS filter, for our example system).
Figure 6: Federated Rescale (RS) Mode
mode. Here, each LF passes some but not all of its information to the M F at each fusion update (the fraction m,),
and retains the remainder (the complementary faction ai,
= 1-mi).
With mi= 0.5, the MF accumulates and retains over half of the total system information, while the
LFs collectively retain the remainder. (The MF contains
1/2 of the LF's current-cycle information, 3/4 of the previous-cycle information, 7/8 of the next-previous, etc.) The
LF resets are relatively simple to mechanize: the LF rescales its common covariance partition by a factor of yi,
= l / a i , = 2, and retains its state vector as-is.
A useful way to characterize these different informationsharing strategies is as follows. In the NR and FR modes,
each LF tells the MF "what I know now," and the MF
combines those results to obtain "what we all know now."
Conversely, in the ZR Mode, each LF tells the MF "what
I have learned since my last report," and the MF adds that
new information to its total of "what we all know now."
The RS mode is similar to the ZR mode, except that the
LFs pass along what they have learned more gradually.
With regard to constrained filter applications, the NR mode
is best suited for use with existing LFs, since it requires no
additional software functions beyond standalone operations.
(The common process-noise multipliers I / @ , are often unnecessary, given conservative process noise models.) The
RS mode may also be usable with existing LFs, if they can
be modified to accommodate periodic rescaling of the covariance matrix. The ZR mode is not directly usable with
existing LFs, but may be approximated by restarting each
LF (at a large, initial covariance value) whenever a zeroreset is required. Last, the FR mode is not usable with
existing LFs, since resetting the LF state and covariance to
MF-supplied values is generally infeasible.
One final point should be noted here. These four federated
filter modes involve exactly the same mechanization equations for all operations except the MF and LF resets (see
Appendix A). They can all be implemented in a single
software package, with a code branch for each reset option. Even a centralized filter can be implemented with the
federated filter software [ 191, by assigning all the sensors
to a single LF or MF and turning off the other filters.
6. REAL-TIME IMPLEMENTATION FEATURES

Four alternate federated filter designs were presented in the
previous section, for the example integrated navigation system. This section examines those alternate designs more
closely from the viewpoint of real-time implementation
features, including processor and sensor fault tolerance.
When maximum MF accuracy is not required, and a reduction in computation burden is desirable, the MF fusion rate
can be reduced to well below the LF measurement rates.
Thus, the relatively large fusion computation burden can be
reduced to, say, one-fifth or one-tenth of the fully optimal
burden. For example, if the TAN LF performs measurement updates twice a second, and GPS does so once a second, the MF might perform fusion updates only once every
ten seconds. In this way, the TAN LF compresses 20
measurements (20 x 1) into a single state vector, and the
GPS LF compresses 80 measurements (10 x 8) into a single
state vector, both used by the MF to obtain its fused
solution.
The Zero-Reset (ZR) and Rescale (RS) modes are most
flexible in this regard, since they permit fusionheset
processing to occur at different times for different LFs.
For example, in the ZR mode, an LF can "dump" its information to the MF and reset itself to zero information at
any time, regardless of when the other LFs do so. In contrast, the No-Reset (NR) and Fusion-Reset (FR) modes require all LF solutions to be propagated to the same time
point (a fairly simple control process), and then to be
passed to the MF where they are all combined to yield the
total solution. Since the total system information is distributed across the LFs, all of the LF solutions must be collected at the same time, to obtain the best total solution.
Another real-time software issue relates to resets from the
MF to the LFs. The NR mode is the simplest in this regard, for the obvious reason: there are no resets. The ZR
and RS modes are relatively simple, since each LF can
reset itself, after sending its solution to the MF, without
waiting for a response from the MF. The FR mode is the
most complex, since each LF has to wait for a reset solution from the MF, before it can continue its normal operations.
Next, consider processor fault tolerance -- the ability to
detect and recover from filter dropouts caused by processor
or bus failures. All of the federated filter modes provide
enhanced processor fault tolerance in a way not possible
with a single, centralized filter (provided the federated
filter components are running on separate processors.)
Each component sends out periodic status messages to the
other components, and monitors their status messages in
return. For example, if an LF disappears, due to a processor or bus failure, the federated controller (FC) detects
its absence almost immediately, via the missing LF status
message. The FC then modifies the information shares for
the remaining filters, and tells the MF to use only those
LFs in the fusion process. Thus, the federation gracefully
reconfigures to the remaining set of available LFs. If the
LF later reappears, the FC reconfigures the federation
again, to re-include the LF.
Finally, consider sensor fault tolerance -- the ability to
detect, isolate, and recover from sensor failures. Here, the
federated filter designs support sensor fault detection,
isolation and recovery (FDIR) at both the local and master
filter levels. For purposes of comparison, it is helpful to
begin by considering the sensor FDIR capabilities of a centralized filter (CF). The CF maintains a single, globally
optimal solution, and incorporates measurements from all
local sensors. It has good capability to detect and isolate
some local sensor failures via measurement residual tests.
27 1
because each measurement is compared against all of the

accumulated prior information.
One disadvantage of the CF is that, like any optimal filter,
it attempts to make its data inputs agree in a weighted
least-squares sense, thereby suppressing their differences.
Hence, an undetected failure in one sensor will be distributed into all of the navigation state and sensor bias estimates, so that they all tend toward agreement. Thus,
while measurement residual tests can readily detect sudden
"hard" failures, they may completely miss gradual "soft"
failures. A second disadvantage of the CF is that, if it
does use bad data from any sensor, its full solution
(accumulated information from the whole system) becomes
irreversibly corrupted. When the faulty sensor is subsequently identified, the only safe means of recovery is to
reinitialize the entire CF solution (state and covariance),
and wait for it to reconverge to its normal accuracy level.
In comparison, the federated filter method provides several
advantages with regard to sensor FDIR. Although specific
advantages vary with the operating mode, there are a
number of common features. First, the individual LFs
accumulate sensor information over several cycles between
MF fusion updates. Thus, a sensor soft failure has more
time to reach detectable magnitude before it meets residual
tests in the MF. Second, each LF maintains bias states for
its local sensor alone. Other sensor bias states are maintained separately by other LFs, and are not drawn into
agreement with the faulty sensor (except in the case of
fusion resets). Third, once a sensor fault has been detected and isolated, uncorrupted LF solutions may still
exist (depending upon the reset mode), so that recovery to
a good M F solution can be quickly achieved.
More specifically, the federated No-Reset mode is highly
fault-tolerant. Since there is no feedback of the MF fused
solution to the LFs, there is no possibility of LF-to-LF
cross-contamination. Each LF has good sensor FDI capability through measurement residual tests against all of the
prior local information, except for certain types of soft
failures. Most important, the MF has excellent LF:LF
FDI capability, since each LF solution contains full accumulated information from its own sensor. Thus the effect of
a sensor soft failure will be increasingly visible 'in one LF
solution only, greatly increasing its probability of detection
by the MF. The NR mode also provides a significant advantage for fault recovery. When the faulty LF is identified, a new M F solution can be generated immediately,
from the remaining good LF solutions.
The other federated filter modes are less attractive with
respect to sensor fault tolerance. The Fusion-Reset mode
provides good measurement-residual screening at the LF
level, since each LF contains a share of the full system
information. However, the MF FDI capability is very
weak, since the fusion-reset process causes all LF solutions
to agree except for their most recent sensor measurements.
Also, if faulty data from one LF is accepted by the MF,
the fused solution is corrupted, and then the other LFs
become corrupted via the MF resets. Thus, the MF and
LF solutions all require reinitialization.
The Zero-Reset mode provides poor measurement-residual
screening at the local level, since the LFs are reset to large
covariance values after each fusion update. The MF FDIR
capabilities are comparable to those of the CF. The MF
has all of the prior system information for use in LF fusion-residual checks, but will still have difficulty detecting
some types of soft failures. If faulty LF/sensor data is
accepted by the MF, the total system solution held by the
M F is corrupted, and must be reinitialized. The unfailed
LFs do not hold any long-term information useful for quick
MF recovery.
The Rescale mode provides fair measurement-residual

screening in the LFs, since the. LF residual tolerances
increase by no more than I/olii" (e.g., 1.414) in sigma
units, after each fusion update. The MF FDI capabilities
are similar to those of the ZR mode, and comparable to
those of the CF; hence, certain types of soft failures will
be difficult to detect. The recovery capability is slightly
better than that of the ZR mode, since the unfailed LFs
retain some accumulated information.
In summary, the No-Reset mode stands out from the others
as having superior sensor FDIR capabilities. It also has
the simplest real-time control requirements, since it requires no LF resets. Its only disadvantages are a slight
degradation in estimation accuracy, and the need for all LF
solutions to be combined by the MF at the same time (generally not a problem). The Zero-Reset and Rescale modes
permit LF solutions to be processed at different times by
the MF, but are not attractive from a sensor FDIR viewpoint. The Fusion-Reset mode provides the best estimation
accuracy, but exhibits disadvantages with regard to sensor
FDIR, computation and databus loads, and real-time software control requirements.
7. FEDERATED FILTER SIMULATION RESULTS

The federated filter and its several information-sharing
modes have been implemented in computer software to
support performance simulation testing. A general-purpose
federated filter package has been built in FORTRAN-77,
for non-realtime simulation testing on VAX and PC (IBMcompatible 80x86) computers [19]. In addition, a generalpurpose federated filter has been built in Ada, with asynchronous multi-tasking capabilities, for real-time testing in an
avionics simulation testbed [20].
The primary goals of the simulation tests reported in this
section have been to:
a) demonstrate global optimality or near-optimality
of federated filter configurations;
b) examine characteristics of master and local filter
solutions relative to one another, and to that of
the equivalent centralized filter;
c) demonstrate improved capability of federated
filter to detect and isolate sensor faults, and to
recover with minimal delay;
d) demonstrate reduced computation load per processor (improved throughput) of federated filter
compared to centralized filter.
7.1 Performance Simulation Scenario

The federated filter has been tested in both non-real-time
and near-real-time simulation environments. The non-realtime FORTRAN environment (the "DKF Simulator" [19])
contained a high-dynamic aircraft trajectory generator, a
strapdown INS model, a baro altimeter model, a GPS satellitekeceiver model, a SAR model (with landmark imaging
and precision velocity updates), and a TAN model (radar
altimeter plus synthetic terrain-map generator). The nearreal-time FORTRAN/Ada environment (the "DKF-Ada
Testbed" [20]) contained the same elements except for the
SAR model. The sensor truth models were generally of
medium-high fidelity, containing significantly more error
sources than those in the corresponding filter models.
Both environments permitted the filter navigation state
estimates to be compared with the true states, to determine
the estimation errors. Filter computation times were measured in the DKF Simulator.
In the simulation runs, the tactical aircraft trajectory consisted of a climb to altitude, a straight high-altitude seg-
212
CENTRALIZE0
ERRORS
2.5
it021
FEO/FR HF ERRORS
EPOSE
SIGPE
2.5
3021
EPOSE
SIGPE
-2.51
TIME (SI
1000
4000
EPOSE
SIGPE
FT
0.0
-2.51
1000
Y
TIME (SI
TIME (SI
4000
Figure 9: Federated Fusion-Reset Estimation Errors
Figure 7: Centralized Filter Estimation Errors

FEO/NR I F ERRORS
1000
4000
Figure 8: Federated No-Reset Estimation Errors

ment, a descent, and an extended low-altitude segment including several jinking maneuvers (zig-zags) and a large
course change. GPS pseudorange and pseudorange-rate (or
delta-range) measurements were processed by the filter at a
rate of 1/5 Hz. Some runs included two significant GPS
data outage periods, separated by a brief period of tracking.
SAR position and velocity measurements were processed at
a comparatively slow rate of once every 100 sec; the SAR
was turned off near the end of the low-altitude segment,
approaching the "target". TAN radar altimeter height measurements were processed during the entire low-altitude
segment, at a relatively rapid rate of 1 Hz.
7.2 Federated Filter Optimality Results

The results in this subsection were obtained with the nonrealtime DKF Simulator [19]. Figures 7, 8 and 9 demonstrate the global optimality of the federated filter. Each
figure shows the east position estimation error (feet) for five
separate monte carlo runs, plus the filter-computed onesigma uncertainties, over a one-hour period from 1,OOO to
4,600 sec. (The north and vertical errors and sigmas are
qualitatively similar.) Figure 7 shows the performance of
the globally optimal centralized filter, while Figures 8 and 9
show comparable results for two federated filters, operating
in the No-Reset and Fusion-Reset modes, respectively. The
federated filters performed fusion updates at the GPS measurement rate, once every 5 sec, for maximum M F accuracy.
The same random number sequences were used to generate
the truth data for each case.
The three sets of error and sigma plots are very similar to
one another, although some minor differences are visible.
The same levels of agreement -- very similar error and sigma plots -- occur for the velocity and attitude states as well.
In each case, the errors and sigmas are relatively small while
GPS data is available. (Plotting these segments on an expanded scale shows virtually identical results for the three
filters.) The estimation errors grow during the GPS outages,
with periodic reductions due to the SAR position and velocity measurements. In between SAR updates, the filter errors
and sigmas increase somewhat, reflecting the reduced accuracy attainable from TAN radarherrain measurements.
While we could discuss the relative performance attributes of
the three navigation sensors at length here, the real point of
interest is this: the federated filter navigation pevormance
is virtually identical to that of the globally optimal centralizedfilter. Hence, these simulation results confirm the theoretical prediction that the federated filter is a near globally
optimal filter formulation. The federated Fusion-Reset results are nearly identical to the C F results. The federated
No-Reset results are slightly less accurate, as the theory
(Appendix B) predicts.
7.3 Federated Filter Component Performance
The results in this subsection were generated with the realtime Ada version of the federated filter, running in the
VAX-hosted DKF-Ada Testbed 1201. Each figure (10 to 14)
shows the east position estimation error and sigma (feet)
versus time (sec). The runs are for a relatively short 200sec segment of the previous trajectory, including the final
stage of descent, initial low-altitude penetration, and a few
horizontal jinking maneuvers. Only the GPS and TAN sensors (and not the SAR) were employed for these runs; hence
there were only two active LFs in the federated filter.
These real-time simulation results represent single-sample
monte carlo runs, with the same random number sequences
used for each case. The single-sample results are consistent
with the more general results obtained from multi-sample
monte carlo runs in the non-real-time DKF Simulator.
Figure 10 compares the globally optimal centralized filter
(CF) results with those of the federated master filter operating in the Fusion-Reset (FR) mode, with equal (50/50)
information-share fractions pi for the two LFs. Figure 11
makes a similar comparison of the C F with the federated M F
operating in the No-Reset (NR) mode, again with equal LF
share fractions. In both cases, the federated M F results are
nearly identical to those of the CF, and to each other. There
are small differences in fine structure between the MFs and
the CF, due to their different update rates. (The C F incorporates one GPS measurement set every 2 seconds, whereas
the MFs combine two or three GPS measurement sets, via
LF1, into one fusion update every 5 seconds.)
213
CENTRALIZEO
1.2
E+O2[
.
0
FEO/FR MFIL
1.2
r+021
EPOSE
SIGPE
.
0
L F l STO-ALN
EPOSE
SIGPE
LF2 STO-ALN
0.0
0.0
- 1O . 2 2160
2160
TIME
(SI
'
TIME IS1O
2360F
2360
2160
Figure 10: CF vs. Federated MF, Fusion-Reset 50/50

FEO/NR MFIL
CENTRALIZEO
TIME (SI
TIME (SI
2160
2360
2360
Figure 12: Standalone LF1 GPS) and LF2 (TAN)

LF2 FEO/NR
EPOSE
1.2
3021
* EPOSE
SIGPE
EPOSE
EPOSE
SIGPE
* EPOSE
SIGPE
0.0
O
-1.21
.
2160
'
. . . . . . .
TIME (SI
P
.
'
-O
1 . 2 2160
'
TIME
(SI
O
2360
P
-1.6
2360
2160
Figure 11: CF vs. Fede ted MF, No-Reset 50/50

Note that the federated NR and FR modes again yield nearidentical results (right-hand sides of Figures 10 and 1l), just
as they did in the earlier test case (Figures 8 and 9). We
conclude that, even though the NR mode is theoretically less
optimal than the FR mode, it is very nearly as accurate as
the FR mode (and the CF) for typical navigation systems.
TIME
(SI
2360
T I M E (51
2160
2360
Figure 13: Federated LF1 and LF2 (NR 50/50)

LF1 FEO/FA
1.2
Et021
~____
LF2 FEO/FR
EPOSE
SIGPE
Figure 12 shows the two LFs (GPS, TAN) operating as

standalone (locally optimal) filters. These standalone LF
results are of interest primarily for comparison with federated LF results (later). However, comparing these results to
those of the C F (Figure 10) also indicates the accuracy improvement in the C F obtained by using both sensors instead
of just one.
EPOSE
1.2
Et021
SIGPE
i'
Figure 13 shows similar results for LF1 and LF2 operating

as components of the No-Reset federated filter with equal
(50/50) information-share fractions. Here, the LF common
process noise strengths are 1.414 (sqrt-2) times as large as
those in Figure 12. However, virtually no difference in
performance is evident in the position errors and sigmas of
either LF. For the velocity errors (not shown), there is a
very small increase in sigma growth rate between measurements, but the overall results are still virtually the same.
Hence, we conclude that the federated process noise multipliers have very little effect on LF performance relative to
the standalone case.
each other. They are also virtually identical to the corresponding M F results (Figure l l ) , except that the LF sigmas
are larger by a factor of 1.4, due to the l / p l = 2 fusionreset multipliers. We conclude that fusion resets dominate
the performance of the component LFs in the FR mode, at
least when the fusion interval is not much larger than the
primary measurement interval.
Likewise, Figure 14 shows results for LF1 and LF2 operating as components of the Fusion-Reset federated filter,
again with 50/50 information shares. In contrast to the NR
case, both LFs here show errors and sigmas significantly
different from their standalone results. Except for the tine
structure, the two sets of LF results are virtually identical to
7.4 Sensor Fault Detection Results

Federated and centralized filter performance results regarding sensor fault tolerance have been obtained in the non-realtime DKF Simulator environment. The results reported here
are for a GPS satellite clock failure consisting of a frequency
2
-1
.-
2
-1
.2160
TIME (SI
2360
2160
TIME (SI
2360
Figure 14: Federated LF1 and LF2 (FR 50/50)
274
CF NORM MSMT RESIOS
GPS
oGPS
GPS
* GPS
it00
PRl
PR2
PR3
PR4
TAN L F l
ONE-SIG
ND
0.01
0 0
TIME
1500
3000
Figure 15: CF GPS PR Residuals after Failure

CF NORM MSMT RESIOS
3000
Figure 17: MF Fusion Residuals after Failure

FEO/NR MF POSN ERRS
GPS PAR1
o G P S PRR2
0 GPS PRR3
AGPS PAR4
* ONE-SIG
4.0.
Et00
TIME IS1
1500
3.0
Et02
EPOSE
EPOSN
EPOSV
SIGPE
* SIGPN
0 SIGPV
D
FT
0.0
1 .o
0.0
1500
A .
-3.0
TIME
3000
1500
.
TIME
(SI
MOO
Figure 16: CF GPS PRR Residuals after Failure
Figure 18: MF Position Errors after Failure
bias shift of 0.5 ft/sec, starting at time 1,700 sec (late in the
high-altitude outbound cruise segment of the flight profile).
The clock frequency failure introduced an 0.5 ft/sec bias into
the subsequent pseudorange-rate measurements, and an 0.5
ft/sec ramp into the subsequent pseudorange measurements,
from the faulty satellite. This scenario represents a "soft"
(slowly-growing) failure in the most accurate navigation
sensor -- generally the most difficult type of failure to detect.
failure, the C F horizontal position estimates exhibit a steadily growing radial error that reaches 425 ft by time 3,000
sec (1,300 sec after the failure).
Figures 15 and 16 show the centralized filter (CF) rms measurement residuals for the GPS pseudorange (PR) and pseudorange-rate (PRR) measurements, over the period from
1,500 to 3,000 sec; the failure occurs at the A mark, 200 sec
after the initial time shown. These rms residuals are normalized, with an expected value of unity, and are smoothed via
a ten-point moving average, to make the trends more visible.
The PR residuals show no effect of the satellite #3 failure, in
spite of the ramp in PR #3. The smoothed PRR residuals for
satellite #3 rise to about 3.5 sigma, while the unsmoothed
PRR #3 residuals (not shown) rise to just over 5.0 sigma.
The CF rejects the PRR #3 measurements while their residuals exceed the specified 3 a threshold. However, the sequence of PRR #3 residuals peaks, then falls back down to
acceptable levels (due to filter covariance growth between
measurements). Thus, the CF loses its ability to detect the
failure, and from that point on, uses the faulty data with no
clear indication that it is bad.
The C F measurement residuals from the less accurate SAR
and TAN sensors likewise provide no clear indication of the
GPS failure. (One of the SAR measurements does eventually show a ramp in its residual sequence; however, the CF
then rejects those good SAR measurements.) After the
The enhanced FDI capability of the No-Reset federated filter

is evident in Figure 17, which shows the M F fusion residuals. In each fusion cycle, the M F starts with the TAN LF
solution, then adds the SAR LF solution, followed by the
GPS LF solution. The SAR fusion residuals (SAR:TAN inconsistencies) are well behaved, whereas the GPS fusion
residuals (GPS:SAR/TAN inconsistencies) ramp off and provide an unmistakable failure indication. (These are normalized rms fusion residuals, with an expected value of unity.)
Figure 18 shows the corresponding M F position errors.
With a residual threshold set at 3a, the M F rejected the GPS
LF solution at about 2,500 sec (800 sec after the initial
fault). At this time, the radial position error had grown to
about 300 ft; this value is consistent with the SAR/TAN
accuracy level of about 100 ft, and the 3a threshold. (Note:
the jumps in the M F errors near 2,500 sec were caused by
noise in the GPS fusion residual as it crossed the 3a threshold, such that the GPS LF solution was rejected, accepted,
then rejected again.)
Note that, within the GPS LF of the No-Reset federated filter, the measurement residuals are qualitatively similar to
those of the CF. The PR residuals give no indication of a
failure. The PRR residuals give a temporary indication, then
decay back to normal levels. This temporary indication is
even weaker in the GPS LF than in the CF, since the LF
covariance and computed residual variances are larger than
those of the CF.
215
'
1,4 .......................................................................................................................
..........
11.2 ......................................................................................................................
.........
,4 ....................................................................................................................................................
...............................................................
..........................................
LFGPS LFSAR LFTAN MF/NR
7.5 Relative Computation Time Results

Computation times have been measured in the non-real-time
DKF Simulator [19], for the federated filter No-Reset mode,
and for the equivalent centralized filter. The various filter
sizes and processing rates for these timing studies were:
Filter
States
Operation
LFGPS LFSAR LFTAN MFMR
CF
Figure 19: Peak Cycle Times, LFs and MF vs. CF
Period
LF 1
18
Propagation
8 GPS msmts
2 sec
2 sec
LF2
17
Propagat ion
1 TAN msmt
1 sec
1 sec
LF3
23
Propagation
7 SAR msmts
5 sec
300 sec
MF
16
Propagation
Fusion (equiv
to 32 msmts)
10 sec
10 sec
CF
27
Propagation
1 TAN msmt
8 GPS msmts
7 SAR msmts
1 sec
1 sec
2 sec
300 sec
In these cases, the filter INS model had 16 states (it had 13
in the previous tests), the GPS model had 2, the SAR model
had 7, and the TAN model had 1. The MF implemented
only the 16 common INS states, since only common information was shared. The M F propagated only its state vector, since the propagated M F covariance is not needed when
the M F retains no information.
Figure 19 shows the peak-cycle execution times (sec) for
each of the filters -- the three LFs and the M F from the
federated filter, and the centralized filter (CF). Each total
time is subdivided into the major functions: propagation,
measurement processing, and fusion. Each of the federated
filter components requires considerably less processing time
for its peak cycle than the CF, ranging from approximately
29 to 56 percent of the C F peak cycle time. Thus, if implemented on separate, parallel processors, the federated filter
would provide approximately a 2: 1 advantage.
Figure 20 shows the average-cycle execution times (sec) for
each of the filters. Here the federated filter components provide an even greater speed advantage, with average computation times ranging from 7 to 40 percent of the CF average
time. Thus, the average per-processor computation loads of
the federated filter are significantly less than that of the CF.
Note that, even if the federated filter average times were
added together (as if they all ran on the same processor),
they would still sum to slightly less than the CF average
time.
These timing results are meant to be illustrative. The speed
CF
Figure 20: Average Cycle Times, LFs and MF vs. CF
advantage of the federated filter over a single centralized

filter is greatest when the number of common INS states is
relatively small, and the number of sensor bias states is relatively large. The illustrative case is not very favorable in
this regard, even though the federated filter still exhibits a
useful speed advantage. (The advantage would be greater,
with fewer common INS states, and/or more sensor bias
states.) These filter timing values were obtained on an
80386/20MHz processor with an 80387 math coprocessor.
(With a different FORTRAN compiler on the same computer, we obtained computation times 50 percent less that those
shown here, because of more efficient object code for the
double-indexed matrix operations).
8. CONCLUSIONS
The new federated filter method presented in this paper

forms a practical and straightforward basis for constructing
integrated multisensor navigation filters with enhanced levels
of fault tolerance. These filters exhibit a number of advantages over centralized Kalman filters, particularly for systems with embedded local filters and/or distributed components:
o theoretically correct partitioned estimation algorithms for cascaded systems with sensor-dedicated local filters (LFs);
o
multilevel fault detection, isolation and recovery

(FDIR) capability; highly fault tolerant in NoReset mode with independent LF solutions;
o increased data throughput due both to parallel

processing of LF operations, and to sensor data
compression in LFs;
o potential reductions in system integration costs
due to modular architecture with sensor-dedicated
LFs and generic, sensor-independent MF.
Relative to previous decentralized filtering methods, the
federated filter also provides one or more of these advantages:
o near globally optimal estimation accuracy of
master filter fused (combined) solution;
o straightforward real-time implementation due to
largely independent LF operations, particularly in
No-Reset, Zero-Reset and Rescale modes;
o minimal data transfer requirements: filter states

and covariances at fusion update times only.
The multisensor navigation system example illustrates the
216
variety of design configurations, or information-management

strategies, permitted by the federated filter method. Of the
several alternate configurations presented, each has attractive
features for some but not all applications.
Numerical simulation results to date confirm the theoretical
predictions regarding near global optimality of the federated
filter method, reduced per-processor computation loads, and
enhanced capability of the federated No-Reset mode to detect, isolate and recover from sensor failures. Additional
simulation test results are reported in [18].
The practicality of implementing the federated filter method
in real-time Ada software has also been demonstrated. The
Ada federated filter has been tested in a near-real-time avionics simulation testbed, and again confirmed the theoretical
performance predictions, including near globally optimal
accuracy, reduced per-processor computation burden, and
robustness in the presence of processor failures.
ACKNOWLEDGEMENTS
The author gratefully acknowledges the contributions to this
research of his colleagues: David Kawa, Daniel Martin and
Clark Neily, Jr. of Integrity Systems; and Michael Berarducci, Peter Howe and Stanton Musick of the Air Force
Wright Laboratory.
APPENDIX A: IMPLEMENTATION EQUATIONS
This appendix presents implementation equations for the
federated filter. These equations are expressed in familiar
covariance form, to facilitate understanding. However, the
equations have been recast in factored U-D (unitary-diagonal) form [18,21] in software implementations where processing speed and numerical precision are important [19,20].
We will use subscript i to refer to LF #i, and subscript m for
the MF. The set of LFs is denoted by i = l..n, and the set
of all filters (MF plus LFs), by k = l..n,m. Each filter is
partitioned into the common INS states and its own unique
sensor bias states. These definitions apply to filter #k:
for all the sensors. When the M F retains information, the

global form may provide somewhat better accuracy; however, it is much larger, slower, and more complex than the
small form.)
The various information-share fractions and multipliers used
by the federated filter satisfy these constraints:
n
The & fractions relate to initial information shares, processnoise shares, and fusion-reset shares (they can be in different
ratios for each operation). The ark, fractions relate only to
the fusion-input and reset operations of the Rescale mode.
The federated filter algorithms will be presented in five
steps: 1) the LF/MF initialization equations; 2) the LF/MF
time propagation equations; 3) the LF/MF measurement update equations; 4) the M F fusion equations; and 5 ) the MF
and LF reset equations for various reset modes. For each
step, we will first indicate the basic operation (marked by m)
in unpartitioned covariance form, then provide the partitioned implementation equations and/or the relevant partitioned
matrix factors. Note that, in Step 2, the prime refers to
values at the beginning of the propagation; in Step 3, superscript + refers to post-measurement values; in Step 4, superscript + refers to values after each LF #i fusion; and in Step
5 , the quote symbol " refers to values after the reset.
1) LF/MF Initialization:
filter k common states
(A-6)
(A-1)
filter k bias-k states

(A-7)
PK =
pcc
"ck]
ILpKkc
(A-2)
2) LF/MF Propagation from time t ' to t:
PKkk]
Fcc
wc PK'
PK =
rn
where "bias-k" means the biases for sensor #k. The full or
global state vector xf contains all of the filter common and
sensor bias states. The filter #k state vector & and the full
state vector xf are related by a simple mapping matrix LK:
+~ GK~
OK G K ~ :
XA= @K xA'
(A-8)
(A-9)
0ckl
1
lo
L
'kk]
(A-10)
Lone
GK =
0nk,1
In this appendix, we will present equations for the "small"

form of the master filter, in which the M F contains only the
common states and any MF-unique sensor biases, as indicated by Eq. (A-1) with k = m. (There is also a "large" or
global form of the MF, in which the M F contains all of the
full filter states xf, i.e., the common states plus bias states
rGcc
lo
'kk]
1;
PK =
3) LF/MF Measurement Update:
rPccYk
IL
1
'kk]
211
BK = P K H K ;
PK+ = PK
AK =
T
HK P K H K + RK
BK AK-'
B K ~
(A-lla)
(A-llb)
PM
(A-25)
PIccXii
(A-13)
p I i chi i
4) M F Fusion UDdate (Common States):
(A-26)
-1 T
+ P I ic P I ccPIic(Xii-l)
PIii
A
k=1
PFcc
Start:
= PMcc
(A-14)
= xm
xf
-C
(no change)
xi
(A-15)
-C
(A-27)
d ) Zero-Reset Mode:
Use Rescale mode (c) with
Xii
l/Ei
>> 1
(A-28)
Do for L F s i = 1..n:
AIcc
= PFcc
PIcchi
PF:c
= PFcc
xfc
-C
+ PFcc
5 ) MF/LF Fusion-Reset
rn
F AI,:
~ ~ P
= xf
=>
PF-'
( A - 16a )
-1
AIcc
xfc)
(xic
~( A - 16b)
(A-17)
("1:
n,m
E LK PIC"-'
k=l
LK~:
a ) No-Reset Mode:
xm"
-C
= Xfc
PI" = PI
A
(A-18)
Use between
fusion times,
then discard
A
J
(no change)
(A-20)
(no change)
(A-21)
b) Fusion-Reset Mode:
-1
-1 T
PKik = PICkk- PKkcPKcc(PKcc- PK;c)PKccPKkc
~<~-i
PF~~PM~:
PM" =
-1
LPM,cPMccPFcc
PM-
PMA
(A-29)
Note that, in the Rescale and Zero-Reset modes, the M F can

incorporate different LF solutions at different times, via Eqs.
(A-14) to (A-17), with resets per Eqs.(A-24) to (A-27).
Also, some care must be taken in resetting the LF covariance partitions to large values representing "zero" commonstate information. For example, attitude error covariances
may need to be limited in size, because of small-angle linearity assumptions, and weak observability of those states.
(The reset multiplier Xi, can be limited in each LF, to prevent reset variances from exceeding specified maximums.)
The federated filter implementation equations presented in
this appendix have been cast in equivalent U-D factored
form, coded in both non-realtime FORTRAN software (all
modes) and real-time Ada software (NR and FR modes), and
validated by extensive simulation testing [ 18-20].
B. THEORETICAL FOUNDATION
This appendix presents a derivation of the federated filtering
method outlined in Section 4. We will formulate the optimal
estimation problem as a sequential, weighted least squares
problem. The full or global least-squares problem will be
partitioned into n local least-squares problems. Then, the
solutions of the n local problems will be combined, via another least-squares operation, to yield the total fused solution. Under certain conditions, this fused solution is globally optimal -- identical to the full, unpartitioned solution.
Under other, more practical conditions, it is near-optimal,
using most but not all of the available information.
Consider the sequential system dynamics and discrete measurement processes described by Eqs. (1) to (8) of Section 2.
First, define S , W and V as square roots of the state error,
process noise, and measurement noise covariance matrices:
S=P%;
c) Rescale Mode:
(A-19)
xA11 = x i
hi = l/(l-ei)
I
J
(A-24)
= PMm- P M m c P M ~ ~ ( P M c cPFcc)PMccP
mc
';
W = Q
V = R
(B-1)
Now, the globally optimal state estimate for this system at

time tk is the value of
that minimizes a weighted least
squares cost function $k+ of all the random (unknown) err?'
terms from step 0 through step k+. The initial estimate 5
is known, with error covariance square root So. The unknown variables are the true initial state 5,
the sequence of
process noises LII, a?d the se uence of sensor #i measurement noises v i j = z i j - H I j 8 c j , where index j refers to
278
successive times t,, adn index i , to the n independent

sensors. Index k + refers to the post-measurement value.
The cost function $k+ can be expressed as follows [7]:
T -T
B = W Z G S
(B-10)
2,
Thus, the optimal pre-measurement value of z is

which
zeroes the first new term in Eq. (B-6). AThesecond term can
be zeroed by choosing g = - E S-' (x-x_). The n e y s is
the covariance square root of the propagated estimate E.
Next, the measurements at step k can be optimally incorporated. We first use Eq. (B-4) to rewrite the measurement
error
in terms of the computed measurement residual
A a and the state error:
~
The sequence of true state vectors z, is also a set of unand the

knowns. These unknowns are linearly related to
process noises
by the state propagation equations:
I,
= e x
j J-1
+ G u
j J
( j = 1..k) (B-3)
a,
At step j , the i = 1. n sensors provide measurements

of the state gi,, containing the independent measurement
errors vi, ma ing up the last set of terms in $k+:
zi
-3
= HI
z +
j
vi
-j
vi
&&-
A&
H I X_ = A&
HI
T A
H I (x_
TA
5) (B-11)
(8-12)
We then rearrange the elements of the squared state and

measurement terms to obtain a new term in x_ alone. The
final cost function at step k + is this:
(j = l..k)(B-4)
The weighted least squares solution is the set of values for

the unknowns 5,
gj (or equivalently z,), and vi, that
minimizes the valueAof the cost function $k+, given the
initial state estimate 5 and the sequence of measurements
&. The process of determining those optimal values (or
estimates) can be performed sequentially. To do so, we rewrite the full cost function $k+ in terms of its accumulated
residuals A$k-l at the previous step, and the new terms
introduced at step k :
where, for i = l..n, the new terms in Eq. (B-13) can be

computed in covariance square root form as follows:
FI = S
HI
A & = & - H I
The valye of
that minimizes this cost function at step
k - 1 is
the computed estimate after step k - 1 . sk-lis
h:e covariance square root of the error in that value. Both
and sk-1 are assumed to have been propagated from
their initial values by recursive step-wise operations, as will
be outlined below for the step from k - 1 to k + .
Now, we can determine the best estimate of
after the
time propagation and before incorporating the measurement.
We use Eq. (B-3) to eliminate a-lin the first term of (B5), and then rearrange the elements of the first and second
terms to produce a new first term in a alone:
(B-14)
T XA_
x-+ = z +
JI CI
-1
(B-17)
A&
(B-18)
Thus, the new optimal estimate of x_ is +;, which zeroes the

first new term in (B-13). S+ is the new covariance square
root. The process noise can still be choszn to zero the
second term, given the optimal value x = z+. The third
term is an irreducible least-squares residual; it is composed
entirely of measured and computed values, and contains no
unknowns. In summary, Eqs. (B-7) to (B-10) and (B-14) to
(B-18) represent the recursive solution of the globally optimal estimator.
With this foundation, we can proceed to develop the partitioned optimal estimator. First, we divide the original cost
function $+ in Eq. (B-5) into n partitions, each containing
the measurement term from one unique sensor:
In Eq. (B-6) we have dropped the subscript k for simplicity,

using a prime (') to denote prior values at k - 1 . New terms
on the right side of Eq. (B-6) have these values:
$+ =
$l+ +
..
+ $n+
$i+
(B-19)
i 11
(B-7)
B = W [I
F ( P F +S S
-T - I F I X
(B-9)
where the S I , G I , W I , and
terms represent the ith parti-
279
tion. The original terms in Eq. (B-5) are exactly equal to

the sums of the corresponding terms in Eq. (B-20) if these
"conservation of information" conditions apply:
The underlying conservation of information principle can be

stated as follows: the total information in the partitioned
estimator must sum to the same value as the total information in the unpartitioned global estimator. This e uali
must hold true for a) the state information P-' = S %s ,
b) the process noise information Q-' = WTW-', and c) the
measurement information R-' = V T V - ' . The measurement
information is naturally divided according to the n independent sensors. The state and process noise information can be
divided as follows, where hi and ri are diagonal sharematrices whose squares sum to the identity matrix:
-?
ated filter. Now, the question arises as to whether $f and

SF are equal to the globally optimal solution given byEqs.
(B-15) to (B-18). The answer is "yes," under certain limited

conditions, and "almost," under other more practical conditions. The answer revolves around the second, process-noise
term in Eq. (B-25). The globally optimal solution requires
that the sum of these partition terms also be minimized. In
principle, this sum could be rearranged such that there were
only two squared terms in the unknowns and.5:
i =1
A 2
- wall
llJAx_
^ 2
llawll
The first term in Eq. (B-30) can be zeroed by proper choice

of L,I given the optimal value of g. The third term is an
irreducible residual. The second term has the same form as

a vector measurement, i.e., the
term in Eq. (B-11). Because this term involves g,it could in principle be combined
(via a final least-squares operation) with the earlier solution
WI-'
= ri w-';
E r2 = I
(B-24)
i
given by Eq. (B-28). However, doing so is generally not
necessary, since the information content of this remainder
Now, each partition of the global estimation problem repreterm is generally negligible. Doing so is also quite undesirsents a local estimator, or local filter. Each cost function Jli able, since the additional computations are very burdensome.
is minimized by the corresponding local filter operations.
After local filter #i is propagated from time t ' to t,and its Now, this remaining information term is identically zero
unique sensor #i measurements are incorporated, its cost under some conditions, meaning that the federated filter
solution is then globally optimal. For example, suppose that
function Jlihas the following form, similar to Eq. (B-13):
the total information is divided proportionately among the n
local filters at time t', via the use of scalar share fractions
-1
2
Jli, = A q i ' + IISI+ (xi, - x_)II
in Eqs. (B-23) and (B-24), and with yi = Xi:
2
E hi = I
= hi S - ' ;
SI-'
(B-23)
IIBI-'[g
BI SI
-1
(xi A
x_)]1I2 ( 8 - 2 5 )
where the terms in (B-25) are defined analogously to those in

the total cost function (B-13). Now, the globally optimal
estimate of x_ is the value that minimizes the sum of the Jli
partitions, i.e, the total cost. Each partition contains g in its
first term, and both g and g in its second term. We can
minimize the sum of the first terms in Eq. (B-25) as follows:
(B-26)
where A$ is an $educible fusion residual (a function only:f

known terms in xi). The full, combined state estimate fx
and its covariance square root SF can be computed either by
orthogonal transformations on ~ 7 - [7,21],
l
or by the equivalent reduction operations on S similar to (B-15). The results
can be expressed as the following information sum:
-1
SI-'
= Xi S
WI-I
= Xi W-';
EX:
= 1
2
EXi = 1
(Ai
= IXi)
(8-31)
(ri
= IXi)
(B-32)
This construction causes each i-term in Eq. (B-30) to be

exactly the same, except for different scalar multipliers Xi.
Thus these terms readily reduce to a single term (having a
= l ) , with no residual. Hence, the
scalar coefficient A,,
remaining information term is identically zero for this case,
and the fused solution (B-22) is globally optimal.
However, if each local filter takes several steps including
measurement updates, then the i-terms in Eq. (B-30) are no
longer simply related by scalar multipliers. In this case, the
remaining information term is not zero, and the fused solution (B-26) is not globally optimal. If we examine the information content of this remainder term, we find that it has an
equivalent measurement information (R-') value as foIIows:
RA-'
-T
VA
VA
-1
P
- + P-
A[(P
-1- -1
Q
P)
1 (B-34)
n
SFaTSF-'
SI-TSI-l
(B-27)
i =1
sF
-T - l A
sF xf =
n
I:
i =1
sr-Tsr-lk
(8-28)
These operations represent the "fusion" process in the feder-
Eq. (B-34) represents the additional information "lost" in the

multi-step process, where we have assumed G = I for simplicity. The A implies a difference across the several local
filters. The _value P equals apeT,before adding Q . Thus,
when Q c c Pover each local propagation step, as is generally the case for navigation filters, the information content of
the unused remainder term is much less than the filter infor-
280
mation P-' that is used in the fusion process:
(However, if Q 2 i ,which is not usually the case, then the

lost information could be significant.) Thus, operating the
local filters independently over several steps is equivalent to
ignoring an available but relatively weak (inaccurate) vector
measurement at each such step. The resulting fused solution
is quite valid, but somewhat suboptimal, in that not all of the
available information has been incorporated. (This loss is
minimal, compared to the information loss in a single large
filter that can't keep up with all the measurements.)
A similar small information loss occurs for the case where
the local filters contain only the common states plus their
own sensor bias states. Here, the information-share matrices
Ai and ri for local filter #i contain only two non-zero partitions, where Aci2 is the common-state share fraction:
FAci
(B-36)
5.
Castanon, D.A., and D. Teneketzis, "Distributed Estimation Algorithms for Nonlinear Systems," IEEE Transactions on Automatic Control, AC-30, May 1985.
6.
Kerr, T.H., "Decentralized Filtering and Redundancy

Management for Multisensor Navigation, " IEEE Transactions on Aerospace and Electronic Systems, AES-23: 1,
January 1987.
7.
Bierman, G.J., and M.R. Belzer, "A Decentralized

Square Root Information Filter/Smoother," Proceedings of
24th IEEE Conference on Decision and Control, Ft.
Lauderdale, FL, Dec. 1985.
8.
Bierman, G.J., and D.W. Porter, "Decentralized Tracking via New Square Root Information Filter (SRIF) Concepts," Business and Technological Systems, Inc. and
Bierman and Associates, May 1987.
9.
Belzer, M.R., and Y.M. Cho, "Micro-Computer Network

Architecture for Range Instrumentation Applications,"
Mentor Technologies, Inc., March 1988.
10 Carlson, N.A. and Neily, C.M. Jr., "Distributed Kalman

Filter Architectures," Phase I Final Report, AFWAL-TR87-1 181, Avionics Laboratory, Wright-Patterson AFB,
OH, June 1987.
11. Carlson, N.A., "Federated Square Root Filter for Decentralized Parallel Processes, IEEE Transactions on Aerospace and Electronic Systems, AES-26:3, May 1990.
"
n
2
E h i = I
i =1
n
C
2
Aci
= 1;
i =1
(B-37)
The previous cost-function partitioning and fusion operations

still apply, given 'these values of Ai and ri. The information remainder term for the fusion process -- the second term
in Eq. (B-30) -- can also be expressed in partitioned form.
This remainder is identically zero (meaning the result is
globally optimal) in the hypothetical case of a system with
no common process noises. When common process noises
do exist (the usual case), the lost information is proportional
to the ratio (B-35) for the common states. In navigation
system applications, this value is generally quite small,
meaning that the federated loss of information relative to the
globally optimal filter is also quite small.
Note that Eq. (B-35) also implies that the one-step information loss is exactly zero when the LFs are reset to zero
information (Ai = 0) after each fusion update, such that g-'
= 0, while the M F retains all the information (A,,, = I).
Hence, the federated filter Zero-Reset mode is globally
optimal in the single-step fusion case, even when each LF
implements only the common and local bias states.
REFERENCES
1.
Speyer, J.L., "Computation and Transmission Requirements for a Decentralized Linear-Quadratic-Gaussian

Control Problem, IEEE Transactions on Automatic Control, AC-24 :2, Apr. 1979.
I'
2.
Chang, T.S., "Comments on 'Computation and Transmission Requirements for a Decentralized Linear-QuadraticGaussian Control', IEEE Transactions on Automatic
Control, AC-25:3, June 1980.
I'
3.
Willsky, A S . , M.G. Bello, D.A. Castanon, B.C. Levy,

and G.C. Verghese, "Combining and Updating of Local
Estimates and Regional Maps Along Sets of One-Dimensional Tracks," IEEE Transactions on Automatic Control,
AC-27:4, August 1982.
4.
Levy, B.C., D.A. Castanon, G.C. Verghese, and A S .

Willsky, "A Scattering Framework for Decentralized
Estimation Problems," Automatica, 19:4, April 1983.
12. Carlson, N.A., "Federated Filter for Fault-Tolerant Integrated Navigation Systems," Proceedings of Position
Location and Navigation Symposium, PLANS, Orlando,
FL, November 1988.
13. Loomis, P.V.W., N.A. Carlson, and M.P. Berarducci,
"Common Kalman Filter: Fault-Tolerant Navigation for
Next Generation Aircraft, " Proceedings of the Institute of
Navigation National Technical Meeting, Santa Barbara,
CA, January 1988.
14. Covino, J.M. and B.E. Griffiths, "A New Estimation
Architecture for Multisensor Data Fusion," Proceedings
of the International Symposium and Exhibition on Optical
Engineering and Photonics, Orlando, FL, April 1991.
15. Griffiths, B.E. and J.M. Covino, "Net Information Approach Phase I Final Report," Synetics Corp., Wakefield,
MA, May 1990.

16. Watanabe, K. "A Decentralized Multiple Model Adaptive
Filtering for Discrete-Time Stochastic Systems," Journal
of Dynamic Systems, Measurement and Control, Vol. 111,
September 1989.
17. Martin, D.P. and N.A. Carlson, "Distributed Model
Adaptive Estimation," Phase I Final Report, TR-91-003,
Integrity Systems, Belmont, MA, August 1991.
18. Carlson, N.A., "Distributed Kalman Filter Architectures
Phase 11," Final Report, to be published as a WL. technical report, Wright Laboratory, Wright-Patterson AFB,
OH, January 1995.
19. Carlson, N.A., "User's Manual for the Distributed Kalman Filter Simulator (DKFSIM)", Integrity Systems,
Belmont, MA, July 1994.
20. Carlson, N.A., D.P. Martin, and D. Kawa, "Distributed
Kalman Filter (Ada) Real-Time Testbed," Final Report,
WL-TR-93-1045, Wright Laboratory, Wright-Patterson
A m , OH, Mar 1993.
21. Bierman, G.J., Factorization Methods for Discrete Sequential Processes, Academic Press, New York, 1977.
28 1
GPSlINERTIAL INTEGRATION OVERVIEW

Richard L. Greenspan
Charles Stark Draper Laboratory, Inc.
555 Technology Square
Cambridge, M A 02139
1. INTRODUCTION
GPSAnertial integration is the process whereby a superior system

navigation solution is produced by properly combining outputs
from a GPS user equipment (UE) and from an inertial navigation
system ( I N S ) . This process is receiving much attention because it
is perceived to be a cost-effective means to satisfy navigation
requirements that could not be met by either CPS or by an INS
acting by itself. Some expectations levied on integrated systems
arerealistic; others, including hopes for mass-market commercial
applications, will be delayedpending the development of suitable
low-cost inertial technology.
The technical basis for considering GPS/INS integration is the
complementary nature of the navigation errors for each system
operating stand-alone. The GPS solution is relatively noisy; the
noise-driven variance of GPS positioning errors is on the order of
ameter per axis, per positiondetermination. However, GPS enors
are bounded, whereas inertial navigation errors are dominated by
a low-frequency component that grows in proportion to the
mission duration.
high-frequency content of inertial errors is
very small, amounting to a few centimeters (rms) over tens of
seconds.) One expects that an integrated navigation solution
would perform like aninertialnavigator whoseerrorswerebounded
by the GPS solution. This performance is actually achieved using
one of the least aggressive approaches to integration; further
benefits achieved using more aggressive integration options are
discussed in the following sections.
me
The first applications of CPS/INS integration have been in military aviation. This isnot surprising because inertialnavigation has
been almost exclusively applied in military systems. One might
object that commercial airliners have carried inertial navigation
systems since the 1970s. Although this is hue, civil aviation
authorities have only recently granted approval to install CPS
receivers on aircraft and their use will be limited to provide a
supplemental navigation aid until at least the mid 1990s.
In contrast, military airborne applications emphasize the ability of
an integrated system to meet requirements for precision delivery
of weapons or materials on target, even during effective radio
countermeasures that cause outages in GPS availability. This
chapter will focus on the technical issues that arise in satisfying
those requirements. The key issues for civilian applications are
not yet as clear but the following are likely to be primary:
System Integrity - The use of internal or external signals and
measurements to immediately detect and prevent out-ofspecification measurements or data from corrupting the
navigation solution.
ii) Partial Outages - Providing the means to use all available

GPS measurements even when astand-a1oneGPS solution
cannot be formed.
iii) Low COS - Developing technology that can be sold at
commercially viable prices. This is especially a concern for
inertial components.
Item (i) is not discussed in this chapter. However, there is already
an extensive literature on integrity techniques. Ref (1)provides a
good survey of receiver autonomous integrity monitoring (for
GPS); numerous other techniques are being discussed that exploit
the integrationof several independent sensors (Ref. 2-4). Item (ii)
is addressed in Section (3) in connection with the discussion of
deeply coupled integration. Item (iii) is raised in the discussion
of error characteristics of inertial navigators also in Section 3.
This chapter is organized in three expository sections. These
provide, 1) an overview of the benefits and trade-off issues of
integration, 2) an overview of integration architectures and algorithmic concerns. and 3) sample results from performance evaluations of proposed integrations. In all of these discussions, the
properties of GPS UE and inertial navigation systems cited here
are generic rather than specific and are representative of technology circa 1992.
2. BENEFITS O F GPSmVERTIAL INTEGRATION

The design of any complex navigation system for civilian or
military markets reflects the designers judgment of the best tradeoff among the following factors:
cost
a)
b)
development (non-recurring)
life-cycle (recurring)
Installation Constraints
a) volume, weight, power consumption
b) interfaces
Performance
a) mission requirements/mission environment
b) reliability/graceful degradation
c) options for improvement
The following remarks will emphasize the performance considerations because that is the area where the benefits of GPSAnertial
integration are most evident. However, cost and installation
factors are often decisive. These are raised throughout the chapter
wherever they may be a significant differentiator between alternative integration techniques. Ultimately, the system designer must
justify his design as being the best way to satisfy the design
282
problem. It is of critical importance that the authorities who are

managing the design team surface all requirements and constraints, both present and anticipated, so that informed and timely
choices can be made among the alternatives.
The GPS system can provide a suitably equipped user with a
Position Velocity, and Time Solution (PVT) whose errors are
generally smaller than those of any alternative navigation system.
This performance is achieved in all weather, at any time of the day.
and under specified conditions of radio-frequency (RF), signal
availability, and vehicle dynamics. Why then would one undertake the cost and complexity of integrating GPS User Equipment
(UE) with any other navigation sensor, and in particular, with an
inertial navigator?
The goal of integration is to provide more robust, and possibly
more accurate, navigation service than is possible with standalone sensors. In particular, integration may be the only way to
achieve the following (Ref. 5-10):
Maintain a specified level of navigation performance during
outages of GPS satellite reception.
Provide a complete six degree-of-freedom navigation solution (translational and rotational motion) at a higher output
rate than is conventionally available from GPS alone.
Reduce the random component of errors in the GPS navigation solution.
Maintain the availability of a GPS solution in the presence of
vehicle dynamics and interference.
2.1 Operation During Outages

A stand-alone GPS receiver typically incorporates current measurements to four or more satellites to update its most recent PVT
solution. Deadreckoning thatmay incorporaterecent estimates of
vehicle acceleration is used to propagate the current PVT solution
in between measurement updates. A GPS outage occurs when
fewer than four valid satellite measurements are available at each
update. During a partial or complete outage, the software for a
stand-alone receiver can continue to produce a navigation output
. if it mechanizes one of the following options, albeit with reduced
accuracy:
a) Compute the least-squares solution with fewer measurements than there are unknowns. Ref. (1 1)
b) Constrain one or more navigation outputs to be fixed, such
as the UE clock bias or the vehicle altitude, or constrain the
navigation solution to lie along a great-circle path. Ref.
(12,131
c) Incorporate measurements from an external sensor. A
barometric indication of altitude is commonly available in
military UE as are radar or pilot inserted position updates.
Ref. (13,14)
During an outage, the navigation solution becomes less accurate
the longer the outage and the greater the vehicle dynamics since
the last full set of measurements. The key factor to be specified
when deciding whether an auxiliary sensor is required is the
maximum acceptable error growth during the outage. In aconser-
vative design, maximum error growth is calculated under worstcase conditions of vehicle dynamics.
Outages may be a concern even for UE that track more than four
satellites at a time. For example, a GPS antenna mounted on top
of an aircraft will only see a limited number of satellites during
a banked turn, and the Dilution of Precision (DOP) parameters
for that visible constellation may be unacceptably high. In more
extreme cases, a vehicle passing through a tunnel may see no
satellites for an extended period, and a military UE may be
jammed as it approaches its target.
Combining GPS with an independent navigation sensor (item c,
above) is aparticularly attractive means to maintain the quality of
the navigation service during a GPS outage. In effect, the
independent sensor can act like a flywheel to provide continuous high quality navigation outputs. Inertial navigators are commonly considered for this role because they are passive, selfcontained, widely available, and are not subject to the causes of
GPS outage. However, they are generally more expensive to buy
and integrate than other radio-navigation sensors such as LORAN
or Omega. Their use has generally been limited to military and
commercial aircraft. However, low-cost, low-performance inertial sensorsimplementedusing mass-production microelectronics
technology are emerging from research laboratories. These may
provide the technological basis for an economical solution to GPS
outages in civil aviation and commercial applications such as
bucks and automobiles. Ref.(15,16)
With respect to GPS/INS integration performance during outages,
the key questions are:
(1) What quality INS is required?
(2) How complex is the integration required to exploit the

inherent INS quality to achieve mission objectives?
The resulting performance must then be weighed against the cost

to determine whether to implement the optimum integration or to
accept a less expensive, lower performance solution.
2.2 Providing All Required Navigation Outputs
GPS UEs routinely estimate only the translational motion of a
point referenced to the GPS antenna. Interferometric processing
of GPS signals received at multiple antennas can also provide
rotational(attitude) information. Ref. (17-19) However, weassert
that inertial sensing is preferable to interferometry for terrestrial
users whenever it is available. This preference is based on three
considerations in addition to superior performance:
(a) The inertial system is self-contained and is not vulnerable

to outages (except those caused by equipment failure).
(b) Installation of an inertial system on an aircraft is less
demanding than an interferometer, and is probably less
demanding on ships and vehicles too.
(c) The theoretical noise floor on the accuracy of a shortbaseline (1-5m) interferometer has not yet been achieved.
It appears that multipath is the culprit, and that is it
premature to expect that anti-multipath techniques will be
effective and practical. Ref. (20)
283
In addition to attitude indication, the inertialnavigator is desirable

because its accelerometers typically sense velocity changes at up
to a 1.0 KHz rate, with a 200 Hz output rate beiig commonly
available. Therefore the INS routinely outputs navigation solutions one to two orders of magnitude more often than a GPS UE.
This high output rate allows the INS to provide accurate inputs to
vehicle control subsystems, platform-stabilization systems, or to
pilot displays, and velocity aiding inputs to GPS tracking loops.
It follows that an integration in which GPS is used to bound the
error growthofanINS-basedsystemnavigationsolutionwouldbe
very effective whenever GPS was available, and the availability of
a calibrated INS may be the only means to maintainnearly as good
performance during an outage.
23 Reduced Noise in GPS Navigation Solutions

In a stand-alone GPS receiver, the navigation processor usually
impl&ents a linear filtering algorithm in which the previous
navigation solution is propagated to the current measurement
epoch. Because GPS does not directly sense acceleration, the
propagated solution is sensitive to errors in the previous accelerationestimateor tojerk that changes the trueaccelerationduring the
propagation interval. In contrast, an inertial system measures
position change very precisely in the interval between GPS
updates. This property can be exploited by a well-tuned Kalman
Filter, using GPS measurements to estimate errors in the INS
output. Since these errors change slowly, the filter can smooth its
update over many GPS measurements thereby reducing the effect
of additive noise on any one update. The result is that the
integrated navigation solution appears to be much smoother
than the stand-alone GPS solution. Ref. (21,22)
2.4 Increased Tolerance to Dynamics and Interference

The INS velocity solution may be fed back to the GPS UE to
reduce the apparent dynamics of the input to the GPS code and
carrier loops. This has two effects; Ref. (5.8.10)
(1)
A fixed bandwidth aided tracking loop can maintain lock on

GPS signals in the presence of dynamics that would cause
the unaided receiver to break lock.
(2)
The tracking loop bandwidths can be reduced to the minimum amount required to track the errors in the INS aiding
signals.* (INS position errors are mostly low-frequency).
The net result of these actions is that the INS aided GPS receiver
can maintain lock and provide GPS measurements over a much
wider range of vehicle dynamics and radio-frequency interference
than the unaided, stand-alone receiver. In particular, there is
provision to operate both RCVR 3A and the MAGR in the INS
(aided mode), or the DRS (Doppler Aided Mode) when improvements in GPS availability at high levels of jamming or
dynamics arerequired. In the following remarks, weconsider only
the INS mode because there is almost no experience with operating any GPS receiver in the Doppler mode. Operation in the INS
mode provides three benefits to a GPS receiver.
(1) The INS velocity output can be used to aid the MAGR codetracking loop when the carrier loop has lost lock. This in turn
means that the code loop tracks only the errors in the INS aiding.
The bandwidth of these errors is much less than the bandwidth of
* This
feature breaks down for unauthorized users in the

presence of selective availability clock dither.
the raw dynamics, hence the tracking loop bandwidth can be

reduced. This bandwidth reduction gains additional jammer
suppression. Thus the specification for State 3 (code loop)
tracking is increased from J/S = 46dB to J/S = 54 dB in the INS
aided mode.
(2) In the INS mode the INS position and velocity are treated by
theMAGR as areferencenavigationsolution that is tobecorrected
by GPS measurements. A direct consequence of using the INS as
a reference in the MAGR is that interpolation/ extrapolation/
synchronization errors in the receiver output are much less sensitive to acceleration and jerk than in the PVA mode. (i.e. the INS
mode foreground solution is almost as good as the background solution.) Another consequence is smoother output. It is
aproperty of the INS reference that the INS errors, though usually
muchlargerthanMAGRperformancespecifications. aresmoothly
growing functions of time: i.e.. the short term quality of the INSbased reference (noise level. precision) is much better than the
PVA solution maintained by an unaided receiver. Software in the
receiver (a Kalman filter) is used to produce an output that
combines the best features of the INS with the GPS: i.e., the
receiver navigation solution output in the INS mode is smoother
than in the PVA mode and (in the presence of valid measurements)
has errors that satisfy the MAGR performance specifications over
a wider range of dynamics and jamming.
(3) The combined INS/GPS solution is continuously available

during any interruption of GPS signals (outage). It can be used to
reduce the uncertainty in platform location during a GPS outage
to a small enough uncertainty that certain rapid reacquisition
strategies built into the MAGR can be exercised. This has the
effect of reducing the duration of a GPS outage by hastening
reacquisition.
3. GPS INTEGRATION ARCHITECTURES AND ALGORITHMS
The degree of complexity of the integration should reflect the
mission requirements; it may also be limited by the investment
that can be made to obtain those objectives. Integration strategies
and mechanisms may be very simple (for example: choose the
GPS UE position and velocity as the integrated solution when
GPS is available with a given precision, or else choose INS
position and velocity as the integrated solution) or relatively
complex (for example: optimally combine GPS UE measurements with INS outputs, Doppler radar outputs, baroaltimeter
signals, true airspeed, and other sensor data). However, in the
following remarks we limit our attention to alternatives involving
only GPS integrated with an inertial system.
3.1 Integration Architectures

Figure 3-1 illustrates three generic functional architectures for
GPS INS integration. The GPS receiver and the INS are treated as
navigation systems in architectures a and b. with GPS supplying
a Position, Velocity, and Time solution (P,V.T) and the INS
supplying a Position, Velocity. and Attitude solution (P,V,0),
respectively. In architecture c, the GPS and INS are treated as
sensors producing line-of-sight measurements
(P, P) and accel-
erations and angular rates (AV, A8)respectively. In addition to the

GPS and INS units, each one includes various data paths and a
processor unit that mechanizes the integration algorithm. These
alternatives are distinguished by the data that passes between the
sub-system components. The proper interfacing and control of
284
these components may incur the largest part of the cost of an

integrationproject, but those concerns are not within the scope of
architectural considerations.
3.1.1 Stand-Alone Mode
Figure 3-la illustrates the configurationin which GPS UE and an
INS produce independentnavigation solutions with no influence
of one on the other. The integratednavigation solution is mechanized by an external integration processor that may be as simple
as a selectoror as complexas amulti-modeKalmanFilter. Alldata
busses are simplex (unidirectional). The characterization of
Figure 3-1a as astand-alonemode is based on the independence
of the GPS and INS navigation functions. Note that, in principle,
the hardware could all be packaged in one physically integrated
(embedded)unit yet the functionality would be that of a standalone architecture.
The potential benefits of a stand-alone integration are:

a) It is the easiest, fastest, and potentially the cheapest approach
when an INS and GPS are both available.
b) It provides some tolerance to failures of subsystem components (except in the embedded configuration, see 3.3).
c) Using an integration processor as simple as a selection algorithm can provide en route navigation at least as accurate as
available from an INS.
3.13 Loosely Coupled Mode
Figure 3-lb illustrates a configuration in which there are several

data paths between the integrationprocessor and the GPS and the
INS equipment. Among these, the provision of the system
navigation solution to the GPS UE is the most important for
getting the maximum benefit from the integration filter. The
inertialaiding of GPS tracking loops is of next greatestbenefit,and
feedback of error states to the INS is of secondader benefit.
There may also be some benefit for system reliability to the extent
that individual components are mature and have been well tested
compared to emerging technology.
3.13.1 Reference Navigation Solution
GPS UE generally employ a Kalman filter mechanization to

compute(P.V,T) updatesbased on current tracking loopmeasurements. A stand-alone UE does not directly sense acceleration; it
must use relatively noisyaccelerationestimatesbased on recent
velocity measurementsfor a deadreckoningpropagationof the
previous navigation solution forward to the epoch of the current
tracking loop outputs. The situation changes dramatically when
the System Navigation solution is fed back to perform that
propagation. In effect, the GPS measurements can now be used
(within the UE navigation filter) to correct the system navigation
solution. Over short periodsof time, that solution is very accurate
because it incorporates INS data based on acceleration sensing.
The UE filter is then mechanized to estimateINS (or System)error
states having relatively low dynamics and low uncertainty (process noise). The filter can be tuned to have a longer time
constant (filter memory) thereby increasing the effective averaging of each noisy GPS measurement. The result is a great
reduction in noiseness of the GPS P,V,T solution.
(P,V,T,e)s
A) STAND-ALONE MODE
B) LOOSELY COUPLED MODE
AIDING
SENSOR
INTEGRATION
PROCESSOW
ENSOR
C) TIGHTLY COUPLED MODE

Figure 3-1. Generic GPSllNS integration Architectures.
3.13.2 Inertial Aiding of GPS Tmcking Loops

As mentioned in Section 2. the availability of a GPS navigation
solutioncan be significantlyincreasedwhen inertial aiding is used
to reduce the vehicledynamicsthat are tracked by the UE code and
carrier loops. In principle, this aiding could be applied directly
from the INS to the GPS UE, but it is shown as an output of the
integration processor in Figure 3-lb because:
GPS tracking loops must be aided by the projection of vehicle

velocity along the line-of-sight (LOS)to each satellite being
tracked. The conversion from inertial coordinates to GPS LOS
coordinates is most appropriately done in the integration processor or in the GPS UE itself. In either case, INS velocity
information is available within the processor hence aiding can
be part of the data flow to the UE. This avoids the expense and
risk of developing a custom interface from the INS to the GPS
UE.
Executing the coordinate transformation external to the INS
retains flexibility in the selection of INS equipment and avoids
the need to develop custom GPS/INS interfaces for each application. However, this raises a concern for data latency (i.e.
feeding delayed data to the tracking loops) as discussed below
in Section 3.3.
3.133 Error-StiateFeedback to the INS
Most inertialnavigationsystemshave the means to acceptexternal
inputs to reset their position and velocity solutions and to adjust
the alignment of their stable platform. The adjustment may be
executed by a mathematical correction in a strapdown inertial
system, or it may be realized by torquing a gimballedplatform. In
285
either case the use of feedback can maintain inertial navigation

errors at a level for which their dynamics are accurately modeled
by thelinearizederror statepropagationequationsusuallyembodied in the integration filter. However, this is aminor consideration
because position and velocity errors would have to exceed many
km and many m/s before non-linear effects become important.
3.1 3 Tightly Coupled Mode
Figure 3-lc illustrates the so-called tightly coupled integration
mode. It differs from the loosely coupled mode in that both the
GPS receiver and the inertial components are limited to their
sensor functions. They are treated as sources of GPS code and
carrier measurements and inertial indications of acceleration
(velocity change) and angular rate respectively. These sensor
outputs are then combined in one navigation processor that may
mechanize an appropriately high-order integration filter. Ref.
(23-25)
In the tightly coupled mode, there is only one feedback from the
navigation processor. Figure 3-lc illustrates the use of velocity
aiding to the GPS tracking loops. Acceleration aiding could also
be effectively used, but we are not aware of any particular
mechanization using other than velocity aiding. The other paths
used in loosely coupled architectures are not needed here because
all computations involved in navigation processing are now internal to one processor.
The concept of tightly coupled integration is often raised in
connection with embedded GPS receivers. These are not
necessarily synonymous. However, it is reasonable that one
would choose to mechanize a tightly coupled integration algorithm if one had already taken the effort to design a GPS receiver
that is physically and electrically integrated with an inertial sensor
or with a powerful navigation processor. We return to this point
in Section 3.3.
As of January 1993, it was uncertain whether the option to
implement deeplycoupled integraton would be denied to authorized GPS users. Plans to develop tightly-coupled systems such as
the GPS Guidance Package and the HoneywelWexas Instruments
Model 764-C3 were well underway when the GPS Joint Program
Office introduced a new GPS security implementation called the
SAASM (Selective Availability, Anti-spoofing Module). Comments from industry were due in January; there was a widespread
concern that the amount of integration processing (e.g. Kalman
Filtering) that could be supported by a prospective SAASM
mechanization would not be adequate to achieve the full benefits
attributable to deep integration using tightly coupled sensor outputs. The jury is still out on this issue.
this solution is within acceptable bounds on its accuracy (via the

GPS Figure*f-merit (FOM)). INS data may be used to interpolate between successive GPS updates when a higher output rate is
needed than can be provided by the UE. During GPS outages the
INS solution extrapolates from the last valid GPS solution. (The
process of forcing the INS solution to equal the current GPS
indicated velocity and/or position is known as a reset if that
correction is actually fed back to the INS).
3.2.2 Filrering
The general filtering problem involves trying to estimate timevarying states whose evolution is characterized by known laws
of propagation, which are usually taken to be a coupled system of
linear differential equations driven by white noise.
States cannot usually be measured directly but are inferred from
measurable quantities to which they are related. These measurements may be made simultaneously, or sequentially at a series of
distinct points in time. The filter will generally incorporate
knowledge of the statistics of the measurements.
Knowledge of the way the states change (propagate) in time,
knowledge of the way the measurements are related to the states,
measurement statistics and measurement data are all used in each
state update. The most common update algorithms use linear
filters, e.g.. ones in which the updated state is a linearly weighted
s u m of the measurements and the previous state value.
Position and velocity of an aircraft are examples of quantities that
may be chosen as states in a filter (these are referred to as wholevalue filter states). For whole-value position and velocity states,
the propagation equations are simply the equations of motion of
the aircraft. To make the whole-value filter propagation equations
a better reflection of the real world, acceleration states could be
added (otherwise, by its omission, acceleration must be treated as
noise, driving the derivative of velocity). GPS indicated position and velocity are examples of memuremetus that might be
processed by an integration filter with whole-value states. At one
extreme, the integration filter could ignore everything except
the GPS receiver position and use this as the integrated position.
This degenerate case is the selection mode cited above in which the
state propagation equations and any other availablemeasurements
would be ignored. In general, some rule must be used in order to
determine how much weight should be put on a measurement and
how much weight should be put on the propagated states. For the
above degenerate case, the weight on the GPS UE position is one
and the weight on the propagated state is zero. The weight on the
measurement is referred to as the filter gain.
These are listed in order of increasing complexity and optimality.

Eachonecanbeusedwithanyoneofthearchitectureslistedin3.1
but the incremental pay-off of a more complex filter is directly
related to the quality of the input information.
Another choice of states are the errors in position and velocity

indicated by the INS (these are referred to as error states). For a
filter whose states are INS errors, accurate representations and
linear approximationsof the propagation equations arewell known.
As in the case of whole-value states, additional INS error states
(for example, states for azimuth and tilt errors, accelerometer bias,
and gyro drift) could be added to the filter in order to make the
propagation equations a better model of the real world. Of course,
the degree to which the filter must reflect the real world is a
function of the estimation accuracy required, and that is a reflection of the mission requirements.
33.1 Selection
A selection algorithm chooses the GPS indicated (P,V,T) as the
system navigation solution whenever the GPS UE indicates that
For a GPS/NS integration filter with INS error states, the measurements would actually be the differences between GPS position and INS position and the differences between GPS velocity
and INS velocity. As with the case of whole-value states, some
3.2 Integration Algorithms

The basic choices for GPS integration algorithms are:
(I) Selection, with or without INS resets
(U) Fixed Gain Filter
(UI) Time-varying filter
286
rule must be used in order to determine how much gain should be

put on the measurements and how much weight should be put on
the propagated states when computing state updates.
We mention in passing that the optimum filter may require an
impractically large number of states. Separation of the 3dimensional filtering problem into a vertical estimator (possibly
mechanized as a fixed-gain fiter) and a horizontal estimator is
a common strategy in GPS inertial integration. Options to decompose a high dimensional estimator into combinations of lower
dimensional filters have also been described in the literature.
Distributed filtering and federated filtering are the terms under
which these options are usually cited as in References (24,25).
Both are believed to be more robust than the optimum filter when
the design must be tolerant to imperfect information about the
estimation problem. It is claimed that federated filters are more
fault-tolerant. These details are beyond the scope of this survey.
332.1 Fixed Gain Fibers

In a fixed filter, the propagated estimates are combined with new
measurement datausing predetermined gains. The gains are fixed
in the sense that they have been loaded into computer memory a
priori, so that the fdter selects from a short list of gains, rather than
computing them. Different gains may be used for different sensor
status and operational status, reflecting the uncertainties in the
propagated solution and in the measurements. In general, the
gains in a fixed-gain filter can have any value (they should at least
properly reflect therelationships among themeasurements and the
states).
If the state dynamics and their uncertainty are limited, and there is
negligible variation of measurement noise during the interval of
interest, it may turn out that the optimum fiiter gains will not vary
very much during the mission. In that case the performance
penalty of mechanizing one fixed set of gains (or a few selectable
sets of gains) compared to optimum time-varying gains may be
acceptably small. The benefit to the integrator is a vast decrease
in computational burden and memory required to implement the
filter. It may even be effective to precalculate and store timevarying gains for use during standard missions.
33.23 Time Varying Gain (Kalman Fiber)
In the Kalman filter, new gains are computed every time measurements are available. The Kalman Filter is a recursive implementation of the optimum least-squares error estimation algorithm.
Itisoptimuminthesensethatitstrikesthecorrectbalancebetween
uncertainly in the presumed dynamics of the states being estimated (process noise), uncertainly in the measurements (measurementnoise) and the observability of individual states (sensitivity),
required to minimize the figure-of-merit. See References (26-28)
for a detailed discussion of Kalman Filtering. In the present
context, we note that the updating of N states by M measurements
involves substantial matrix manipulations, propagation of difference equations. and memory to store the matrices. Current
technology can handle updates of around 20states at up to a few
times per second in a reasonably cost-effective processor. Since
upwards of 100 error sources may influence an integrated GPS
INS solution, the brute force approach to real-time integration is
not yet computationally feasible. Each designer must complete
detailed design studies to determine the minimum number of
states and the update rate that will result in an acceptable navigation error using the available processor resources and with sufficient design margin. Given the rapidly changing computational
capabilities available to avionics integrators, questions of computational feasibility should be reconsidered every few years.
333.3 Custom Tuning the GPS Internal Navigation Filter
Custom tuning is an action that can be taken internal to the GPS
UE. It does not require an external integration filter to derive some
degree of more robust performance in the presence ofhighvehicle
dynamics or jamming.
For example, the standard MAGR or RCVR 3A internal Kalman

filter is a complex computer program which embodies a mathematical algorithm for combining the INS reference with GPS
measurements in a manner that is statistically optimal. About two
dozen parameters in the program characterize the dynamics,
timing, and statistics of the INS and GPS. Adjusting these
parameters is called tuning. The standard tuning in the MAGR
assumes a very low quality INS (large errors, large error growth,
very noisy, and very unstable calibration coefficients) apoorclock
in the UE.and very high noise in the carrier-tracking (delta-range)
measurements. Using these assumptions, Rockwell Inc. produced
a navigation solution that is impervious to bad UE clocks, carrier
loop tracking cycle slips, and substandard INS quality.
Although thestandard tuningoftheMAGRproduces averyrobust
solution in the presence of GPS measurements, this solution may
deteriorate within a few minutes when no GPS measurements are
available. Custom tuning tries to take advantage of a navigationgrade INS by matching the tuning parameters to the expected INS
performance. One effect achieved is similar to what happens
when an external integration filter is used The INS is better
calibrated using the nine INS error states in the retuned MAGR
internal Kalman filter so that when aGPS outage occurs (e.g. from
jamming) the MAGRs navigation solution error growth is much
slower than with the standard tuning. Thus. the uncertainty in
vehicle position that must be searched to reacquire GPS signals
after an outage isreduced. This is the mechanism whereby custom
retuning can result in faster reacquisition after jamming. This
result was observed in a recent comparison between a MAGR
integrated with an RLG and a tightly-coupled embedded navigator.
Another advantage of custom tuning, not available to an external

integration, is that the effective code tracking bandwidths in state
3 can be made more narrow. The adjustment is made using certain
tuning parameters. The procedure is successful because the INSaided loop is known to be tracking the errors of a higher quality
INS than is assumed by the standard filter tuning parameter
selection. Taking full advantage of the INS aiding means that the
loops are more resistant to jamming and high dynamics and stay
in lock longer.
Under no circumstances will custom tuning improve the positioning accuracy of the MAGR when GPS code measurements are
available from four satellites. When satellite measurements are
available, accuracy is limited by the 3 to 5 meter bias errors in the
measurements and by the satellite geometry (GDOP). In summary, the benefit of custom tuning is not to improve accuracy in
the benign case, but to increase the receivers resistance to jamming and high dynamics and to obtain faster reacquisition of
specified performance after code loops lose lock for any reason.
Negative factors of returning include the cost of requalifying the
set, and the on-going logistics budendcosts of stocking replace-
281
ments for many different varieties of sets at depots. The use of a

few predefined sets of strapable coefficients might reduce this
burden at the cost of adding safety procedures to ensure that the
wrong coefficients could never be loaded. Some additional risks
are also incurred because the custom tuned MAGR assumes a high
quality INS. but unlike an external integration program, no explicit checks to monitor INS performance can be made to identify
substandard INS performance. Thus, the retuned receiver is more
sensitive to INS quality than the standard tuned receiver.
3.23 Discussion
The stand-alone mode is inferior (in performance) to the loosely
coupled integration mode. The stand-alone and loosely coupled
integration models are inferior (in performance) to a tightly
coupled mode because information inherent in the sensor measurements is lost in the receivers mechanization of its K h a n
filtered (P,V,T) solution, i.e., it is not always possible to backtrack
from a (P,V,T) solution to the raw GPS measurements with
sufficient bandwidth and precision to support a tightly coupled
integration..
The feedback of the system navigation solution to the INS (via

resets) is a second order improvement in the loosely coupled
mode (and is inherent in the tightly coupled mode).
The current generation of military high dynamics GPS UE (Receiver 3A. MAGR) isoftencriticizedbecauseitmustbeintegrated
in the loosely coupled mode. This criticism is valid, but it really
addresses a cost tradeoff in which the integration filter in RCVR
3A was limited to twelve states, of which nine represent very
generic INSerrors (P,V,8). The decision for RCVR 3A was based
onunit cost andthedesireto produce agenericstandardequipment
that did not burden any user with features that were not justifiable
in his application. In principal, a modified RCVR 3A operating in
the INS mode with an expanded internal filter and appropriate
software could perform nearly as well as a tightly coupled
integration.
In design studies of elaborate GPS/INS integration filters as many

as eighty inertial error states are modeled, in addition to GPS error
states related to delay measurement bias, tracking loop errors,
propagation errors, and user clock errors. In some ultra-precise
systems, it may even be useful to incorporate additional states that
model multipath effects. Nevertheless, many studies have shown
that most of the benefit of expanded error state formulations is
gained with 25-30 states, and that adequate performance can
usually be obtained from 14-17 states. Ref. (29.30)
The tightly coupled mechanization does avoid one problem commonly attributed to loosely coupled integration, namely the possibility of instability (in state estimates) arising when the GPS
navigation errors become highly correlated with INS navigation
errors. This situation may occur at low input signal to noise ratios
when GPS code loops remain in lock only because inertial aiding
allows the loop bandwidth to be reduced thereby reducing the
effective levels of noise and interference. Now, the narrower the
loop bandwidth, the more the loop error approximates the error of
the aiding signals so that the correlation cited above becomes
significant. See References 3 1and 32 for further discussion of this
point.
3 3 Embedded Systems
As GPS approaches its operational status, therehasbeen amassive
increase in investment in civil GPS technology which has led to
smaller, lower power consuming, higher performance UE than
were dreamed of as recently as the late 1980s. One consequence
of this trend is that GPS UE can be packaged on a single card that
can be embedded in other systems. As noted in Section 3.1. the
concept of GPS embedded in an INS is one such application that
is being prominently discussed at present, with several efforts
underway to demonstrate the concept. Ref. (33-36)
Setting aside the valid claims of savings in size,power weight that
a m e from embedding, it is reasonable to ask whether there is any
functional or performance payoff directly attributable to embedding. The answer is a qualified yes. There are potential performance improvements, but the system may be vulnerable to a
single-point failure, such as a power supply or the processor.
33.1 Tight Coupling
There is no inherent reason to claim that embedding implies tightcoupled integration. An embedded receiver could be stand-alone,
loosely-coupled or tightly coupled. However, developers of
embedded systems have tended to mechanize tight coupling as a
performance feature.
3 3 2 Carrier Loop Aiding
Standard military UE use inertial aiding only for code loops, and
only after carrier loops have lost lock. The decision to limit the
INS aiding goes back to the late 1970swhen it was argued that the
latency (time delay) between the sensing of inertial velocity and its
receipt at a GPS receiver could be as large as tens of milliseconds,
even with the high speed data busses that were available. With this
much delay, it was argued that errors in the aiding signal during
accelerations or turns could be large enough to drive the carrier
loop out of lock.
There are at least two ways to mitigate this concern in an embedded system. The most common approach is to customize the data
link between the INS and the GPS carrier loop to reduce the
latency to a few tens of microseconds and to minimize the
uncertainty in the latency. With that small a delay, the maximum
error of the aiding signal is negligible even for an aircraft rolling
as fast as 1 rev/s. and moving towards a satellite with a relative
velocity of 2000 fdsecond. Under those conditions the error
caused by a 20 ps delay in attitude indication would be approximately VE, where
VE < 2rr (20 x l(r6) radian x 2000 f/s < .25f/s

which is well within the acceptable range for GPS receivers. An
alternative that has not yet been mechanized is to delay the GPS
signals by an amount that matches the latency, before the aiding
signal is applied. For modem pre-correlation digital GPS receivers one could store tens of milliseconds of GPS samples in a buffer
mechanized by a single memory chip.
3 3 3 Tmcking Fewer Than Four GPS Satellites
The loosecoupling approach (see Figure 3-lb) integrates a GPS
(P,V,T) solution with an inertial (P.V.8) solution. When the GPS
solution is unavailable, the integrated solution flywheels using
the inertial solution as corrected at the start of the GPS outage. In
contrast, the tightly coupled solution uses raw GPS measure-
288
ments, which are available as long as one or more satellites are

being tracked. Thus it is a more robust solution vs. outages that
could prevent a GPS navigation solution from being formed in the
loosely coupled configuration. Reference (29) gives a good
insight into the potential performance improvement.
However, there is one caveat to consider. Conventional GPS UE.
such as GPS RCVR 3A can continue to provide a navigation
solution (albeit degraded) with only two or three satellites. Thus
it is inappropriate to claim that a loosely coupled integration mrcst
convert to a free-running inertial solution in the presence of one or
more satelliteoutages. Theperformancewilldepend onthe details
of the GPS UE mechanization.
3.3.4 Quantization
All calculations within an embedded system are more likely to be

executed as full-precision quantities than in a system wherein
the GPS and INS and navigation processor are connected by data
busses. These busses are usually so heavily used that data must be
coarsely quantized for data transmission (compared to their internal precision) in order to satisfy communication bandwidth constraints and to conform to data transmission protocols. An
alternative that has not been explored in GPS navigation data
communications is to use data-compression to increase the information content of the message structure. This would require
reworking interfaces and message protocols, but the effort might
be cost-effective in high precision applications.
4. INTEGRATION CASE STUDIES
mounted at righi angles), and tilting the plane to null the accelerometer outputs* The perpendicular to this plane defines UPin
an East, North. Up coordinate system.
The error signal that drives the leveling loop is proportional to the
instantaneous value of the tilt error, i.e., to approximately 1.0 g
times the alignment error (in radians). In mathematical notation.
A
the output for each horizontal accelerometer is
g=(l+s,)ge,+b
where
g is the magnitude of the local gravity vector
e, is the angular error of the platform normal to the accelerometer

axis (i.e.. the tilt error)
b is the accelerometer bias

sf is the accelerometer scale factor error
A
For an aircraft at rest g should be zero when the platform is level,

however, the accelerometer bias causes thenull to occur for anonzero value of tilt error, For accelerometers used in contemporilry
fighter aircraft b is about 150 pg and sf is about 500 ppm. For
g=1.0, the steady state tilt error e is then approximately
We shall consider two case studies that illustrate performance

benefits of GPS/INS integration. The properties to be addressed
include:
a)
In flight INS alignment
b)
Reduced error growth of a calibrated INS during a GPS

outage
4.1 Using GPS for Inflight Alignment

Alignment is the process that ties inertial platform coordinates to
the geographic frame in which the host vehicle navigates. Alignment establishes the conditions necessary for the proper integration of accelerometer outputs to accurately estimate thechanges in
user position and velocity.
Gyrocompassingalignmenttypicallyextendsovera10-15minute
period prior to a standard aircraft takeoff. We need to look at the
factors that contribute to alignment time in order to understand the
opportunities for using GPS to speed up the process, or to allow it
to proceed in-air in addition to on the ground.
4.1 .I Conventional Alignment Procedures
Alignment consists of platform leveling and establishing a reference bearing. The following discussion is based on the alignment
of a gimballed platform INS. but the general conclusions and
timing estimates are good approximations to the performance of
a strapdown INS.
Lateling
Leveling is the process of establishing one plane of the M U

instrument package perpendicular to the local gravity field. This
is typically accomplished by mounting two accelerometers on that
plane so that their input axes are not colinear (they are usually
9, where
- 150prudian = 30arcsec
(4.1)
where we assume that angle quantization effects in the leveling

feedback loop are negligible. This 30 arcsec is a bias error; the
random component of leveling errors is on the order of a few
arcseconds.
North Seeking
North Seeking is the process of establishing a reference direction
(azimuth) in the leveled plane containing the east and north,
accelerometers. The most widely used scheme for self-alignment
of the platform is gyrocompassing.
Gyrocompassing exploits the following properties of gyroscopes.
for a local level north-oriented system at latitude 1:
(I)
If the gyro input axes are physically aligned with the

geographic axes (North, East, Up), then the system will
remain aligned if each gyro is individually torqued at a rate
equal to the projection of the earth rate on its input axis.
(n)
If the platform is level, but not aligned to north it will rotate

around the east axis, causing the north axis to rotate from
horizontal. This rotation produces a level error than can be
sensed by the north accelerometer, and used to drive the
azimuth error to zero.
* Alignment for strapdown inertial systems is a mathematical

operation that does not involve physical motion of the sensor
platform.
289
The rate of rotation of the north axis in response to an azimuth

misalignment of radians is given by
where
neis the earth rate.
produces net level error (tilt) of
8,
If this rate acts for t seconds it
8,
(radians)
(4.2)
From Eq. 4.2 we see that for gyrocompassing to convert a
misalignment error
8,
into an equally large observable tilt error
(i.e., for 8, = 8,)- then the effect ofthe error m u s t h integrated

for at least b seconds, where
with the opportunity to use aircraft maneuvers (turns, climbs,

dives) to apply acceleration in the horizontal navigation plane
(North, East), in addition to the vertical plane (up), and the
potential use of GPS for in-flight alignment is clear.
The issues for in-flight INS alignment using GPS are related to the
noisiness of GPS velocity measurements compared to inertial
measurements. Unless the INS misalignment is so large that, at the
vehiclevelocity. it produces avelocity error that exceeds the GPS
noise level, the alignment will require several minutes, or more, of
observations. Thus the time to align is a function of the accuracy
goals, the magnitude of acceleration that can be applied during
maneuvers, and the noise level of GPS measurements.
Figures 4- 1to 4-3were generated in a linear covariance simulation
of INS alignment for a Fighter Aircraft using a standard (1 .O nm/
h) inertial navigator. The integration filter mechanized eleven
states (3 position, 3 velocity, 3 misalignment, 2 clock), and the
inputs are GPS pseudo range and deltarange and velocity changes
output by the INS.
Two takeoff profiles were investigated:
Standard Takeoff
(1) Accelerate at 0.3 g to 160 knots indicated airspeed (KIAS)
using MIL (standard) acceleration.
At mid latitudes, (say )c = 45) and for
ne= 2Hradians I day

we have
t o
==
days = 5.4hours
(4.3)
In practice, the requirement for alignment accuracy is substantially less than the requirement on leveling accuracy, and this
allows the alignment to proceed faster than indicated by Eq. (4.3).
We can estimate this speedup using the approximation that the
standard deviation of leveling errors is inversely proportional to
the time spent in leveling. For medium accuracy inertial sets that
are found in contemporary aircraft the random noise in leveling is
on the order of a few arcseconds. and the uncertainty in azimuth
alignment after gyrocompassing is on the order of 160-200
arcseconds. The ratio of these standard deviations is about 50.
Substituting 8, = 50 8, in Eq. (4.2) yields to = b /50. which is
approximately 7 minutes. This estimate is consistent with Air
Force specifications for the time to achieve standard accuracy
alignment by gyrocompassing under favorable conditions. Ref.
(37)
4.1.2 GPS AidedAlignment
Leveling is essentially an instantaneous process (especially in a

strapdown system) because it seeks to drive two directly sensed
gravity indications to be equal.. This gives the clue as to the
benefits of in-flight alignment. Velocity changes sensed by the
inertial system are compared to velocity changes sensed by GPS
and the alignment parameters are adjusted to drive the residuals to
zero. Sensing INS alignment errors from velocity residuals is
significant because it eliminates the delay required to integrate a
velocity error into a position error of detectable magnitude for
gyrocompassing alignment. Now, combine this observability
(2)
Begin rotation to 5 pitch angle and accelerate to 170 KIAS

takeoff speed.
(3)
Climb at 5. and accelerate to 385 KIAS, on 45 heading

(northeast).
(4)
Continue climbing at 5. and constant velocity to cruise

altitude at 20,000 ft. and 385 KIAS.
Alert (Scramble) Takeoff

(1) Accelerate at 0.4 g to 165 KIAS using maximum afterburner thrust.
(2)
Rotateto 12pitchangleandclimbat0.4g to415 ktat2500

ft altitude.
(3)
Climb from 2500 ft to 20.000 ft altitude at constant speed

and 12 pitch, arriving at 20.000 ft approximately 180
seconds after takeoff.
(4)
Level flight at20,Wfeet then accelerate to Mach0.85 (533

kt).
Navigation for standard takeoff assumes that the INS has been
aligned by gyrocompassing and that GPS is continually available.
Navigation for alert takeoff assumes that the INS is aligned by
astored heading reference, and that GPS is not available until five
minutes after takeoff. In both cases, different turn, climb, and dive
maneuvers after takeoff were considered for their effectiveness in
reducing alignment errors after takeoff.
The values cited in Table 4-1 were used to characterize the
standard errors (1 - s levels) remaining after gyrocompassing and
stored heading alignment respectively. These are the initial
conditions for subsequent Kalman filtering of GPS plus inertial
sensor outputs.
290
Table 4-1. Alignment Errors at Takeoff (arcseconds)
Stored Heading
Gyrocompassing
Level
40
37
Azimuth
450
225
Three individual test cases are presented here. These are listed in
Table 4-2.
Table 4-2. Test Cases for lnfllght Alignment Studies
Case
1
2
3
Takeoff
Mode
standard
alert
alert
Alignment
Mode
gyrocompass
stored heading
stored heading
GPS
OnAt
0s
300 s
300 s
Alignment
Maneuver
2/39 turn
1 g turn
2 g S-turn
Case I
Case 1 illustrates the potential benefits of using GPS aided
alignment in a standard mission, where the INS is gyrocompassed
prior to takeoff. and GPS is available throughout takeoff. A 2/3g turn that produces a 45 heading change beginning at t = 495
seconds is included to illustrate the additional improvement available from in-flight maneuvers.
Figure 4-1 illustrates the history of RMS alignment errors. As

expected, the availability of GPS during the period of acceleration
at takeoff reduces the alignment error; it appears to decrease
asymptotically towards about 50 arcsec. The additional acceleration after t = 495 seconds produces a modest further improvement.
The final azimuth error is approximately the same as the level
error, which signifies that accelerometer bias is the limiting factor
in improving alignment by this procedure. An analysis of the error
sensitivity of the tilt estimates confirms this intuition, with
nonorthogonality of the horizontal accelerometers with respect to
the up direction being the only other significant error source.
Case 2
Case 2 illustrates potential benefits of GPS for a scramble
takeoff, when GPS is available only after 300 seconds. The
aircrafttakesoffwithastoredheadingalignmenthavinganominal
azimuth uncertainty of 450 arcseconds. The aircraft executes a 1 g turn that produces a 45 heading change starting at t = 495
seconds. (The heading angle for Case 2 is essentially the same as
for Case 1. but the turn is completed faster.)
Figure 4-3 illustrates the rms alignment error. As expected,

Figures 4-2 and 4-3 are identical until the maneuver begins, and
then the higher gs involved in Case 3 reduce thealignmenterror
below the level for Case 2.
423 Discussion
The simulations reported above support the following performance conclusions:
Inflight GPS measurements during periods in which lateral
acceleration is on the order of l g can be used to reduce
azimuthal alignment errors to the same magnitude as level
errors. Larger accelerations lead to smaller errors.
Even in the absence of lateral acceleration the availability of
GPS measurements can be used to reduce the errors of a
stored-heading alignment to the level of gyrocompassing.
This reduction exploits the correlation between INS alignment errors and the INS navigation errors that are uncovered by comparison with GPS observations.
Simple maneuvers such as l g coordinated turns over as little
as45, andlastingfornomorethan30secondsareadequate
for realizing the benefits of inflight alignment.
Figure 4-1. INS alignment errors with GPS availableat takeoff

and gyrocompassedalignment. (standard takeoff)
Figure 4-2 illustrates the history of RMS alignment errors. The

error is constant until t = 300 seconds when GPS becomes
available. The availability of GPS makes adramatic improvement
in alignment even without maneuvers because the position and
velocity errors that have accumulated in the INSnavigation results
are highly correlated with misalignments. We see that, following
an initial decrease to 280 arcseconds, the alignment errors fall to
an asymptote of 200 arcseconds which is comparable to gyrocompass alignment accuracy. The execution of 1-g lateral maneuver
drives the alignment errors to the level predicted by accelerometer
bias.
Case 3
Case 3 involves an alert takeoff with nominal stored heading
azimuth errors and a2-g "s" turnmaneuver at t 4 6 0 seconds. GPS
is not available until t=300 seconds (corresponding to a fiveminute warm-up and acquisition time).
Figure 4-2. INS alignment errors with GPS available 5 minutes

after an ALERT takeoff.
29 1
and trajectory. Five minutes after the loss of GPS, the east
component of horizontalpositionerrorshas grown to 80 feet (24.4
meters rms), exclusive of the low-frequencyGPS bias. Figure 45 shows the contribution of various error sources to the net
navigation error. For at least ten minutes, the accelerometer and
gravity disturbance terms are dominant. Gyro errors do not
become dominantuntil more than aquarter of a Schuler period (22
minutes) after the outage.
Figure 4-3. INS alignment errors with GPS avallabie at 300

seconds after an alert takeoff with stored heading
alignment.
4.2. Integrated Navigation Solutions During a GPS Outage

The previous examples illustrated certain benefits of GPS/INS
integrationin a benign environment,with low to moderatevehicle
dynamics. The use of integrationto lengthen the interval that GPS
data is available, and to ride-out the period of outages caused by
the interference or high dynamics is the focus of the following
remarks.
42.1 lntegratwn Case Study Overview
Figure 4-4b illustrates the growth in east errors for a suboptimum21-statefiiter,whosestatesarelistedinTable4-5. These
results are almost indistinguishable from Figure 4 4 % for the
optimum filter. Finally, Figure 4 4 c gives the east error for a
15-state horizontal filter whose states are listed in Table 4-4.
These results too are nearly indistinguishablefrom optimum for
about the first 420 seconds of GPS outage.
Table 4-3. Truth Model States
Description
State #
Horizontalposition error
12
3
Wander angle error
Platform misalignment
456
7,8,9
10
11-13
In a recent military aircraft avionics upgrade, GPS was to be

integrated with a navigation-grade gimbaled-platform INS.
14-16
The principal performanceobjective was to limit the error growth

after five minutes of GPS outage, providing that the INS had been
calibrated by at least seven minutes of GPS measurements prior to
the outage. Based on extensive simulations, it was determined
that:
17-18
1.
2.
19
21-23
The horizontal CEP after five minutes of GPS outage could

be held to 30 meters. (Recall that the free-running INS is
specified as a 1.O nmi/hr system.)
24-26
A 21-state filter mechanized as two independenthorizon-
27-29
30-32
tal and vertical filters would provide close enough to

optimumperformance to be acosteffective,computationally
effective approach.
These results were derived from a covariance analysis that incorporated a 73-state truth model for the INS,plus additional error
states for GPS.
Table 4-3 lists these truth states. Table 4-4 characterizes the
mission segments of the standard aircraft flight path used in the
simulation.
42.2 Navigatwn Pegormanee of the Integrated System
Figure 4-4a illustrates the growth in rms east positioning navigation error versus time after the loss of GPS at approximately
2550 seconds. In this case, the GPS and INS position and velocity
measurementswereprocessedby afulloptimal73-state filter with
dynamics,plant noise parameters and measurement noise parameters that exactly match those in the truth model. In other words,
this represents the best that any filter can do given this truth model
33-35
36-41
42-44
45-47
48-50
51-53
54-56
57-59
60-65
66-71
72
73
Comment
100 arcsec initial horizontal; 6 degree initial vertical
Velocity error
Vertical position error
Auxiliary baro-inertial states Used in describing barcinertial loop
,002 degreelhr rms horiMarkov gyro bias
zontal .005 degree/hr rms
vertical; Correlationtime =
5 minutes
Markov accelerometer bias 3 microg rms; 10 minute
correlation time
100 feet rms
Markov bar0 bias
35 microg rms; 20 nautiGravity disturbance
cal mile; correlation distance
.01 degreelhrrmshorizonGyro bias
tal; .022degree/hrrmsvertical
0.0002 rms
Gyro scale factor error
Gyro misalignments about
3.3 arcsec rms
spin axes
Remaining gyro
20 arcsec rms
misalignments
.015 degree/hr/ rms
Gyro g-sensitivity
Gyro g-squared-sensitivity .02 degree/hr/g rms
150 microg rms
Accelerometer bias
0.0002 rms
Acc. scale factor error
Acc. scale factor asymmetry 0.0001 rms
Acc. nonlinear scale factor
IO microg/g2rms
asymmetry
Accelerometer non linearity IO micro-g/g*rms
Accelerometer orthogonal
30 micro-g/g2rms
g-squared sensitivity
Accelerometer
20 arcsecond 4ms
misalignments
300 feet rms
Baroaltimeter bias
Baroaltimeter scale factor
.04 rms
error
292
TOTAL
Table 4-4. Test Trajectory

Time (sec)
Description
0-300
Gyrocompass at true heading = 120 degrees
300-490
Taxi and turn left approximately 75 degrees to prepare

for takeoff at a heading of approximately 45 degrees.
490-540
Takeoff. Speed = 434 f/s, altitudeapproximately 1WO

feet.
540-840
Climb, accelerate to 655 f/s, altitude approximately

9660 feet.
m.0-1
/ /
840-1020
Level off at approximately 9660 ft. Cruise at 655 fls.
am.0
zmoa
.
z100.0
m m
2900.00
moon
sio0.oo
TlYE
SECOND8
Figure 4-5. Contribution to INS Error Growth During a GPS

Outage.
1020-1100 Descend to approximately 200 feet.

Table 4-5. States of the 15-state HorizontalFilter
1100-2720 Level off at 200 feet, cruise at 655 f/s.
Description
Horizontalposition error
Wander angle error
Platform misalignment
2720-2780 90 degree right turn.

2780-2806 Pop-up, dimb to approximately 12,000 feet, accelerate to approximately 820 f/s.
2838-2858 Level out at 635 feet, decelerate to 655 fls.

Weapons delivery.
Horizontalvelocity error
2858-2918 90 degree right turn.

9-11
12,13
3226-3276 Turning climb to 9900 feet, speed = 655 f/s.

14.15
3276-3346 High altitude combat.

3346-5086 Turn towards home. High altitude cruise at 9900 feet,
655 f/s.
Plantnoiseof9.4E-l7rab/
sec (horizontal)and 5.9E-
2806-2838 Dive to 635 feet.
2918-3226 Low altitude combat, 300 to 2700 feet, 485 to 655 f/s.
Comment
16 rad%ec (vertical)used
to account for unmodelled
gyro errors
Plant noise of 9.4E-12 ft2/
sec3 used to account for
unmodelledaccelerometer
errors
Gyro bias
Horizontalgravity
disturbance (also absorbs
other accelerometer errors)
Accelerometer scale
factor error
In an actual implementation, wander angle error would be combined into a single state with vertical misalignment.
5086-5446 Descend to 3000 feet. accelerate to 820 f/s.
5.0 CONCLUSIONS
5446-5876 Loiter, decelerate to 434 f/s.

5876-6126 Land.
SGUA b t m.0-
We have presented a description of the processes whereby the

combination of GPS with an inertial navigator can produce a
system performance that is superior to either one acting alone.
These benefits include:
(a) 73-state optimum filter
(b) 21-atatelilter
(e) 15-state horizontalfilter
(FEW
m.01a.o-
zsno.0
2ma z100.0
morn
zso0.m
1.
Smaller random errors than seen in stand-alone GPS navigation solutions.
2.
Improved availabilityof GPS operations during maneuvers

and in the presence of radio frequency interference (RFI).
3.
A navigation solution whose position and velocity errors

are bounded by the errors in the GPS navigation solution.
4.
A calibratednavigation solution whose errors grow slower

than those of a free-running uncalibrated INS during GPS
outages.
m a a1m.m
TIME
SECONDS
Figure 4-4. RMS East Position Error vs. T h e : GPS measurements Stop at 2550 seconds.
293
They are available because the long-term (low frequency) content

of INS errors is negligible.
How can the filter be maderobust against variations in the error

characteristics of individual INS or GPS units?
All four benefits are available to GPS users who are authorized to
obtain the GPS Precise Positioning Service (PPS). Only the f i s t
and third are guaranteed to GPS users who are vulnerable to the
selective availability clock dither that corrupts the GPS Standard
Positioning Service (SPS). Clock dither prevents the tracking
loop bandwidths from being reduced to the bandwidth of the INS
errors, and corrupts the use of GPS measurements to calibrate the
INS error states.
How can the filter be used to detect the onset of anomalous

conditions that may indicate a failure in the GPS or the INS subsystems?
GPS receiver technology is evolving rapidly in response to pressure from the civilian market. This trend is evident in the
miniaturization of full-capability receivers that can be physically
embedded in a host system such as an INS or a mission computer.
This in turn makes it practical to obtain even higher performance
levels by treating both GPS and the INS as sensors that produce
measurements to be optimally combined by anavigation filter into
an optimum navigation solution.
For many users, the primary benefit of embedded configurations

is not improved performance. These users focus on savings in
volume. weight, power consumption, and cost that are predicted
for the embedded system. Some cost savings are non-recurring
(e.g. the one-time investment in developing integration software
may be borne by the vendor of the embedded system rather than
by the integrator), whereas others are recuning (e.g. reduced
production cost of the system hardware).
In addition to these cost and performance benefits, the integrated
system may be able to support functional capabilities that were
previously not available to the user. For example, the use of inflight INS alignment canmake it possible for an aircraft to takeoff
without waiting for the 5-10 minutes routinely reserved for INS
alignment. This may be a life-saving capability for military
aircraft that have to take off immediately after an alert that hostile
forces are incoming. Another example is the inclusion of error
states for gravity anomalies in the integration filter. This is the
basis for balloon-borne GPS/lNS instrumentation that will allow
for more extensive gravity mapping than has ever been possible.
(Ref. 38)
The design of a Kalman integration filter that mechanizes the
navigation solution must also address certain standard questions
such as:
These design questions are addressed in more detail in other

chapters of this book.
REFERENCES
1.
Brown, R.Grover,ABaselineGPS RAIM Scheme,NAVICATION, Vol. 39, NO. 3, Fall 1992, pp. 301-314.
2.
Van Graas. Frank, Sole Means Navigation and Integrity

Through Hybrid Loran-C and NAVSTAR GPS,Proc. ION
NationalTechnical Meeting, SantaBarbara,CA, January2629, 1988, pp. 294-301.
3.
Brown, A.K., Schmid. T.. Integrity Monitoring of GPS

Using a Barometric Altimeter, h o c . ION National Technical Meeting, Santa Barbara, CA, January 26-29, 1988. pp.
245-253.
4.
Tsang, Wai L., and Stein, Barry A., A Robust RAIM

Scheme Using GPSIGLONASS Systems, Proc. ION GPS89, Colorado Springs, CO, September 27-29.1989. pp. 373378.
5.
Cox, D. B.. Integration of GPS with Inertial Navigation

Systems, Reprinted in Collected GPS Papers Volume I.
Institute of Navigation, Alexandria, VA.
6.
Johannessen, R., Asbury M. J. A., Towards A Quantitative

Assessment of Benefits INS/GPS Integration can offer to
Civil Aviation, NAVIGATION, Vol. 37, No. 4, Winter
1990-1991, pp. 329-346.
7.
Greenspan, R. L., et al. The GPS Users Integration Guide,

Proceedings, ION National Technical Meeting, Santa Barbara, CA, 26-29 January 1988. pp. 104-112.
8.
Wiederholt, L., Klein, D., PhaseIII GPS Options for Aircraft

Platforms, NAVIGATION, Vol. 31, No. 2, Summer 1984,
pp. 129-151.
9.
Brown, A. K., Bowles, W. M., Interferometric Attitude

Determination Using GPS, Proceedings, Third Geodetic
Symposium on Satellite Doppler Positioning, Las Cruces,
NM. February 1982. pp..1289-1302.
How many states to estimate?

How often should the filter be updated?
How should correlated measurements be treated?
By how much can the computational burden of the filter be
reduced by exploiting sparseness of the states dynamics matrix
or the measurement covariance matrix?
How should the filter accommodate transient events such as
changes in the constellation of satellites being tracked?
How should the filter be tuned to provide robust performance
versus unknown aspects of the design problem?
10. Eller, D., GPS/IMU Navigation in a High Dynamics Environment, Proceedings, First International Symposium on
Precise Positioning with GPS, April 15-19,1985. Rockville,
MD. pp. 773-782.
11. Bridges. P. D., Influence of Satellite Geometry Range, Clock
and Altimeter Errors, on Two-Satellite GPS Navigation,
Proceedings, ION GPS 88. Colorado Springs, CO, September 19-23, 1988, pp. 139-146.
294
12. Kalafus, R. M., and Knable, N., Clock Coasting and Altimeter Error Analysis for GPS. NAVIGATION, Vol. 3 1, No. 4,
Winter 1984-85.pp, 289-302.
13. Bartholomew. R. G..et al. Software Architecture of the
Family of DoD Standard GPS Receivers. Proceedings First
Technical Meeting of ION Satellite Division, Colorado
Springs, September 1987, pp. 23-24.
14. Dellicker. S. H. and Henckel, D., F-l6/GPS Integration
Test, Proceedings of ION GPS-89, Colorado Springs, September, 1989, pp. 295-303.
15. Bar~ur,Neil.etal,InertialInstruments-WheretoNow7,
Proceedings AIAA GN&C Conference, Hilton Head, S.C.,
10-12 August 1992. AIAA-92-4414-CP.
16. Boxenhorn. B., et al, The Micromechanical Inertial Guidance System and its Application, Proceedings Fourteenth
Biennial Guidance Test Symposium, Holloman AFB, New
Mexico, 3-5 October 1989, pp. 97-112.
17. Ward, P., Rath, J., Attitude Estimation Using GPS, Proceedings, ION National Technical Conference, San Mateo, CA,
23-26 January 1984, pp. 169-178.
18. Kruczynski, L. R., and Li, P. C.. Using GPS to Determine
Vehicle Attitude. Proceedings, ION GPS 88. Colorado
Springs, CO, September 19-23.1988. pp. 139-146.
19. Jurgens, Richard, Real Time GPS Azimuth Determining

System, Proceedings, ION National Technical Conference,
San Diego, CA, 23-25 January 1990. pp. 105-110.
20. van Graas. F., andBraasch, M., GPS Interferometric Attitude
and Heading Determination; Flight Test Results, Proceedings, 47th Annual Meeting (ION) Williamsburg, VA, June
10-12, 1991, pp. 183-191.
21. Graham, W.R. and Johnston, G . R.. Standard Integration
Filter State Specification and Accuracy Predictions, NAVIGATION, Vol. 33. No. 4, Winter 1986-1987,pp. 295-313.
22. Cunningham, J.. and Lewantowicz, Zdzislaw, H., Dynamic
Integration of Separate INS/GPS Kalman Filters, Proceedings, IONGPS 88,Colorado Springs, CO, September 19-23,
1988. pp. 139-146.
23. Diesel, John W., Integration of GPS/INS for Maximum
Velocity Accuracy, NAVIGATION, Vol. 34. No. 4, Fall
1987.
24. Berman,G. J., andBelzer,M.,ADecentralizedSquareRoot
Information Filter/Smoother, Proceedings 24th IEEE Conference on Decision and Control, Ft. Lauderdale, Florida,
December 1985.
25. Carlson, N. A., Federated Square Filter for Decentralized
Parallel Processes, IEEE Trans. Aerospace and Electronic
Systems, Vol. 26, No. 3, May 1990, pp. 517-525.
26. Gelb, A., (ed) Auulied Outimal Estimation, MIT Press.

Cambridge, MA (1974).
27. Maybeck, P. S., Stochastic Models. Estimation and Control
(Volume 1) Academic Press, Orlando, FL, 1979.
28. Bletzacker, F., et al., Kalman Filter Design for Integration of
Phase III GPS with an Inertial Navigation System, Proceedings.IONNationalTechnicalMeeting.Long Beach, CA, 2123 January 1986, pp. 113-129.
29. Lewantowicz. Z. H., and Keen. D. W., Graceful Degradation
of GPS/INS Performance with Fewer than Four Satellites,
Proceedings, ION National Technical Meeting, Phoenix.
AZ,January 21-24,1991, pp. 269-276.
30. Widnall, W. S.. Stability of Alternate Designs for Rate
Aiding of a Non-coherent Mode of a GPS Receiver,
Intermetrics Corp., Cambridge, MA, 25 September 1978.
31. Buechler,D..andFoss, M.,IntegrationofGPS and Strapdown
Inertial Subsystems into a Single Unit, NAVIGATION,
Summer 1987, Vol. 34, No. 2, pp. 140-159.
32. Franklin, M.. Pagnucco, S.,Development of Small Ernbedded GPS/INS RLG and FOG systems, Proceedings, ION
National Technical Conference, San Diego, CA, 27-29 January 1992, pp. 3-12.
33. Tazartes. D. A., Mark, J. G.. Integration of GPS Receivers
into Existing Inertial Navigation Systems, NAVIGATION,
Spring 1988, Vol. 35.No. 1, pp. 105-120.
34. Homer, W. C.. An Introduction to the GPS Guidance Package (GGP) Proceedings, 15th Biennial Guidance Test Symposium, CIGTF Guidance Test Division, Vol. 1, pp. 12-15,
Holloman AFB, New Mexico, September 24-26, 1991.
35. Donna, J., and Greenspan, R.. Measurement Errors in GPS
Observable, NAVIGATION, Vol. 33, No. 4, Winter 198681, pp. 319-334.
36. Soltz, J. A., et al., An Option for Mechanizing Integrated
GPS/INS Solutions, NAVIGATION, Vol. 35, No. 4, Winter
1988-89.pp. 443-458.
37. S cification for USAF Standard Form, Fit and Function

(F )Medium Accuracy Inertial Navigation Unit, SNU-841, Rev. 3, Amendment, 13 February 1987, USAF Aeronautical Systems Division (now Aeronautical Systems Center).
38. Jekeli, C.. et al., Instrumentation Design and Analysis for
Balloon-Borne Gravimetry and Attitude Determination USing GPS and INS, Proceedings ION National Technical
Meeting, San Francisco, CA, 20-22 January 1993 (To Be
Published).
295
Design Considerations for a Suboptimal Kalman Filter

D.J.DiFippo
Defence Research Establishment Ottawa
Ottawa, Canada
K1A 024
1.
INTRODUCTION
In designing a suboptimal Kalman filter, the designer must

decide how to simplify the system error model without
causing the filter estimation errors to increase to
unacceptable levels. Deletion of certain error states and
decoupling of error state dynamics are the two principal
model simplifications that are commonly used in suboptimal
filter design. For the most part, the decisions as to which
error states can be deleted o r decoupled are based on the
designer's understanding of the physics of the particular
system. Consequently, the details of a suboptimal design
are usually unique to the specific application.
In this paper, the process of designing a suboptimal Kalman
filter is illustrated for the case of an airborne transfer-ofalignment (TOA) system used for synthetic aperture radar
(SAR) motion compensation. In this application, the filter
must continuously transfer the alignment of an onboard
Dopplerdamped master inertial navigation system (INS) to a
strapdown navigator that processes information from a less
accurate inertial measurement unit (IMU) mounted on the
radar antenna. The IMU is used to measure spurious
antenna motion during the S A R imaging interval, so that
compensating phase corrections can be computed and
applied to the radar returns, thereby preventing image
degradation that would otherwise result from such motions.
The principles of S A R are described in many references, for
instance [11, [2].
The primary function of the TOA Kalman filter in a S A R
motion compensation system is to control strapdown
navigator attitude errors, and to a lesser degree, velocity
and heading errors.
Unlike a classical navigation
application, absolute positional accuracy is not important.
The motion compensation requirements for SAR imaging
are discussed in some detail in [3]. This TOA application is
particularly appropriate as a vehicle for discussing
suboptimal filter design, because the system contains
features that can be exploited to allow both deletion and
decoupling of error states.
In Section 2, a high-level background description of a S A R
motion Compensation system that incorporates a TOA
Kalman filter is given. The optimal TOA filter design is
presented in Section 3 with some simulation results to
indicate potential filter performance. In Section 4, the
suboptimal Kalman filter configuration is derived.
Simulation results are also shown in this section to allow
comparison between suboptimal and optimal filter
performances. Conclusions are contained in Section 5 .
Throughout this paper, it is assumed that the reader is
familiar with basic Kalman filter theory. If not, a treatment
of this subject can be found in [4], [SI or [6].
2.
SAR MOTION COMPENSATION SYSTEM
DESCR1J"I'ION
2.1 sensors
A functional block diagram of a S A R motion compensation
system (SARMCS) that utilizes a TOA Kalman filter is
shown in Figure. 1.
Figure 1. Functional Block Diagram of a SARMCS

The principal motion compensation sensor is a small inertial
measurement unit (IMU) mounted directly on the antenna
structure. It provides raw measurements of the angular
rates and linear accelerations of the antenna in the standard
form of angular and velocity increments. The antenna
structure is assumed to be physically stabilized in roll and
pitch. Consequently, the strapdown IMU is exposed to only
residual angular motion about the aircraft's roll and pitch
axes resulting from imperfect antenna stabilization, while it
is fully exposed to angular motion caused by aircraft
heading changes. It is assumed that the IMU contains two
dry tuned-rotor gyroscopes and three accelerometers with
the specifications shown in Table 1.
The onboard master inertial navigation system (INS) is
remotely located from the IMU. It is assumed to be a
medium-accuracy system (1 nm/hr, lo) with a baroaltitude
damped vertical channel. It employs three ring laser
gyroscopes and three accelerometers, which have the
characteristics indicated in Table 2.
The Doppler velocity sensor (DVS) employs a standard
strapdown lambda 3-beam configuration [q to measure
aircraft velocity in aircraft body coordinates relative to the
terrain surface. The primary timecorrelated error in the
measurement of the forward velocity component is a scale
factor type error in the order of 1-2% ( l a ) , depending upon
the terrain. The error in measuring the lateral velocity
component principally results from an antenna boresight
azimuth misalignment which is typically less than 1' (lo).
In addition to these timecorrelated errors, there is also a
296
significant noise component to the DVS measurements of

velocity; the level of noise is typically 1 mls (lo) for the
forward velocity measurement, and 1.5 mls (lo) for the
lateral velocity measurement. It is assumed that the DVS is
only operated over land, so that additional errors introduced
by the effects of sea currents and waves can be ignored.
The air data sensors (ADS) include a pressure transducer to
measure static air pressure, and an air temperature probe to
measure outside air temperature. This information is used
to compute baroaltitude.
Characteristic
Performance
150 ppm (lo)
0.2 Olhr (lo)
0.2 "lhrlg (lo)
Scale factor repeatability

Bias repeatability
G-sensitive drift repeatability
Characteristic
Performance

Bias repeatability
I
Table 1. IMU Gyro and Accelerometer Performance

Characteristics.
INS Gyro Performance Characteristics

Characteristic.

Bias repeatability
Random drift
The strapdown navigator estimates of antenna phase centre

motion are provided to a targetting algorithm during
intervals of S A R imaging. The targetting algorithm initially
computes a lineof-sight vector from the antenna phase
centre to the target, using target information supplied from
other parts of the S A R system. This lineof-sight vector is
then updated throughout the imaging interval. At each point
in time, the targetting algorithm calculates the component of
the antenna phase centre velocity lying along the lineofsight vector, and integrates this velocity component in time
to obtain phase centre displacement along the lineof-sight.
From this displacement, an appropriate value of phase
correction is computed for each point in time, and applied to
the radar signal returns to compensate for the spurious
antenna motion.
The outputs of the strapdown navigator, along with
information from the master INS and DVS, are supplied to a
Kalman filter which estimates various system and sensor
errors associated with these devices. The filterestimated
strapdown navigation errors and IMU instrument errors are.
fed back to the strapdown navigator algorithm and used
there to correct the relevant quantities in a closed loop
fashion. The net result of this Kalman filter implementation
is that alignment of a Dopplerdamped INS is continually
transferred to the strapdown navigator.
3.
OPTIMAL TOA KALMAN FILTER DESIGN
3.1 Statevector
The state vector in the optimal TOA Kalman filter design
contains 26 states, representing all significant t i m e
correlated errors. The state vector can be expressed as
5 PPm (lo)
0.01 "lhr (lo)
0.003"1d hr (lo)
Performance
Characteristic
Bias repeatability
Performance
the point of interest for S A R motion compensation. The

computed baroaltitude is used in a classical third-order
damping loop to stabilize the strapdown vertical channel.
50 ppm (lo)
50 Pg (10)
Table 2. INS Gyro and Accelerometer Performance

Characteristics.
2.2 Processing Overview

Velocity and angular increments from the IMU are
processed in a strapdown navigator algorithm which
mechanizes the navigation equations in a wander azimuth
frame. Since the IMU is mounted directly on the antenna,
the strapdown algorithm fundamentally computes antenna
attitude and heading, and velocity and position of a
particular point at the IMU location. A lever ann correction
is applied within the navigator to obtain estimates of the
velocity and position of the antenna phase centre which is
where the subvectors xM and xs contain master INS system

error states and strapdown navigator system error states
respectively, xlNSand xlMUcontain states that represent INS
and IMU instrument errors respectively, and xDvs contains
augmenting states that represent errors in the DVS. These
26 states are listed in Table 3. The error states modelled in
xM and xs are defined as system-indicated values minus true
values, while the instrument error quantities modelled in
xINS,xlMUand xDVSare defined as factory-calibrated values
minus true values.
3.2 Initial Covariance Matrix

The initial covariance matrix Po for the state vector is
expressed in terms of the initial estimation error
Po= EIZoZ: 1 ,
E,
as
291
where "E" denotes expectation value, the superscript "T"

indicates matrix transpose, and the initial estimation error is
defined by
-,.
I o "I o - I o
where
PM,= E[iM,i:
with io denoting the estimated value of ~ 0 the

, true value of
the initial state vector. The initial covariance matrix for this
TOA system is expressed in terms of submatrices as
-T
PLNh
= E [ ~ L N s1 a ~ ~ ~
9
Ps,= E[isagl 1 ,
P
,
Subvector
State
-
'DVS
XS
X~~~
1
2
E[i,,i,T ]
Description
Coordinate Frame
master INS position error along X axis

master INS position error along Y axis
master INS velocity error along X axis
master INS velocity error along Y axis
master INS platform misalignment about X axis
master INS platform misalignment about Y axis
master INS platform misalignment about Z axis
Wander Azimuth'
G%
G
G Z
master INS x accelerometer bias

master INS y accelerometer bias
master INS x gyro bias
master INS y gyro bias
master INS z gyro bias
Aircraft Body'
6S
DVS forward scale factor error
Aircraft Body
DVS azimuth boresight error
AMX
x~~~
(3)
strapdown navigator position error along X axis

strapdown navigator position error along Y axis
strapdown navigator velocity error along X axis
strapdown navigator velocity error along Y axis
strapdown navigator platform misalignment about X axis
strapdown navigator platform misalignment about Y axis
strapdown navigator platform misalignment about Z axis
IMU x accelerometer bias

IMU y accelerometer bias
IMU x gyro bias
IMU y gyro bias
IMU z gyro bias
Wander Azimuth
IMU Body3
Wander azimuth frame: X,Yaxes level, Z axis up

Aircraft body frame: x axis forward, y axis out right wing, z axis down
IMU body frame: x axis forward, z axis down, y axis oriented to form right-handed coordinate system
Table 3. Description of States for the Optimal TOA Kalman Filter.
298
The submatrix PI,,is given by

.
a:,o
O u : , O
0
0
0
0
0
0
0
O u : , O
o u : , o
0
0
&
O
u
02, o
0
0
o u : , o
0
0
o u h o
0
0
0
0
0
0
0
0
O
0
0
0
0
O U L h
0
0
0
0
In Eq.(13), U:+ is defined as the variance of the error in the
o u g
initial estimates of 6% and #+.
where the variables are defined as follows:
6RMx and 6RMv,

U
The submatrices PIN,, PI,,, and PDwoare expressed as
- the variance of the error in the initial estimates of
&, -
U+ ;
U
:
,
the variance of the error in the initial estimates of

6VMxand 6VMv,
0
0
0
Pmh= 0
0
OM, and 4My

9
- the variance of the error in the initial estimate of
U$
#MZ
In this system, position, velocity, wander angle and heading

in the IMU strapdown navigator are initialized to master
INS values for these quantities. This implies that at filter
initialization, the following relationships between estimation
errors are true:
0
0
0 0
ULn
.LE
U L n
o u : , o
- the variance of the error in the initial estimates of
Wz
Pm=
0
0
0
0
0
0
0 0
U&
0
0
U L E
ULE
(15)
(16)
, i = X,Y
SV =-SV, , i = X,Y
SR = SR,
si
(12)
#sl=#MZ
Strapdown navigator roll and pitch are initialized

independently from the master INS values because the IMU
is mounted on the stabilized antenna while the master INS is
mounted on the airframe. This last point, together with the
equations in Eq.(12), results in the following expressions for
P and PMB0:
80
P%=
~
u
o
o
o o o
o
o
o u ; + o
o u : +
U:,
0 U i h
In the above expressions, the variables are defined as:
- variance of the master INS accelerometer bias, as
specified in Table 2,
U
LE- variance of the master INS gyro bias, as specified in

Table 2,
f-
variance of the strapdown IMU accelerometer bias,

as specified in Table 1,
LE-
variance of the strapdown IMU gyro bias, as

specified in Table 1,
&, -
variance of the Doppler forward scale factor error,

as specified in Section 2.1,
f-
variance of the Doppler antenna azimuth boresight

misalignment, as specified in Section 2.1.
299
3.3 State Dynamics

The continuous model for the error state dynamics has the
standard form applicable to linear systems:
i ( f ) = F(t)X(t) + ~ ( t )
This yields
(18)
In the above equation, WO) is a 26 element vector of zeromean white noise processes. This noise vector has the same
structure as the state vector, namely
0 0
R,
0
F,
4,
4
41
4,
41
0
--vz
R,
IgZ
R,
--vz
f,
-f,
--l
--@ffiz
-a,
.a*
R,
a
nvx
a*
and FO) is a 26 by 26 element matrix with the following

structure:
f,
RX
-fz
-hffiz
--
%ffiz
affiz
-ow
am,
-aI"X
ffiz
rFM
FMIINS
F= 0
Fms
0
0
0
FDVS0
(20)
with
0 Fs F,,
O O F ,
,
where F, and Fs are 7x7 matrices, F,,,
is a 7x5 matrix,
FsnMUis a 7x5 matrix, F,, is a 5x5 matrix, FIMU
is a 5x5
matnx, and FDW is a 2x2 matrix. Using the expressions
from Eqs.(l), (19) and (20) in Eq.(l8), the system dynamics
can be expressed in terms of the state subvectors as
XM
FMX, + F",sxmS
+ WM
>
(21)
Rx=rx+h ,
XDVS
X, =
F, xS + Fw, , X
FDVS 'DVS
X,=F,X,+W,
wDVS
(23)
RT=rT+h
+ W,
(25)
The elements of F, correspond to a true frame inertial error

model [8] mechanized for a wander azimuth coordinate
system that has its origin at the master INS,and that has X,
Y,Z axes and wander angle a as depicted in Figure 2.
In the above equation, the variables are defined as follows:
Vx, V,V,-
the X,Y,and Z components of V, the

master INS velocity vector,
fx,frfi -
the X, Y,and Z components o f f , the master

INS specific force vector,
aE., aEy,aEg
rwl
Figure 2. Orientation of Wander Azimuth Frame
(24)
@IwX*%#y"Iw;
the X, Y,and Z components of

the angular velocity vector of the earth
with respect to inertial space,
the X,Y,and Z components of a,,,,,
the angular velocity vector of the
wander azimuth coordinate frame
with respect to inertial space,
h-
altitude of the master INS above the earth

reference ellipsoid,
rx, r y -
local radii of curvature of the earth reference

ellipsoid along the X and Y axes respectively.
300
Vertical channel error states for the master INS are not
modelled in the Kalman filter because vertical channel
errors are controlled with a third order ftxed-gain damping
loop implemented internally in the INS.
The elements of Fs have exactly the same form as that
expressed in Eq.(26) for F,, except that the wander azimuth
frame of interest has its origin at the strapdown IMU, and h,
V, and f apply to the strapdown IMU rather than the master
INS.
The submatrix F,
has the standard expression for a
strapdown mechanization:
0
0
Cll
0
0
CIZ
cz, czz
0
0
0
0
0
0
0
0
0
0
0
0
zero, since the IMU is mounted on the antenna which is

physically stabilized in roll and pitch.
The dynamics of the error states in xIM, x,,", and xDvs are
modelled as fvst order Markov processes, resulting in the
following submatrices:
FINS
0
0
(27)
-ell -clz-c13
-c1, -czz-c1,
Fm
-8,
0
0
-8,
0
0
-BsorI
0
0
-P,
-c,, -qz-q3-
where C,,is the element in the ith row andjth column of the
direction cosine matrix
which expresses the
transformation from the aircraft body frame to the master
INS wander azimuth frame. The aircraft body frame is
defined in Table 3.
is computed as
-COSACOSR- C O S A S ~ R sin Asin P sinR sin Asin PcosR (28)

-cosPsinR
-cosPcosR
In Eqs.(29) to (31), the variables are defined as:

inverse of the correlation time for the master INS
accelerometer bias error,
cosAcosP -sinAcosR+ sinAsinR+

cos A sinP sin R cos Asin P cos R
= -sinAcosp
(31)
with
inverse of the correlation time for the master INS

gyro bias error,
inverse of the correlation time for the strapdown
IMU accelerometer bias error,
inverse of the correlation time for the strapdown
IMU gyro bias error,
inverse of the correlation time for the DVS forward
scale factor error,
inverse of the correlation time for the DVS azimuth
boresight error.
A=a+H ,
and the variables defined as:
a - wander angle for the master INS wander angle frame,
R - aircraft roll,
P - aircraft pitch,
The vector of random forcing functions in Eq.(18), denoted

by w, is described in terms of its covariance matrix which is
given in the continuous formulation by
P,= ~ w ( ~ ) w ( T ) Q~ (]f =) S ( f - 7 )
H - aircraft heading from North.

The submatrix FsnMu
has exactly the same form as F,
in
Eq.(27) except that in the case of FSnMU,the C,, are
(32)
where 60-r) is the Dirac delta function and QO) is a spectral

density matrix. For this system, Q is expressed in terms of
submatrices as
enw
which describes
elements of the direction cosine matrix
the transformation from the strapdown IMU body frame to
the wander azimuth frame centered at the IMU. The
enw
strapdown IMU body frame is described in Table 3.

is
given by the same expression as in Eq.(28) with roll, pitch
and heading interpreted as being that of the IMU. In
general, the roll and pitch of the strapdown IMU is close to
(33)
30 1
In the above equation, the submatrix Q, is given by:
qm+
qw+
qyoN
Qs
Q,
qlurr
qM.4sr
(34)
%Df
qm+
o q , , o
oq,,
4SGr
qSGr
0
0
q~lw'
QJarr
where qvD is as defined previously, and qqsWu and

qSaF are the spectral densities of white noise models that
0
0
0
0
0
0 qMoN
account for the effects of strapdown IMU accelerometer
scale factor errors, IMU gyro mass unbalances excited by
aircraft accelerations, and the z gyro scale factor error
where q, and qmF are the spectral densities of white noise
respectively. Of course, gyro mass unbalances are also
models to account for the effects of vertical deflections and
excited by the constant gravitational field, but this
INS accelerometer scale factor errors respectively, and qMGN
component of the mass unbalance effect is indistinguishable
is the spectral density of the white noise associated with the
from a gyro bias effect and so is lumped in with the
random drift of the INS ring laser gyros, as specified in
modelled gyro bias states. It should also be noted that the
Table 2. The spectral densities q, and qmF are given by
effects of the IMU x and y gyro scale factor errors are not
the expressions
modelled because in this case, these errors do not contribute
to inertial platform misalignments; since the x and y gyro
input axes are physically stabilized in the horizontal plane,
(35)
they do not experience any significant angular velocities that
would excite scale factor error effects. On the other hand,
the z gyro input axis, which essentially points along the
local vertical, does experience a significant angular velocity
during an aircraft heading change, so that during a turn, the
z gyro scale factor error causes a platform misalignment
where the variables have the following definitions:
about the Z wander azimuth axis.
the nominal value of gravity,
gThe spectral densities qwp qsGMvand qsGsFare given by
0
QVON
d,
the correlation distance of the random deflections of

the vertical,
a jD
the variance of the deflections of the vertical,
aa
the expressions:
aircraftgroundspeed,
magnitude of the aircraft acceleration vector,
(39)
hF- the variance of the master INS accelerometer scale

factor error, as indicated in Table 2.
A derivation for the expressions in Eqs.(35) and (36) is

given in [9].
where the variables have the following definitions:

a
&, -
LW - the variance of the strapdown IMU gyro mass
The submatrix Qs is written as:
the variance of the IMU accelerometer scale factor

error, as specified in Table 1,
unbalance, as specified in Table 1,
- the variance of the strapdown IMU gyro scale
factor error, as specified in Table 1,

Iw,I
the magnitude of the z IMU body axis component

of the IMU angular rate vector.
302
Derivations for the expressions in Eqs.(38) to (40) are found

in [9].
The submatrix Q,,
,
I
Q,,(f>S
is defined by the equation
(t-
7 )
E[~,(t)w,(~)~l
(41)
and is given by
Q,,=
o q , , o
0
0
0
0
0
0
0
0
0
0
0
0
qvD 0
0
0
0
0
0
0
0
0
0
0
0
0
(42)
In the above expression, the two non-zero terms along the

diagonal represent the correlation that exists between
elements of w, and ws as a result of modelling vertical
deflection effects as a white noise process that drives both
the master INS and strapdown navigator velocity error
states.
The submatrices QINS,QIMU,
and Q,
are
(45)
The spectral densities in the expressions above are

computed from the following expressions:
where all variables have been previously defined. The

spectral density expressions in Eqs.(46) through to (51)
represent the level of driving white noise for each Markov
process model that results in the steady-state variances for
the models being equal to the initial estimation error
variances described in Section 3.2.
From the description of the state vector and state dynamics
model given in this section, it is clear that even this socalled "optimal" design is not truly optimal, in the sense that
certain simplifying assumptions have already been made to
restrict the number of modelled states.
The model
simplifications include the following:
1) Vertical position error, vertical velocity error and vertical
accelerometer bias states have not been modelled in the
Kalman filter for either the master INS or the IMU
strapdown navigator because these errors are kept small by
separate fixed-gain baroaltitude damping loops.
2) The effect of the scale factor error in the master INS ring
laser gyros has not been modelled because it is very small (5
ppm) and its effect is expected to be insignificant compared
to the effect of other master INS gyro errors.
3) Certain time-correlated errors in the strapdown IMU
sensors, such as gyro mass unbalances and gyro scale factor
errors, have only been modelled, if at all, as white noise
processes driving the inertial system error states, because
for this configuration where the IMU is physically stabilized
in roll and pitch, their effects are too small to allow these
errors to be observable as separately modelled states in the
Kalman filter.
4) Vertical deflection effects have been modelled as a white
noise process driving velocity error states because these
effects are expected to be too small to warrant modelling as
filter states.
Whether or not a filter designer feels confident in making
such model simplifications right at the outset of the design
process depends largely on the designer's experience and
understanding of the system at hand. Typically, it is
desireable to make some initial model simplifications simply
to reduce the development time required to amve at the
final Kalman filter configuration. However, if there is any
doubt as to the significance of certain errors, it is better to
err on the side of caution and include them as modelled
states in the initial filter design. Later, through further
analysis, detailed simulations and field trials, the designer
gains additional insights into the system error behaviour,
which may then allow him to reduce the model complexity
with a higher degree of confidence. The initial filter
formulation then serves as the designer's "optimal"
configuration against which the performance of other more
suboptimal designs is compared. In this respect, the state
dynamics model presented in this section is considered the
optimal Kalman filter model for the TOA system, despite
containing the model simplifications described.
3.4 Measurements
Two different sets of measurements are constructed for use
in the Kalman filter. One set is a velocity matching
measurement ~0 between the DVS and the master INS. This
303
measurement type is used to estimate velocity and attitude

errors in the master INS. The other set is a velocity
matching measurement + between the master INS and the
strapdown navigator, which effectively accomplishes the
transfer of alignment.
The measurement vector z, is computed as
where C& and C k are the direction cosine matrices

describing respectively the transformation from the
strapdown IMU wander azimuth frame to a specified earthfmed frame, and the transformation from the master INS
wander azimuth frame to the earth-fmed frame. The matrix
C I is calculated as
where the variables are defined as follows:

2;
the master INS/Doppler measurement vector

resolved in the aircraft body frame, denoted by the
superscript "A",
CY-the direction cosine matrix computed from Eq.(28)

by using the roll, pitch, heading and wander angle
indicated by the master INS,
v,"
- the velocity vector computed by the master INS and

resolved in the master wander azimuth frame, as
denoted by the superscript "MW",
v; a;
the aircraft velocity vector indicated by the DVS and

resolved in the aircraft body frame,
- the angular velocity vector of the aircraft body

frame with respect to an earth-fmed frame, as
computed by the master INS, resolved in the aircraft
body frame,
d; -
where a, L and I are the wander angle, latitude and

longitude computed by the IMU stripdown navigator. The
matrix C k is calculated with exactly the same expression
as in Eq.(55) except that the wander angle, latitude and
longitude used are those indicated by the master INS.
The Kalman filter actually processes only zvx and zVy, the
horizontal components of
zY , and zDxand zDy,the x and y
components of z t .
Also, the master INS/Doppler
measurements are processed only during nominally straight
and level flight to avoid additional inaccuracies in the DVS
that result when the aircraft is steeply rolled o r pitched. The
measurement vector z that is processed by the filter is then
written as
the lever arm vector from the master INS to the

DVS, resolved in the aircraft body frame.
The measurement vector
+ is calculated by the expression
where the variables not previously defined are :
3.5 Measurement Models
z r - the master INSlstrapdown measurement vector,
The linear model for discrete measurements is given by
resolved in the master INS wander azimuth frame,
V," - velocity vector computed by the IMU strapdown

navigator, and resolved in the strapdown IMU
wander azimuth frame, denoted by the
superscript "SW",
d: -
the lever arm vector from the master INS to the

strapdown IMU, resolved in the aircraft body frame,
- the direction
cosine matrix describing the

transformation from the strapdown IMU wander
azimuth frame to the master INS wander azimuth
frame.
The matrix
is computed by
cz
CZC,., ,
(54)
z=Hx+v ,
(57)
where H is the measurement matrix with dimension 4x26

and v is a 4 element vector of measurement errors that are
assumed to be zero-mean sequences uncorrelated in time.
The measurement matrix is derived by perturbing the right
hand side of Eqs.(52) and (53) and expressing the result in
terms of the modelled error states.
The following
assumptions are made in this derivation: 1) the effect of
emrs in computing the o x d terms and the transformation
matrix
can be neglected (these small effects are
included in the measurement noise model), 2) the x and y
axes of the aircraft body frame are nominally in the
horizontal plane when master INS/Doppler measurements
are constructed, which allows the measurement error terms
containing vertical velocity errors to be neglected, and 3)
vertical velocities are small enough (i.e. < 10 m/s) to allow
measurement error terms containing vertical velocity to be
neglected. These last two assumptions are valid since, as
304
pointed out earlier, master INS/Doppler measurements are

processed only during nominally straight and level flight.
The resulting measurement matrix is expressed in terms of
submatrices as:
.;R
The submatrices H,
%V
IS
H,
(58)
U;
- the variance of the measurement noise for z
Dy
- the variance of the measurement noise for z X and
The main contributor to the measurement noise for z4 and

zD, is the noise on the DVS velocity measurements, which is
and Hbm, which have dimensions of
2x7 and 2x2 respectively, are given by
specified in Section 2.1. The measurement noise modelled

for z., and
I,
includes the effects of quantization of master
INS velocity data, and the effects of airframe flexing which

causes errors in computing relative velocity between the
master INS and strapdown IMU.
The fact that the matrix R is diagonal implies that the
elements of v are uncorrelated with each other. Strictly
speaking, this is not true. There is a correlation between the
measurement noise for z4 and zD due to the fact that the
noise associated with the DVS forward and lateral velocity
measurements is somewhat correlated.
However, by
assuming that the noise elements are uncorrelated, there is
only a small loss in optimality, while there is a significant
advantage gained in that the scalar components of the
measurement vector can b e processed sequentially by the
filter at a given update time, rather than processing the z
vector as a whole. This allows matrix inversion to be
avoided, thereby increasing computational efficiency.
where the variables are defined as:

the elements of
5DX
y,
the x component of V,^
V ,V -
the X and Y components of V,,
V ,V -
the x and y components of V; which is

computed as
Mx
Mx
MY
My
v; = y v ,
The submatrices HvMand Hvs, both of dimension 2x7, are
given below:
0 0 - 1 0 0 0 0
H,.=[
0 0 0 - 1 0 0 0
0 0 C,
%=[
0 0
c;,
where the C; are the elements of
c;,
0 0 0
c;, 0
0 0
c.
(63)
The measurement noise vector v is characterized by its

covariance matrix R which is given by
R = E [ v v ~=]
In the above equation, the variables are defined as

U
h-
the variance of the measurement noise for z
OX
A high-fidelity computer simulation is an extremely useful

tool in evaluating a Kalman filter design and predicting filter
performance. Such a simulation was performed for the
optimal TOA filter described in this section.
The simulation package that was used generates simulated

master INS, strapdown IMU, DVS and air data, which are
then used as inputs to a processing package. The simulated
sensor data corresponds to a particular user-defined flight
profile, and includes the effects of all the sensor errors
discussed in Section 2.1, as well as other errors l i e
quantization of the sensor output data. The simulation
package also provides files containing true position,
velocity, attitude and heading as a function of time for both
the master INS and strapdown IMU. These files serve as
the reference against which the processing package outputs
are compared in order to evaluate Kalman filter estimation
errors. The processing package consists of the Kalman
filter, strapdown navigator and baroaltitude algorithm shown
in the block diagram of Figure 1.
For the Kalman filter processor, the Kalman filter equations
are implemented in their discrete form using Biermans U-D
factorization algorithm [lo] to efficiently propagate and
update the error covariance matrix . Other features that are
used to enhance computational efficiency include sequential
processing of measurements, sparse matrix multiplication
techniques and exploitation of the block structure of certain
matrices to increase the execution speed of matrix
operations. In the discrete Kalman filter formulation, the
model for the error state dynamics as expressed in Eq.(18)
is converted to a discretetime model of the form:
305
'k+I
@kXk +
w,
(65)
where w, is a vector of zero-mean white noise sequences

with covariance matrix Q,, and x, and
refer to the
values of the state vector at times tk and tk+, respectively.
The interval A C = (fb+, - f k ) is the interval between Kalman
filter measurement updates. The update interval for this
simulation is 10 seconds. To obtain the transition matrix
computed by the strapdown navigator in the processing

package.
The dashed lines show the filter-predicted
standard deviation bounds for these estimation errors. The
sharp decreases in the standard deviation bounds generally
occur right after turns, when DVS error states and inertial
platform misalignments about the Z axis become separately
observable from the other error states. It can be seen that
the time histories of the estimation errors are fairly
consistent with the predicted standard deviation bounds.
a,, a transition matrix a j is first computed over a smaller

subinterval, referred to as the propagation interval, by
retaining the first three terms of the matrix exponential
expression below:
a1= eXp(FjAf,)
=
where
Af,
At,
I + RjAf, + -F,2(At,)2
1+ -Fj3(Af,)3+...
12
6
is the propagation
= (tbI - f,) , and
Fj
interval
,(66)
defined by
is the average value of F(r) over
1
-63
43
-84
At,, with FO) given by the expression in Eq.(20). In the

simulation, the basic propagation interval is 2 seconds. The
'k
ajt4aj+3aj+2aj+1aj
(67)
'
-81
-80
-60
Figure 3. Simulated Flight Profile
transition matrix a, is recalculated every propagation

interval, and ak is then calculated as
-82
LONGITUDE (deg)
TIME
-
DESCRIPTION OF MANOEUVRE
(8)
The covariance matrix Q, is computed from the expression
Q,
p"
e'~*~Q(7
d7
(68)
by expanding the matrix exponential terms, multiplying out

and integrating the products with the assumption that QO),
the spectral density matrix given in Eq.(33), is constant over
the interval A <
of F(r) at tk+,.
In the above expression, Fk+lis the value
For this simulation, the elements of the submatrices PMmand
Ps,in the initial error covariance matrix Powere assigned

values consistent with expected system errors in the INS
following a stationary ground alignment. The flight profile
used for the simulation is depicted in Figure 3. A
description of the various manoeuvres included in the
simulated flight profile is given in Table 4. It should be
noted that the DVS data is not used to construct Kalman
filter measurements until cruising altitude is reached. The
velocity matching measurements between the strapdown
navigator and master INS are constructed and processed
right from time zero.
The performance of the optimal TOA Kalman filter is
indicated in Figures 4 to 8. The solid traces in the plots
indicate the strapdown navigator velocity, attitude and
heading estimation errors over the course of the flight,
assuming closed loop correction of strapdown navigator
quantities with the filter-estimated error states.
The
estimation errors are obtained by subtracting the "true"
velocity, attitude and heading values, as generated by the
simulation package, from the corresponding values
Stationary; heading is 150 degrees
Start accelerating down runway
75
Takeoff, aircraft pitch increases to 6 degrees
116
Achieve and maintain cruising speed of 108 m / s
173
Reach and maintain cruising altitude of lo00 m;

aircraft pitch returns to zero
330
Enter racetrack; aircraft roll increases to 30 degrees
392
Complete first 180 degree turn of racetrack
500
Enter second turn of racetrack
562
Complete second 180 degree turn of racetrack;

aircraft heading returns to 150 degrees
870
Start turn to heading of 90 degrees
891
Complete turn to heading of 90 degrees
941
Start S-manoeuvre
1067
Complete S-manoeuvre; aircraft heading returns to 90

degrees
3400
Start S-manoeuvre
3506
Complete S-manoeuvre; heading held at 150 degrees
4500
Complete profile
Table 4. Description of Manoeuvres in Simulated Flight

Profile.
306
1000
2000
3000
TIME ( 8 )
4000
5' 10
Fgure 4. Strapdown Navigator North Velocity Error and

Predicted Error Standard Deviation (Optimal
Filter).
Figure 5. Strapdown Navigator East Velocity Error and

Filter).
Figure 6. Strapdown Navigator Roll Error and Predicted

Error Standard Deviation (Optimal Filter).
Fgure 7. Strapdown Navigator Pitch Error and Predicted

Error Standard Deviation (Optimal Filter).
TIME (s)
TIME
(8)
Fgure 8. Strapdown Navigator Heading Error and

Predicted Error Standard Deviation
(Optimal Filter).
307
4.
SUBOPTIMAL TOA KALMAN FILTER DESIGN
4.1 StateDecoupling
If certain parts of a system state vector are only weakly
coupled, it is often possible to split the one Kalman filter
into two o r more lower order filters. This can reduce the
computational burden by a significant amount since the
number of multiplications required to perform the most
intensive filter task - the solution of the error covariance
equations - varies roughly as the third power of the number
of states.
In this section, it is shown that the optimal TOA Kalman
filter described in Section 3 can effectively be split into two
filters.
The Doppler damping of the master INS is
performed in one filter, while the transfer of alignment from
the master INS to the strapdown navigator is performed in a
second filter which is mathematically decoupled from the
first.
The means by which the original Kalman filter can be
decoupled becomes more apparent if one considers an
alternate filter formulation for which the differences between
the absolute errors in the master INS and strapdown
navigator are modelled, instead of the absolute strapdown
navigator errors themselves. Such a filter formulation is
based on the fundamental insight that measurements
constructed by comparing equivalent information from two
systems with the same error dynamics will only allow
differences between the two systems to be observable,
rather than the absolute errors in each system. Thus the
velocity matching measurement between the master INS and
strapdown navigator provides direct observability only for
x; where
and it has been assumed that F

, F,. This assumption is
valid as long as the axes of the master INS and strapdown
navigator wander azimuth frames are nearly coincident (i.e.
I , the identity matrix) and the relative motion

between the master INS and strapdown IMU is small
compared to the nominal values of velocity and specific
force appearing in the F matrices. Both these conditions are
true in this case.
The measurement model for the master INS/strapdown
velocity matching measurements is obtained .in terms of X;
by first writing out the following expression, using Eqs.(56),
(57) and (58):
where HvY and Hvs are given by Eqs.(62) and (63).

Substituting xS = x; + xY into the equation above and
assuming again that
I yields
(73)
The form of the above equation clearly c o n f m s that only
x; is fundamentally observable from the z+ measurement.

In order for the optimal TOA filter to be decoupled, the
estimation of one set of states x1 must be mathematically
decoupled from the estimation of the second set x2, where in
this case, it is desired that
(74)
x;=xs-x
(69)
and not for xs o r x,, themselves, whereas the DVS

measurements, as defined in Eq.(52) do allow elements of
xM to b e observable. From the viewpoint of observability
then, an alternative selection of inertial system error states
for the original fully coupled filter could just as well be xM
For x1 to b e decoupled from 3, there must b e no modelled

correlation between x1 and 3. In other words, the crosscovariance matrix of x1 and 3 must be zero at all times.
There are generally three ways that correlation betwken x1
and 4 can be introduced into the Kalman filter:
1) through the measurement model,
2) through the model of the system dynamics,
3) through the initialization of the covariance matrix.
The measurement model for z,,, and z,,, as expressed in
and x; instead of xM and xs. In that case, the dynamics of
Eq.(73), is only a function of 3, and the measurement
the state vector x; are derived by subtracting Eq.(21) from
model for z4 and zD discussed in Section 3.5 is only a
Eq.(24) to give
function of xl. So these measurement models are already in

a form that avoids correlation between x, and 3.
i; = Fsx; + (Fs/hcuxm - FnclmsXms)

+ w;
(70)
where
Wb=Ws-W,
(71)
Examination of the error dynamics models in Eq.(70) and

(21) indicates that correlation will develop between x, and 3
because both x,, and x; are driven by a common error state
term Fmsxns. However, the error covariance for the states
308
in xIMUis modelled as being significantly larger than the
i, = F,x,+ W , ,
i , =F,x,+ W, ,
error covariance for xIm, so that the effect of Fmsxmson the

estimation of x; is expected to be much less than the effect
where
of the Fmxm term, and can therefore be neglected. There

is also a certain amount of correlation between w,,,and w; .
The cross spectral density matrix for these two process
noise vectors, using Eqs.(34) and (42), is found to be
F, =
and
0
0
0
0
0
0
0
0
0
0 -4-,
0
0
0
0
0
0
1 0
0
0
0
0
0
0
0
-4-,
0
0
0
0
0
-qMGN
0
0
0
0
0
0
0
0
0
(75)
The initial cross-covariance matrix between x1 and 5 is
expressed as
-qMGN
-qmNJ
However, consider the spectral density matrix for W; . This

matrix Qi is expressed in terms of previously defined
spectral density matrices as
Qi =
+ QS-zQM,
0
0
0
0
0
0
0 q.%.W+
0
4 ,
0
0
qsw+
0
0
0
0
%Mu+
4WsF
(76)
77)
qMGN
4sonnr+
qUGN
The principal terms in the above expression that affect the

estimation of
(81)
All of the terms in the above expression are found to b e

x,,, and x;.
This term is evaluated, with the help of
Eqs.(ll) and (14), as
0
0
PIS = E[%6%'4 ] =
1
~[%ms~%&
I~
E[iDmaf&
1
H%Ma%&
zero except for the initial cross-covariance matrix between

QM
which yields
1
H%msa%LaT
I
4'Dm,'LaT
1
E['Ma';aT
14
are qsGMuand qsGsFwhich, in general, are
significantly larger than the qMGNterms that appear in the

same locations in the matrix Q;,
assumed that Q;,
Consequently, it can be
0 with little effect on the estimation of
x; . It may be noticed that qn, does not appear in Eq.(77).

This is consistent with the expectation that vertical deflection
errors, which affect both xg and xMin the same way, should
not affect the difference states. Based on the above
discussion, the state dynamics models for the two decoupled
filters can be written as
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0 -ut
0 0
0
0
0
0
0
0
0
-U&
0
0
0
0
0
0
0
(82)
The magnitude of the non-zero terms in the above

expression is much smaller than the initial variance
assigned to the estimation error for $& and .4;
U:,
This
variance is large to reflect the large uncertainty in the

knowledge of the initial roll and pitch of the IMU. Under
these conditions then, it can b e assumed that Puo=O, with
little consequence for the estimation of 5.
In summary then, it can safely be assumed that the three
conditions to allow state decoupling, as stated previously in
this section, are satisfied for the two vectors x, and 5, and
consequently, the optimal TOA filter can be mechanized as
two decoupled filters.
309
It should be noted that for the purposes of closed loop

correction of the strapdown navigator parameters, it is still
an estimated value of xs that is required, not x;.
Consequently, for this decoupled filter configuration where

xs is not explicitly estimated, it must be constructed when it
x s = x M + x;
(83)
where the symbol denotes estimated quantities.
0.02
e:
0
K
is needed as
0.03
et
e:
0.01
z
0
3E
0.00
w
rn
9 -0.01
m
0
* -0.02
K
4.2 State Deletion

If certain error states that are being modelled in the filter are
too small to be observable with the given measurement
models, then they can often be eliminated with little impact
on the filter performance. For example, consider the plots
in Figures 9 to 11 obtained from the simulation run with the
optimal filter. The solid traces in these plots show the time
histories of the master INS gyro bias state estimation errors,
and the dashed lines show the filter-computed standard
deviations for these estimation errors. It is obvious that the
predicted standard deviations do not change significantly
from their initial values over the course of the run. This is a
clear indication that these states are too small to be
observable by the filter.
x
-0.03
10
Fgure 9. Master INS X Gyro Bias Estimation Error and

Filter).
0.03
f
\
t
z
0.02
3K
K
When states are eliminated, it is usual to attempt to

compensate for their loss by adding a component to the
process noise. For example, from the expression for F,,,
in Eq.(27), the first-order effect of a gyro bias state is to
cause a linear growth in inertial platform misalignments.
Thus, the variance growth for the platform misalignment
estimation error as a result of a modelled gyro bias state is
given approximately by
0.01
z
0
ljc
0.00
B
v)
f -0.01
U -0.02
a;+@)=
ULBf 2 .
(84)
If the effects of a gyro bias were modelled by a white noise

process with spectral density qMGB,
then the variance growth
for the platform misalignment estimation errors is expressed
as
0
;p = qmB
(85)
-0.03
Fgure 10. Master INS Y Gyro Bias Estimation Error and

Filter).
0.03
One criterion that the filter designer may use to assign an

appropriate value to qMGB
is to select a value of qMGB that
results in the
U;+
buildup being equal for both cases at a
0.02
e:
chosen time T. This condition is expressed as
aLBT2=qWBT,
5s
w 0.01
(86)
which yields
qWB=ULBT
(87)
The choice of T is itself fairly arbitrary, but a suitable

choice is the filter settling time, which is in the order of
about lo00 seconds. The quantity qMGB
computed from the
above expression is then added to the qMGN
terms in the
spectral density matrix Q, to compensate for the deletion of
the gyro bias states.
Another case for deleting states can be made when it is not
important to know the state estimate, and the state does not
rn
9 -0.01
m
0
K
*0 -0.02
N
-0.03
Fgure 11. Master INS Z Gyro Bias Estimation Error and

Filter).
3 10
have a significant effect on the estimation of other states of

interest. A good example of this is the position error states
modelled for the master INS and the strapdown navigator.
The primary requirement for the TOA filter is to control
attitude errors and velocity errors in the strapdown
navigator.
Absolute positional accuracy is not a
requirement. Furthermore, deletion of the position error
states does not impact on estimation of the other states.
This can be seen by considering the measurement model and
error dynamics model for the TOA filter. Since the
measurement model does not include any terms that contain
the position error states, then obviously it is not affected by
the absence of these states. With regard to the error
dynamics, from examination of the expression for F, in
Eq.(26), it can be seen that position error terms which drive
the velocity errors and platform misalignments are weak
compared to other driving terms for these states. The
dominant term that would be omitted as a result of deleting
w
the' position error states is of the form A 6 R . To first

R
order, this term causes a linear growth of platform tilts.
Following a similar approach to that taken for the gyro bias
states, an appropriate selection for the spectral density of an
equivalent white noise process is
where a t is chosen to be the expected variance of the

position error over the entire !light, and T , as before, is
selected to be the filter settling time. This new spectral
density term is included in QM by summing it with the qMGN
terms that affect the estimation of platform tilts. A similar
spectral density term qsp can be incorporated into Q6 to
compensate for the deletion of position states from the s;
vector.

An identical simulation run to that performed for the optimal
TOA filter was conducted with the suboptimal filter design.
For this run, the modifications to the measurement and error
dynamics model discussed in this section were implemented
to realize two decoupled filters without position error states
or master INS gyro bias states.
The results of the run are shown in Figures 12 to 16 which
depict the velocity, attitude and heading estimation errors of
the strapdown navigator, along with associated filterpredicted standard deviations, after closed loop correction
with the filterestimated error quantities. From comparison
to the corresponding plots in Figures 4 to 8, it is clear that
the suboptimal filter performance is very nearly the same as
that of the optimal configuration. These results confirm the
validity of the model simplifications discussed in this
section.
5. CONCLUSIONS
The process of designing a suboptimal transfer-of-alignment
(TOA) Kalman filter for a S A R motion compensation
application has been detailed at some length. While a
certain amount of the information presented is fairly unique

to this particular application, the criteria used in making the
filter model simplifications provide useful guidelines for
general suboptimal filter design. Also, much of the
discussion regarding inertial system behaviour is of course
relevant to aerospace navigation applications involving
inertial technology.
Further information on general
principles of suboptimal filter design and implementation
can be found in [6].
Specifically, two principal techniques of Kalman filter model
simplification, namely state deletion and state decoupling,
have been demonstrated for the TOA filter example.
Starting with a description of the optimal Kalman filter, the
process of designing modifications to accommodate state
deletion and decoupling has been discussed, along with the
rationale for these modifications. Simulation results for the
optimal and suboptimal filter configurations were presented
to confirm that the suboptimal filter design provided similar
performance to the optimal filter.
6. REFERENCES
1. J.J.Kovaly, Synthetic Aperture Radar, Artech House
Inc., Dedham, Massachusetts, 1976.
2. M.I.Skolnik, ed., Radar Handbook, Chapter 23,
McGraw-Hill Book Company, New York, 1970.
3. D.J.DiFilippo,
G.E.Haslam,
and
W.S.Widnall,
"Evaluation of a Kalman Filter for S A R Motion
Compensation", proceedings of PLANS 88, Orlando,
Florida, Nov.29-Dec.2, 1988.
4. R.E.Kalman, "A New Approach to Linear Filtering and
Prediction Problems", J. Basic Eng., March 1960, pp.35-
46.
5 . R.E.Kalman and R.S.Bucy, "New Results in Linear
Filtering and Prediction Theory", 1. Basic Eng., March

1961, pp.95-108.
6. A.Gelb, ed., Applied Optimal Estimation, The M.I.T.
Press,Cambridge Massachusetts, 1974.
7. M.Kayton and W.R.Fried, eds., Avionics Navigation
@stems, Chapter 6, John Wiley and Sons, Inc., New York,
1969.
8. D.O.Benson, "A Comparison of Two Approaches to
Pure Inertial and Doppler-Inertial Error Analysis", IEEE
Transactions on Aerospace and Electronic Systems, Vol.
AES-11, NO.4, July 1975, pp.447-455.
9. D.B.Reid and J.Hepburn, "Final Report for S A R Motion
Compensation Study", Report H8402-03/SARMC/DBR,
Huntec (70) Ltd., February, 1984.
10. G.J.Bierman, Factorization Methocls for Dkcrete
Sequential Estimation, Academic Press, New York, 1977.
31 1
TIME
Fgure 12. Strapdown Navigator North Velocity Error and

Predicted Error Standard Deviation (Suboptimal
Filter).
Figure 14. Strapdown Navigator Roll Error and Predicted

Error Standard Deviation (Suboptimal Filter).
(8)
Fgure 13. Strapdown Navigator East Velocity Error and

Predicted Error Standard Deviation (Suboptimal
Filter).
Figure 15. Strapdown Navigator Pitch Error and Predicted

Error Standard Deviation (Suboptimal Filter).
IO
Figure 16. Strapdown Navigator Heading Error

and Predicted Error Standard Deviation
(Suboptimal Filter).
312
VERTICAL CHANNEL DESIGN CONSIDERATIONS
J . Stanley Ausman
Litton Guidance and Control Systems Division
5500 Canoga Avenue
Woodland Hills, CA 91367-6698
USA
1.
Introduction
The vertical channel of an inertial

system is unstable. This
instability is caused by the gravity
compensation fed back to the
vertical accelerometer output. The
gravity compensation, computed as a
function of altitude after doubly
integrating the output of that
accelerometer, creates an unstable,
positive feedback loop. The time
constant of this instability is
about 560 seconds near the earth's
surface. For ballistic missiles and
rockets this does not pose a problem
because the guidance is completed
before the instability becomes
serious. For aircraft systems,
however, one must augment the
inertial measurements, typically
with barometric altimeter
information, in order to stabilize
the vertical inertial channel.
Earliest mechanizations of the baroinertial loop employed second-order
feedback with constant gains
(References 1 and 2). The next step
was to add integral feedback in
order to bias the vertical accelerometer, thus creating a thirdorder system. Widnall and Sinha
(Ref 3) investigated the third-order
loop to find the optimum set of
fixed gains. Not surprisingly, they
found that the optimum set of gains
depended on the values assumed for
the noise characteristics of the
accelerometer and the barometric
altimeter. Because the noise
magnitudes will vary as a function
of the aircraft's flight regime, the
baro-inertial feedback gains should
not be constant, but should also
vary.
Litton first mechanized a thirdorder variable gain baro-inertial
loop in CLASS, an all-weather close
air support system, successfully
demonstrated in 1972 (Ref. 4). They
gradually improved upon that basic
design over the years as successive
systems, principally ARIS (Refs. 5

and 6); LW-33 (Ref. 7 ) ; and LN-39
(Ref. 8), uncovered more and more
barometric altitude error
characteristics which had to be
accommodated. The culmination of
this evolutionary development is the
baro-inertial loop currently
mechanized in the LN-93 and LN-94
systems for the USAF Standard RLG
INU (Ref 9).
Following a discussion of barometric
and inertial errors, we will take a
detailed look at the LN-93/94
conventional vertical channel
mechanization, the reasons behind
the loop design, and some simulation
results illustrating the loop
performance when subjected to
certain flight maneuvers and barometric errors.
These simulations show that a slow
descent (approximately 60 fps) is
the most difficult situation for the
baroinertial loop to handle because
it is hard to discriminate between
an inertial velocity error and a
barometric scale factor error. In
the final section we see that a
Kalman filter mechanization can
alleviate this difficulty by
including barometric scale factor as
one of the filter states. The
resulting mechanization is a 5-state
Kalman filter (inertial altitude,
vertical velocity, accelerometer
bias, bar0 scale factor, and bar0
bias) which uses barometric altitude
updating. One can also augment the
updates with GPS altitude or radar
altimeter measurements, if
available. The performance of the
Kalman filter mechanization is superior to that of the conventional
mechanization of the vertical
channel.
In order to properly design the
vertical channel, one must first
understand the various error inducing mechanisms. The following
paragraphs discuss the various error
313
sources that may occur in the

barometric altimeter and in the
vertical channel of the inertial
system.
Barometric Error.
There are two general categories of
barometric errors. These are (1)
modeling errors in mathematically
characterizing the real-world
atmosphere as a function of
altitude, and (2) sensor errors in
measuring the static pressure and
temperature of the atmosphere.
Typical air data computers model
pressure and temperature as a
function of altitude according to
the U.S. Standard Atmosphere
(Ref 10). Nonstandard day
conditions will cause the air data
computer to compute the wrong
altitude even if the sensor
measurements are correct.
Nonstandard day atmospheric pressure
generates a bias-like altitude
error. Near sea level, a 10 mm Hg
change in pressure corresponds
approximately to a 120m change in
altitude. Hence, a 10 mm Hg
pressure deviation from standard day
pressure (7560 mm Hg) would cause a
120m altitude error.
Because the weather and its
associated parameters, pressure and
temperature, are subject to change,
the error due to nonstandard day
pressure may also vary with time and
geographic location. The
correlation time associated with
diurnal variations in atmospheric
pressure would be 2412
= 3.82 hr.
The movement of weather systems over
the earth adds more uncertainty and
tends to decrease the correlation
time. Based on these
considerations, a correlation time
of 2 to 3 hours seems reasonable for
atmospheric pressure variations.
Blanchard (Ref 11) analyzed
deviations in mean heights of
pressure surfaces listed in (Ref 12)
and concluded that the RMS day-today variability is 380 ft. This
data covered latitudes from 20 deg N
to 80 deg N and altitudes from
10,000 ft to 50,000 ft.
In addition to temporal variations
in pressure at one location, changes
may also occur due to spatial
changes as the aircraft moves from

one location to another. Barham and
Manville (Ref 13) give a value of 1
ft per nmi as the RMS slope of a
constant pressure surface. This
value is in general agreement with
the conclusions of Asbury et. al.
(Ref 14) in their study of the same
problem.
Nonstandard day atmospheric density
can cause an apparent altimeter
scale factor error as the aircraft
changes altitude. Deviations in
atmospheric density from standard
atmosphere are altitude dependent
(Ref 10). Nonstandard density at
low altitudes (below 8000m) is
negatively correlated with nonstandard density, at high altitude
(above 8000m). Nonstandard day
temperatures have a similar negative
correlation between low and high
altitudes. At the "isopycnic" level
of 8000m, the atmospheric density
stays constant to within 2 percent
year round and over all latitudes
(Ref 10). At sea level, air density
can 'reach extreme deviations from
standard day values by as much as
-+ 15 percent. For an RMS value,
5 percent deviation is appropriate.
Because air density is subject to
the same diurnal and weatherchanging patterns as air pressure,
the correlation times should be the
same. Hence, a 2- to 3- hour
autocorrelation time for air density
variations appears reasonable.
The principal errors encountered in
air data sensing systems arise from
the static pressure ports, which may
either be on an air data probe or
flush-mounted on the aircraft skin.
Errors in the air data pressure
sensing device and computer are
generally small in comparison with
the errors in the porting and tubing
system, which transmits air pressure
to the sensing device.
Static pressure ports attempt to
pick up the pressure of undisturbed
(static) air. However, on an
aircraft moving through the air, the
pressure picked up may be
contaminated with dynamic pressure
(q = 0.5
Vz)
One refers to the
amount of contamination as "static
pressure defect". Static pressure
defect is principally a function of
Mach number, angle-of-attack, and
314
side-slip. All new aircraft undergo

an elaborate calibration procedure
to measure the static pressure
defect, which can then be
compensated for in the air data
computer. Nevertheless, the
calibration is imperfect, and
residual static pressure defect
errors remain. These can be
especially large as the aircraft
approaches Mach 1 or actually
transits the transonic region.
False, instantaneous altitude jumps
of several hundred feet are not
uncommon in these instances.
Time lag is another type of pressure
sensing error. It is caused by the
volume of air contained in the input
end of the pressure sensor and the
tubing which connects it to the
static pressure port. As the port
pressure changes, air must flow in
or out of the tubing and pressure
sensor input chamber in order to
reach a new equilibrium. The larger
the volume of air and the smaller
the tubing (which restricts air
flow), the greater the time lag.
Typical air data installations have
time lags of a few tenths of a
second. The effect of this time lag
is to cause the barometric altimeter
to read low during a climb and high
during a dive.
Inertial Errors.
Inertial system errors in the
vertical channel are primarily
caused by the vertical
accelerometer. Gyro errors tend to
be small and temporary by
comparison, their predominant effect
being to couple a small amount of
any horizontal acceleration into the
vertical channel. Medium-accuracy
inertial navigation systems have
accelerometers with bias
repeatability of 50 to 100 ug and
scale factor accuracies of 50 to 100
ppm. These may vary somewhat with
time due to temperature changes.
Typical autocorrelation times for
such changes are of the order of
1000 seconds in ring laser gyro
(RLG) systems.
correlation distance of 20 nmi.

2.
Conventional Baro-Inertial
Loops.
Figure 1 is a software block diagram

of the LN-93/94 vertical channel.
Note that it is divided into two
sets of computations: a fast loop
and a slow loop. The iteration rate
for the fast loop computations is 64
Hz; the slow loop computation rate
is 1 Hz. The function of the fast
loop is to integrate vertical
acceleration to generate vertical
velocity and short-term dynamic
changes in altitude. The slow-loop
computations contain the barometric
stabilizing feedback paths and their
gain-setting logic.
Fast LOOK).
The fast-loop (64 Hz) computations
include the integration of inertial
acceleration and velocity to compute
vertical velocity and altitude
changes, respectively. In these
fast-loop mechanization equations
the terms ABZ and AFB (see Figure 1)
are treated as constants, but they
actually change at a once-per-second
rate determined by the slow-loop
computations.
After the double integration, which
generates inertial altitude HI there
is a first-order lag filter with a
time constant 1/KF equal to 1
second. This matches the 1-second
low-pass filter on the raw baroaltitude input, which is needed to
accommodate applications providing
only very coarse (100 ft) resolution
of the baro-altitude data.
The last item computed in the fast
loop is the difference between the
filtered inertial and baroaltitudes, HL and HB. The slow loop
computations use this difference as
an input only once per second, but
computing it at 64 Hz ensures that
the HL and HB used to form the
difference are synchronized to
within 1/64 second.
Slow Loop.
Gravity anomalies can also cause

errors which look like accelerometer
bias errors and can change values as
the aircraft moves over the earth.
A typical RMS value for gravityinduced bias noise is 25 ug with a
The slow-loop (1 Hz) computations

contain; (1) the gravity
compensation as a function of
altitude, ( 2 ) the third-order
feedback stabilization loops, ( 3 )
315
the gain variation logic for setting

the feedback gains, (4) a baro-bias
compensation differentiatorintegrator combination, and (5) a
16-second bleedoff of the baro-bias
compensation integrator. The
following paragraphs discuss each of
these functions in detail.
Gravity Compensation and
Stabilization Loops.
The H part of the feedback term
(Kl/a) (H-S2) to the acceleration
node mechanizes the variation in
gravity with altitude. This is the
positive feedback which necessitates
the stabilizing influence of the
barometric altimeter. The -S2 part
of that term is the negative
feedback necessary to just bring the
loop to a neutrally stable
condition. The other three paths
for 52 through KB2, KB3, and KB4
provide a stabilizing third-order
feedback control system. The
quantity 52 is the error signal for
the third-order feedback loops. The
gains KB2, KB3, and KB4 are chosen
so that there is one real solution
with a root at -K/T, and a pair of
complex roots at -K/T, 2 jK/T,.
Thus, the sinusoidal solution has a
0.707 damping ratio and decays at
the same rate as does the real
solution to provide good setting
characteristics. The quantity Tv is
the nominal loop time constant,
while K is a parameter that can vary
between 0 and 1 in order to vary the
loop time constant from infinity to
Tv, respectively.
Gain Variation Loqic.
The original gain changing logic
(Ref.4) for setting K was as
follows:
K = 1 -KB5
where
KB5 = Vz2/(Vz2
and
DO = 30 fps
(1)
+ DO2)
(21
The above logic makes K approach

zero, effectively opening the
feedback loops, whenever the
aircraft is in a steep climb or
dive. Under these conditions, the
barometric altimeter input is
suspect, and one should rely more,
heavily on the inertial data.
Subsequently, several problems arose
(Ref.5) which caused us to alter the

logic somewhat. The first of these
was the observation that sudden step
changes of several hundred feet
could occur in the barometric
altimeter input when the aircraft
approaches Mach 1 even though the
aircraft is in straight and level
flight. These steps can result from
shock waves sweeping across the
static pressure ports of the air
data system. A s the aircraft is in
level flight, Vz=O, and the feedback
loops are closed (K=l). Hence,
these steps introduced large
transients into the vertical loop.
To circumvent this problem, we first
tried to replace Vz in Eq.(2) with
HBdot, the rate of change of
filtered barometric altitude. This
effectively opened the loops when
the steps occurred. However, this
mechanization, employed in the
LN-39, ran into difficulty during
long, slow climbs or letdowns in the
presence of nonstandard day lapse
rate (barometric scale factor error)
and large quantization (100 ft)
steps in the barometric altimeter
input. The final solution,
therefore, was to return to the Vz
formulation of Eq.(2), but to make a
separate test comparing HBdot with
Vz in order to reject step changes
in barometric altitude. This test
takes the following form.
IF
1HB - Vzl >2DO, then KB5 = 1, else

use eq(2).
(3)
Originally, the parameter DO (Eq.3)
was fixed at 30 fps. This has the
effect of opening the feedback loops
whenever HBdot differs from Vz by
more than 60 fps, Eq. ( 3 ) , or
whenever Vz is much larger than 30
fps, Eq.(2). During air combat
maneuvers or repeated bombing runs,
however, the aircraft may be
climbing and diving almost
continuously for long periods of
time. Because of the unstable
nature of the vertical inertial
channel, we cannot leave the
barometric feedback loop open for an
appreciable length of time relative
to the 560-second time constant of
the instability. Otherwise, time
variations in accelerometer bias and
spatial variations in gravity
anomalies would be sufficient to
316
cause unacceptable divergence in the

vertical channel.
Consequently, in the LW-33 and LN-39
vertical channels, we caused DO to
increase gradually whenever Vz
exceeded DO. The rate of increase
in those systems was linear with
time and at such a rate that DO
doubled in either 4 seconds (LW-33)
or in 30 seconds (LN-39). These
systems worked well for either
severe vertical maneuvers or level
flight, but long slow climbs and
letdown in the presence of
barometric scale factor errors still
caused problems because the loop
tended to track the barometric error
rate too closely.
To alleviate this behavior in the
LN-93/94 vertical channel, we delay
the increase in DO so that it
doubles in 300 seconds.
Furthermore, the increase occurs as
a parabolic function of time,
essentially flat to begin with and
increasing rapidly after 300 seconds
to ensure loop closure. Figure 2
displays this nonlinear increase in
the parameter DO, which occurs
whenever the magnitude of Vz exceeds
DO.
Once loop closure is accomplished,
as recognized by the magnitude of Vz
being less than DO, then DO
decreases back to its minimum value
of 30 fps at a rate of 0.5 ft/sec.
Hence, if it took 300 seconds to
close the loop by increasing DO to
60 fps, it would take only 60
seconds to bring DO back to 30 fps
once the magnitude of Vz diminished
and stayed below DO.
Baro-Bias Compensation.
Reducing the feedback gains during
climbs and dives protects the
feedback loops by blocking out, or
at least greatly attenuating,
changes in barometric altimeter
error, but only during the climb or
dive. When the aircraft levels out,
the feedback loops will reclose, and
any change in barometric error which
still persists, as caused for
example by a scale factor error in
the altitude measurement, would now
introduce a new loop transient into
S2, which would take several minutes
to damp out. The baro-bias
compensation path through KB5
(Figure 1) overcomes this difficulty

by accumulating the changes in
barometric error when the feedback
paths are open and subtracting them
from the inertial-bar0 difference
signal DHBO before forming the
feedback error signal S2. The tacit
assumption here is that during
climbs and dives the inertial data
is more reliable than the barometric
data and most of the difference
between the two represents
barometric error.
Observed changes in DHBO are
multiplied by KB5 and then summed in
an integrator. The gain KB5 is the
complement of K, the feedback gain
parameter. When K is low, KB5 is
high and vice versa. The output of
the integrator represents the
accumulated changes in baro-bias
error and is subtracted from the
inertial-bar0 difference signal,
thus canceling out the change in
baro-bias.
To see how this works, let us
consider the case where the
barometric altitude error HB
suddenly changes by 400 ft in level
flight. This causes KB5 to be set
to 1, because lHBdot - Vzl exceeds
60 fps. With KB5 = 1 and K = 0, the
feedback loops open up and the full
amount of the 400 ft step change in
the bar0 input becomes stored in the
summing integrator following KB5
(see Figure 1). This 400 ft, when
subtracted from DHBO to form S2,
exactly cancels the 400 ft step in
DHBO so that S2 sees no change at
all. Now when the feedback loops
reclose, there is no transient to
ring through the system. In the
paragraph under Response to larqe
Step in Barometric Altitude Error we
will see simulation results of a
similar example complete with time
histories of several of the loop
variables.
Bias Compensation Bleedoff.

When we first mechanized the barobias compensation scheme described
in the previous section (Ref.4), we
soon found that it tended to
rectify. That is, during repeated
climbs and dives the integrator
following KB5 would build up an ever
increasing compensation value
because air-craft climb rates tend
to be different than their descent
317
rates. As a result, KB5 would

typically have higher values during
dives than it would during climbs.
The net effect was to cause the
inertial altitude H to gradually
diverge from the barometric
altitude. This is not a serious
problem from a system standpoint
because the vertical velocity Vz is
still valid, and the sole purpose of
the baro-inertial channel is to
produce an accurate vertical
velocity. One should never use the
altitude output from a baro-inertial
loop when precision is required. In
weapon delivery systems, altitude
change should be obtained by
integrating Vz directly outside the
baro-inertial loop.
Nevertheless, for display purposes
the vertical channel does output
"inertial altitude,'' and the users
are disturbed if this "inertial
altitude" diverges from the
barometric altitude. Hence, to
prevent this divergent behavior we
added a bleedoff path around the
baro-bias compensation integrator
with a 16-second time constant.
This is the K/16 path from DHHB
(Figure 1) back to the integrator's
input. The value K/16 is really
(1 - KB5)/16 wherein the 1/16
portion is the direct 16-second
bleedoff. We shall soon see the
reason for the KB5/16 portion.
When we bleed off DHHB, we have to
do it in such a way that S2, the
feedback error signal, remains
undisturbed. We accomplish this by
also feeding -1/16 of DHHB into AFB
from whence it travels through the
altitude integrator back to S2, thus
canceling at the S2 node the direct
bleedoff of DHHB back through its
own integrator. The effect of this
bleedoff cancellation is to drive
the filtered inertial altitude HL
back to the filtered barometric
altitude HB. However, this bleedoff
cancellation path can also go up
through "s" and KB5 to the input of
the DHHB integrator. To cancel this
effect, an additional bleedoff
KB5/16 is added to the DHHB direct
bleedoff, making its total (1
KB5)/16, which is the same as K/16.
There is still another secondary

bleedoff path from the altitude
output, H - S2, back to the gravity
compensation which enters the Kl/a
block (Figure 1). We cancel this

path by feeding DHHB/16 through an
integrator (to match the vertical
velocity to altitude integrator in
the other path) to the Kl/a block.
This forestalls any change in
gravity compensation which would
otherwise occur when the inertial
altitude is forced to track a
possibly erroneous barometric
altitude.
Vertical Velocitv and Altitude
Outputs.
As previously stated, the primary

purpose of the baro-inertial loop is
to provide accurate vertical
velocity. To this end we output Vz
and not Hdot (see Figure 1) for
vertical velocity. The ,quantity
Hdot can be corrupted by barometric
altimeter noise coming directly
through KB2 in 1-second steps,
whereas Vz is isolated by the
smoothing effect of at least one
stage of integration from any bar0
noise. Hdot can exhibit
discontinuities (steps) at 1-second
intervals while Vz at most may have
slope discontinuities at 1-second
intervals.
The "altitude output" H-S2 , tracks
the barometric altitude and should
not be used where precision altitude
changes are required as in weapon
delivery systems. For one thing, HS2 is not necessarily the integral
of Vz. Weapon delivery systems
should only use the Vz output and
integrate it externally to the baroinertial loop. This open loop
integration scheme is the only way
to ensure that changes in altitude
indeed represent the integral of
vertical velocity.
The reason for listing H-S2 as the
altitude output is as follows:
Dynamically, H is a better measure
of altitude, whereas statically the
raw bar0 altitude HBR is preferable.
The error signal S2 indicates the
low frequency difference between H
and HBR. Hence, the quantity
H - S2 provides an altitude measure
both with good dynamic response as
well as good static accuracy.
Figure 3 provides a good example.
This figure shows the loop response
t o an initial S2 error of -400 ft
created by initializing HBR'and HB
to 400 ft but initializing H and HL
318
to zero. Note that H approaches 400

ft with the response characteristics
of the third-order feedback loop.
However, H
S2 immediately achieves
a value only slightly above 400 ft
and then slowly approaches 400 ft in
an exponential fashion.
One undesirable feature of the

H - S2 output is that S2 is computed
only once per second. This can
introduce step changes into H - S2
at one-second intervals. For
display purposes this is not
serious, and as stated earlier
H - S2 should never be used for
precision altitude applications
anyway.
In spite of our protestations that
the "altitude output" is unsuitable
for weapon delivery, some users
insist on using it anyway. To try
to accommodate these people, we
added a pseudo open-loop Vz
integrator in the LN-93. The LN-94
still retains H - S2 altitude
output. Figure 4 is a block diagram
of the altitude output computation
in the LN-93. To prevent a
divergent output due to integration
errors, we wrap a first-order
feedback loop around the integrator
with a very long time constant.
This keeps the altitude output very
loosely coupled to the barometric
altitude so that for short time
intervals the altitude output is
very nearly the integral of the
vertical velocity Vz. This pseudo
open-loop integration process is
entirely outside the baro-inertial
loop and does not affect the
vertical velocity accuracy.
Vertical
LOOP Corrections.
USAF specification SNU 84-1 requires

that the baro-inertial loop accept
corrections to (1) vertical
velocity, (2) inertial altitude, and
( 3 ) barometric altitude. Such
corrections may be made as a result
of GPS updates, for example. Figure
1 shows where these intermittent
corrections will enter the baroinertial loop. In order not to
disturb the loop, it is important
that the inertial altitude
correction and the baro-altitude
correction be equal and applied
simultaneously.
Simulations of Vertical LOOP

Performance.
This section contains simulation
results showing the performance of
the LN-93/94 baro-inertial loop when
subjected to various barometric
disturbances. The first simulation
is the response to a large step
change in barometric altimeter
error. This simulation demonstrates
the effectiveness of the baro-bias
compensation and bleedoff network in
squelching this type of barometric
altimeter error. The second
simulation shows the effect of a
30-second burst of barometric
altimeter noise as might be
encountered when an aircraft
traverses the transonic region.
The third simulation displays the
loop behavior during and following a
steep dive and high-g pullout. The
final set of simulations explores
the most difficult flight regime for
the baro-inertial loop, namely,
long, slow descents in the presence
of barometric scale factor errors.
In all of these simulations, we
presume that the barometric
altimeter input is quantized in
100-ft increments and has a 0.3second time lag. Furthermore, the
altimeter scale factor error is
modeled as being -5 percent at sea
level and varying linearly to
+5 percent at 52,000 ft. This is
fairly representative of density
variations in the real atmosphere
under nonstandard day conditions
(Ref.10).
Response to Larqe Step In Barometric
Altitude Error.
To test the effectiveness of the
baro-bias compensation and bleedoff
networks, let us introduce a step
change in altimeter bias error of
400 ft. This differs from the
earlier simulation (Figure 3 ) in
that, instead of initializing the
feedback error signal S2 to - 400
ft, this time S2 is initially.zero,
the loop is completely settled, and
then we introduce a step into the
altimeter bias. In this simulation,
the aircraft is in level flight
(Vz = 0) and the bar0 bias step
enters at time t = 10 seconds. We
then watch the loop response for
another 190 seconds. Figure 5 shows
319
the Vz and S2 responses as well as

the DHHB (output of the baro-bias
compensation integrator) and H-S2
responses.
Note that neither VZ nor S2 are
significantly affected by the 400-ft
step in bar0 bias and that DHHB
jumps immediately to -335 ft to
cancel most of the step input before
it can reach S2. Thereafter it
bleeds of exponentially with a
16-second time constant. During the
bleedoff, H-S2 builds up to a value
of 400 ft to match the new baro
input value. The reason that DHHB
does not jump all the way to -400 ft
is that the 1-second filter on the
bar0 input reduces the peak in HBdot
somewhat.
This example serves to emphasize the
point that the "altitude output , HS2, is not necessarily the integral
of vertical velocity Vz. In this
case, H-S2 changed by 400 ft whereas
Vz remained essentially zero.
'I
Response to 30-Second Burst of

Altimeter Noise.
When an aircraft penetrates the
transonic regime, large variations
in the static pressure sensed by the
altimeter can occur. Figure 6 shows
the response of the LN-93/94 baroinertial loop when subjected to a
30-second burst of altimeter white
noise of magnitude 800 ft peak-topeak. For simulation purposes, the
altimeter error is randomized
between -400 ft and +400 ft with any
value within that range being
equally likely. Figure 6a shows a
sample of one such 30-second burst.
After 30 seconds, the altimeter
error is zeroed, but we continue the
simulation for another 170 seconds
to observe the vertical velocity
transient effects caused by the
noise burst.
Figure 6b displays a set of 10 Monte
Carlo runs each with a different set
of random numbers to generate the
30-second burst of simulated white
noise. In the worst case of these
10 runs, the vertical velocity error
reaches a peak value of 1.8 fps.
The RMS peak value over these
10 runs is 0.93 fps.
Response to Steep Dive and Pullout.
This type of maneuver frequently

appears in the specification for a
vertical channel, and or this
reason we include it. However, we
shall see shortly that a shallow
climb or dive is a more severe test
of vertical channel accuracy. The
maneuver simulated here starts with
the aircraft flying level at 30,000
ft. It then pulls down or 5
seconds at 4.7 g to a 750 fps rate
of descent.
After 35 seconds the aircraft pulls
up at 4.7 g for 5 seconds to level
off at sea level. Elapsed time from
level flight at 30,000 ft to level
flight at sea level is 45 seconds.
We then continue the simulation for
another 155 seconds to observe any
overshoot which might occur.
Figure 7d shows the vertical
trajectory as a function of time.
Figures 7a and 7b display the vertical velocity error and the feedback
error signal S2. In Figure 7c we
see the gain cutoff parameter HBdotVz in comparison with its cutoff
value 2(DO). We see from this last
figure that IHBdot-Vz :exceeds 2(DO)
in both the high-g pull down and
pullout, so that K=O and the
feedback loops are wide open.
During the dive itself the high rate
of descent makes K very close to
zero, and the feedback paths are
still essentially open. Hence both
S2 and the vertical velocity error
remain very small until the loops
reclose at time T = 45 seconds.
Then we see a small transient in
both S 2 and in Vz, but the vertical
velocity error over-shoot is still
only 0.25 fps.
Response to Shallow Dives.
These simulations also start with
the aircraft flying level at 30,000
ft. We simulated 6 different rates
of descent - 30 fps, 40 fps, 50 fps,
60 fps, 75 fps, and 100 fps - in
order to find the worst case which
yields the largest vertical velocity
error. In each case the aircraft
descends from 30,000 ft to sea level
using 1 second to establish the rate
of descent and 1 second to pull out
at sea level. Because these rates
are of the same order of magnitude
as the gain variation parameter DO,
the feedback loops are still closed,
albeit with a somewhat longer that
320
nominal loop time constant.

Consequently, the vertical velocity
error will tend to track the rate of
change of barometric altimeter
error. If the bar0 scale factor
error is 5 percent and the rate of
descent is 60 fps for example the
bar0 error rate would be 3 fps. For
slower rates of descent the bar0
error rate would be slower, but
would persist for a longer time
because the aircraft takes longer to
reach sea level. Conversely, for
faster descents the bar0 error rate
is higher, but for a shorter time.
We expect the worst-case descent
rate, i.e. the one that causes the
largest vertical velocity error, to
be between 30 fps and 100 fps, and
that is the reason for simulating
six different rates covering this
range.
Figures 8a through 8d show the
vertical velocity error during and
shortly after the 30,000-ft descent.
Note that the worst-case peak value
is almost 3 fps. This is for the
60-fps case wherein the bar0 error
rate at sea level is indeed 3 fps (5
percent of 60 fps). Figure 9 is a
plot of the peak vertical velocity
errors taken from Figure 8 as a
function of the descent rate. We
see from this graph that a 60-fps
rate of descent represents the
worst-case situation for this
vertical channel mechanization.
Because the real culprit here is the
bar0 scale factor error and the
descent times are of the same order
of magnitude as the 560-second time
constant of the basic instability of
the vertical inertial channel, there
is not much more that we can do to
the baro-inertial loop itself to
reduce these errors. One way of
improving the vertical velocity
performance is to eliminate the bar0
scale factor error in the air data
computer. This can be accomplished
through use of a Blanchard algorithm
(Ref.15) which uses air temperature
as well as static pressure to
determine air density and then
integrates the reciprocal of air
density (scaled by gravity) with
respect to pressure to calculate
altitude.
Figure 8. The answer is no, because

the scale factor error is largest at
sea level, diminishes to zero at
26,000 ft. and reverses sign.above
that altitude. Hence, in slow climb
starting from sea level, the
parameter DO is smallest (30 fps)
during the greatest bar0 error rate,
and the baro-inertial loop can do a
better job of combating the error.
In the case of descent, the aircraft
encounters the highest bar0 error
rate at the end of the long, slow
descent and DO has built up to the
same value as the descent rate
itself, making K = 0.5. Hence, the
vertical velocity error after a 60fps descent from 30,000 ft to sea
level is greater than it is after a
60-fps ascent from sea level to
30,000 ft. Figure 10 verifies this
fact by showing the vertical
velocity error during and after a
60-fps ascent from sea level to
30,000 ft. The peak vertical
velocity in this case is -1.9 fps as
compared with the 2.9-fps value
observed in the descent case
(Figure 8d).
3.
Kalman Filter Baro-inertial

LOOQ
The increasing availability of G P S

for altitude updating makes it
attractive to consider a Kalman
filter mechanization of the vertical
channel. The mechanization
presented here is the same as that
published in Reference 16.
Reference 13 provides an approach to
designing a vertical channel Kalman
filter. It uses 5 states: inertial
altitude error, vertical velocity
error, accelerometer bias error,
baro-scale factor error and pressure
altitude error. In this filter,
baro-scale factor is modeled as a
Markov process with a 10,000-second
time constant, which allows for some
slow variation with time.
Reference 10 indicates that air
density deviations from the standard
atmosphere tend to reverse sign at
about 26,000 ft. This suggests that
a better model of baro-scale factor
error would be:
(4)
One might also ask if slow climb

rates produce the same errors as
those seen for slow descents in
where k is the scale factor error at

true altitude h,,k, is the scale
321
factor error at sea level, and

h,, = 26,000 ft.
With this model, the error in

pressure altitude, hp is
where lh, is the error in pressure

altitude at sea level. We see from
Eq (5) that this model requires two
pressure altimeter error states: a
bias state, chb and a state we shall
call the sea level scale factor
error , k,.
In this next section we describe the
resulting Kalman filter
mechanization using Eq (5) as the
model for pressure altimeter error.
Table 1 summarizes the
mechanization.
4.
Analysis
State Equations.
As suggested in Ref 13, we will use
5 states for the vertical channel

error model. Three of these are the
same as in Ref 13: inertial
altitude, vertical velocity, and
accelerometer bias. However, we
will replace the pressure altitude
state with an altimeter bias state,
and instead of a scale factor error
independent of altitude, we will use
Eq (4). That state variable will be
k, the sea level value of the scale
factor error
The state equations are:
hii = cvz
r
=
Fa, + 2(g/a,)hi
[a; -10'~6az
(6)
k;, = -10" (0.5 + v,,/vo)k,
=
(0.5
Vh/vo) fhb
/Vz
where
a, = Earth's Astronomic Radius
in feet
g = Gravity in ft/sec2
vh = Magnitude of aircraft horizontal
velocity in fps
Vo = 500 fps
In Eq (6), the term 2(g/a,)hi is the
positive feedback term due to the
altitude-dependent portion of the
gravity compensation. Eq (6) models
the accelerometer bias error as a

Markov process with a time constant
of 1000 seconds. The -lo4 ( 0 . 5 +
V,/V,)k, term is similar to the Markov
process with a 104-second time
constant used in Ref 3 , but modified
here to be a function of horizontal
velocity. This effectively makes
the change in k, somewhat spatially
dependent as well as time dependent.
The state equation for
h, is
similar and uses the same
correlation time as that for the ko
state.
Equation ( 6 ) leads us to the
dynamics matrix (F-Matrix) given in
Table 1.
Covariance Matrix.
Table 1 lists the initial values for
the covariance matrix. Note that
the off-diagonal element which
represents the correlation between
pressure altitude and inertial
altitude is initialized with a unity
correlation coefficient. This
presumes that inertial altitude is
initially set equal to the pressure
altitude, so that their errors are
equal.
Table 1 enumerates the Q-Matrix
elements which augment the
covariance diagonals at each update
cycle. Small amounts of "plant
noise" or Q are added to the
inertial altitude and velocity
states to model computer resolution.
Additionally, an amount of 0.125 x
( A Vz2) &tk is added to the
inertial velocity error state to
model vertical accelerometer scale
factor error, where V, is the change
in vertical velocity from one Kalman
filter update to the next, and t, is
the time between updates.
To model accelerometer bias shifts,
g 2 errors, and gravity anomalies, we
set Q~~ equal to (~+V,/V,)XIO-~~~~/S~C~.
The constant part of Qa models the
accelerometer errors while the Vh/V,x
ft2/secsportion models gravity
anomalies which may change as the
aircraft moves over the earth.
For bar0 scale factor error plant
noise, we use a value Q- equal to
0.2 x
+ (Vx/V,)' x lo'' sec-'. The
first term is comparable to the
322
value of 0.18 x 10%ec" used in

Ref 13 and the second term allows
for modeling errors in case the real
scale factor deviations from the
standard atmosphere do not conform
exactly to our model with its linear
dependence with altitude.
For pressure altitude error variance
plant noise, QSS, Ref 13 recommends 4
x los6 ft2 per each 10-second update
cycle. This seems unrealistically
small. The same reference states
that the isobaric surfaces typically
are sloped relative to the
horizontal plane with a standard
deviation of 1 ft per nautical mile.
At 500 fps, an aircraft would travel
almost a mile in 10 seconds, and one
should expect to add approximately
(1 ft)' to C,, each 10-second period.
The actual value for Qss listed in
Table 1 is (1 + 0.5 Vh /Vo )ft'/sec.
The 1 ft2/sec portion of this term
accounts for a time variation in
pressure at a fixed place, while the
0.5 V,,/Voportion accounts for the
spatial variation in pressure as the
aircraft moves to another place.
Measurement Equations.
chi
(7)
Similarly, the pressure altitude is:

h, = h, + h, (1-h,/2ho)k,
+ ch,,
(8)
Differencing h, and h, to form the

error signal, we have
E = hi
The R-matrix listed in Table 1

consists of two parts. The first
part, (0.12)' (0.6 hz x lo-' + 27.8)',
comes from Ref (13) and represents
0.12 millibars of pressure noise
converted to altitude noise by the
conversion factor (0.6 hz x lo-' +
27.8). Actually, Ref (13)
recommended using 0.12 millibars of
noise for vertical velocity
magnitude less than 25 ft/sec and
0.32 millibars of noise for
vertical velocity magnitude in
excess of 25 ft/sec. The second
part of the R-matrix, (0.3 V,,)',
adds additional measurement noise
during high rates of climb or dive.
The 0.3 V, value models a 0.3-sec
altimeter time lag. At IVz I= 25
fps, the amount of equivalent
pressure noise added by this term is
0.23 millibar. When added to the
0.12 millibar figure for the first
term, the total is 0.35 millibar,
which is close to the Ref 13
recommended value of 0.32 millibar
for vertical speeds in excess of 25
fps.
Validity Test.
Table 1 shows the measurement matrix

(H-Matrix) for altimeter updating.
Its derivation is as follows: If h,
represents the true altitude, then
the inertial altitude is:
hi = h,+
Measurement Noise.
h,=~hi-h,(1-h,/2ho)ko-~hb
(9)
Because we do not know the true

altitude in flight, we approximate
it with the inertial altitude hi to
form
E = ghi - hi (1-hi/2ho)ko- $h,
(10)
The H-matrix given in Table 1

follows directly from Eq (10) by
inspection.
One problem that arises in baroinertial loops is the possibility of

large (-500 ft) step changes in the
barometric altimeter input. These
changes can occur near Mach 1 as
shock waves sweep across the static
pressure ports of the air data
system. Conventional vertical loops
(Ref 9) handle this situation by
drastically reducing the feedback
gains and by computing a baro-bias
compensation. In effect, when the
baro-inertial error signal is
excessively large, one assumes that
the inertial signal is correct and
that the problem is in the
barometric signal.
We can use the same philosophy to
devise a strategy in case of failure
of the validity test in our Kalman
filter vertical channel. For
example, consider the following
strategy: If E' > HCHT + R then set
HCHT + R = E2 and after the update,
add E' - R to C,, (except during
absolute altitude updates on the
ground).
The reasoning behind the above
approach is as follows: Because the
323
reason for failing the validity test

is presumed to be a sudden change in
the altimeter bias, we acknowledge
this fact by increasing the
covariance of the altimeter bias
state (after the update) by the
amount of the observed error squared
less the measurement noise. During
the update itself, we reduce the
feedback gains b setting the
denominator (HCH + R) in the gain
equation equal to E2, the square of
the observed error. This is
equivalent to adding additional
measurement noise, R I to account for
the observed error difference.
5.
b.
Sea level bar0 scale factor

error of 5 percent
C.
Initial vertical velocity

error of 1 fps
d.
Accelerometer bias error of

0.001 ft/sec2
e.
Accelerometer scale factor

error of 400 ppm
f.
Baro-altimeter resolution of
2 . 5 ft
9. Baro-altimeter time lag of

0.3 sec
Simulations
Baro-scale factor
nonlinearity with altitude
per Figure 13.
h.
Description of Simulation Proqram.
Figure 11 shows an overall block
diagram of the simulation program
used to simulate the Kalman filter
vertical channel. A desktop
computer was used to perform the
simulations. The Kalman filter
update interval was one update every
8 seconds, although some runs were
made with 4-second and 2-second
update intervals. Figure 12 shows
little difference in vertical
velocity accuracy as a function of
update rate, and for that reason we
chose the slowest of the three rates
for the remaining simulations.
To initialize the filter, we had it
perform absolute altitude updates
for the first 98 sec and then switch
to barometric altitude updates. The
absolute altitude updates employ the
same Kalman filter except that the
H-Matrix, observation error signal,
and measurement noise are changed
to:
If we simulated the real-time

barometric scale factor as a linear
function of altitude, the truth
model would match the Kalman model
exactly, and we might achieve
unrealistically good results. To
avoid over-optimism, we introduced a
modeling error by assuming a
nonlinear variation in barometric
scale factor error with altitude.
The simulated nonlinear truth model
for pressure altitude is
h, = (k,t + .04 (3/h0)(3 + dh,
where
= h, (1
h,/2h0)
and k, is true barometric scale

factor error at sea level.
With this formulation for the bar0
error model, the bar0 scale factor
error variation with altitude is
= (k,t + .8@/h,)
H = [ 1 0 0 0 0 ]
E = hi - h,
R = (10 ft)2
This procedure simulates turning the
system on when the aircraft is on
the ground at a known elevation
above sea level
Truth Model.
The simulation truth models included
the following:
a.
An initial baro-bias error

of 500 ft
(14)
dht
,o
+f
and I(? = 1 -h,/h,

Combining the above equations, we
have ,
clhf
._L
=(KO t +.8(3/h,)
(16)
4 ht
A plot of Eq (16) vs altitude
appears in Figure 13.
Simulation Results.
The above error model was subjected
to several different vertical
maneuvers and barometric anomalies.
324
These were: (1) a series of fast

climb and dive maneuvers, (2) a fast
climb followed by a slow descent,
and (3) a series of sudden steps in
baro-bias. A discussion of these
simulations and their results
follows*
Fast Climbs and Dives.
In this simulation, after ground
alignment, the aircraft takes off
and pulls up at 150 ft/sec2 (4.79) to
a 750 fps rate of climb. At t = 140
seconds it pulls down at 150 ft/qec2
(4.7g) to level flight at 30,000 ft
altitude. At t = 200 seconds it
pulls down in 5 seconds to a 750 fps
dive. At t = 240 seconds it begins
a 4.7g pullout to level flight at
zero altitude. This climb and dive
maneuver is repeated every 200
seconds as shown in Figure 14a.
Figure 14b shows both the vertical
velocity error and the covariance
estimate of the vertical velocity
error. The dashed, symmetric curves
are the plus and minus values of the
square root of the vertical velocity
covariance, C
,
.
The solid third
curve is the vertical velocity error
itself, which should and does stay
between the 2 1-sigma error values
calculated by the filter.
The inertial altitude error appears
in Figure 14c along with the plus
and minus values of the filter
estimates of the RMS altitude
accuracy. The initial absolute
altitude updates bring the altitude
error down to about 10 ft. After
700 seconds of climbing and diving
maneuvers, the altitude error
gradually increases to about 50 ft
and remains within the 21-sigma
values calculated by the Kalman
filter.
Figure 14d is a plot of the sea
level bar0 scale factor error, i.e.
the difference between S,, the value
estimated by the Kalman filter, and
the true value, 0.05 in this
example. Also shown (dashed curves)
is the 1-sigma range of the error in
S, as determined by the square root
,
.
After
of the covariance element C
the first climb to altitude, the
filter correctly estimates bar0
scale factor to about one-percent
accuracy, and the error is in
general agreement with the filter
covariance estimates. In the

subsequent maneuvers the error
increases somewhat to about 2.5
percent after 700 seconds of
climbing and diving.
At times, especially during the
level flight legs at sea level, the
error in the filter estimate of k,
exceeds its one-sigma value as
calculated from SQRT (Cu), and the
whole curve is biased to the
positive side of zero. The
discrepancy is due to the
mismodeling of the bar0 scale factor
as shown in Figure 13. The filter
attempts to fit a straight line
through the actual nonlinear
function of altitude. As readily
seen from Figure 13, a best
straight-line fit through the truth
model will intercept the zero
altitude axis at a value somewhat
higher than 0.05. This causes the
bias offset observed in Figure 14d.
In Figure 14e we see the error in
the filter's estimate of bar0 bias
in comparison with its expected RMS
value as calculated by the filter.
The spike-like increases in SQRT
(Css)result from failing the
validity test during some of the
high-g pullup or pulldown maneuvers.
These increases help to maintain the
+1-sigma range outside of the actual
error. This imparts an adaptive
behavior to the filter, which helps
prevent divergence.
Finally, in Figure 14f we see that
the filter reduces accelerometer
bias error from about 100 ug to
about 50 ug, and the actual error
stays comfortably within the 51sigma limits +SQRT (C&)
calculated
by the filter.
Fast Climb and Slow Descent.
This simulation starts out the same
as the last one, but after reaching
30,000 ft, the aircraft then
descends slowly at 60 fps back down
to sea level (Figure 15a). This
slow descent represents the worst
case for conventional baro-inertial
loops (Ref 9) and also for the
Kalman filter vertical channel of
Ref 13. In these mechanizations,
the slow descent case created peak
325
velocity errors of 2.9 fps (Ref 9)

and 1.8 fps (Ref 13). These errors
peaked out near the end of the
descent.
In contrast, the mechanization
proposed herein limits the peak
vertical velocity error to about 0.5
fps, and this occurs early in the
descent (see Figure 15b). The
reason for this exceptional
performance is that the Kalman
filter estimates the bar0 scale
factor error during the climbout and
can then compensate for the bar0
scale factor error during the
descent.
Figure 15c shows the inertial
altitude error. It also peaks out
early in the descent phase at about
80 ft and thereafter diminishes as
the filter compensates for the bar0
scale factor error. In Ref 13, a
similar gradual descent generated
ten times as much altitude error.
Figure 15d illustrates the ability
of the filter to estimate bar0 scale
factor error. Again, it shows both
the actual error as well as the
-+1-sigma values for the error as
calculated by the filter. This
figure shows that the filter
estimates sea level bar0 bias error
to about one-percent accuracy, and
t,he filter l-sigma estimates are in
agreement with the actual error
itself. Once again however, the
actual error is biased somewhat to
the positive side because of the
bar0 scale factor modeling error
(Figure 13).
Figure 15e shows the bar0 bias error
relative to the true value of 500 ft
and compares it with the filter
estimated value of the expected RMS
error, +SQRT(CSs). The bar0 bias
error in this example stays within
the +SQRT(CS5) range throughout the
maneuver. The spike in SQRT(C,,) at
T=140 seconds is caused by failing
the validity test during the high-g
pulldown to level flight at
30,000 ft.
In Figure 15e we see the ability of
the filter to estimate accelerometer
bias. It shows a gradually
improving estimate as judged by the
narrowing range between the 21-sigma
values. The accelerometer bias
error itself remains between the
filter 21-sigma values and

diminishes from almost 100 ug at the
start of alignment to about 30ug at
the end of the fast climb and slow
descent maneuver.
Steps in Baro-Bias.
In this simulation, we start off
with zero baro-bias, but at t=100
sec we suddenly change it to 500 ft.
Then at t=200 sec we reset the barobias to zero. We keep repeating
this pattern every 200 sec as
indicated in Figure 16a.
Figure 16b shows the resulting
vertical velocity error and compares
it with the filter estimated
-+1-sigma values of the velocity
error. The actual velocity error is
less than 0.2 fps throughout this
simulation (after alignment) and
stays within the filter estimated
l-sigma range.
The inertial altitude error caused
by the bar0 bias steps appears in
Figure 16c along with the filter
estimated 51- sigma values for the
inertial altitude error. Note that
the altitude error gradually
increases to about 30 ft after the
eighth bias step. This error is
consistent with the +40 ft l-sigma
range calculated by the filter.
Filter 16d is a plot of the plus and
minus RMS values for the bar0 bias
error as computed by the filter.
The actual bar0 bias error itself
appears in Figure 15e. We separated
these two graphs because it would be
difficult to distinguish one set of
lines from the other if we had
plotted them together as in the
previous figures. Note that the
500-ft error introduced at each bias
step is almost completely corrected
at the very next Kalman filter
update. The 500-ft stepup in the
bar0 altitude RMS magnitude at the
first update following each bias
step is the result of failing the
validity test. This stepup in the
filter l-sigma value persists only
until the next update, at which time
it returns to nearly the same low
value it had before the step
occurred. This shows that the
filter is able to follow and correct
for the sudden jumps in baro-bias.
The error in sea level bar0 scale
326
factor error, ko, appears in Figure

16f along with the filter RMS
estimates, 5SQRT (C4). Because the
actual altitude remains at sea level
throughout this simulation, the
filter cannot estimate ko. However,
because the filter computes H, from
the inertial altitude instead of
true altitude, and the inertial
altitude deviates slightly from zero
(Figure 16c), the filter computes a
nonzero H, and thinks it cal update
ko. Hence, the +SQRT ( ~ 2 ~curves
~)
tend to converge slightly instead of
staying flat at their initial values
of 50.05.
The last figure in this set, Figure
16g, shows the accelerometer bias
error plotted along with its filter
computed 1-sigma values, +SQRT (C33).
The bar0 bias steps have no
appreciable effect, and the filter
is able to update the accelerometer
bias to within 25ug of its correct
value. At the same time the
filter's RMS estimate of the bar0
bias accuracy is consistent with the
actual error.
Conclusion.
6.
The Kalman filter vertical channel

design summarized in Table 1
exhibits much superior performance
in comparison with either the LN93/94 conventional baro-inertial
loop (Ref 9) or the Kalman filter
design proposed in Ref 13. It
obtains this superior performance by
modeling the bar0 scale factor error
as a linear function of altitude, a
relationship which the real
atmosphere tends to follow.
7.
Nomenclature.
AB 2
term
AFB
term
a or a,
a,
C
DVZ
DHBO
Acceleration feedback
Altitude rate feedback
Earth's radius
Vertical accelerometer
bias error
Covariance matrix
Vertical Velocity
increments from
vertical acceleromleter
Difference between
filtered bar0 and
filtered inertial
altitudes
Error signal, hi-hp
E
F
Dynamics matrix
Gravity
9
Gravity at sea level on
go
the equator
H
Inertial altitude in
conventional filter.
Observation matrix in
Kalman filter.
Hdot or l!l
Time rate of change of
H
HB
Filtered bar0 altitude
Hbdot or Hb Time rate of change of
HB
HBR or h,
Raw bar0 altitude input
Inertial altitude in
hi
Kalman filter
Filtered inertial
HL
altitude
Isopycnic level, 26,000
h0
ft
Pressure
altitude ft
hP
True altitude ft
h,
k
Bar0 scale factor error
Value of k at sea level
k0
K
Gain changing parameter
=1 - KB5
K1
2 go
0.005302
K2
KB2
Velocity feedback gain
KB3
Acceleration feedback
gain
KB4
Integral acceleration
feedback gain
KB5
Gain change parameter
computed as
=v,/ (Vz2+DOZ)
1 sec"
KF
Plant noise matrix
Q
R
Measurement noise
S
Laplace operator. S
signifies
differentiation;
1/S signifies
integration.
Represents state
vector in Kalman
filter.
s2
Feedback error signal
t
Time
Time between Kalman
updates
Nominal loop time
T"
constant =20 sec during
alignment and 100 sec
thereafter
Magnitude of aircraft
horizontal velocity
500 fps
Inertial vertical
321
vz
velocity
Change in vertical
velocity between Kalman
1-3
Meeting, Phoenix, AZ,

January 22-24 1991, pp 303-308
10. U.S. Standard Atmosphere, 1976,

NOAA-S/T 76-1562, October 1976
1. K.L. Perkins, R.R. Palmer and

J.S. Ausman, Vertical Velocity
Measuring System, U.S. Patent
3,005,348, October 24, 1961
11. R.L. Blanchard, An Improvement

to an Algorithm for Computing
Aircraft Reference Altitude, IEEE
Transactions on Aerospace and
Electronic Systems, September 1972,
pp 685-687
2. M. Kayton and W.R. Fried,

Avionics Navigation Systems, John
Wiley & Sons, New York, 1969,
pp 317-319
12. S.L. Valley, Scientific Ed.,

Handbook of Geophysics and Space
Environments, New York; McGraw-Hill,
April 1965
3. W.S. Widnall and P.K Sinha,

Optimizing the Gains of the BaroInertial Vertical Channel, AIAA
Guidance and Control Conference,
August 1978, Palo Alto, CA
13. P.M. Barham and P. Manville,

Application of Kalman Filtering to a
BarolInertial Height Systems, Royal
Aircraft Establishment, Technical
Report 69131, June 1969.
4. J.S. Ausman, et al., Close Air

Support System (CLASS) F-4D Flight
Test,, Vol I Summary, Air Force
Avionics Laboratory, TR-73-363,
WPAFB, Ohio, September 1973,
pp 96-102
14. M.J.A. Ashbury, D.A. Forrester,

C.S. Dixon and R. Johannessen,
Probability Distributions that are
Important when Assessing Barometric
Aiding to GPS, Proceedings of
Institute of Navigation GPS-90
Meeting, Colorado Springs, CO,
September 19-21, 1990, pp 445-453
8. References.
5. J.S. Ausman and F.J. Hellings,

Development of a System for Scoring
Simulated Bombing Runs, NATO AGARDCP-198, October 1975, pp 25-1 to
25-19
6. C.H. Leatherbury, J.S. Ausman
and H.E. Thompson, Development and
Test of an Airborne Range
Instrumentation System (ARIS),
Proceedings of the 23rd
International Instrumentation
Symposium, May 1977, pp 269-278
7. M. Weintraub, LW-33 Advanced

Inertial Nav/Attack System for
Improving the Effectiveness of
Combat Aircraft, Section 5
International and foreign
Development, 12th Joint Services
Data Exchange for Inertial Systems,
October 1978, Norfolk, VA
8. LN-39 Standard Inertial
Navigation system AN/ASN-141
Technical Description, Litton
Document No. 155568, September 1987
9. J.S. Ausman, Baro-Inertial Loop
for the Standard RLG INU,
Proceedings of the Institute of
Navigation, National Technical
15. R.L. Blanchard, A New

Alogorithm for Computing Inertial
Altitude and Vertical Velocity, IEEE
Transactions on Aerospace and
Electronic Systems, Vol AES-7, No.
6, November 1971, pp 1143-1146
16. J.S. Ausman, A Kalman Filter
Mechanization for the Baro-Inertial
Vertical Channel, Proceedings of the
Institute of Navigation FortySeventh Annual Meeting,
Williamsburg, VA, June 10-12, 1991,
pp 153-159.
329
SECTION V
REPRESENTATIVE VEHICLE IMPLEMENTATIONS
INTRODUCTION
DR. JOHN NIEMELA
This section describes the

navigation suites of a diverse
though representative spectrum
of aerospace vehicles - fixed
and rotary wing aircraft as
well as spacecraft. Each
aerospace vehicle is
configured to meet dissimilar
performance requirements,
environment constraints and
mission profiles.
However, there is a common

methodology in the requirement
analysis, sensor selection and
system synthesis used in the
development of the navigation
system for each aerospace
vehicle type. It is the
intent of this section to
describe generic classes of
navigation systems with
emphasis on the technical and
operation factors that
influence their design.
330
Aerospace Navigation Systems Requirements for Fixed Wing Aircraft

bY
Carlos A. Bedoya
McDonnell Douglas Aerospace - East
St. Louis, MO 63034
USA
Introduction
The purpose of this paper is fourfold, (1) to
provide the reader with a status report on the
state-of-the-art of navigation system technology as applied to fixed wing aircraft, (2)
recommend a disciplined systems engineering process to be used in determining navigation system requirements as they relate to
navigation, sensor cueing and targeting as
well as weapon requirements (3) examples
of how to functionally decompose each aircraft and analyze its mission requirements
into the navigation system requirements and
(4) A typical set of requirements for a fixed
wing aircraft. The results will allow the
navigation systems designer to have a
process he can follow in determining or verifying the requirements of a particular application as well as a baseline set of
requirements..
1.0 Navimtion Svstems Current Status

Aerospace navigation systems have undergone revolutionary changes in the past fifteen years (References 1, 3, 5, 10, 12, 13,
14, 16, 18, 19. Major developments have
occurred in the inertial navigation Systems
(INS) and satellite based navigation systems
with the advent of the ring laser gyroscope
and the development of the Global Positioning System (GPS) as well as the explosion in
the field of microprocessors and integrated
circuits. Electromechanical gimbaled inertial navigation systems using mechanical gyroscopes have been replaced by Strapdown
systems using ring laser gyros and powerful
microprocessors providing equivalent accuracy and twenty fold increases in reliability.
GPS has begun to provide world wide satellite based navigation to military and civilian
users with unprecedented accuracy in position, velocity and time. It has the potential
to replace all current radio navigation aids
by the turn of the century. In addition to in-
creases in performance, these developments

have significantly reduced the weight and
the cost as well as increased the reliability
and maintainability of state-of-the-art
aerospace navigation systems (Fig 1.0-1).
The increased demands in performance, reliability, maintainability and decreased weight
and cost of fixed wing aircraft have also
brought significant changes to the architecture of avionics and flight control. Federated, independent, dedicated systems have
evolved to more and more integrated systems with common interfaces such as standard multiplexed buses, multipurpose displays and controls, integrated functions and
integrated controls. Standard navigation
systems such as the United States Air Force
Standard Navigation Unit (SNU-84- 1) the
Fighter Navigation Unit (FNU-85), the
United States Navy Carrier Align Inertial
Navigation System (CAINS 11), and the
multiservice Standard GPS Receivers, the
3A and the Miniature Advanced GPS Receiver (MAGR). The miniaturization of
GPS receivers to the size of one electronic
circuit card such as the Rockwell Collins
GEM 111, has allowed the incorporation of
the GPS in the same electronic enclosure or
box as the INS allowing easy retrofit of GPS
into existing aircraft, designs such as the
Embedded GPS Inertial (EGI) and the GPS
Integrated Navigation Assembly (GINA).
This design is finding more and more advocates as it allows a low cost method of incorporating the upgrade to include the GPS
function by reducing the fielded aircraft
structural and wiring modifications while
also providing a more tightly coupled
INS/GPS for better performance (Reference
17). The new GPS receiver shares the
power supply and existing wiring of the INS
while only needing the addition of an antenna wire to the INS box.
33 1
Parameter
SuppllerA
Supplier B
0.005
5
0.003
15.24
0.003
5
0.0015
18.4
16
45
Dithered
0.8
2.5
None
7x7~11
$75k
4000+
19
22
Bias
Scale Factor
Random Walk
Path Length
System Wieght
System Power
Type
Nav Performance
Velocity
Cooling Air
(Ib)
(w)
(nmhr)
(Wsecrms
Slze
(W
cost
(1990 Dollars)
MTBF
Hours
'Alr Data Capability
Radar Altimeter Capability
Common Module Maintainabllity
Yes
Yes
Yes
Yes
Multloscillator
0.8
2.5
None
7x7~10
$75k
4000+
Yes
Yes
Yes
Yes
Figure 1.0-1. Current State-of-the-Art Inertial Navigation Systems
Referenced
Navigation
High .Resolution Radar

(Electronically Scanned Array)
Figure 1.0-2. Integrated Navigation /Inertial Network Concept
332
CORE PROCESSOR #2
INS #l
40 Hz
CbsedLoopKFUl
GC. CV, IFA, NAV NO&
Portomuna0.8 nnJhr
Positin. Velocily
Alltude. AlIitude.
Preprocess External Sensore

Measurement Noise and Residual6
i
Open Loop Kalman Fitter U1
Corrected Nav Data
N w l g a t b n Aldlng
C o n s c t b n A)gor(thms
Performance
Need separate Open Loop Kalman Filters for each INS to Properly Model
Correlations Built up during Independent Alignment Functions
Figure 1.0-3 Dual INS Performance is Improved Over Single System
The current generation of military aircraft
now under development will be using common modules as building blocks of even
more integrated and interdependent navigation system architectures as shown in Figure
1.0-2. The design and development of fixed
wing aerospace navigation systems is based
on operational requirements of the vehicle,
mission scenarios, sensor suites, weapons
suites, reliability and maintainability requirements. Because of the high degree of
integration of current and future systems, it
is imperative that this design and development follow a rigorous analytic process in
order that the establishment of the required
accuracy, signal characteristics, interface requirements, noise, bandwidth, reliability,
physical characteristics and cost of the system be well defined.
The system engineering process allows the
designer to perform trade studies in order to
optimize system design (References 2 , 5 and
6). For example, it may be more cost effective to achieve higher system performance
by using lower cost inertial navigation systems aided by aircraft sensors performing
velocity and position updates. Another ex-
ample could be a configuration using averaged dual medium accuracy INS'S to achieve
high accuracy and higher mission reliability
instead of single high accuracy fault tolerant
INS as illustrated in Figure 1.0-3. A dual
INS installation would also allow placement
of one system at the Radar antenna for better
motion compensation and the other system
at a structurally benign position for flight
controVautopilot reference or located at another sensor. Section 2.0 will describe a
recommended approach for this process.
1.0.1 Definitions
From the previous discussion it can be seen
that significant changes have taken place in
the field of aerospace navigation systems.
This paper uses the definition of navigation
as defined in Reference 4, "the process of
directing the movements of a craft from one
point to another", and will deal primarily
with fixed wing military aircraft such as
fighters, attack, strategic and transports. It
will deal with three main missions, the tactical, the strategic and the transport. At this
point it is necessary to define some new
concepts and terms used in today's aerospace
vehicles. As shown in Figure 1.0.1-1
333
(Reference 3,aerospace vehicle functions

can be divided into three main systems, (1)
the Pilot Vehicle Interface (PVI), (2) the
Mission Management System (MMS) and
(3) the Vehicle Management System (VMS).
The first deals with the aircrewhachine
interfaces, the second consists of all the
mission critical systems, while the third is
made up of all fight critical systems. Figure
1.0.1-1 also indicates that there exists some
overlapping functions between the mission
management and the vehicle management
functions such as the trajectory control of
the vehicle when performing functions such
as terrain following and automatic landings
or other integrated control functions.
loss of the capability of the vehicle to

successfully complete its designated mission.
Aerospace navigation system functions are
critical to both these systems and must address dual requirements in some cases which
in turn will affect system reliability, redundancy, survivability and cost. Current navigation system functions in fixed wing vehicles must provide much more than the classical outputs of position, velocity and time.
Vehicle attitude, attitude rate, acceleration,
acceleration rate in body and geodesic coordinates with time correlation for all outputs
are also required. In addition to geographic
Pilot Vehicle Interface
0
Mission
Functions
0
Vehicle
Functions
............................
WIIU..
COntml
Conbol Interlace
Figure 1.0.1-1. Aerospace Vehicle Functions

In this paper we will define a VMS as an
Integrated System that manages flight and
vehicle critical systems and the interfaces
with those systems that support the flight
and vehicle critical systems. Flight critical
systems being defined as those elements
whose failures cause loss of flight path
control. Vehicle critical are those elements
whose failure or degraded performance
cause loss of the vehicle (Reference 2 and
5). Mission critical systems are defined as
those elements whose failures will cause the
navigation, relative navigation to a fixed

point or one or more aircraft is also needed.
With the above major functional definitions
in place, it is now appropriate to examine the
purpose of aerospace navigation systems in
fixed wing aircraft.
334
systm
Englrwerlng Process
Air Vehicle Level
. .. ..... ...
Figure 2.0-1. Navigation System Development Process
2.0 The S v s t e m n e e f i n g Proass

In order to determine the requirements of
modem fixed wing aircraft navigation systems we need a disciplined process. An
disciplined engineering development process
would have the following characteristics:
7. Top down, requirements driven.

Disciplined and repeatable.
Incorporate concurrent engineering
principles.
4. Incorporate effectiveness and balance
trade-off analyses.
5. Allow for design iteration and refinement.
2.
3.
Figure 2.0-1 provides an overview of the

recommended navigation system development processhethodology which incorporates the needed characteristics (Reference
5). The process is a top down approach
wherein air vehicle level requirements are
flowed down to the navigation system level.
System engineering (Reference 6) is a management process which controls total system
development. Its objective is to achieve the
optimum balance of all system elements.
MIL-STD-499A defines Systems Engineering as, "the application of scientific and engineering efforts to (a) transform an operational need into a description of system performance parameters and a system configuration, through the use of an iterative process of definition, synthesis, analysis, design, test, and evaluation; (b) integrate related technical parameters and ensure compatibility of all physical, functional, and
program interfaces in a manner that optimizes the total system definition and design;
and (c) integrate reliability, maintainability,
safety, survivability, human engineering,
and other such factors into the total engineering effort to meet cost, schedule, supportability, and technical performance objectives".
The requirement to use systems engineering
is invoked by DoD DIR 5000.1 Part I, Paragraph C.1.b dated 23 February 1991. It
specifies that, "Program plans must provide
for a systems engineering approach to the
simultaneous design of the product and its
manufacturing, test, and support processes.
As shown in Figure 2.0-2, systems engineering comprises five major activities, some of
which have been decomposed into smaller
tasks. The output of the system engineering
process is documentation including specifications, plans, functional flow block diagrams, and trade study reports. The
approach is to apply the process of Figure
2.0-2 at each level as outline in Figure 2.0-1.
The first step at each level is to determine
the customer needs as shown in step 1, convert those to mission requirements of step 2,
perform a functional analysis as shown in
step 3 and then allocate the requirements as
shown in step 4. Steps 2.3, and 4 constitute
the system functional analysis which is then
used in step 5 to perform the design synthesis and system integration. The next step is
to define the system for specifications, interfaces, development plans and logistics as
shown in step 6 which results in a description of the system elements. If the design is
still not complete, step 7 is performed in order to optimize the design by performing
trade studies and effectiveness analysis, as
well as other related analyses. This final
step is the evaluation and decision step.
This step may result in further iterations
through steps 1 to 5 until the decision path in
6 is that the design is complete. Section 3.0
will perform some examples of Step 1, input
requirements, step 2, mission requirements
analysis and step 3, functional analysis and
Step 4,requirements allocations in order to
give the reader a methodology to follow and
some results that may be useful in his applications. Section 3.2 will present a table of
typical requirements allocations for a modem fixed wing tactical aircraft that can be
used as a basline as it would be difficult to
define all possible requirements without performing an analysis for all aerospace vehicles. To define all possible requirements
could also result in overspecifing the system
and increased cost, weight and maintenance.
336
ixed W-rcraft
(S~QS
l-tLmuu
This section will examine a top level functional decomposition of possible areas of
coverage for navigation systems in fixed
wing aircraft in order to establish major
baseline functions that may be required
which in turn can be decomposed into sub
functions. The purpose of this section will
be to define the possible functions of the
navigation system which will then be later
used to relate them to aircraft requirements,
vehicle functions, mission functions, mission tasks, sensor suites, weapons suites,
survivability, reliability and maintainability.
Figure 3.0-1 shows the first step in the decomposition of aircraft requirements which
simply shows how mission and aircraft requirements can be mapped into the three
major system requirements.
The navigation system in modern aerospace

vehicles serves for more than directing the
vehicle from point to point. As described in
the previous section, the navigation system
has functions dedicated to support traditional
navigation position and velocity but is also
required to provide aircraft attitude for the
flight instruments used by the pilot, aircraft
references for flight control and autopilot
functions, references for stabilizing and aiding aircraft sensors, parameters for weapon
delivery, references for filtering and improving the air data parameters, references for
integrated control functions such as trajectory control and f i e control.
The increased functional requirements have
in great part resulted from increased operational requirements of the vehicle. Improvements in aircraft performance, weight
Vehicle
Figure 3.0-1. Aircrafl Functional Decomposition
337
and cost have been achieved by reducing the

size of aerodynamic surfaces which in tum
reduce aircraft stability. Stability is then recovered by the use of more accurate inertial
data by the flight control system. Fuel and
time savings can be achieved by optimal
path control during climb out and cruise
mechanized by coupling the flight control
system, the propulsion system, navigation
and inertial data. Weapon systems have also
been improved by the integration of the
flight control, the fire control and the navigation system where the f i e control system
is coupled in order to aid the pilot in air-toair combat or air-to-ground weapon delivery.
..
3.0.1 Militarv Fixed Wine Aircraft
Missions
Due to the previously stated operating

modes, it has become more complex to def i e the navigation system that meets its
operational requirements. In order to define
all functions and ensure that they map into
navigation system requirements and then
that these requirements in turn are mapped
into a correct technical specification, a well
defined process must be used. For military
aircraft this process is very complex as these
aircraft have wide ranging aircraft and
divided into multiple aircraft. These

multirole aircraft must be able to perform
their mission while based at full service facilities, austere bases and aircraft carriers.
Military fixed wing aircraft can be divided
into their tactical, strategic and transport
missions. As seen if Figure 3.0.1-1 the tactical mission can then be subdivided into airto-ground and air-to-air, these in turn, for
example air-to-ground can be further subdivided into (1) close air support, (2) battle
field air interdiction, (3) deep interdiction,
(4) reconnaissance. The air-to-air mission
can be subdivided into (1) offensive counter
air and (2) defensive counter air. The
strategic mission can be subdivided into (1)
hi I lo extremely long range interdiction and
(2) hi I hi extremely long range interdiction.
The transport aircraft mission can be divided
into two main divisions (1) rapid deployment of air and ground forces and (2) long
term supply of air and ground forces.
T
v
D
i
p
Peauirements
A tactical aircraft must be capable of rapid
deployment on both air-to-air and air-toground missions and offer affordability,
availability, survivability, and lethality..
Typical aircraft performance requirements
I
Tactical
Strategic
Air-to-Ground
Air-to-Air
Transport
Figure 3.0.1-1 Militury Aircraft Missions

mission requirements. Military aircraft
typically require that the aircraft flight
envelope range from low subsonic to high
supersonic (mach 2 or 3), from low altitude
(50 to 100 feet) to extremely high altitudes
(70,000 feet). Environmentally they must
operate from temperatures of -40 degrees to
70+ degrees C with reaction times of 30
seconds to five minutes. In addition, the
trend is to multimission I multirole aircraft
that must meet the requirements traditionally
are summarized in the paragraphs below:

Altitude and S R W ~ The equipment shall
operate with the accuracy state herein at altitudes form -1000 feet to 75,000 feet and
over a speed range of 0 to 300 feevsecond
horizontal and minus to plus 1500
feet/second vertical.
Aircraft c
'1' m. The equipment shall
be capable of operation in all aircraft attitudes.
338
Defensive Counter Air Mission
Air Interdiction Mission
combat
combat
(4) MK44 IAMI
(2) AIM*
20 mm (1,ooo Rouna)
0.9 M&h
(4) AIM-120IAMRAAM)
0.8 0.9 M&h
Close Air Support
Offensive Counter Air Mission

Canbat
(4) MK%2 GPBa

(2) A I M 4
20 m (1,mRound$)
Combat
0.8 0.9 Mach
0.9 Mseh
(4) AIM-120 (AMRAAM)
all
Figure 3.0.2-1 Tactical Aircraft Missions
Battlefield
Air
interdiction
1. We-mission Planning
2. Pre-flight and Pre-takeoff
3. Takeoff
4. Departure
5. Cruise (haven)
6. Let Down toTerrain Following
7. Ground Target Acquisition and Attack
8. PostAttack
9. Return to Base
10. Penetration / Approach
11. Land
12. Turnaround
Takeoff I
ingress
Air-to-Ground
Attack
Egress I
Landing
Service and
Maintenance
Figure 3.0.2-2Battlefield Air Interdiction Mission Phases and Segments
dvnamic resuonE. The equipment shall be capable of functioning and
maintaining alignment such that overall
system perf&n&ce will not be degraded by
aircraft maneuvers up to the following limits:
(a)
Aircraft maximum angular rate
Pitch Rate = 2.0 radiandsecond

Roll Rate = 5 radiandsecond
Yaw Rate = 3.5 radiandsecond
Aircraft maximum accelerations

Pitch Acceleration = 3.5 radiandsecond2
Roll Acceleration = 18 radiandsecond2
Yaw Acceleration = 3.5 radiandsecond2
(b)
339
Tactical Aircraft Mission Requirements

The tactical aircraft will be required to
perform multiple missions. The primary
missions are air-to-ground missions and airto-air. A pictorial representation of some
tactical mission profiles is shown in Figure
3.0..2-1.
To further define a typical mission, it is necessary to decompose the mission into detailed tasks. First, the mission is divided
into mission phases. Within each mission
phase there are mission segments, and
within each mission segment there are the
detailed tasks. The Air Interdiction mission
phases and their segments are shown in Figure 3.0.2-2 as an example of decomposition.
porating a low-observable phased-array antenna, an extremely accurate inertial navigation system, a strategic Doppler radar altimeter, and a star tracker. Its defensive
avionics are built around the AN/ALQ- 161
electronic counter measures system, with
extended frequency coverage, and includes a
tail warning radar and expendable decoys
such as chaff and flares.
It has three internal weapons bays with the
capacity to carry up to eight AGM-86B airlaunched cruise missiles (ALCMs) and
twenty-four short-range attack missiles
(SRAMs). In a conventional, non-nuclear
role, it can carry up to eighty-four 500
Figure 3.0.3-1 Representative Strategic Aircmft

3.0.3 Strategic Aircraft Customer Re-
auirements
The strategic aircraft selected as an example
is representative of the B-1B and B-2
bombers. The strategic aircraft is a long
range, multi-role strategic bomber and the
mission/sensor/weapons are considered to be
representative of advanced bombers. It is
capable of carrying, in three weapons bays,
varying combinations of nuclear air-toground missiles, conventional or nuclear
free-fall bombs, and auxiliary fuel. In addition it has electronic jamming equipment, infrared countermeasures, radar location and
warning systems, other advanced avionics
and low observable technology.
The strategic aircraft, shown in Figure 3.0.31, has a maximum gross weight of 477,000
pounds and is capable of supersonic flight at
speeds approaching Mach 2.0. Its primary
role will be high subsonic, low-altitude penetration to the target area. Its variable-geometry wing permits operation from shorter
runways in the upswept position and supersonic flight in the fully swept position. The
avionics systems include advanced forwardlooking and terrain following radars incor-
pound MK-82 or twenty-four 2,000 pound

MK-84 bombs. It also has the capability to
carry, on external stations, an additional
forty-four MK-82s or fourteen MK-84s.
Strategic Aircraft Mission Requirements
The typical strategic mission is normally an
Air Interdiction type of mission entailing
much longer ranges and greater weapon
capacity than the tactical aircraft missions.
In a wartime scenario, the tactical aircraft
would operate out of Forward Operating
Bases where the strategic aircraft would be
operating from Main Operating Bases much
farther removed from the area of conflict.
Strategic missions are longer duration, more
autonomous, and have different target assignments and associated weapon loads than
a tactical mission. Reference 3 provides
additional information on strategic missions.
The nuclear mission profile includes two
different profiles. One mission, carrying
ALCMs, involves a high altitude launch of
the ALCMs outside of enemy airspace and a
return to base. The other mission profile,
involving delivery of nuclear bombs, is executed at a low altitude and high subsonic airspeed during penetration into enemy
airspace. This mission. profile would re-
340
cruise missile is aligned to the reference INS

prior to launch. Alignment of the cruise
missile is allocated sufficient time for convergence because of the longer mission duration, as compared to tactical missions.
After launch, the cruise missile descends to
200 feet, flies over water for 500-1000 NM,
and ingress's to the target at 0.7 Mach which
is 2350 NM from launch point (a total of
about 5 hours flying time from launch). Approaching the target, a single three-fixed update is made to improve probability of kill.
semble the tactical air interdiction profile in

Figure 2.2.2- 1 except with longer cruise-toletdown distances. The weapons would be
delivered in a level delivery (one g-loading)
so the g-loading on the aircraft during the
entire mission would likely range between
one and three g's. Delivery of its nuclear
cruise missiles (one g-loading) would be
preceded with a transfer alignment of the
cruise missile inertial reference system. The
terrain correlation matching system in the
missile is used for enroute and terminal position updates. Terrain reference navigation
(TRN) compares real-time sensed terrain
variations with onboard reference maps to
determine the position of the missile.
The non-nuclear mission profile is a high altitude profile for the entire mission thus surGreat
Circle
Strategic Alrcrafl
Crulse Mlssle
(1) Takeoff
(2) INS Air Start'
(3) Start Ovenvater
Flight (8 hr)
(4) IRS Alignment
to Missle Launch
Missle Launch
(5) Launch
(6) Landfall
(7) Update (Option)
(8) At Target
35
Attitude
Terrain Avoidance - Mach 0.7
1,000 ft
9
10
Flight Time - hr
(7) (8)
I
11
12
13
I
14
1
15
tttrttt t t t t t t
* Doppler updatesevery 30 seam& during overfandflight
L+v
** Each stellnr update consists of 3 star lixes
6 Position Fixes
Stellar Updates''
CC24-0157-33-D
Figure 3.0.3-2 Strategic AircraftCruise Missile Mission Profile
The profile for the cruise missile mission is

depicted in Figure 3.0.3-2. The events table
shows tasks of both the strategic aircraft and
cruise missile. The reference INS is started
in flight using Doppler updates every 30
seconds and radar position fixes to align the
INS. During flight over water (up to eight
hours), stellar updates (three star fixes) are
performed every 30 minutes to better calibrate the INS. During captive flight, the
face-to-air and enemy air threats are not significant factors. Carpet bombing is accomplished by taking advantage of the large
numbers of conventional weapons, (MK-82
or MK-84 low drag general purpose bombs),
that can be carried on the strategic aircraft.
34 1
3.0.4 Transport A ircraft Customer Reguirements

Examples of this type of aircraft are: C-17
and C-130 Advanced Theater Transport, and
Special Operations Aircraft. The mission
requirements for a transport aircraft are
outlined as follows:
Covert operation
Combat zone
Austere basing
Low level (below 200 ft.) flight
All weather / night operation
Long / extended range
Cooperative tactics
Aerial delivery
Medical evacuation
Transport Aircraft Mission Requirements
Typical mission segments for a transport

aircraft are shown in Figure 3.0.4-1. The
most critical segment for a VMS from a performance point of view is the low altitude
parachute extraction segment. This requires
holding altitude at 4 to 6 feet above the
ground while off-loading cargo of up to
60,000 pounds. Although this altitude-hold
requirement is more stringent than typical, it
does not impact the VMS architecture. The
most significant architectural impact is due
to the length of the mission. The long duration, medical evacuation long range requirement drives the safety and redundancy
requirements for its' VMS architecture.
1 Pre-mission Planning
2 a Preflight and Pre-takeoff
3. Takeoff
4. Departure/Cruise
5 a Medium Altitude Airdrop
6 a Outbound Flight
7 a Rendezvous
8. Returning Flight
9. Low Altitude Parachute Extraction
l o a Return To Base
I 1 Descent And Land
12. Turnaround
a
Figure 3.0.4-1 Mission Segments For Transport Aircraft
342
3.1
-F
A functional analysis must be performed in

order to examine the navigation system performance requirements of the aircraft. Requirements addressed should include those
for initial alignment, basic navigation, sensor cueing and designation, Inertial Navigation System (INS) drift requirements, and
Vehicle Management System (VMS) flight
critical functions. Weapon delivery systems
impose additional requirement on the navigation system to provide updating to improve weapon delivery functions. "Aided"
inertial sensing performance requirements
should also be addressed. VMS functions to
be examined are inner-loop flight control
(basic flight) and integrated flight modes.
bitial Alignment Analvsis Inertial navigation performance during flight is greatly
dependent on the quality of alignment received before takeoff or during an in-air
alignment. Performance parameters are
usually based on a full-up, ground, gyrocompass alignment. Shortened reaction time
alignments can be mechanized resulting in
reduced accuracy or in special parking and
handling of the aircraft prior to starting the
alignment. Several alignment modes are required in order to provide flexibility in reaction time and resulting accuracy.
Other alignment requirements are (1) automatic interrupted ground alignment detection using measurements of vehicle velocities, and (2) alignment quality indication
(approximate CEP drift rate in nm/hr).
Normal Ground Alignment Analvsis During normal alignment the INS will autonomously determine the correct reference
frame. The vertical reference is established

by sensing the earth's gravitational acceleration. The true north heading reference is
determined by sensing earth rates and equating those to the known earth rates for the
latitude at the aircraft position. The alignment time required for full performance is
typically four minutes.
Alignment requirements usually specified
for a navigation system are alignment time
and accuracy. These two requirements primarily affect the quality of gyros required.
Although detailed gyro specifications are not
usually included in a procurement specification, knowledge of operation is essential to
evaluate the feasibility of candidate systems
and alignment requirements.
Typical Allocation of Error Sources During Alignment
Error Source
Allocated Emw
1. Gyro WRN
0.36
2. Gyro Bias
0.13
0.0015"hr
3.Vel Disturbance
0.13
0.02 ftlsec
4. All Other E r r m
0.03
mrads
IFigure 3.1-1 Typical Allocation
of Error
0.41
(about 0.4 NMVhr)
Sources During Alignment

Figure 3.1-1 shows a typical error budget for
a Strapdown INS alignment (Reference 7).
The largest single contributor to the residual
alignment error is the gyro wide band random noise (WRN) or random walk which
typically dominates for alignment times
greater than 2-3 minutes. The relationship
for wide band random noise and residual error is shown in Figure 3.1-2 (Reference 7).
rl
&=
___-_.-_-__-_--__1_1___
RxSqrt(T)xCosO.)
where,
E = heading error allocated to wide band random noise
q = gyro wide band random noise specification
12 = earth rate (15.04"hr) (sidereal time)
h = latitude
T = alignment time
Figure 3.1-2 Random Wide Band Noise and Alignment Error
343
Stored Heading Alignment Analysis A

stored heading ground alignment is provided
to minimize INS alignment time. During
this mode the INS heading is set to the value
that was stored during the previous ground
alignment or at the end of the last flight.
This mode is usable only if the aircraft has
not been moved since the last alignment or
flight. Reaction time is typically 30
seconds.
Attitude And Heading Reference (AHRS)
Ground Alignment Analysis The AHRS
ground alignment mode provides a 30
second alignment without stored heading
available. In the AHRS mode the INS provides heading, attitude, angular rates, and
acceleration with a lower accuracy. Subsequent in-flight aiding or in-flight alignment
results in full navigation performance.
Inflight Alignment Analysis The inflight
alignment mode enables the INS to perform
an inflight start up or improve system
performance if less than a normal full
alignment was performed. The INS will use
the best present position, velocity, and heading data available from the non-inertial navigation aids.
Basic Navigation Analysis The basic
navigation functions for all of the tactical
and the strategic mission are similar. The
strategic mission places a different requirement on the basic navigation function due to
extended mission duration. Inertial navigation is the process of calculating position
and velocity based solely on inputs from
self-contained acceleration sensing instruments.
Accelerometers provide the
acceleration magnitude sensing function.
Gyros provide the acceleration direction
sensing function (i.e. define the direction of
the accelerometer sensing axes). The basic
navigation concept is to integrate the sensed
acceleration vector data (corrected for
gravity) to determine vehicle velocity, and
then integrate the velocity data to determine
vehicle position.
Basic navigation parameter outputs required
are latitude, longitude, altitude, vertical velocity, horizontal velocities, true heading,
roll angle, pitch angle, and load factors
(linear acceleration). Accuracy's needed for
basic navigation are indicated in the following paragraphs.
Latitude And Longitude During normal

navigation, aircraft latitude and longitude
are required for on-course flight to a way
point or rendezvous point. Autonomous
position accuracy available with current,
standard inertial navigation technology (0.6
to 1 nm/hr CEP diift rate) have been and are
still adequate for basic navigation present
position. Worldwide operation (*9000)latitude, f 18000 longitude) of the inertial present position function is required.
Present position requires fault tolerance and
survivability for safety at night, in weather,
and for return to friendly temtoryhase. To
increase system availability, a fail-op continuous inertial navigation capability is provided. Survivability is provided by use of
radio navigation aids in case battle damage
completely disables the fault tolerant inertial
navigation function.
Heading And Attitude True heading, roll

angle, and pitch angle are required for attitude control, coordinated turns, landing and
takeoff, autopilot and proper display orientation. Autonomous true heading and attitude accuracy available with current inertial
navigation sets (better than 0.1 deg, 1 s)
have been and are still adequate for safe
navigation. Range required is all possible
heading and attitudes, usually expressed as 0
to 360 degrees heading, f 90 degrees in
pitch and f 180 degrees in roll.
In previous aircraft, the INS, Attitude
Heading Reference System (AHRS), Magnetic Azimuth Detector (MAD), and backup
instruments provided the aircraft true heading, magnetic heading, pitch-roll attitudes,
and rate of turn for navigation and safe flight
at night and in weather. Consolidation of
these equipment sets requires at least a fault
tolerant INS or two sources of attitude data
(2 INSs or INS plus AHRS). In addition, a
survivable source is required for attitude
when coupled directly into the flight control
system for low altitude operations such as
short take off and landing (STOL) and Terrain Followinflerrain Avoidance (TFRA).
Survivabie heading and attitude requires at
344
least two (2) physically separated sources

(such as IIRA, 2 INSs, INS plus AHRS, or
INS plus backup in flight control system).
The configuration which meets these requirements in a most cost effective manner
should be part of a trade study in the design
process.
Horizontal Velocities Analysis Navigation

frame (North/South and EasWest) velocities are required for on-course/on-time flight
to a way point or rendezvous. Body frame
(nose, right wing, belly) are required for the
display of the aircraft velocity vector during
low level maneuvers, landing, and weapon
delivery. Autonomous velocity accuracy
available with current inertial navigation sets
(2.0 to 2.5 ft/sec (1 sigma) navigation or
body frame) have been and are still adequate
for navigation. Range required is -3000
ft/sec to +3000 ft/sec in either frame.
Velocity fault tolerance (fail-op), and survivability requirements are similar to those
for present position. Velocity survivability
will be provided by air data computations
and radio navigation aids in case battle damage completely disables the inertial navigation function.
Altitude And Vertical Velocity Analysis
Inertial altitude and vertical velocity are required to supplement the altimeter during
large variations in altitude. Pressure-altitude
can be invalid during and after steep dives or
climbs. Short-term inertial vertical velocity
is required to be accurate to 2 ft/sec (1
sigma) during this period primarily to provide accurate velocity vector calculations.
The long-term inertial altitude is required to
track the pressure-altitude within 2 feet (1
sigma) during steady state conditions and to
track true altitude within 150 feet during
maneuvers. Range required is - 1000 feet to
+75,0000 feet for altitude and + 1500 ft/sec
for vertical velocity.
Altitude and vertical velocity fault tolerance
and survivability are required due to the requirement for valid velocity vector display
during landing. Fail-op inertial outputs will
be provided to increase availability and
safety during large altitude variations.
Linear Acceleration Analysis Longitudinal, lateral, and normal linear accelerations

(expressed as load factors) are required for
aircraft load warning, recording, and limiting. Normal accuracy and ranges for a tactical fighter are 0.05 g's (1 sigma) accuracy, +
3g's range for longitudinal and lateral accelerations, and -4g's to +12g's range for normal accelerations.
Cooperative Functions Reference Analysis Relative position between cooperating

aircraft, and each aircraft's heading are required for triangulation during some cooperative operations. INS heading accuracy of
0.05 deg (1 sigma) is required to achieve a
ranging accuracy of 3% under nominal
conditions. Relative position can be derived
using INS data or by several radio navigation aids. The required accuracy of a few
hundred feet is not obtainable with an INS in
autonomous operation, but requires frequent
position updates or navigation fusion.
During lengthy missions, the aircraft will require inflight refueling. Rendezvous with
the tanker aircraft is required prior to refueling. Based on F-15 and F-4 experience a
position accuracy of 1.5 nm has been
demonstrated to be adequate in the existing
cooperative refueling environment.
Sensor Management Analysis The navigation system must provide present position
and attitude on data with sufficient accuracy
to ensure that the way point or target is
within the sensor field of view (FOV) or
mapped area when expected.
The INS position and position drift requirements to support sensor cueing in this example will be based on the air interdiction
mission discussed in section 3.0. The position error allowed for successful cueing is
based on a cueing probability of 0.99. The
cueing probability is a function of the way
point or target location within the sensor
FOV or mapped area is calculated as shown
in Figure 3.1-3. The major contributors to
way point or target position error within the
sensor FOV or mapped area are the target
location uncertainty and the aircraft position
error.
345
(1) Position Uncertainties
-1
Half Field Along LOS (Fa)

If Field Cross LOS (Fc)
Line
Target Location With Position Error

Ideal Patch Location
Patch Location With Position Error
(2) P (Target in Patch) = P (Target in Along Track) * P (Target in Cross Track)
= P (FdAlong Track Error) P (FdCross Track Error)
-3
-2
-1
0
X
-3
-2
-1
Figure 3.1-3 Calculating the Cueing Probability
3
CC24-0157-87-D
346
1. Perform
Velocity
Update
2. Create
A
SAR
Map
3. Designate
Way point
cc240157-83-0
Figure 3.1-4 Cueing and Waypoint Detection with Radar
1.o
Patch Size
3.34 x 3.34 NM
0.8
0.6
Probability
Waypoint in
SAR Patch
0.4
1 . 3 4 ~1.34 NM
0.2
0.67 x 0.67 NM
2,000
4,000
6,000
Patch Error - ft, CEP

CC24-0157-64-D
Figure 3.1-5 SAR Map Cueing Probability
347
Map Size
(NM x NM)
Target Location
Uncertainty
(Note)
(ft, CEP)
Aircraft Present
Positio n Uncertainty
(ft, CEP)
0.67 x 0.67
850
295
800
1.34 x 1.34
1705
295
1680
3.34 x 3.34
4250
295
4240
4.67 x 4.67
5945
295
5935
10 x 10
12.7k
295
12.7k
20 x 20
25.5k
295
25.5k
40 x 40
50.9k
295
50.9k
80 x 80
102k
295
102k
CC24-0157-65-D
Figure 3.1-6. Position Error Allowed for Cueing Radar with P=0.99
Wide FOV 5" x 5" or

Narrow FOV 1.25" x 1.25"
Target Location
Uncertainty = 295 ft CEP
to 5,000 ft
Altitude
L-Detect
Range
_I
CC24-0157-66-0
Figure 3.1-7 Cueing and Way point Detection with FLIR
348
Ground
Range - NM
5" x 5" FOV 5,000 ft altitude AGL
1.o
20
15
0.8
10
0.6
Probability
Waypoint in
FLlR FOV
0.4
0.2
1,000
2,000
3,000
Position Error - ft, CEP

CC24-0157-67-0
Figure 3.1-8. Position Accuracy for Cueing Wide FOV F U R

1.o
1.25" x 1.25" FOV 5.000 ft altitude AGL
0.8
0.6
Probability
Waypoint in
FLlR FOV
0.4
0.2
0
0
3,000
1,000
2,000
Position Error - ft, CEP
CC24-0157-68-D
Figure 3.1-9. Position Accuracy for Cueing Narrow FOV F U R
349
Maximum i'arget Location

Aircraft Present
Field-of-View Allowed
Uncertainty
Position Uncertainty
[degs x degs: (ft, CEP)
(ft, CEP)
(fl, CEP)
20
15
10
5
3
Wide
2427
1822
1217
614
377
5x5
295
295
295
295
295
2409
1797
1180
539
235
295
295
295
295
295
530
347
73
20
15
10
5
3
Narrow
1.25 x 1.25
606
455
304
154
94
Conditions: 1. PFOV = 0.99

2.5000 ft altitude
3. INS to sensor Boresight better than 4 mrads
CC24-0157-69-D
Figure 3.1-10 Position Accuracy for Cueing F U R with P=0.99

Procedure Applles to Waypolnt or Target Acqulsltlon Wlth Radar SAR Map or FLlR
Avlonlcs
Mlsslon
I
Mechanization
Equations
Preliminary
Error Budget
Determine Sensitivity
Equations by Differentiating
Equations wrt Error Sources
Individual
Errors
Aircraft Velocity
Range to Target
Height Above Target
Designation Error
Requirement
(From Update or
Weapon Delivery
Requirements)
Equipment Specs
Established
CC240157-88-D
Figure 3.1-11 Equipment Specifications From Designation Error Analysis
350
Radar Sensor Cueing Figure 3.1-4 shows

a typical Synthetic Aperture Radar (SAR)
mapping procedure. First a precision velocity update is performed to limit designation
errors. Next the map is created for the land
area assumed to contain the way point or
target. The probability that the way point is
in the map is based on map size and aircraft
position error as shown in Figure 3.1-5 for
several example map sizes. The maximum
position accuracy allowed for a given probability are shown in Figure 3.1-6 for a probability of 0.99. The maximum position accuracy allowed consists of aircraft position
error and target position uncertainty. The allowable aircraft position error is based on a
295 ft, CEP target (map) uncertainty which
has been subtracted in a Root-Sum-Square
(rss) fashion.
FLIR Sensor Cueing Analysis The
methodology for determining the position
error allowed for FLIR cueing is the Same as
that used for radar cueing once the FLIR
along and cross LOS fields are found as illustrated in Figure 3.1-7. This is done by
projecting the FLIR FOV onto the earth surface. The cross line of sight (LOS) half field
is simply equal to:
Rc = Rs x Tan(q /2)
where:
Rc = Cross LOS Field
Rs = Slant Range
q = Sensor Field of View
The probability that the way point is in the

along LOS FOV is assumed to be unity due
to the geometric enlargement along the
ground track as shown.
The probability that the way point (or target)
is in the projected field is shown in Figures
3.1-8 and 3.1-9 for wide and narrow FOV's
respectively. As the FOV gets smaller and
the distance to way point gets smaller, the
projected area reduces thus reducing the
probability as shown in the figures.
The computed allowable position errors are

shown in Figure 3.1-10 for a 0.99
probability of cueing. Thus, from the figure
the aircraft position error allowed for a 5
degree by 5 degree FOV FLIR at 20 nm is
2409 ft, CEP (P=O.99). The requirement for
the narrow FOV FLIR is only 530 ft, CEP
(P=0.99). Some conditions shown in the
figures are not realizable even with perfect
aircraft position accuracy due to the way
point position uncertainty. In these cases the
target must be found first using the wide
FOV. Once the target is detected and
stabilized in the wide FOV, the FLIR can be
switched to narrow FOV for tracking and
subsequent target state estimation.
The INS position accuracy requirement for
sensor cueing is defined by the most stringent position accuracy for the Radar or
FLIR. This is found to be 235 ft in the case
of FLIR cueing at 3 nautical miles in the
wide FOV. Attitude and heading accuracy
required for sensor cueing were defined in
Reference 10 and 15 and are specified as 0.1
degrees (1 sigma).
Designation Management Analysis The
inertial navigation system must provide the
necessary platform references to the appropriate sensor to establish relative location of
the platform with respect to target or way
point. The relative data is necessary to
compute ballistic solutions for bombing or
to provide an accurate platform position fix
relative to a known map feature. The
methodology used to determine the contribution of inertial reference parameters to
designation errors is outlined in Figure 3.111. The procedure applies to way point or
target designation with radar SAR map or
FLIR.
Sensitivity equations are determined by differentiating the mechanization equations
with respect to the error sources of interest.
Error values and aircraft parameters are then
entered into the sensitivity equations. The
resulting error contributions from each sensor source are RSS'ed together and compared with the required designation error.
The error budgets are iterated until the requirement is met and a suitable compromise
is made between equipment specifications
and equipment producibility.
35 1
Along LOS Designation Error

Aircraft
velocity
lt
1P-
Cross L(IS Designation Error
Angle
Aircraft
7-l
Ideal Patch Location
Aircraft
Patch Location with

Designation Errors
Depression
Height
. _.a - -
Ground Range
CC24-0157-70-0
Figure 3.1-12. Radar SAR Map Geometry for Designation Error Determination
Sensitivity Equations
Error Source
Operator Designation
Cursor Related Ranging
Doppler Process
Aircraft Altitude
Velocity - x
Cross LOS
Along LOS
dR*Npd
*Ah
R S C W
TncosO *AVx
dA
Npd
Ah
R,tanQcos'@
1 [&
+ TnsinO]
AVx
Velocity y
Velocity - z
V sin0 cos@
R,
AVz
AV
d, = Aiimuth Distance Resolution
Rs = Slant Range
v = Azimuth Angle
dR = Range Resolution
h = Height
T, = Nav Time
N pxl = Number of Pixels
e = Squint Angle
V = Aircraft Velocity
$I = Depression Angle
CC24-0157-71-D
Figure 3.1-13. Radar SAR Sensitivity Equations
352
Error Source
Units
Cursor Related
Ranging - Doppler Process
Aircraft AItitude
Velocity - x
Velocity - y
Velocity - z
Heading
REP
DEP
Standard
Deviation
1.50
2.10
0.30
100
0.80
0.80
2.00
0.96
Along
LOS
ft
Cross
LOS
ft
12.8
18.0
2.5
5.2
8.5
0.5
0.0
0.0
17.0
12.8
18.0
2.5
5.2
94.8
77.8
16.0
87.5
104
105 ft
Conditions: 15 nmi grd range, 4,800 ft alt, 500 kts,

45" squint, 15 sec nav, 8.5 ft resolution
CC24-0157-72-D
Figure 3.1-14. Radar SAR Designation Error for 8.5feet Resolution
Error Source
Units
Standard
Deviation
Cursor Related
Ranging - Doppler Process
Aircraft Altitude
Velocity - x
Velocity - y
Velocity - z
Heading
REP
DEP
Conditions: 15 nmi grd range, 4,800 ft alt, 500 kts,
45" squint, 15 sec nav, 127 ft resolution
1S O
2.10
0.30
100
0.80
0.80
2.00
0.96
Along
LOS
ft
Cross
LOS
ft
190.5
266.7
38.1
5.2
8.5
8.5
0.0
0.0
190.5
266.7
38.1
5.2
94.8
77.0
16.0
87.5
330
363
408 ft
CC24-0157-73-D
Figure 3.1-15. Radar SAR Designahit Errorfor 127feet Resolution
353
10
15
25
20
30
Range to Update Point NM

cc2+0157-7+0
Figure 3.1-16. Position Update Accuracy with Radar SAR Map

350
I
AI I
300
250
200
150
100
50
0
GPS
PVU
3
Autonomous INS
Velocity Error fps, rms

CC24-0157-75-D
Figure 3.1-17. Designatioa Error Sensitivity to INS Velocity Error
354
CC244157-76D
Figure 3.1-18. F U R Geometryfor Designation Error Determination

Sensitivity Equation8
Error Source
Along
LOS
Cros8
LOS
Height
cos$ ARS
sin$ ARs
Laser Range Timing
V cosy co2g At
V cosy cos$ sine At
Laser Alignment
Rs cos$ cote AQ
Rs cos) Ai$
Laser Range
Gimbal Readout Timing
~sin%*At
Gimbal Readout
Tracking Loop
Gimbal Boresight
FLlR Boresight
Flexure Uncertainty
Rs sine A@
Rs A y
Rs cos$ A$
Azimuth Reference
Rs cos$ AV
Elevation Reference
Rs sin@ ABp
Rs cos$ ABp
RS = Slant range
y l = Azimuth angle
$
I = Depressionangle
A yl = Azimuh angle uncertainty
V sin4 cos$ At
\ = Aircraft velocity
A I = Timing uncertainty
AOp = Pitch angle uncertainty
Figure 3.1-19. F U R Sensitivity Equations
CC240157-770
355
Level, 1,000 ft
Error Source
Unlts
Laser Range
Laser Range Timing
Laser Alignment
Gimbal Readout Timing
Gimbal Readout
Tracking Loop
Gimbal Boresight
FLlR Boresight
Flexure Uncertainty
Azimuth Reference
Elevation Reference
ft
ms
mr
ms
mr
mr
mr
mr
mr
mr
mr
Standard
Deviation
15
33.7
0.15
9.6
0.985
0.141
0.1
0.37
0.33
0.96
0.71
Subtotals (ft, 1 Sigma)

Horizontal Clrcular Error Probable (ft)
Along
Track
ft
Cross
Track
ft
14.9
28.3
22.1
0.1
1.o
0.1
0.1
0.4
0.3
0.0
0.7
38.9
0.0
Helght
ft
1.3
2.4
1.8
0.7
11.8
1.7
1.2
4.5
4.0
0.0
8.4
16.2
0.0
0.0
0.0
11.8
1.7
1.2
4.5
4.0
11.5
0.0
17.7
33.4
Conditions: 12k ft Slant Range, 1,000 ft Altitude, 0" Squint, 500 Ms
CC24-0157-780
Figure 3.1-20. F U R Designation Error

24a
V = 500 k t ~
20a
160
Update
Accuracy
CEP
120
ft
80 H = 1,000ft
12
Slant Range to Update Point - 1,000 ft
Figure 3.1-21. Position Update Accuracy with F U R
16
20
CC24-0157-70-0
356
Radar SAR Designation Analysis In the
S A R mode the radar constructs a map of the

cued area. Position updates or target
designation is accomplished by identifying a
known landmark and indicating its relative
position to the map coordinates. Figure 3.112 shows the radar SAR map geometry for
determining the designation error. The designation error consists of along LOS and
cross LOS errors. The sensitivity equations
used to find these errors are shown in Figure
3.1-13 for the significant contributors.
Derivations of these equations can be found
in References 10, 15 and 17 .
I
I
A typical SAR error budget is shown in Figures 3.1-14 and 3.1-15 for two different map
resolutions. (All error contributors discussed
in this section are 1 sigma values.) The dependence of position update accuracy on
map resolution and update range shown in
Figure 3.1-16. The parametric curve shown
in Figure 3.1-17 shows the dependence of
S A R map designation errors on INS velocity
error. Even at 0.8 ft/s velocity error the
designation error is increased significantly
over what it would be if no horizontal velocity error was present. A typical INS velocity
specification of 2.5 ft/s would more than
double the designation error with 0.8 ft/s velocity errors. This dependence explains the
need for performing a precision velocity update (typically a radar mode) immediately
prior to construction of a S A R map or the
use of GPS velocities to update the INS.
FLIR Designation Analysis The FLIR

provides a relative position update or target
designation by acquiring a landmark and
indicating its relative position to the aircraft.
The landmark must be recognized and
designated by the pilot unless automatic target recognition is provided. If the latitude
and longitude of the landmark is known,
then a new aircraft absolute position can be
calculated. Depending on the aircraft and
electro-optical system one of two techniques
is typically used for performing navigation
updates and target designation. One technique uses a FLIR in combination with a
laser rangeddesignator. In this system the
relative pointing angles to the target, in
conjunction with slant range data from the
laser ranger, are used to derive the aircraft
position. The other system uses track angles

combined with measured angle rates and
known aircraft velocity to derive aircraft
position. This technique is often referred to
as an angle rate system. Figure 3.1-18
shows the FLIR geometry for determining
the designation errors for either FLIR sys-
tem.
The sensitivity equations for the laser designator system for along LOS, cross LOS, and
height are shown in Figure 3.1-19. A typical
error budget for this system is shown in Figure 3.1-20. The parametric curve in Figure
3.1-21 shows the dependence of position
update accuracy on slant range to target and
altitude. (These update accuracy do not include any way point location uncertainties.)
357
@ 125 NM to Target
TRN Data Base not Availabl
150 ft, CEP
15 min. @ 500 kts

GPS, TRN, Radio
Nav Still not Availat
@BlindCue
Bomb
Radar
Target
to 7
@ No Updates for 0.5 hr
GPS not Available

TRN not Functional Ove
No Intermediate Waypo
No Radio Nav Available
No Star Tracker on Airc
Use 0.67 x 0.E

Map Size for Bl
Bombing Accui
CC24-0157-86D
Figure 3.1-22. Scenario I: Radar Cueing

Cueing Requirements
0 (99%) Prob
Radar 20 x 20 NM Patch
(255 ft Resolution)
3
Radar 10 x 10 NM Patch
(1 27 ft Resolution) 2-
1
Radar 4.67 x 4.67 NM Patch a
Radar 3.34 x 3.34 NM Patch
F L l R 5 ~ 5 ~ 0 2 0 N+
M
2
3
Time Since Last Update - hr
(Last Update Accuracy = 150 ft, CEP)
5
CC24-015741-D
Figure 3.1-23 INS Drift After 5 Hours Autonomous Flight
358
Cueing Requirements
0 (99%) Prob
2,000
F L l R 5 x 5 " Q 15NM
1,500
Position Error
( f l l CEP)
F L l R 5 x 5 " Q 10NM1,000 Radar 0.67 x 0.67 NM Patch

FLlR 5 x 5" 8 5 NM
FLlR 1.25 x 1.25" 0 20 NM
FLlR 1.25 x 1.25" Q 15 NM
FLlR 1.25 x 1.25" 0 10 NM
I
I
Time Since Last Update - min

CC24-0157-83-D
Figure 3.1-24. INS Drgt After Autonomous Flight (Scenario I )
INS Drift Analysis Example The INS drift

requirements for sensor cueing are defined
relative to the air interdiction mission defined earlier. Two scenarios are used as
examples for analysis, one using the Radar
and one using the FLIR.
Example scenario I The first scenario,
Figure 3.1-22 uses the radar for both way
point and target acquisition. The assumptions for this scenario are that the navigation
system will cue a 3.34 nm x 3.34 nm SAR
radar map after a .5 hour flight with no
navigation aids available. With an update
accuracy of 170 ft (consistent with the SAR
map size), the navigation system must cue a
0.67 nm x 0.67 nm radar map to a target after an additional 15 minutes of unaided
flight.
The graphs in Figure 3.1-23 show aircraft
position error propagation after an initial position update accuracy of 150 ft. CEP (an
initial position fix from GPS or terrain refer-
ence navigation has been assumed). The

initial error (150 ft, CEP) is combined in an
rss manner with the product of INS drift and
time. Shown are drifts from 0.1 nm/hour to
1.0 ndhour, CEP.
The INS drift requirement for the first update using the 3.34 nm x 3.34 nm map is
found by using the position accuracy requirement for the 3.34 nm x 3.34 nm from
Figure 3.1-6 above (4240 ft) and the curves
of Figure 3.1-23. The required INS drift is
found on Figure 3.1-23 at the intersection of
the allowable position error for the 3.34 nm
x 3.34 nm map and the time of the last update. The first update requirement is found
to be in excess of 1 nm/hr. The drift requirement for cueing the .67 nm x .67 nm
S A R map for target acquisition is found in a
similar manner. From Figure 3.1-24 the allowable drift rate is found to be 0.45 nm/hr
CEP for .99 probability.
359
Last Update
with TRN,
150 ft, CEP
@ No Update for 10 min
- GPS Not Available

- Radar Alt Turned Off for Last 80 nmi
- Radar Not Used for Acquisition
Acquire Target
- Cue FLlR to Target at 5 nmi
- Use 5 O x 5 FOV
- See Weapon Delivery
Analysis for CEP Achieved
CC24-0157-82-D
Figure 3.1-25. Scenario II: F U R Cueing

Cueing Requirements
0 (99%) Prob
2,00(
FLlR5x5"e 1 5 N M a
1,50(
FLlR5x5"O 10NM1,OO(
Radar 0.67 x 0.67 NM Patch a

FLlR5x5"OSNM
FLlR 1.25 x 1.25" 0 20 NM
FLlR 1 . 2 5 ~
1.25" 0 15 NM --c
FLlR 1.25 x 1.25" 0 10 NM
10
15
Time Since Last Update - min
20
CC260157-84-D
Figure 3.1-26. INS Drift After Autonomous Flight (Scenario I I )
Example scenario I1 The second scenario

(Figure 3.1-25) is the same as the first except in this case the FLIR is used to acquire
and track a ground target to support a bombing mission. This scenario requires curing
the FLIR to a target after 10 minutes of un-
aided flight. Figure 3.1-26 shows the

requirement for cueing the 5degree x 5
degree FLIR wide FOV. From this illustration the INS drift allowed is 0.51 (P=O.99)
nm/hour, CEP.
360
Map; 8.5 ft, 10 NM 45' Squlnt, Release; 540 kts, 1,000 R Level
Error Source
Unit.
Standard
Devlatlon
1 Operator
2 Cursor
Pix
1.5
2.1
3 RangeDoppler
pix
0.29
11
100
Pix
4 Altitude
5 Radar Velocity x
6
Radar Velocity - y
11s
11s
11s
11s
11s
INSVelociy-x
INS Velocity y
8 INS Velociy Vertical
9 Attitude Angle Reference (El)
7
mr
mr
X
mr
(Az:
Radar Slant Range
Beam LOS Position
TrueArspeed
Pilot Steering
14 Bomb EjedionVelocity
I5 Bomb Release Time Dday
I B o m b Dispersion
I7 Ballistic Fa
IC
I1
12
I?
0.8
3.0
mr
mr
1
R
R
2.0
0.33
12.75
18.00
2.45
5.23
92.51
14.16
12.79
5.28
9.87
0.18
10.27
43.48
3.3
ms
Track
ft
0.96
mr
lh
Croea
12.75
18.00
2.45
5.23
39.17
39.17
9.04
9.04
53.91
17.39
0.8
2.5
2.5
2.0
0.713
1.78
1.69
2.50
1.15
11s
:zE
0.38
30.76
2.73
51.27
3.47
REP = 78
17.63
14.24
DEP = 66
CEP = 126
Figure 3.1-27 Radar SAR Blind Bombing Error Budget
Level, 1,000 ft
Error Source
Aircraft Position at Release
Slant Range
Ranger Boresight
FLlR El Angle
FLlR Az Angle
Pilot fiz Steering
Weapon Range
Bomb Ejt Velocity
Release Time Delay

Bomb Dispersion
INS Velocity Reference
INS Vert Velocity
INS Elevation Ref
INS Boresight (Elevation)
True Airspeed
SideSlip Angle
Ballistic Fit
Total 1 Sigma Error
Units
Standard
Deviation
!:;
Cross
Track
ft
21
0.2
1.2
1.3
3.0
10.01
14.00
27.90
0.00
0.00
0.0
0.0
8.7
20.4
fps
2.0
52.50
0.0
ms
10.0
3.0
3.0B.O
2.0
1.5
1.5
2.7
1.5
0.33
9.10
75.40
22.80
52.90
38.20
38.20
0.47
0.00
15.50
0.0
20.6
22.8
0.0
0.0
0.0
0.0
0.1
0.0
126.80
37.9
mr
rnr
mr
rnr
mr
fps
fps
mr
rnr
kcs
mr
mr
ft
0.0
Figure 3.1-28. F U R with Laser Ranger Bombing Error Budget
36 1
CC240157-81-D
Figure 3.1-29. Angle Rate Bombing Geometry
Standard
Devlatlon
~~
~~
mr
mr
mrh
FCS Roll Ref
INS Angle Ref
Pilot A2 steering
Neapon Range
Bomb E# Velocity
Release Time Delay
Bomb Dispersion
INS Velocity Reference
INS Vert Vebdty
INS Boresight
True Airspeed/AOA
AOA Boresighl
Ballistics R
rad 1 Sigma ~ n w
mrls
mr
mr
mr
fps
ms
mr
fps
fps
mr
ldslmr
mr
mr
1.Y1.5
7.Y1Sll.5
0.2
0.1
10.0
1.4Y1AY1.45
3.0
7.7
7.7
30.9
0.0
0.o
3.2
0.0
2.0
10.0
3.0
3.0B.O
2.0
3.011SJl.5
2.7141
2.1
0.33
50.7
9.4
71.1
12.3
0.0
3.3
11.8
5.9
15.5
96.8
0.07
7.8
0.0
5.0
10.4
0.2
20.5
0.1
o..o
20.5
0.8
0.0
2.8
0.8
0.0
0.0
32.3
~~
:lrcular Error Probable
74.3 ft
022441677-02
Figure 3.1-30 Angle Rate & d i n g Error Budget
362
On the basis of the scenarios and sensors

analyzed the requirements placed on the
navigation system to support sensor cueing
and way point/target designation are drift
rate and velocity accuracy. For the specific
cases presented the constraining condition
for drift rate is the long duration flight without position updates. For this case hand-off
to a 3.34 nm x 3.34 nm SAR map with a
0.99 probability after a 15 minute flight requires an INS with a drift rate of less than
0.45 nm/hr, CEP. Obviously, changes in the
particular aspects of the scenario will change
the limiting conditions.
Weapon Delivery Bombing Modes

Analysis The navigation system provides
the velocity and attitude reference required
to compute weapon ballistic trajectory.
Errors in the navigation parameters
propagate into bomb impact errors.
The methodology used to determine the
contribution of error sources to bomb impact
error is analogous to that used for navigation
and designation updates. First, sensitivity
equations are determined by differentiating
the mechanization equations with respect to
the error sources of interest. Weapon release conditions (mode, altitude, dive angle,
and velocity), and error source values (target
sensor errors, inertial reference errors, pilot
steering error, bomb dispersion, etc.) are
entered into the sensitivity equations. The
resulting contributions from each error
source are combined together to compute the
bomb impact error. The impact error is then
compared to the requirement. Requirements
are iterated until a suitable compromise is
made between equipment specifications and
equipment producibility.
Blind Bombing Analysis The most

stringent velocity reference requirements
come from blind bombing. The target is acquired using the radar SAR mode as discussed above. The bomb impact error is
equal to the rss of (1) the target designation
error, (2) relative navigation error incurred
from target designation to weapon release,
(3) the release error due to uncertainty in the
release conditions, and (4) bomb dispersion
after release.
Figure 3.1-27 shows a blind bombing error

budget. The velocity reference uncertainties
dominate. Most of the level velocity contribution is due to the designation error. The
velocity accuracy required to limit designation error to a reasonable value, and thus
limit bomb impact error, is 0.8 fps, 1 sigma.
Conventional Bombing Analysis
Conventional bombing with a head up display or FLIR sensor requires less velocity
accuracy. Designation with the FLIR is
similar to way point designation as previously discussed except that the target is
locked on and tracked until weapon release.
Figure 3.1-28 shows an error budget for
bombing with a FLIR and laser ranger. The
error contribution due to velocity is significant but not dominant.
Angle Rate Bombing Analysis In angle
rate bombing (ARB), a FLIR tracks the
target, and provides line-of-sight angles and
angle rates to the target relative to the
aircraft. These angles and angle rates are an
implicit measure of range to the target and
target relative velocity normal to the line-ofsight. The ARB geometry and relevant
equations are shown in Figure 3.1-29. The
advantage of angle rate bombing is that the
first order effects of aircraft INS velocity errors cancel. The nullifying effect results
from the fact that both target range and
weapon range are nearly linear functions of
aircraft velocity. A positive error in velocity
would indicate both an increased target
range and a compensating increase in
weapon range therefore nullifying the impact of the error. Figure 3.1-30 shows an error budget for angle rate bombing. The inertial velocity and attitude contributions are
small compared to the total error.
From examination of the budgets the navigation system requirements to support
weapon delivery are primarily for aircraft
velocity. Of the three modes reviewed
"blind bombing" is the only mode that delivery accuracy is directly dependent on velocity accuracy. It is for this reason that blind
bombing is usually supported by a precision
velocity update either from radar or GPS. In
the other modes of automatic weapon delivery the sensor (FLIR or radar) provides most
of the precision targeting data with the navi-
363
gation system supporting the sensor. The error budgets in Figures 3.1-28 and 3.1-30
show that the relative impact of navigation
system velocity error on the ARB system is
less than for conventional bombing.
Vehicle Management System (VMS)
Analysis Basic Flight Overall, the basic
flight functions do not require highly accurate inertial state vector information. The
inertial sensing requirements can be readily
met with an Attitude Heading Reference
System (AHRS). Pilot relief modes also require velocity information which can be
supplied by the reference INS. However,
inertial data must be provided with enough
fault tolerance and redundancy to must meet
safety-of-flight issues (Reference 5).
Air Data Backup Analysis If the Air Data

System fails, inertial data can be used to
provide an air data backup capability. Inertially derived velocity data can be combined with aircraft attitude to estimate angle
of attack and side slip angle needed for
flight control. These estimates ignore the effects of winds, and therefore would support
a degraded mode of operation (Reference 5).
Pilot Relief Modes Analysis Autopilot
functions are provided through the VMS
computer using inputs from the aircraft motion sensors and air data sensors. Autopilot
modes provided are navigation steering
mode for great circle course guidance,
ground track steering mode, attitudeheading
hold, and barometric or radar altitude hold.
The pilot relief modes are non-critical both
in terms of flight safety and mission success
(i.e. the mission can still be completed with
this mode of operation although a greater
workload may be placed on the crew) and
are readily met with a medium accuracy inertial navigation system or Attitude Heading
Reference System (AHRS) (Reference 5).
Vehicle Management System (VMS)
Analysis Integrated Modes Overall, the
integrated flight modes of the VMS do not
require highly accurate inertial state vector
information. The inertial sensing requirements can be easily met with an aided
medium accuracy INS. Inertial data must be
provided with enough fault tolerance and re-
dundancy to must meet safety-of-flight issues (Reference 5 ) .
Terrain Following/Terrain Avoidance

(TF/TA) Analysis TF/TA mode requires
both absolute navigation data and relative
navigation data. Absolute position accuracy
of 0.5 nm CEP is required for initial way
point acquisition. Velocity accuracy of 2.0
ft/sec (1 sigma) is required for use in
computation of commands for control of the
aircraft velocity vector. Relative navigation
data will be provided by use of radar and
radar altimeter to achieve low level TF/TA.
Terrain following systems that blend radar
and DTED or rely solely on DTED for terrain definition require navigation position
accuracy better than 300 ft and a heading
accuracy of 0.1 deg (References 5, 10, 15).
364
5.120fi
Steering
Required
at Breakout
Point
/ //
/
1. Velocity
Update
. Fly to
Breakout Point
(200 ft Ceiling)
3. Designate
Touch Down
Point
CC240157-84-0
Figure 3.1-31. Autonomous Precision Landing

INS Velocity
INS Velocity
INS Velocity
Error = 1.O fps
Error = 2.0 fps
Error = 0.5 fps
Cross
Along
Cross
Along
CrOSS
hack (ft) Track (ft) Track (ft) Track (ft) Track (ft)
Operator Designator Error
13
13
Cursor Related Errors
18
18
Ranging Doppler Filter Process Errors
Output Lag Error
Aircraft Altitude at MAP
44
78
Radar Velocity - Cross-Track

.______--------------INS Velocity - Along Track
Radar Velocity Along-Track
98
-----.
11
Same
Same
Same
Same
22
--------- ----- - - - - - ----11
22
22
44
44
INS Velocity Cross Track
11
11
22
22
44
44
18
36
INS Velocity Vertical

CEP - f t
117
122
141
Note 1: Pilot makes steering corrections upon breakout

cc24-0157-93-D
Figure 3.1-32. Autonomous Precision Landing Touchdown Point Accuracy
365
Autonomous Precision Landing Analysis

Navigation parameters required to support
an autonomous precision landing capability
are present position, velocity, attitude, and
heading as illustrated in Figure 3.1-31. A
present position accuracy of 2 nm CEP is
required for navigation to the landing area.
On board target sensors then locate and
designate the touchdown point. A position
error subsequently propagates due to several
error sources. This position error must be
corrected by the pilot prior to touchdown.
The integration of INS velocity errors over
time after sensor designation contribute to
3.2 Tactical Fixed W ing A ircraft Reguirements Allocation (Reference 10)

Vehicle Management System Requirements Allocation
The basic flight functions do not require

highly accurate inertial state vector information. The inertial sensing requirements can
be readily met with an Attitude .Heading
Reference System (AHRS). Based on the
results from the Advanced Vehicle Management System Architecture Study
(AVMS), a triplex fly-by-wire flight control
Error Source
Units
Refresh Rate
Body Axis Rate

Body Axis Acceleration
Ai rcraft Attitude
True Heading
Ground Speed
Ground Track Angle
Velocity (NED)
Velocity (Body Axis)
Acceleration, Vertical
deg/sec
ft/sec/sec
deg
deg
knots
deg
ft/sec
ft/sec
ft/sec/s ec
200 Hz
200 Hz
200 Hz
200 Hz
50 Hz
50 Hz
200 Hz
50 Hz
50 Hz
Figure 3.2-1: Typical INS Outputs to Flight Control Systems
position errors as shown in Figure 3.1-32.

Based on the total touchdown errors
presented in Figure 3.1-32, an INS velocity
accuracy requirement of + 1 ft/sec is
established. Attitude and heading accuracy
of 0.1 degrees (1 sigma) is required for final
approach and landing (Reference 10).
system was assumed to the most cost-effective means for satisfying flight safety requirements. The VMS must provide aircraft
stability and performance which satisfy the
requirements of MIL-F-8785 and MIL-F9490D for Class IV aircraft with Level I flying qualities in the A, B, and C flight phase
categories. Aircraft body rates and acceleration sensing must be fail-op/fail-op, survivable, and have a mission reliability of 1.0
E-10. Typical flight control data are shown
in Figure 3.2- 1.
366
CC240157-I0 5 C
Notes: All values are RMS except as noted

(1) All data shorn is besed upon a full performancegyrocompass alignment
(2) Requires IIN Navigation Updating pressure altitude aiding is used to stabilize the verlical channel
(3)lndudes digital 'body bending' filter to remove airframe cycle effects
(4) Velocities mbe lead compensated using accelerationto minimize delay (c 2.5 msec)
(5) True Heading. Platform Azimuth. and Wander Angle accuracies shall comply with the fdlwing equation
True Heading Acaaacy
>=I
[(Wander Angle
(Platform Azimuth Accuracy)*]
=d
(6) X,Y.Z Acceleration accuracy = 0.5 W d s e c

I(O.064 W d s e c y t (0.49W d s e c f 1
(7) Accuracy shall be consistent with source data (Wander Azimuth (Alpha) and Latitude)
(8) The refresh rate shall be a multiple of 25 Hz
(9) The refresh rate of the time tag is the rate of the associateddata
(10) The Aircrafl c.g. State Estimation (Reference Nav) Vector Definesthe Airwafl Rigid Body Axis
CC240167.1WO
Figure 3.2-2: Aircraft c.g. State Estimation Performance Allocations

(0.8 N d H r INS)
361
Mission Related
Top Level Functions
Events &Tasks
Flight Related
Vehicle Management Functions
Position
I FPC
Integrated Modes
0 Aided Inertial
Horizontal Vertical
Altitude velocity
Velocity
2000 ft CEP
- Flight Path Optimization

Note: All values RMS except as noted
(1) Requires pressure altitude aiding
(2) DTED position accuracy 300 ft /cell
(3)Relative altitude (AGL Above Ground Level)
(4) Driven by radar altimeter accuracy requirements
(5) Cue 5 deg x 5 deg (WFOV) FLlR at 3 NM from target with Pa.99
(6) Cue 0.67 NM x 0.67 NM SAR map with Pa.99
CC24-0157-108-D
Figure 3.2-3: "Aided" Aircraft c.g. State Estimation Performance Allocations

Table 3.2-I: Flight Control Pitch and Yaw Gyro Performance Allocations
Range
Linearii"
-+200 deglsec
Zero Offset
Threshold
Resolution
Hysteresis
Natural Frequ-ncy
Damping
Cross-Axis Sensitivity
Residual Oscillation
and Noise
0.10 deglsec
0.02 deglsec
0.02 deglsec
0.05 deglsec
>30Hz
0.5 to 1.O (-20 to +160 OF)
0.05 deglsecldeglsec
Scaling Tolerances
-+0.5% Full Scale (060% Full Scale)
-+3.0% Full Scale (50-100% Full Scale)
0.05 deglsec (0-5 Hz)

0.10 deglsec (5-20 Hz)
+5%
Linearity is defined as deviation of actual output at any point from the least
square straight line fit of measured input/out relationship
CC24-0157-58-D
368
Table 3.2-2: Flight Control Roll Gyro Performance Allocations
Range
Linearity*
Zero Offset
Threshold
Resolution
Hysteresis
Natural Frequency
Damping
Residual Oscillation
and Noise
Scaling Tolerances
+300 deg/sec
2 . 5 % Full Scale (0-50% Full Scale)
+2.0% Full Scale (50400% Full Scale)
0.80 deg/sec
0.02 deg/sec
0.02 deg/sec
0.30 deg/sec
>30 Hz
0.5 to 1.0 (-20 to +160 OF)
0.05 deg/sec/deg/sec
0.05 deg/sec (0-5 Hz)
0.1 0 deg/sec (5-20 Hz)
+5%
Linearity is defined as deviation of actual output at any point from the least
square straight line fit of measured input/out relationship
CC24-0157-59-0
Table 3.2-3: Flight Control Lateral and Longitudinal Accelerometer Performance Allocations
Range
Linearity*
Zero Offset
Threshold
Resolution
Hysteresis
Natural Frequency
Damping
Residual Oscillation and Noise
Scaling To1erances
k3G
+1% Full Scale
0.01 G
0.001 G
0.001 G
0.001 G
>30 Hz
0.5 to 1.O
0.002 G/G
0.004G
+2.5%
Linearity is defined as deviation of actual output at any point from the least square
straight line fit of measured inputlout relationship
CC24-0157-60-D
369
Table 3.2-4: Flight Control N o m 1 Accelerometer Performance Allocations

~
Range
Linearity*
Zero Offset
Threshold
Resolution
Hysteresis
Natural Frequency
Damping
Residual Oscillation and Noise
Scaling Tolerances
k12G
+1% Full Scale
0.05G
0.005G
0.005G
0.005G
>30 Hz
0.5to 1 .O
0.002G/G
0.002G
+2.5%
Linearity is defined as deviation of actual output at any point from the least square
straight line fit of measured inputlout relationship
Table 3.2-5: AHRS Accuracy Allocations

Parameter
Heading
Heading Drii
CC24-0157-61-D
Attitude
Horizontal
Velocities
I
I
I
Units
degs, RMS
degdhour,
RMS
degs, RMS
fDs.RMS
. .
I
I
Requirement
with Aiding
0.25
ria
I Requirement I
I withoh Aiding *+
I 0.25 Initial
1
3
1.5
16
INS and air data aiding

** AHRS must meet requirement for 15 minutes after loss of aiding.
Aircraft cg State Estimation Performance

Allocations
The aircraft c.g. state estimation performance allocations are shown in Figure 3.2-2,
which is representative of medium accuracy
0.8 n d h r inertial navigation system. Aided
requirements are shown in Figure 3.2-3
which require navigation updating with an
external navigation aiding sensor.
The VMS must provide aircraft stability and
performance which satisfy the requirements
of MIL-F-8785 and MIL-F-9490D for Class
IV aircraft with Level I flying qualities in
the A, B, and C flight phase categories. Aircraft body rates and acceleration sensing requirements to provide this performance are
listed in Tables 3.2-1 through 3.2-4. Inertial

sensing must be fail-op/fail-op, survivable,
and have a mission reliability of 1.0 E-10.
Equivalent AHRS allocations needed for the
flight control function is shown in Table 3.25.
370
Inertial Senslng
Functional Requlrements
Attitude (1:
Mlsdlgnment
Compensation
CC24-0157-1090
Notes:
(1) Flexure Results from Changes in the Aircraft's Reference Frame (Rigid Body) Caused by Factors such as
Maneuvers, Structural Bending Modes and High Frequency Vibration Effects
(2) Static Boresight Improves Manufacturing Boresight Tolerances
(3)Quasi-Steady State Flexure Bandwidth (0.01 - 1 Hz), Vibration Flexure Bandwidth (> 1 Hz)
(4) Transfer Alignment Requires an Inertial Reference Source, Alignment is with Respect to Inertial Navigation Frame
(Weapons Include: AGM-65 MMW Maverick, JDAM, JSOW, SLAM, AGM-86, ALCM, AIM-120 AMRAAM)
(5) Misalignment is Taken with Respect to Aircraft Rigid Body Axis (.i.e. Local Attitude = N C Attitude + AAttitude)
(6) Collocatedon Airframe
(7) Weapon Attachment Tolerances are 0.5Deg in Pitch and Yaw, 1.0 deg in Roll
(8) Vibration Stabilizedvia Local Gimbal Platform
(9) N C c.g. State Estimation (Reference Nav) Vector Definesthe Aircraft Rigid Body Axis Reference Frame
CC24-0157-110.0
Figure 3.2-4: Local State Estimation Functional Allocations
371
Latitude
Longitude
Baro-Inertial Altitude
I
I
deg
deg
ft
I+/-
180
I+/-180
I - l k t075k
25 Hz
25 Hz
50 Hz
50 Hz
AuL
-
. . .
notspecified
[ 6
notsnecifiad I
50 Hz
50 Hz
50 Hz
50 Hz
200 Hz
200 Hz
3Q.k
0.5
0.5
0.49
0.49
0.49
0.5
50
0.5
f9ll
0.0078
0.0078
0 0156
50
200 Hz
200 Hz
200 Hz
50Hz
50 Hz
50Hz
200Hz
23msec23msec
,
I
38-
notspecified
8
8
8
I 8
l e
Notes:
(1) Different vendors and motion compensation techniques require different inertial requirements, therefore some parameters
show a range of values rather than values for a particular mechanization
(2) All values are RMS except as noted
(3) Precision Velocity Updates (PVU) improves horizontal velocity accuracy to 0.5 Wsec. 1.0 Wsec needed for High Resolution
Mapping (HRM). GPS improves velocity to 0.3 Wsec. Accurate velocity reference improves SAR resolution by reducing
pointing errors under acceleration and the along track error.
(4) Accuracy = square root ( mean accuracy2 + jitter2)
(5) Requirements translate into accelerometer accuracy: 100 pg bias, 200 ppm scale factor
(6) Required for cueing
(7) Required for designation
CC249157-115-D
(8) Required for motion compensation
(9) Also known as radar harmonization
Figure 3.2-5: SAR Inertial Sensing Allocations
Local State Estimation (LSE) Allocations

Local State Estimation provides the local
motion/attitude/position inertial state vector
at sensor and weapon locations which supports static boresight, weapodsensor dynamic alignment, motion compensation,
stabilization, and the transfer alignment of
weapon Inertial Measurement Units (IMUs).
Figure 3.2-4 shows these allocations.
SAR Motion Compensation

One of the basic radar motion compensation
functions is the control and stabilization of
the antenna beam pointing. This control
function is required in all radar modes to
provide sufficient antenna illumination gains
for target detection and terrain imaging purposes. Figure 3.2-5 shows this requirement.
Electro-Optical Sensor Stabilization

The targeting FLIR provides a high resolution IR display with provides accurate
pointing information. Using other onboard
sensors, the targeting FLIR can be cued to a
designated point, or it can be manually positioned. After a specific target is identified,
the targeting FLIR can track the target by
processing LOS commands to ground stabilize the electro-optics via a gimbaled platform and using the LTD/R laser designate if
required. The Electro-Optical (EO) sensor
suite has the capability to hand-off a single,
tracked target to an IR MAVERICK, requiring only aircrew consent for launch. It can
also be used to update the reference INS by
supplying accurate position information.
Figure 3.2-6 summarizes these allocations.
372
Notes:
(1)
. . Different vendors and stabilization techniques require different inertial requirements, therefore some parameters show a range of
values rather than values for a particular mechanization
(2) All values are RMS except as noted
(3)Accuracy = square root ( mean accuracy *+ jitter )

(4) Accuracy shall be consistent with source data (wander azimuth (alpha) and latitude)
(5) Required for cueing
(6) Required for designation
(7) Required for motion stabilization
CC24-0157-1~D
(8) Requirements for gimballed platform gyro translates into 0.004Ohr (North,East). 0.015"hr (Down), 500 ppm Scale Factor
Figure 3.2-6: EO Inertial Sensing Allocations
313
Notes: All values are RMS except as noted

(1) Compensation is required to meet alignment accuracies shown (static andlor dynamic flexure)
(2) Collocated on airframe
(3)Misalignment is represented as a A pitch, A roll and A yaw taken with respect to aircraft rigid body axis
(.i.e. local attitude = A/C attitude + A attitude)
(4) Data rates, data latency, bandwidth are driven by weapon/sensor location, compensation technique
and overall system implementation. Simulation/analysis will update parameters.
(5) Static boresight is a constant by definition and only needs to be estimated once at power on
(6) Does not require static boresight compensation
CC24-0157-112-0
Figure 3.2-7: WeapodSensor Alignment Allocations

Weapon/Sensor Alignment (Pointing)
yaw and 1 degree in roll. The overall misAllocations
alignment can easily reach 10 milliradians
Mission sensor performance is dependent on
accurate inertial information and the precise
alignment (boresight) of the system sensitive
axis. Manufacturing installation tolerances
are cited at 7.5 milliradians, but advanced
mission sensor alignment requirements can
be as stringent as 0.25 milliradians. Manual
static boresighting techniques are currently
used to align mission sensors. These techniques utilize complex optical-mechanical
procedures (expensive) and ground support
personnel. This approach is only partially
effective and does not address dynamic
alignment as the aircraft structure flexes in
flight, especially during severe maneuvers.
Weapodstore manufacturing attachmw (01erances are typically 1/2 degree in pitoh and
(mils) or more. The values above are all

static, that is, they do not include the effects
of flexure (wing bending and twisting relative to a zero g condition). Wing flexure is
most pronounced at the outboard stations
and at higher g levels. Flexure can displace
the weapon by over 10 inches, change the
relative pitch angle by over 0.75 degrees,
and change the relative roll angle by over 5
degrees. The angular displacement levels
can be larger than the weapodstore attachment tolerances themselves.
Figure 3.2-7 shows the sensors and weapons
requiring angular alignment or local attitude.
The local attitude accuracy allocations are
based on experience on F-15, F-18, AV-8B,
A- 12.
314
Figure 3.2-8 Weapon IMU Transfer Alignment Allocations
375
Weapon IMU Transfer Alignment

Allocations
Weapons with Inertial Measurement Units
(IMUs) for midcourse and/or terminal guidance usually require prealignment before
launch. Transfer alignment procedures have
been developed which prealign the weapon
IMU by dynamically matching time histories
of the aircraft reference INS data with the
weapon slave IMU data. Such procedures
The challenge of transfer alignment is to

utilize the physical error models of the INS
and IMU to separate and identify the various
alignment and instrument error terms so that
they can be compensated prior to weapon
launch. Certain error parameters (notably
the alignment errors) enter the error model
as terms multiplied by the aircraft acceleration or rotation rate. This dictates the need
to provide an aircraft maneuver with sufficient acceleration amplitude and time varia-
Table 3.2-6 Error Contributors to Weapon Position Error
Initialization
Final Checkpoint
Master System
(Position Drift)
Transfer Alignment
Measurement Noise
Master System
(Attitude and Velocity
Errors)
Missile System
- Position Error at Launch Due to

Designation Errors
- Position Error at Launch Due to
Aircraft Navigator Errors
- Position Uncertainty
Between
Aircraft and Missile IMU Due to Flexure
- Attitude and Velocity Errors Transferred
From Aircraft to Missile IMU
- Quality of Missile Inertial Measurement

Unit (IMU)
External
Gravity Anomaly
Mapping
- Vertical Deflection
- Taraet to Identification Point Uncertaintv
CC24-0157-15643
are mature with extensive research and successful applications. Exceptions to aligning
the IMU prior to launch include weapon systems with Global Positioning System (GPS)
which allow an inflight alignment capability.
However, weapons with GPS still require
coarse initialization to ensure quick acquisition of satellites. Table 3.2-6 shows other
error contributors to weapon IMU position
errors.
tion so that terms can be separated. Higher

acceleration and/or rotation rates can lead to
better estimation accuracy and faster estimate times but the higher accelerations and
rates also increases aircraft flexure. Aircraft
flexure can limit transfer alignment accuracies and convergence times since the uncertainty of the "true" weapon position increases. Weapon IMU transfer alignment
performance allocations are shown in Figure
3.2-8.
376
Supports Aircraft c.g. State Estimation 'Aided' Performance Requirements

Supports lnflight Alignment Function
Supports Navigation Fault Detection, Isolation 81Recovery
Provide Measurement Residuals
Provide Kalman Filter Covariance Values
Update Types
GPS Velocity
GPS Position
Radar Velocity
Radar Position
Generalized Velocity
Generalized Position
Other Capabilities
Reasonableness Checks (i.e. Chi-Squared)
Adaptable/Flexible to Allow Cost-Effective Integration of New Sensors or Upgrades
CC24-0157-127-0
Figure 3.2-9 Navigation Updating Functional Allocations
Updating Modes
Update Accuracy
Advantages
Disadvantages
105ftCEP (6)
0.5Wsec CI)
Good Range, Relatively

Unaffected by Weather, Velocity
Update Available, Autonomous
Emissions Required
FLlR Update
33ftCEP
Good Accuracy, No RF
Emissions, Autonomous
Limited Range, Affected by

Weather
Global Positioning
System (GPS)
30 ft CEP
0.3 Wsec
(6)
Best Accuracy, No Emissions,

Jam-Resistant, All Weather,
World-Wide
Not Jam-Proof, Satellites and

Ground Stations Subject to
I Attack
~~
Terrain Referenced
Navigation
150to300ft
Low Power Emissions,

' Jam-Resistant, All Weather,
~
TACAN
1.3Kft CEP 0 10 NM
llKftCEP0100NM
Limited Map Availability, Not

Operational Over Water, Cities,
Autonomous
Widely Deployed, Available for
Rendezvous
Available in Enemy Terriiory
97ftCEP
(6)
No Emissions, Limited Avionics

Required, Autonomous
Affected by Weather, Limited

Range
Notes: All values RMS except as noted

(1) Based on 15 NM ground range, 4,800 ft altitude, 500 kts, 45 deg squint, 15 sec navigation time, 8.5 ft
(2) Based on 12K slant range, 1,OOOft altitude, 500 kts, laser ranging
(3)Based on good satellite geometry

(4) Based on accuracy of 1.5 deg one sigma in bearing, 600 ft one sigma range
(5) Based on 6K ft slant range, 100 ft altitude, level
(6) Waypoint mapping emrs not included
(7) Accuracy of Precision Velocity Updates (PVU)
Figure 3.2-10 Navigation Updating Accuracies
resolution
CC244157-12&0
371
Navigation Updating Allocations Navigation aiding sensors include Global Positioning System (GPS), pressure altimeter,
S A R , Targeting FLIR, Tactical Air Navigation (TACAN), Head-Up Display (HUD),
and a terrain referenced navigation with a
digital map such as SITAN (Sandia Inertial
Terrain-Aided Navigation) or TERPROM
(Terrain Profile Matching).
Navigation "fusion" updating consists of integrating information from multiple navigation sensors in a statistically optimum way.
This is accomplished by a Kalman filter.
Although it is a filter in the digital processing sense, it may more fundamentally be
considered a statistical algorithm which
provides an optimum estimate of the values
of certain parameters associated with a dynamic process. In a navigation system, these
parameters (or "states" of the system) consist
of various navigation errors, so that the filter
provides an optimum, ongoing estimate of
such errors, which are then used to correct
the navigation system outputs. Historically,
the software and processor for accomplishing this has resided in the INS and has been
provided by the supplier as part of the system. As navigation becomes more dependent on information from multiple sensors,
and as other systems become more integrated with the INS, it is not obvious that
this location will continue to be appropriate
for the Kalman filter.
The navigation system updating functional
allocations are shown in Figure 3.2-9. The
ability to obtain periodic navigation system
updates during the course of the mission is
extremely important to overall mission
effectiveness. However, mission scenario
and crew workload considerations may place
constraints on the frequency and quality of
the navigation update. Several update
modes have been quantified in terms of
accuracies, advantages and disadvantages as
shown in Figure 3.2-10.
4.0 Conclusions
This paper has reviewed the current state-ofthe-art of navigation system technology for
fixed wing aircraft, suggested a
methodology to decompose the mission of
the particular aircraft in order to determine

requirements and a methodology to apply
systems engineering to determine the
detailed requirements of the system to the
particular set of mission segments, sensors
and functions required of the system. Some
examples of the analysis required for several
functions such as sensor queing, targeting
and weapon delivery have been shown. The
results of the example analysis have been
shown in tables throughout section 3.1 and
may be of some use to the reader for
application to his or her particular need. It
would be difficult to establish one set of
requirements for all applications, therefore
the goal is to provide the reader a basis for a
methodology to follow in his or her analysis.
Section 3.2 provides a . typical set of
requirements allocations for fixed wing
aircraft as a baseline system that can be used
as a reference by the reader. The references
called out in the text and outlined in the
reference section should serve as a more
detailed basis for the reader to use in his
application.
378
Definitions
Several error measurement terms are used in this paper for specifying accuracy requirements
and/or bombing errors. Definitions of each are given below. The error distribution is assumed to
be linear and normal (References 1,4,9, 10 and 17).
Standard Deviation (Sigma): The square root of the mean of the squares of the deviations (68.27% of all errors occur within the standard deviation limits of + or - 1 sigma for
a normal distribution).
Mean Deviation or Probable Error (PE): The sum of the absolute deviations divided
by their number. There is equal probability that the error will be either larger or smaller
than this value(50% of all errors will not exceed this value). Probable error is commonly
used to express weapon impact error.
Range Error Probable: The probable error measured along the range axis (along the
weapons flight path).
Deflection Error Probable: The probable error measured perpendicular to the weapons
flight path (cross axis).
Circular Error Probable: A two dimensional measurement of error defined as the radius of the circle for which 50% of the errors fall within its boundary. In practice, the
circle is centered at the true position of the target, which is equal to the mean in the absence of systematic errors.
319
References
1. Kayton, Myron and Fried, Walton, "Avionics Navigation Systems", John Wiley and Sons
Inc., New York, NY. 1969
2. Thompson, Dan and Schrank Michael, "Advanced Vehicle Management System Architectures Study", Digital Avionics Conference, October 1991, IEEE Proceedings.
3. Kelley, Ronald, "Integrated Inertial Networks", Joint Services Data Exchange Conference,
October 1992, JSDE Proceedings.
4. US Naval Oceanographic Office, "Navigation Dictionary", Pub. No 220, Washington, DC.,
1956.
5. Gault, Kenneth, et al, "Advanced Vehicle Management System Architecture Studies Final Re-
port", TR-WL-92-3106, September 1992.

6. Defense Systems Management College, "Systems Engineering Management Guide", January
1990.
7. Savage, P. G., "Laser Gyros In Strapdown Inertial Navigation Systems", IEEE Position Location and Navigation Symposium, San Diego, California, Nov. 1-3, 1976.
8. Bevington, J. E. and Marttila, C. A., "Precision Aided Inertial Using S A R and Digital Map
Data", IEEE Position Location and Navigation Symposium, IEEE Proceedings, 1990.
9. Greenwalt, C. R., "Principles of Error Theory and Cartographic Applications", Aeronautical
Chart and Information Center Technical Report No. 96, Feb. 1962.
10. Kelley, Ronald et al, "Integrated Inertial Network", WL-TR-94-1012, Volume 1, Final
Report, January 1994.
11. Martin, E. H., "GPS User Equipment Error Models," Journal of the Institute of Navigation, Vol. 25, No. 2, 1978.
12. SS-GPS-300, "System Specification for the NAVSTAR Global Positioning System"
13. SNU 84-1, "Specification for USAF Standard Form, Fit and Function Medium
Accuracy Inertial Navigation Unit", Aeronautical Systems Division, Air Force
Systems Command, Wright-Patterson AFB, OH, 28 February 1986.
14. "Procurement Specification For Inertial Navigation Set", PS 68-870204, McDonnell

Aircraft Company, Revision B, 30 November, 1987.
15. Burns, R. C., et al, "Strapdown Performance Study," AFWAL-TR-82-1074,
Volume 1, July 1982.
16. Interface Control Document (ICD-GPS-200), "NAVSTAR GPS Space
SegmenVNavigation User Interfaces, December 1986.
17. Beck, G. R., et al, "GPS Exploitation for Precision Targeting Study Final Report"', WL-TR94-1068.
380
18. "Specification for Global Positioning System Integrated Navigation Assembly (GINA)",
Naval Command, Control and Ocean Surveillance Center, Warminster, PA.
19. SME-EGI-92, "Systems Requirements Document for an Embedded Global Positioning
System Receiver in an Inertial Navigation System (EGI)",Rev-2,20 January, 1994.
38 1
REPRESENTATIVE VEHICLE IMPLEMENTATION
ROTARY WING AIRCRAFT
JOHN NIEMELA
DAVID F. LIANG
U.S.
NATIONAL DEFENCE HEADQUARTERS
ARMY COMMUNICATION-ELECTRONICS COMMAND
RESEARCH & DEVELOPMENT CENTER
OTTAWA, ONTARIO
FORT MONMOUTH, NJ
CANADA
USA
1. INTRODUCTION
Rotary wing aircraft have seen an ever expanding role in
support of land and sea military operations including:
search and rescue, scout, attack, troop transport, antisubmarine warfare, anti-surface ship targeting, cargo and
electronic warfare. They are ideally suited for operation
in confined and unprepared areas where no other form of
aerial transport is suitable.
For a helicopter to survive on the modem battlefield,

flight profiles are mandated that impose unique
requirements on the navigation system. These include
contour and nap-of-the-earth flight as well as hover in
defilade in close proximity to obstructions. Critical
mission segments are conducted in nap-of-the-earth flight
during which the pilot conceals the helicopter with
terrain, foliage and buildings. These missions must be
accomplished in all weather and visibility conditions. In
the course of these mission, the rotary wing aircraft crew
must maintain accurate self-location to maintain
geographic orientation and situational awareness relative
to friendly and hostile forces. A typical tactical mission
profile overlaid on a contour map is shown in Figure 1,
illustrating the non-linear flight path which takes
advantage of terrain masking [l].
For the anti-submarine warfare missions the helicopter

navigation system must maintain stable and accurate
tactical plots over long periods of time. In the antisurface ship targeting role, a high degree of absolute and
relative navigational accuracy are vital to rapid and
successful action. There are further complicating factors
as well. Operations must often take place under radio
silence and shorebased or satellite navigation aids may
be destroyed or jammed during wartime. The small
crew of the helicopter must not be burdened with
monitoring the functioning of, or updating, the
navigation system.
2. TACTICAL HELICOPTER NAVIGATION

In general, two accuracy related operational aspects drive
the tactical helicopter configuration. The first of these is
for the helicopter aircrew to maintain geographic
orientation during low altitude flight. Geographic
orientation assures point-to-point navigation, rendezvous

with friendly forces, and avoidance of known enemy
threats. To maintain geographic orientation the
helicopter aircrew must be able to identify natural or
cultural features and correlate them with a hand-held or
electronic map. Failure to do so may jeopardize the
mission, in that the helicopter may not a m v e at its
destination in a timely manner, inadvertently overfly a
hostile threat or impact unexpected terrain or obstacles.
A hundred meter navigation error, when displayed on a
paper or electronic map at an appropriate scale
(nominally 1:50,000),could direct the helicopter aircrew
up the wrong streambed or draw when operating at
nap-of-the-earth altitudes.
The metric, height above terrain divided by forward
velocity, has been applied to aircrew geographic
orientation and is proportional to the time available for
the aircrew to futate, identify and correlate a ground
feature. Considering the limited number of features that
the aircrew sees at any one instance when operating at
nap-of-the-earth height (5-10 m AGL) and their nominal
velocities (20-50 kn) the metric is comparable to a low
level tactical fighter at near supersonic velocity.
The second accuracy driver for the tactical helicopter is
for target location. Many scout and attack helicopters
have weapon direction systems that locate targets relative
to the aircraft. It is the function of the navigation system
to provide data to register these locations to target
coordinates on a geographic grid. The navigation
position error of the helicopter adds to the target location
error. The system error must meet ordinance related
miss distances after being handed over to weapon system
elements. Twenty-five meter position accuracy meets a
large class of such requirements when considered in the
context of system error budgets.
The nominal position accuracy requirements that arise

from consideration of the above discussed requirements
for tactical helicopter are 25 meters when external aids
are available or 0.5% to 1.0% of the distance traveled
for the self-contained system. The accuracies are
typically specified at the ninety-five percent (95%)
Probable Error Level. Note that if the external
navigation reference is not available, the aircrew must
382
manually update the self-contained navigation system

fairly frequently (every 5-10 km) to maintain geographic
orientation as well as maintain position accuracy
sufficient for weapon direction.
aircraft and may be unavailable during wartime.
** It is recognized that these performance levels can be

improved on by a large margin if, as expected, a GPS is
part of the navigation system configuration.
3. MARITIME HELICOPTER NAVIGATION

3.3 ALIGNMENT CONSIDERATIONS
For maritime military helicopter applications, fight

duration tends to be rather extensive, therefore the
performance accuracy of the navigation system depends
on the flight profile specified and the type of radio
equipment available. In general, maritime helicopter
operations can be represented by the following two flight
profiles:
For maritime military helicopter applications, one of the

key navigation requirements is the appropriate alignment
of the INS, which requires initialization of position,
velocity and attitude data. When GPS is available as an
alignment reference, both the fmed base and moving base
alignment can be readily accomplished, since GPS will
provide an excellent position and velocity reference.
3.1 CONVOY SCREENING MISSION PROFILE

The helicopter searches a moving sector ahead of a
convoy of ships. The profile is normally characterized
by alternative short cruise phases and hover phases. The
hover phases represent periods in which the helicopter is
using a dipping sonar to detect threats in the vicinity of
the convoy. Figure 2 shows the plan view for the
convoy screening mission.
3.2 SONOBUOY MISSION PROFILE
This mission includes a constant heading cruise from the

takeoff point to the vicinity of a target. The mission
profile undergoes a sonobuoy deployment phase,
followed by a Magnetic Anomaly Detection (MAD)
phase and a weapon delivery maneuver. Return to the
mother ship consists of another extended cruise phase.
Total mission duration can exceed several hours. Figure
3 shows the plan view and figure 4 shows the altitude
profile for the sonobuoy mission.
With these two mission profiles in mind, the operational
accuracy requirements can be specified as follows:
Radial Position Error (95%):
- with external aids* 2.0 nautical miles (nm)
When GPS is not available, during the initialization

process, the initial position can be inserted manually or
automatically. For a fmed base initialization, the initial
velocity can be assumed to be zero; but the same cannot
be assumed for a moving base initialization. For a
moving base, a velocity reference is required or else it
must be computed from a sequence of reference
positions. Attitude alignment can be inferred from the
accelerometer measured specific force. However, the
most difficult task is the initial heading alignment, which
can be accomplished by several alternative methods.
Heading alignment can normally be divided between the
so-called coarse and fine alignment. For a fixed base,
the coarse alignment can be accomplished with wide
angle gyrocompassing, which attempts to calculate body
roll, pitch and heading from gyroscope and
accelerometer measurements in body coordinates. For
an in-motion coarse heading alignment, the true heading
can be derived from a magnetic compass or
magnetometer.
In gyrocompassing alignment, reference velocity or
position measurement noise error is a fundamental source
of error that determines the time required for alignment.
Therefore for a rapid alignment, the reference
measurement noise must be very small, which is the
condition normally achieved with a fixed base alignment.
- without external aids 1.5 nm/hr

Radial Velocity Error (95%):
- with external aids**
3.0 fi/sec
- without external aids 4.0 ftlsec

Attitude Error (95%): 0.5 deg
Heading Error (95%): 0.5 deg
* External aids are those systems such as Omega, Loran

and the Global Positioning System (GPS) which rely
upon transmitters which are located external to the
For in-motion alignment, the most rapid alignment is

achieved with transfer of alignment from the ships
inertial navigation system. For this, very small short
correlation measurement error must be achieved, which
is difficult to accomplish for a smaller size ship. The
next rapid alignment method requires the ship to
maintain steady course and speed and uses inserted
course and speed as the velocity vector reference.
However, because of the surge and sway of the ship, the
inserted constant velocity has significant short correlation
error. The accuracy of this method is limited by the
gyro drifts and by the bias error of the inserted velocity.
383
When doppler aiding is available, the most prominent

source of error is the doppler velocity long term
correlation error due to waves and current effects.
However, when shipboard coarse alignment is used in
conjunction with infight doppler aiding, it is not
necessary to have long duration shipboard alignment
to achieve excellent integrated navigation performance.
surface vessels requires that the navigation system must

be compatible with the electromagnetic interference
experienced in that environment.
4. PILOT VEHICLE INTERFACE
The system must be capable of providing a compatible

navigation input to an instrument letdown in Instrument
Meteorological Conditions (IMC) acceptable to civil
authorities for operation at civil airports.
The means by which the navigation data, generated by

the sensors and processed by computer configuration to
the required accuracy, is displayed to the aircrew is
subject to many constraints. Typically navigation
displays provide the following capabilities:
(1) Position in geographic

reference (latitude,longitude)/UTM/tacticalgrid
coordinates.
For rapid response missions, the tactical helicopter,

including its navigation system, must be operational
within two minutes of power application.
In addition to the above, as with all military avionics

systems it is desired that the helicopter's navigation
capability be realized with minimum impact on aircraft
system cost and gross weight as well as be modular in
construction to facilitate system growth and maintenance.
6. ARCHITECTURE
(2) Steering, bearing, track display, distance and

time to go.
(3) Look ahead to destination, and drift angle.
(4) Groundspeed and track.
(5) Windspeed and direction.

(6) Position updating including "on-top", manual and
offset.
(7) Manual slew of map image

(8) Display freeze.
(9) Map marking.
(10) Ability to permit the aircrew to view a continuous

tactical map display of various scales (typically
1:50,000)over which the mission is being flown.
A typical rotary wing requirement definition considers

the several factors discussed in the introduction and
includes extensive analysis and simulation to determine
the class of sensors and means of integration. With a
fair degree of consistency, a set of navigation sensor
types and integration concepts are widely used to meet
helicopter mission requirements. Aircrews require a
position location and navigation system that will permit
accurate navigation worldwide when operating at terrain
fight altitudes under all visibility and meteorological
conditions. The system must b e accurate, low-cost,
lightweight, and provide anti-jam protection as well as
positional information in Latitude/Longitude and in the
Military Grid Reference System (MGRS) o r the
Universal Transverse Mercator (UTM) coordinates. The
system shall have the capability to store multiple
waypointddestinations and display azimuth and distance
to these locations.
(13) Allow for map orientation as desired by the

aircrew.
In general, the desire to have at least one mode that is

not dependent on external radio navigation aids has led to
the requirement for a self-contained navigation mode - a
mode in which the navigation system is not dependent on
external radio navigation aids. The need for a
self-contained mode is driven by the operational and
technical concerns:
(14) Alternative displays (e.g., heads-up, etc.) providing
' Externally referenced systems may be susceptible to
(11) Portray three dimensional topographic perspective.

(12) 100 waypoint entry and storage.
required tactical information shall also b e

acceptable.
(15) Displayed navigation information should be

compatible with the aircrew's night vision and target
acquisition systems.
5 . OTHER MILITARY REQUIREMENTS
Operation off the deck or in the proximity of NATO
enemy electronic countermeasures and physical

destruction. External sensor input may be susceptible
to jamming o r spoofing and in times of crisis or war,
may not be fully usable o r continuously available.
' Externally referenced systems may be affected by

natural phenomena and man-made obstructions.
Ground based navigation aids that require lineof-sight
reception do not provide the required capability at
terrain fight altitude.
384
Hence, helicopter navigation requirements are typically

met by the integration of self contained and externally
referenced navigation sensors and modes. The
combination of these two navigation sensors create a
hybrid navigation system (e.g. GPSlinertial,
GPS/doppler/ AH RS , GPS/doppler/inertiaYterrain
referenced navigation (TRN). The sensors which
comprise such hybrid navigation systems have
complementary error spectrums and/or data availability
characteristics which yield a more accurate and reliable
navigation system.
interrupted o r lost.
In the complete absence of GPS, the INS performance
can be aided using Doppler velocity subsystem.
However, the INSlDoppler performance over the sea
depends on initial alignment; sea bias effects calibration;
as well as on the surface wind effects compensation.
For maritime military helicopter applications, the hybrid
navigation systems can provide the following additional
alignment capabilities:
7. NAVIGATION MODES AND SENSORS
(1) Fixed-base ground alignment mode,
7.1 HYBRID NAVIGATION MODES
(2) Shipboard Transfer alignment mode,
Typical hybrid navigation systems aboard military

helicopters are comprised of a set of self-contained
(doppler and inertial) and externally referenced
navigation sensors (GPS and TRN). These sensors are
configured via standard data busses and Kalman filters to
yield system performance and integrity appropriate to the
particular helicopter mission.
(3) Shipboard GPS-aided alignment mode,
The benefits of hybrid navigation systems are numerous.

The sensor data integration software can perform the
following tasks.
The self-contained mode is typically obtained with a

doppler navigation radar coupled with a accurate heading
reference. In instances where a more accurate heading
reference is needed, such as for weapon direction, a
dopplerlinertial hybrid configuration is employed. The
doppler navigation system operated with an accurate
heading reference, is particularly suitable for low
altitude, slow flying air vehicles as the tactical
helicopter, as compared to a stand-alone inertial
navigation system. This is because a doppler navigation
error propagates in proportion to distance traveled
whereas inertial propagates error in proportion to time.
Considering the respective stateof-the-arts of doppler
and inertial technology, and the relatively low speed that
a tactical helicopter achieves over the battlefield, leads to
the conclusion that doppler navigation is a more accurate
means of providing self-contained navigation for this
category of aircraft. For example, a doppler properly
integrated with a good quality heading reference can
yield one-half percent of distance traveled navigation
accuracy independent of the helicopter's velocity. To
achieve an equivalent level of navigation accuracy with a
stand-alone inertial would require one-quarter of a
nautical mile per hour performance - a more costly
alternative. This trade-off is portrayed in Figure 5 which
shows that for the tactical helicopter critical mission
segment of 20 - 50 kn at nap-of-theearth flight, one
percent of distanc,e traveled performance of a
dopplerlheading reference is equivalent to a 0.3 NM/HR
inertial system. As respective advancements have been
made in doppler, heading reference and inertial system
technology the performance of all have improved.
However, the general observation that a doppler with a
good heading reference can outperform a stand-alone
inertial system in the helicopter environment is as valid
a. Integrate the available navigation input data from

the sensor subsystems. Appropriate blendings of
sensor data can lead to rapid alignment sensor
calibration. Optimum navigation accuracy
and enhanced reliability through redundancy and
navigation.
b. Continuously calculate the appropriate gains to weight
the information provided from the different sensor
subsystems to generate real-time gain calculation.
c. Calculate the measurement of uncertainty (covariance)

in the estimate of the system navigation states and the
individual subsystem states.
d. Perform system failure detection and isolation as
well as "outlier" editing to detect and isolate both
hard and soft failures in the sensor andlor computer
subsystems.
e. Perform automatic system reconfiguration to remove
the failed subsystem while integrating the remaining
navigation sensors.
When the on-board navigation systems include both the

GPS and INS,the integrated navigation solution can
provide excellent position, velocity and attitude accuracy.
As soon as GPS has achieved its position fm,the
integrated position and velocity solution will be excellent,
and the well calibrated inertial system accuracy will
remain sufficiently accurate, should the GPS reception be
(4) Airborne GPS-aided alignment mode, and
( 5 ) Airborne Doppler-aided alignment mode.

7.2 SELF-CONTAINED MODES
385
today as it was when noted by T. J. Penfold in 1975 [2].
7.5 GPS RECEIVER.
Additional characteristics for helicopter doppler

navigation technology that have made it an attractive
means of navigation is that the velocity data is very
useful for air to ground fire control and for hover
stabilization. This feature is of particular importance for
the anti-submarine warfare mission where the helicopter
must hover over a field of sonabouys. Also its rapid
reaction time, typically less than two minutes, is a useful
characteristic for tactical operations.
This element senses the helicopters position, velocity and

time in earth centered - earth ftxed coordinated and
provides a statistically stationary navigation accuracy (not
time or distance dependent) to bound the error of the
self-contained portion of the navigation system. Tradeoffs and consideration on GPS integration for tactical
helicopter is given in Ref 2.
Extensive experience with doppler navigation systems

have shown them to be a reliable and cost effective
means of providing a self-contained navigation capability
for tactical aircraft, particularly helicopters.
Vulnerability to ECM is minimal due to the frequency of
operation (13.3 Ghz), low radiated power, and very low
side lobe radiation resulting in a low RF signature
relative to other signatures such as infrared and acoustic.
This element senses height above terrain for display to

the aircrew and for use by on-board TRN equipment.
Nominal accuracies of 0.5 - 1.0 m RMS bias plus a 2 3 % altitude dependent scale factor error.
7.6 RADAR ALTIMETER.
7.7 NAVIGATION PROCESSOR.
A typical configuration is shown in Figure 6 which

portrays the principle sensing, processing, and control &
display functional elements as well as primary data
interfaces. The functions of each of these elements are
as follows:
This element combines the nav sensor data through use

of a recursive statistical filter. In addition, it computes
bearing and distance to waypoints, waypoint storage,
magnetic variation and navigation moding. For a hybrid
navigation system, the functions of the navigation
processor expand considerably to include sensor detection
isolation and system reconfiguration.
7.3 ATTITUDE & HEADING REFERENCE.
8. CONCLUSION:
This element senses the roll, pitch and heading of the

helicopter which when combined with aircraft body axis
velocity from the doppler provides the self-contained
navigation solution. A secondary function is to provide
data to aircraft flight instruments, flight and fire control
systems. Ring laser gyros have been found to be
superior to spinning mass gyros, particularly for
strapdown AHRS application due to greatly reduced scale
factor and gyro axis alignment stability errors [3].
Typical navigation suites aboard tactical helicopters are

significantly different than those of fmed wing fighter or
transport aircraft. This is due primarily to the difference
in flight profile between these categories of aircraft. The
lower and slower flight profile of the helicopter favors a
low cost inertial - doppler ensemble for the self-contained
navigation function whereas a medium to high accuracy
inertial system is in general favored for fixed wing
aircraft. In addition, the translational and rotational
degrees of freedom of the helicopter are typically an
octave more benign then that of high performance fighter
aircraft. This allows adjustments to be made in
integration of the Kalman filter to achieve improved
performance in the helicopter environment.
For utility and cargo helicopters this function is typically

realized with a magnetically slaved heading reference or
pendulously erected AHRS. Nominal accuracies are
1" RMS - heading; 0.5" RMS - pitch & roll. For scout
and attack helicopters in which heading accuracy can be
directly related to targeting accuracy, a strapdown
AHRS, which employs gyrocompassing techniques for
alignment, is typically employed. Nominal accuracies
are 0.3" - 0.5" RMS heading; 0.15" - 0.25" RMS pitch
and roll.
7.4 DOPPLER VELOCITY SENSOR.
This element senses the helicopter translation velocity
over land or water for use by the navigation processor in
generating the navigation solution. A secondary function
is to provide ground reference velocity data for air to
ground fire control, and hover stabilization. Nominal
accuracies are 0.25% of velocity plus a bias of 0.1 kn.
386
REFERENCES
[l] Coffee, J. N. "GPS - INS Integration Requirements
for Robust NOE Helicopter Navigation: american
Helicopter Society 47th Annual Forum, 6 8 May 1991.
[2] Penfold, T. J. "Medium Accuracy Low cost

Navigation System for Helicopters", AGARD Conference
Proceeding No 176 on Medium Accuracy Low Cost
Navigation, September 1975.
[3] Niemela, R. J., et a1 "Helicopter Flight Test of a
Ring Laser Gyro Attitude and Heading Reference
System". Proceeding of the Institute of Navigation
National Technical Meeting, San Diego, CA
17 -19 Jan 1984.
387
Nap-of -the-Earth
Flight
Area
-5
i
-5
15
10
20
2s
Distance East (Km)
FIGURE 1: TYPICAL TACTICAL HELICOPTER MISSION PROFILE
f
-IZr - 3
N'--l
N
CPF
f
/
FIGURE 2:
PLAN VIEW OF CONVOY SCREENING MISSION
388
N4-
I
I
I
Sea King Replacement (SKR) Course
-- -
CanadIan Patrol Frigate C P F l Course
- - - - ~ub-rurlaceContact Course
4
FIGURE 3:
Sonobouy
Torpedo Attack Zone
NM
Naullcal MI19
PLAN VIEW OF SONOBUOY MISSION
SKA Altltud. Profile
c=]
FIGURE 4 :
Canadian Patrol Frigsta
ALTITUDE PROFILE FOR SONOBUOY MISSION
389
21
\ \
e 1.2
Z
l - 1
;0.8
LL
5 0.6
W
0.2
----.-
\ E/
1 '
VELOClTY
SITUATION
SENSOR
E+
VERTICAL
SITUATION
HDG REF
INDICATOR
POSITION
COWL
DATA LOADER
DlGlTAL DATA BASE
DISPLAY
ON/OFF
MOOING
RAD ALT
SENSORS
FIGURE 6:
PRDCESSlNG
TYPICAL TACTICAL HELICOPTER NAVIGATION SUITE
390
Spacecraft Navigation Requirements

Judy L. Miller
Charles Stark Draper Laboratory, Inc.
555 Technology Square
Cambridge, MA 02139
USA
1. S U M M A R Y
Spacecraftoperation depends upon knowledge of vehicular position and, consequently, navigational support has
been required for all such systems. Technical requirements for different mission trajectories and orbits are
addressed with considerationgiven to the varioustradeoffs
which may need to be considered. The broad spectrum of
spacecraft are considered with emphasis upon those of
greater military significance (i.e., near earth orbiting satellites.) Technical requirements include, but are not
limitedto,accwacy;physicalcharacteristicssuchas
weight
and volume; support requirementssuchaselecmcalpower
and ground support; and system integrity. Generic navigation suites for spacecraft applications are described. It
is shown that operational spacecraft rely primarily upon
ground-based tracking and computational centers with
little or no navigational function allocated to the vehicle,
while technology development efforts have been and
continue to be directed primarily toward onboard navigation suites. The military significance of onboard navigators is shown to both improve spacecraftsurvivabilityand
performance (accuracy).
2. INTRODUCTION
The diversity of space missions and spacecraft presents a
wide range of technical requirements, and it is useful to
consider requirements in terms of spacecraft types. After
a brief description of the space environment, spacecraft
will be categorized by mission characteristics and flight
regime. Flight trajectories are typically comprised of
some combination of powered flight, coast, and aerodynamic flight segments. Technical requirements fordifferent mission orbits will be presented and applicable navigation suites identified. A brief discussion of orbital
mechanics is included to introduce concepts and riomenclature related to orbital navigation.
Following a description of spacecraft classes and missions, technical requirements and figures of merit for
requirements specification are presented. With an understanding of the space environment, vehicle classes designed to operate within that environment, and technical
requirements, we will then be in a position to describe
state-of-the-artgeneric navigation suites for spacecraft
Finally, the relationship betwcen technical requirements
and generic navigation suites is characterized with tradeoffs considered.
3. SPACECRAFT
For most space missions, the spacecraft physical configuration changes at discrete time intervals during the mission as hardware is separated and jettisoned. Using an
example of boosting a surveillance sensor (the payload)
into earth orbit aboard a satellite, the final spacecraft
configuration consists of sensor and spacecraftbus where
the spacecraft bus provides support such as attitude con-
trol and electrical power for the payload. Initially, however, the spacecraft configuration as erected at the launch
pad is a stack consisting of launch vehicle, upper stage,
spacecraft bus, and payload. While the launch vehicle is
active, its effective payload consists of the upper stage,
spacecraft bus, and sensor. Similarly,afterlaunch vehicle
scparation and upper stage activation, the effective payload is spacecraft bus and sensor. Therefore, in this
example there are three vehicle configurations involved
with support of this single payload and, as we shall see,
navigation requirements and suites are considerably different among these vehicles. Prior to considering technical requirements and navigation suites for spacecraft, it is
useful to characterizethe space environment and vehicles
designed for space operation.
3.1 The Space Environment
To describe the space environment. we consider fifst the

boundary of space. The question of where space begins
does not have a unique answer. Propulsionengineersmay
consider space to begin at 45 kilometers since this is the
altitude at which a vehicle must provide its own oxidizer
along with fuel. Aeronautical engineers may consider
space as the altitude above which aerodynamic forces
acting on a vehicle are considered negligible, which
occurs above 100 kilometers. At approximately 160
kilometers altitude, there is insufficient air to transmit
sound waves and scatter light so the region remains in
silence and darkness. For navigational purposes, an
altitude of 160 kilometers may be selected.
The abundance of space enables a wide range of trajectories which may be categorized according to: (a) surfaceto-surface, @) surface-to-space,(c) space-to-space, and
(d) space-to-surface. These mission trajectoriesare suffciently different to have resulted in development of several space vehicle classes. These classes, in turn, possess
distinct technicalrequirementsandemploy differing navigation suites. We initially consider a broad spectrum of
space vehicles and subsequenlly focus primarily upon
those with greater military significance.
3.2 Space Vehicles
As indicated previously, the term spucecrufi refers to that
collection of unmanned and manned vehicles designed for
operation within the space environment. Sounding rockets have surface-to-surfacetrajectories. Surface-to-space
trajectory spacecraft consist of launch vehicles for boosting satellites from the earth surface into orbit, although
booster components themselves fall back to the earth
surface. Trajectoriesremaining within the space environment include upper stages for transfemng satellites from
parking orbit to final mission orbit, earth orbit satellites
which remain in stable orbit around the earth, and interplanetary spacecraft. Vehicles with space-to-surfacerrajectories include descent, landing, and recovery vehicles.
39 1
Surface-to-surfacetrajectory vehicles are characterized

by booster rocket stages to provide high lhrust over a short
time interval (near-impulsive) for payload delivery into a
high altitude ballistic trajectory. Sounding rockers pm
vide a short duration space environmentand are typically
used for environmental data acquisition while in the upper
atmosphere OT short-term ( 4 5 minutes) microgravity
environmentalconditions for scientific experiments. The
trajectory objective is primarily to establish a contfolled
vehicle attitude so sounding rockets are typically stabilized but usually unguided. These vehicles can be tracked
by ground stations and additional terrestrial-based assets
such as aircraft, and do not typically require onboard
navigational data
b u n c h Vehicles provide propulsion and trajectory control

to boost a payload into a specified orbit. These vehicles
impart nearly all (but usually not quite all) the impulse
required and are usually ballistic. Although typically
ignited from a vertical orientation at a futed site launch
pad, one recently developed vehicle (Pegasus) for smaller
payloads is released horizontally from an aircraft undercarriage. Several missions flown to date have released the
launch vehicle from a B-52 aircraft at approximately
40,000 feet Following release, Pegasus employs a two
stage rocket to propel the payload into low earth orbit
Boosters are either expendableor reusable. Current U. S.

expendablesare derivatives of ballistic missiles originally
dating back to the 1950s. and this trend has been continued with e n t refurbishment of Titan I1 missiles as
launch vehicles for military satellites. Consistent with
their ballistic missile heritage, most expendable launch
vehicles use inertial guidance systems. Radio guidance is
still in use by some older Atlas vehicles (Atlas E4F
vintage), but these vehicles are no longer in production.
The sole reusable vehicle in the U.S.inventory is the
NASA Space Transporntion System (Shuttle) which is

a h the only U. S. manned spacecraft in operation. The
Shuttle is launched into orbit in the traditional manner but
atmospheric reentq is as a winged glider using aerodynamic surfacecontrolfor cross range maneuver capability
of approximately 2OOO km during unpowered descent to
runway landing. The Shuttle mission profile is shown by
Figure 1.
During ascent, three redundant gimballed inenial measurement units (IMU) are used for navigation. Theat are
two orbital navigation methods depending upon whether
vehicle operation is powered flight or orbital coast. During coast, a ground uplinked state vector is propagated
forward in time with force models for gravity, aerodynamic drag, reaction control system jet firings, and waste
vents. In the presence of thrust, the navigation method is
switched to propagate the trajectory using accelerometer
data and a less sophisticated gravity model. Since aII state
vector data are derived from the IMU,disturbing acceleration models, and propagating the initial conditions, position accuracy degrades over time and groundderived
state vector data must be uplinked periodically. Scheduling of these uplinks depends upon the accuracy needed to
support the mission timeline and flight operations.
The reentry navigation suite consists of IMU,tactical air
navigation (TACAN). microwave scan beam landing system (MSBLS), and radar altimeter. Navigation aid usage
during descent and landing is shown by Figure 2. Shuttle
flight test has demonstrated an ability to detect TACAN
from low altitude orbit, and on-orbit navigation software
could be modified to incorporate this measurement type
onboard. There is also interest in integrating a Global
Positioning System (GPS)receiveronbaardShuttlewhich
could be used during both on-orbit and descent mission
phases.
SOURCE: Van Leeuwen. A. Rosen E.,and Carrkr L he Global IbsLtlOnSystem and Its
Appflcatbn fn SpQcecmj N a u Q a t W Navlgatlon. ION Global PoslUonlng System Papers. Vol. 1.
Figure 1. Shuttle mission profile.
392
L-band
C-burd
15
10
phases of Shutlleoperation. Deployment from the Shuttle

payload bay is typically scheduled for 10 to 35 hours after
launch so the IUS inertial navigation system is subject to
accuracy degradation due to instrument errors and disturbing accelerations. The navigation state is typically
updated prior to deployment by the Shuttle to eliminate
error resulting from uncompensated atmospheric drag
effects on the Shuule vehicle. The IUS attitude can be
updated predeployment with a star scan technique. The
Titan trajectory is direct orbit insertionof IUS so the coast
period is short and inertial navigation is adequate.
Ku-band
I
20
I5
A l l i l U d s @In)
SOURCE:
ShutUo A "Ionla s
r.N A S A SP-504. 19BS.
Figure 2. Shuttle descent and landing navigation aids.

Vehicles used to transfer satellites from one orbit to
another are referred to as upper stages or, equivalently,
orbit transfer vehicles (OW). Upper stages provide
transfer from the launch vehicle established orbit to final
mission orbit. Most upper stages are inertially navigated
and guided. An exception is the Payload Assist Module
(PAM-A and -D) which is unguided. The PAM provides
payload transfer from parking orbit to geosynchronous
transfer orbit (GTO).The absenseof radio guided upper
stages may be attributable to poor geometry between
futed, ground-based control centers and upper stages
during the bum. Geostationary satellites are a primary
user of upper stagesas are interplanetaryspacecraftwhich
rely upon the transfer vehicle to impart sufficient velocity
for escape from the earth sphere of influence.
The InertialUpper Stage (IUS) is compatible with Shuttle
and the Titan 111 and IV launch vehicles. For Shutlle
configuration, the IUS inertial system propagates prelaunch alignment through launch and during on-orbit
J4;
SOURCE:
-&
Earth orbital spacecrafi operate at altitudes sufficiently

above the sensible atmosphere to maintain stable orbit.
We shall use the term satellite to refer to earth orbital
spacecraft. Satelliteshave historically performed support
rolesfor strategicmilitary systemsandmorerecentlyhave
emerged as a force multiplier for tactical operations Wef.
11. Military satellite missions include surveillance, meteorology, navigation, and communication. Mission scenarios for typical low earth orbit (LEO)and geosynchronous earth orbit (GEO) satellites launched by an expendable launch vehicle are shown by Figures 3 and4, respectively.
Satellite navigational state informationrequirements are
derived from two sources. The first requirement is the
need for predictive navigational state informationto support operations such as scheduling of ground contacts,
radiofrequency (RF) signal acquisition and uacking, and
command and control. These operational requirements
can be satisfied with moderatenavigationalstate information accuracy. The other requirement is for support of
payload data processing such as data location and analysis. Payload users may require high accuracy such as for
image data registration in geographic coordinates, but
calculation of satellite position at the time of data collection may be accomplished long after-the-factusing precision orbit redetermination techniques as described later.
Hock 5D-2*tern
Anatysh & p a t
RCA b h - E k r h h . May
Figure 3. Mission scenarioesfor typical low earth orbit launched on an ELV
1883.
393
Figure 4. Typical mission for GEO satellite launched on an ELV.

Onboard navigation for satellite missions is generally
limited to the transportation vehicles (launch and orbit
mnsfer). Once on-orbit, h e satellite is tracked by the
ground segment using multi-program and/or programdedicated ground stations with support from computational centers. Several features of uacking and computational facilities which have been developed by North
American AerospaceDefenseCommand(NORAD),U.S.
Air Force (USAF), U.S.Navy (USN), and NASA are
shown in Table 1. Tracking data are used in conjunction
with orbital mechanics U, determine the satellite orbit
The determined orbit parameters correspond to an epoch
within the tracking data arc and provide the basis for
prediction to future time using orbital mechanics and
dynamicsmodels. The tracking arc length may be as short
asseveral minutes(foraLE0satellite with limitedground
station visibility) or as long as several days (for CEO
satellites with continuous line+f-sight visibility).
Thissolutiontothesatellitenavigationproblem hasproved
practical because in contrast with terresuial-based systems, the nature of orbital kinematics and dynamics permits adequate prediction of vehicle navigation state (e.g..
inertial position and velocity) over meaningful time intervals from knowledge of current vehicle navigation state.
There are two methods by which satellites make onboard
use of the predicted orbit parameters. First, the orbit
predictioncanbeappliedoveradesired timeinterval(e.g.,
a day) in the form of an ephemeris which is essentially a
catalog of position and velocity vectors at discrete times
(e.g., every 12 minutes). For spacecraft requiring higher
accuracy than can be provided with the ephemeris, interpolation between the onboard vectors may be performed.
The other method is prediction of the determined orbit U,
a desired time with uplink of this time-tagged state vector
as initial conditions for onboard extrapolation. With the
ephemeris technique, e m r remains bounded over the
applicable time interval, but a new ephemeris must be
generated and uplinked by the ground segment for accuracy is to be maintained. With the initial condition
extrapolationapproach, accuracy degrades with time due
to unmodeled errors in the disturbing accelerations. The
ground-based tracking,orbit determination,and orbit pre-
diction technique is shown by Figure 5.

There are numerous reasons for this dichotomy between
onboard inertial navigation for transportation spacecraft
and ground-based mcking, orbit determinatiodprediction approach for satellites. One major reason is because
most satellites have not required real-time navigation
state knowledge. For those systems in need of onboard
OROANIATw1(
P * a M
CIANRIEI
I wwunmu8
Table 1. Tracking and computational featues.

position and velocity data, accuracy requirements have
been satisfied with ground-based techniques. Another
major reason is that technology to support on+rbit navigation (primarily significant digital computational capability) has not been available to satellites. This does not
imply, however, that onboard navigation suites for satellites have not been of interest Since the 1960's there has
been continuing interest in onboard navigation to enhance
h e survivability of military satellites and reduce operating costs for civilian satellites. Consequently, numerous
studies of autonomous navigation concepts have been
conducted and reported in the literature with sporadic
technology developmentand demonstration efforts interspersed. These concepts and the resulting technology
394
efforts are considered in detail in a subsequent section.
Figure 5. Ground track, orbit determination/predibn

technique.
The enormous technological evolution in inertial navigation for aircraft. missile, and naval systems over the past
twenty-five years has had limited impact upon satellite
navigation because the orbital navigation problem is inherently different from these other navigation problems.
The basic goal of all non-orbital navigation problems is to
measure specific force (the non-gravitational accelerations) which is compensated for gravity effects and then
integratedto obtain velocity and integrated again to oblain
position. In the orbital navigation problem there are
essentially no specific forces acting on the vehicle (aunospheric drag, solar radiation pressure, and spacecraftgenerated effects are typically five orders of magnitude
smaller than gravitational effects), and the primary requirement is to characterize the complex gravitational
field including the effects of a non-spherical earth and
multi-body (e.g., earth, sun, and moon) interactions. In
their simplest forms, non-orbital navigation is a problem
of measurement whereas orbital navigation is a problem
of modeling. Hence, the need for significant digital
computationalresources is driven by the need for complex
onboard models.
Interplanetary spacecraft escape the earth gravity field
and travel into deep space for encounter with other planets
in the solar system. These vehicles are characterized by
extremely long mission durations which may exceed ten
years of continuous, non-serviceable operation at distances from the earth which can result in round trip
communication time delays measured in hours. During
the long interplanetary coast period, spacecraft orbital
dynamics are extremely low and ground-based tracking,
orbit determination/prediclion techniques are employed
using the Deep SpaceNetwork(DSN). From Figure6, the
DSN consistsof three complexesof tracking sites located
approximately 120 degrees of longitude apart to provide
worldwide tracking coverage. Spacecraftdesigns to date
have employed celestial, fixed attitude hold during interplanetary coast and inertial equipment used during other
mission phases such as descent and landing has been
turned off to conserveelectricalpower and extend operating life times. The purpose of the interplanetary missions
which have flown to date has been to acquire basic
scientific knowledgeoftheuniverseandthesevehiclesare
not directly relevant for military missions.
Descenf Md landing vehicles perform controlled de-orbit
from space and land on a planet or moon surface. These
vehicles use propulsion and aerodynamic surfaces to
control the descent trajectory during a combination of
maneuvering and coasting periods. Typically, these vehicles navigate with inertial navigation systems and radar
altimeters. Probably the most well known of this class of
vehicle is the Apollo Lunar Module. The mission profile
for the more recent landing of Viking on the surface of
Mars is shown by Figure 7.
Spacecraft return into the earth atmosphere at a known
time is sometimesreferred to as reentry. The reentry time
is noted and the vehicle will either impact theearth or burn
up in the atmosphere. Physical retrieval of a spacecraft
which has reentered and landed on the earth surface is
sometimes termed recovery.
U'
Descent Sequence
24
I
-.
48
226
40
47
46
LONGITUDE, deg WEST
SOURCE: ONeil, W.J., An Overview of Viking Navigation

Proceedings of the ION National Aerospace Meeting,
Denver, Colorado, 1977.
Figure 7. Profile of Viking Lander
395
3 3 Mission Orbits
Satellites operate in mission orbits selected to establish
and maintain an acceptable environment for the payload.
Orbital payloads perform data collection or data relay
missions. Meteorology, surveillance, and scientific mission payloads observe and collect data then transmit the
data to the ground segment for processing and analysis.
Payload data analysis and post-processing may require
precise orbit reconstruction using tracking data,extensive
force models, and more sophisticatet! data reduction techniques than those applied during the original orbit determination process used for prediction. For example, The
Ocean Topography Experiment (TOPEX) satellite mission (1334 km altitude and 63.1 degree inclination) will
carry a radar altimeter with subdecimeter accuracy. As a
goal, the precision orbit determinationprocess will incorporate data from global GPS ground terminals and an
onboard receiverwhich will be edited,calibrated,archived,
and processed to generate precision orbits using techniques such as tuning the gravity model U, fit the observation data.
Communication mission payloads receive data from a

transmitter and retransmit the data to the recciver. Navigation satellites act as a data relay by broadcasting navigational informationreceived from the ground. Data relay
missions capitalize upon geometrical advantages offered
by high altitudes of space to overcome the line-of-sight
limitations inuoduced by curvature of the earth and,

therefore, are typically earth-viewingpayloads. Data
collection missions are a mixture of earth- and spaceviewing payloads. Storeand forward payloads blend data
collection with relay U, collect msmiued data and store
it onboard until properly positioned in space for retransmission to a cooperative but remotely located earth-based
receiver.
Figure 8 indicates orbits of interest for space missions.
Low earth orbit (LEO) is typically of low eccentricity,
high inclination, and altitudes less than lo00 km. Mid
Earth Orbit @EO) is usually with low eccentricity, mid
inclination,and altitudes between 10,000and20.000km.
GeosynchronousEarth Orbit (GEO) is nearly circular (ec
0.1, typically), low inclination (ic10 degree, usually and
often zero), approximately 40,OOO km altitude, and an
orbital period approximating one mean sidereal day (0.9
revolution/day In l 1.I revolution/day usually.) Molniya
orbits are typically highly elliptical orbits at mid-inclinations. Super SynchronousOrbits (SSO)are those above
40,000km and typically to 200,000 km. Earth escape
requires velocity in excess of 11.2 k d s . Of these, orbits
of military interest are primarily LEO,MEO,GEO,and
SSO.We introduce these orbits because both navigation
requirements and genericnavigation suites are sensitiveto
orbit characteristics.
CURRENT SYSTEMS SUPPORTED BY AIR

FORCE SATELLITE CONTROL
BOOSTER I BAUlSTlC MISSILE TESTS

SPACE MISSIONS
70 SATELLmS CURRENTLY SUPPORTED

90+ SATELUTES PROJECTED BY FY 07
SOURCE: U. S. Air Force
Figure 8. Mission orbits of interest.
396
3. ORBIT MECHANICS
Orbital mechanics is the study of trajectoriesand orbits of
space vehicles. Numerous books on the subject of orbital
mechanics have been published [e.g., Refs. 2-51 and
should be consulted for comprehensive development of
the topic including the derivation of equations. This
section is intended to introduce the basic kinematics of
elliptical trajectoriesas applied to satellites in Earth orbit
The ellipse orientation relative to an earth-centeredreferewe system can be specified with three angles (a,w, i) as
shown in Figure 9. These angles are the longitude of the
ascending node (a),
argumentof perigee (w), and inclination (i). Specification of three additionalconstantsrelated
to the size and shape of the ellipse and time provide a set
of six elements which completely specify the Keplerian
motion. Theseconstantsaresemimajoraxis(a,orbit size),
eccentricity (e, shape), and time of perigee passage (7).
satellites,andthe reader may pursuedevelopmentofthese

approaches in the cited references. The equations shown
below are known as Lhe Gaussian form of Lagrange's
planemy perturbation equations. Several recommended
references which derive these equations are Wefs. 2-41.
The nomenclatureshown below is most closely attributed
to Jerardi. The method involves resolving the perturbing
force into three orihogonal components. The height
component (H)is along the radius vector, the along track
component(L)is in the orbital plane along the directionof
satellitemotion and perpendicularto H, and thecrosstrack
component (C) is normal to the orbital plane. First order
perturbations to the orbital elements can then be formulated in terms of velocity changes (AV) along this local
vertical local horizontal (LVLH) coordinate system centered at the satellite. The chosen set of orbital elements is
comprised of the semi-major axis (a), eccentricity (e),
inclination angle (i), longitude of the ascending node
(I$$,argument of perigee (0,
and mean anomaly at
epoch oc)such that M = n t + x .
Figure 9. Orbital coordinates.
The six elements are uniquely related to position and

velocity in Cartesian coordinates,and one useful form of
this relationship is given by:
x = r (cos R cos (U + f ) - sin R sin (a+ f) cos i)
y = r (sin R cos (U + f) + cos R sin (a+ f) cos i)

z = r (sin i sin ( o + f ) )
where
a (1 -e*)
r=
l+ecosf
and angle f is the true anomaly measured from perigee.

The relationshipbetween7 and f is through mean anomaly
(M), eccentricanomaly (E),and Kepler's equation according to:
M =n (t
-7)
= E - e sin E
where n is mean motion. True anomaly is related to

eccentric anomaly as:
tan(
i)=
i E t a n ( 5)
In practice, forces other than the central force which act

upon satellitescausedeviationsfrom a Keplerianorbit and
variations in the orbital elements. Several approaches
have been developed to describe the perturbed motion of
The AV,,, AVc, and AV,correspond to velocity increments

in H, C, and L directions, respectively. These velocity
increments may represent disturbancesources or control
inputs. Note that q.
and i are dependent only upon C
while a, e, and x are independentof C, and the argument
of perigee is influenced by all lhree components.
The astute reader has by now observed the potential
numerical problems associated with small eccentricity (e
= 0) and low inclination (i = 0 degree) orbits. These cases
can be solved through a change of variables [Ref. 51.
4. TECHNICAL REQUIREMENTS
Technical requirements imposed on spacecraftnavigation
systems m a y be derived from the vehicle physical characteristics, natural physical environment, natural operating
environment,and mission orbit. Additionally,there may
be requirements tailored to specific mission objectives.
Although military and civilian weather satellites share
many requirements, military systems will likely have
survivability and tactical data distribution requirements
which are unnecessary for civilian systems. Technical
requirements for earth satellites will be emphasized as
these vehicles currently possess military roles, with expectations for expanded roles in the future, and yet the
navigation capabilityof these vehicles remains extremely
limited.
391
Wedefineperformanceparameletsasacc~y,operating
life, and computational loading. Accuracy is dependent
upon the functions which use the navigation state data.
For satelliles, navigation state data generally supports
functions which do not impose stringent accuracyrequirements such as magnetic momentum management, solar
array pointing, and mission data annotation. Table 2
illustrates functions which utilize satellite navigational
data in terms of both current functions performed by the
ground and space segmenu as well as future functions
which could be performed onboard. A wide range of
operating life exists for spacecraft. Launch vehicles may
operale for 10minutes while satelliteswith S-yearoperating life must function continuously for 43,830 hours, and
interplanetary spacecraft with a 10-year mission operate
for 87,660 hours.
S m t e l I I i ~collwm
-y
HL..lon &U
-- .
h l c n n a panlln#
UJUUnUldmd&.u
newly-launched
Design life
(NiCd)
160 w-hrkg
200 w-hr/kg
2 Y3 Y-
80 W-hr/kg (GEO)
15 W-hr/kg (GEO)
5 W -hr& (LEO)
30 W-hr/kg (GEO)
25 W-hr/kg (LEO)
5 Y-
(LEO)
2 Y W (LEO)
Fuel Cells
primary
-SUllOnLCpl~
- c-it
nycal
-- nenduvow
t n s l v c nuncuvrn
Specific power
110 W k g
Radiothermal generators (RTG)
Specific power
5 w/kg
Loaplum roorm.r1oa-
flfiw
Rsdtn U I C I I I l e
mentry locaitm
De-
L.LM
Secondary baueries
Energy density
(LiX, NAX)
(NiCd)
(Niw
lk m-um
M d m r ulcIIl18
hulth and S U t U .
Primary batteries
Energy density
(Agnn)
(LiSOCL2)
Design life
(Agnn)
(LiSOCL2)
lllh platform8
- Won-coopcraoVc
;
C
SOURCE:
R
C
~
Table 2. Functions using navigation data.
Figure 10. Capabilities of satellite power sources.
Finally, computational loading depends upon whether

spacecraft translational kinematics are driven primarily
by measurable, specific forces or complex gravitational
effects which must be modeled. since models tend to be
more demanding of computational resources. As indicated previously, acceleration is dominated by specific
forces for sounding rockets, launch vehicles. and upper
stages so the gravity model need not be highly complex.
For non-maneuvering satellites. the gravitational effects
and, to a much lesser extent, other non-contact forces
dominate acceleration. Therefore, it is necessary to include higher order terms in the gravity model. During the
process of a poslerwri precision orbit determination,the
gravity model may be "tuned" to the specific saellite orbit
in such a manner as to provide a better fit of the computed
orbit with the tracking data. Interplanetary spacecraft
must accountfor gravity fields of other bodies such as sun,
moon, and other planets depending upon location within
the solar system.
The vehicle physical characteristics place consmints on
the navigation suite. Primary parameters of interest are
mass, power, and volume. Tolal electrical power available to satellites is limited by onboard power sources as
shown by Figure 10. Satellite mass is limited by the
launch vehicldupper stage mass-to-orbit capability as
shown by Figure 11. An upper limit on satellite physical
dimensionswill be introducedby a numberof factors such
as launch vehicle, upper stage. and fairing (payload covering to reduce drag during ascent) dimensions.
NO!=:
1. LEO Low Eaflh O M
GTO Geosynchronws Translef Orbn
Esc Eanh Escape
POWER SOURCE
Solar arrays
Specific power
Design life
CAPABILITY
2. Mass in kibgrane
3. Varklbm due lo wtit anitude and indinal'im
36 W/kg
IO years (GEO)
Figure 11. Mass to orbit capabilities of U.S. launch

vehicles.
398
The natural environment imposes requirementson elecmnics and surfaces to withstand space radiation effects,
thermal extremes, and the vacuum of space. Space radiation imposes hardening (i.e. promtion of electronics
against radiation effects which can result in tempomy or
permanent device failure) requirements against total and
maximum dose exposures. Repeating sunlit and eclipse
conditions to which the satellite is exposed result in
significant temperature variations which must be accommodated Materials and coatings are effected by space
vacuum conditions.
The navigation suite must withstand the operating environment from launch through orbital operations. Launch
operationintroducessubstantialforcesandvibrationwhile
steady-stateorbital operations may impose field-of-view
and angular rate constmints on strapdown optical sensors
thus influencing device performance characteristics,
mounting location, and shading devices. Laslly, remote
satellite operation precludes repair and maintenance opportunity for all but a fraction of satelliteorbits limited by
the "reach-ability"of the Shuule.
Although requirements m y widely among missions and
vehicles, a set of ideal technical requirements for an
advanced,earth-viewingsatellite are postulated by Figure
I
I
12.
w
loo
POSITION ACCURACY
POWER
<
meters
< 20 warn
IO kilograms
<IO00 cu. in.
> I years
-5 10 45 deg C
> 10' rad-Si
MASS
VOLUME
LIFE EXPECTANCY
TEMPERATURE
RADIATION
GROUhT) SUPPORT
5 1 updaiehno.
Figure 12. /dealtechnical requirements for satellite
navigation.
4.1 Military Requirements

Since the early 1960's there has been concern regarding
military satellite dependency upon overseas tracking stations which are perceived as vulnerable to a regime of
threats including political climate, sabotage, and direct
auack as well as natural disasters such as volcanic eruptions and hurricanes. There has also been concern regarding vulnerability of space-to-ground data links to denial
from intentionaljamming, nuclear blackout, and scintillation. Satelliteautonomy has been regarded as a means to
improve military asset survivability, but the belief that
autonomy adds complexity and, therefore, reduces reliability has been an obstacle to autonomoussatellitedevelopment. Typical life expectancy of military satellites
ranges from 3-4 years at lower altitudes to 7-10 years at
higher altitudes. At extremely low altitudes where atmospheric effects are more severe,operating life expectancy
may be substantially shorter.
Many satellite requirements and designs reflect the fact

that military satellite development and operation have
only recently been transfmed from the research and

development community to military operations through
establishmentof U.S. Space Command. Satellites have
had low production rate (often less lhan one per year) and
volume (typically, a buy of six or less for a given design)
with negligibletechnologytransfer across p g r a m boundaries,resulting in custom-designed satellite systems. The
GPS Navstar satellites represent the largest production
base to date with procurement of 28 Block I1 vehicles and
planned procurement of 20 Block W replenishment vehicles. Even within a given program, satellites are operated as isolated assets which may relay data from anolher
satellite to the ground in "bent pipe" fashion to enhance
datalinksurvivability,but whichdonot makeonboard use
of the data. In the next century,the GPS Block IIR system
is intended to actually exchange data among the constellation satellites and perform onboard data processing to
provide autonomousnavigationcapability in theevent the
ground station becomes unable to provide ephemeris
updates.
There are indications that future generationsof spacecraft
will have requirements reflecting enhanced operational
capabilities and direct suppon for tactical as well as
strategic operations [Ref. 11. Emerging requirements
levied upon satellites may include survivability against
hostile threats and direct (and timely) reporting and/or
relay of mission data to tactical users. In tun,implications
for the navigation system might include:
(a) Direct dismbution of collected data to tactical users
without going through ground stations will increase the
need for accurate mission data annotation and location so
tactical users can receive and interpret data locally.
Onboard navigation may need to suppon the location of
tactical users near hostile territory such as via geographic
latitude and longitude coordinates. As transmitted signal
beamwidths become narrower to mitigate the potential for
in ten tional jamming and unintentional se1f-jamm i ng
among multiple satellites and users, precision pointing
requirements for the satellite transmitting antenna will
become more demanding.
(b) Performanceof autonomousorbit maneuvers ranging
from routine orbit adjustment maneuvers (e.g., drag compensation, stationkeeping) to t i m e 4tical, responsive
maneuvers for threat evasion and collision avoidancemay
be required. Autonomous return U, normal operating
conditions following maneuver completion has implications on response time requirements to damp out the
maneuver effects on satellite attitude and rate.
(c) Intra-constellationdata-sharingmay be required. As
indicated earlier, this trend has been initiated by the U.S.
DoD GPS Block IIR program which intends to crosslink
navigation data among the satellites to enhance survivability.
(d) Navigation accuracy may have to be maintained
through conflict. This has historically been of concern for
satellites supporting strategic missions but as the boundary between strategic and tactical becomes less distinguishable, satellites supporting tactical missions may expect to encounter sophisticated threats during regional
conflicts.
399
4.2 Civilian Requirements

Civilian satellites provide science data such as stellar
information and earth observation data such as land resources. An important civilian system is the Geostationary Opcrational Environmental Satellite (GOES)for detection of damaging weather systems such as hurricanes.
Science and earth observation data post-processing may
require precise orbit determination after the fact. Generally, the approach has been to reconstruct the orbit, analyze the residual error to identify improvements to the
models, and re-process the orbit in an iterative manner.
Future requirements for improved accuracy and shorter
response time may impose new requirements which can
only be satisfied with an onboard navigator.
stations,(b) batch estimator.and (c)on-IineKalmanfilter.

Briefly, the batch estimator includes detailed force models to gencrate a batch-weighled least-squareseslimate of
the ephemerides. which are then predicted forward in time
to generate an ephemeris reference for the Kalman filter.
Both theestimamr and filter operateon identical measurement data obtained from ground monitor stations. The
filter fine tunes the reference ephemeris and provides a
best estimate ephemeris as well as estimates for clock
errors. The batch estimator models effects of the
geopotential, atmospheric drag, solar radiation pressure,
satellite thrusting, tidal forces, and multi-body force effects of the sun and moon.
6. REFERENCE SYSTEMS
4 3 Crew Requirements
Manned vehicles such as Shuttle are concerned with crew

safety requirements. Safety-critical systemssuch as guidance, navigation, and control must provide an independent backup capability during flight-critical operations.
Abort scenarios are needed to enable safe crew return
throughout all mission phases. Navigation data must
interfacewithcrew commandanddisplaysystems. Manned
spacecraft have stringent reliability and redundancy requirements for navigation during flight-critical operations. With rhe exceptionof Skylab, manned US missions
flown to date have been relatively short duration (5 2
weeks) in comparison with unmanned missions.
4.4 Interplanetary Requirements
Interplanetary spacecraft must be capable of autonomous

operation during the long signal delay caused by the round
trip light time at long ranges between ground control and
spacecraft. Other special requirements are exceptionally
long lifeexpectancy and planet-relativenavigation during
planet encounters.
4.5 Commercial Requirements
Commercial space applications are primarily geostationary communication satellites which may be continuously
tracked and controlled from a single earth station. Since
service interruption equates to revenue loss, satellite design is frequently kept as simple as possible. The combination of satellite simplicity, favorable orbit characteristics, and strong relationship between payload mass and
revenue are all drivers for minimizing satellite onboard
requirements.
5. A SPECIAL CASE: NAVIGATION OF NAVIGATION SATELLITES
Kepler's laws of planetary motion are inadequate for
precise orbit determination as needed for navigation satellites, and a two-steppmedure is used. First, numerical
integration of force models for external and internal disturbing accelcrationsprovides a "computed orbit". Then
a least squarcs fit of the computed orbit to a series of
observations from ground tracking stations is performed
[Ref. 61.
The ephemeris determination approach dcveloped for

GPS Navstar satellitesis described in Reference 7. Major
components of the determination process are: (a)
pseudoranging measurement data from ground monitor
Navigation state data requires amordinate frameand time

reference. Two coordinate systems of interest for spacecraft navigation are earth centered inertial (ECI) and earth
centered earth fixed (ECEF) frames. The most recent
ECEF frame is DoD World Geodetic System 1984 (WGS
84). and most recent ECI frame is the Conventional
Inertial System (CIS)known as the FK5 System of Epoch
J2000.0. The WGS 84 provides a reference frame, geometric earth figure. gravitational field model, and a means
for relating positions on various local geodetic systems to
an ECEFcoordinatesystem. The J2000.0 defines an ECI
frame at the new standard epoch J2000.0 (2000 January
1.5). Associated with this system are new theories for
precession and astronomic nutation and new definition of
universal time as adopted by the InternationalAstronomical Union. Transformationbetween these frames is welldefined and real-time transform requires earth orientation
prediction (i.e., earth rotation and polar motion).
The WGS 84 system is described in detail by Reference8.
The origin is located at the center of mass of the earth,the
Z-axis is parallel to the direction of the pole, the X-axis is
the intersection of the reference meridian plane and the
plane of the equator, and the Y-axis completes a righthanded. earth-fixed orthogonal system. The coordinate
system definition is shown by Figure 13.
Two time systems are universal and sidereal time -both of
which are based upon diurnal rotation of the earth. Sidereal time is derived from Earth rotation with respect to the
stars while universal time is related to diurnal motion of
the sun. These time systems include effects of nonuniform rate of earth rotation and thus are non-uniform
time systems. One mean sidereal day is equivalent to 23h
56m Wsof mean universal (solar) time. When corrections
are applied to universal time to account for variations due
to observed polar motion, the time scale is called UTI. A
time scale whose epoch is periodically adjusted by integral seconds to maintain synchronization within 0.9 seconds of UT1, but whose rate is based upon atomic time
(which is continuous) is known as Coordinated Universal
Time (UTC). This time is provided by the U.S. Naval
Observatory and is appropriate for annotation of events,
observations,and other occurrences requiring a time tag,
and for the independent variable in the equations of
motion. Sidereal time is used in the transformation
between WGS 84 and J2000.0.
400
AUTONOMOUS NAVIGATION
An orbital navigator is considered autonomous under the
following conditions:
- Its long-term mor growth is bounded

- The navigation measurements do not require any
-
(cooperative or uncooperative) artificial suppofi

external to the system containing the navigator
It is tolerant of moderate initial condition m r s
Figure 14. Definition of autonomous navigation.
Ground-based systems may be thought of as tracking

stations and computer programs which perform orbit
analysis. Therefore, in a strict sense of the terminology
these systems are not truly navigation systems. However,
this technique is so pervasive for satellite position/velocity vectordeterminationthat itcannotpracticallybeomitted from discussion of spacecraft navigation.
Figure 13. WGS coordinate frame axes.
Ephemeris time is related to the orbital motions of the
earth,moon, and planets in the solar system. The ephemeris second is fixed by definition so ephemeris time is
theoretically uniform. The relationship between ephemeris and universal time is given as:
E.T. = U.T. + AT
where AT is a yearly increment provided in Reference 9.
Thus, the relationship between ephemeris time and sidereaVuniversal time must be determined empirically. Reduction formulae for conversions among these time systems are included in detail in Reference 10.
7. NAVIGATION SUITES
The two primary components of navigation suites are the
sensor complement and computationalprocessor. Generally, for ground-based navigation, both components are
located on the ground and similarly, for onboard navigation both are physically located onboard the satellite. One
exception is ground-based processing of optical data from
interplanetary spacecraft. As interplanetary spacecraft
approach a planet, optical sensors can image the planet
against the stellar background. This data is processed on
theground to estimate spacecraft flight path relative to the
planet.
The type of observation measured by the sensor is sometimes used to classify an onboard system as either autonomous or aided. Autonomous navigation relies upon measurement of naturally occurring phenomena such as lineof-sightto astronomical bodies. Aided navigatorsprocess
measurements from artificial sources such as GPS and
ground beacons. This concept has evolved primarily from
concerns of measurement denial and corruption which
could adversely effect navigation system integrity. A
formal definition for autonomous navigation is given by
Figure 14. Conceptually, the space to ground interfaces
among these techniques are shown by Figure 15. The
aided navigator shown in the figure assumes a navigation
satellite.
Figure 15. Ground-based, aided, and autonomous

navigation conceptual interfaces.
7.1 Ground Systems

The ground-based approach to navigationalstate information currently used to support orbital satellites can be
divided into four major functional areas: (a) data collection, (b) orbit determination,(c) orbit prediction, and (d)
ephemeris generation. The former two areas are usually
accomplished by shared facilities such as the NORAD
SpaceSensorNetwork, US AF SCN,USN NAVSPASUR,
and NASA STDN systems. The latter two areas are
typically handled by program-dedicatedfacilities. There
are examplesof systems,however, such as GPS where the
end-to-end process relies primarily upon program-dedicated facilities.
7.1.1 Measurement Techniques
Common observables include line-of-sight(LOS)angles,
range, and range-rate. Each observable may be collected
cooperatively or noncooperatively (to varying degreesof
accuracy) with existing sensors. The DoD Space-Ground
Link System (SGLS) provides accurate range and rangerate measurements via two-way S-band radio links. The
inherent accuracy of sensors in use varies but is sufficient
to preclude measurement errors from becoming a dominant error source during steady state operation. Opponunities for observation vary from approximately 10% for
LEO satellites (<IO00 km altitude) using only ground
40 1
stations to 10096 for satcllites in geosynchronous orbit

Noncooperative observationscan be obtained from radar
(skin tracking) and optical sensors. Radar measurements
often consistof time-taggedazimuthand elevationangles,
range,and range-rateobservables. Passiveoptical Sensors
are typically limited to angular observations.
Two-way ranging uses a coherent forward and mum link
to measure the elapsed time between transmission of the
forward pseudorandom noise (PRN) code epoch (or an
equivalent technique) and reception of the same epoch by
the receiver channel. Delays introduced by signal propagation through ground station equipment and cables are
subtracted from elapsed time by computer to calculate
actual time delay. This measurement type requires the
satellite to synchronize thereturn to the forwardlink code
and requires a stable clock. If a satellite possesses a
transceiver with proper ratio between forward and return
carrier frequencies, then the ground system can measure
two-way Doppler. One-way Doppler measurements can
be made for any satellite.
For interplanetary spacecraft, telemetered optical data
from celestial sensors onboard the satellite can be ground
processed for planet-relative navigation data.
7.1.2 Models
Orbit determination techniques are well-established for

stable orbits. Epoch state accuracies of less than 100
meters are achievable for a wide range of orbits and
cooperativetracking can be used to reduce these errors by
at least an order of magnitude. Since ground-based
models need not execute in real-time, complex environmental and vehicle disturbancescan be incorporated. As
a minimum, higher order gravity terms are usually included. Other disturbances which may be modeled include aerodynamic drag, solar radiation pressure, lunarsolar attraction. and disturbances resulting from thruster
firings. The magnitudeof these disturbancesis shown by
Figure 16 as a function of satellite altitude. For satellites
demanding higher accuracy navigational state information, the disturbance models and, in particular. model
parameters,may be derived from empirical data acquired
over long periods of time.
10.1
7.1.4 UplinkJDownlink
The ground system uplinks the predicted satellite navigational slateasatime-taggedstatevectororeqhemeris. The
frequency with which this must be performed is ondemand and depends upon the accuracy required so may
occur as often as several times per day or as infrequently
as once per month. For satellites with short passes over
transmitting ground stations, it may be necessary to limit
the amount of uplink data and ephemeris compression
techniques may beemployed. Additionally, commands to
fire thrusters may be uplinked to effect orbit maneuvers
such as drag adjustment and stationkeeping.
Downlink of computed position and velocity vectors

enables ground operators to monitor the difference between onboard estimates and ground-based calculations.
This allows unscheduled state updates to be uplinked to
the satellite should the error become excessively large.
7.2 Flight Systems
Onboard navigators estimate satellite position and velocity vectors in real-time based upon a set of initial conditions, an apriori system model, and a sequenceof external
observations of the navigation state. Conceptually, the
navigator consistsof two basic elements: state propagator
and measurementprocessor. Navigator inputs are a set of
initial conditions and sequence of measurements which
relate to the navigation state. The current estimate of the
state vector is the primary navigator output The basic
structure for the onboard satellite navigator is depicted by
Figure 17 Wef. 121.
Figure 17. Conceptual diagram of onboard navigator.
--
The function of the state propagator is to extrapolate the

initial state vector forward in time by numerically integrating a non-linear, vector differential equation which
describes the motion of the satellite in a perturbed orbit
The state propagator is characterized by a state vector, a
set of disturbing acceleration models and a numerical
integration algorithm.
750
LEO
YE0
GEO
* & m m s I.10 dcpwldd
Figure 16. Disturbances as a function of orbital altitude.
7.13 Clocks
Ground-based systems employ precision oscillators of

cesium beam and hydrogen maser atomic oscillator types.
These oscillators exhibit high short-term (10. T = 1
second) and long-term (lOI4, T = 1 day) stability. Clock
drift is of the order of 10over one day and 1013over one
year [Ref. 111.
When specific force measurementsare available, the state

propagator may incorporate these as well. For example, if
performance characteristics of inertial instruments are
capable of measuring the drag force, these measurements
could be incorporated to improve accuracy.
7.2.1 Measurement Techniques
Onboard measurementsof navigation state may be optical
or radiofrequency (RF). Astronomical optical sources
includeearthhorizon,sun, moon.and stars. The measured
402
phenomenology is the angle between two optical sources

(typically,a star and near body such as stadearth horizon).
An optical sensor oriented toward the earth surface may
use natural surface features for landmarks. Optical navigation aids include artificial ground landmarks and laser
passive satellites. Radiofrequency aiding sources include
GPS,TRANSIT,TDRSS, and RF ground beacons. Note
that the measurements described are useful for attitude
determinationor communicationssystems and, therefore,
such equipment may be shared with the navigator.
system. Use of TDRSS for onboard navigation would be

most appropriate for sakllites already requiring a TDRSS
antenna for the data relay capability.
Ground beacons at known locations can provide range
and/or range-rate measurements. Transmitters provide
one-way range and/or range-rate while tranqmders provide tweway range and/or range-rate. For survivability
purposes, ground beacons for U.S.satellites should be
located within the continental U. S., which limits measurement coverage.
7.2.2 Measurement Sources
There are three existing satellite systems with navigational missions which provide radiometric measurements.
GPS provides one-way range and range-rate data and
broadcasts ephemeris data. TRANSIT provides rangerate data and broadcasts orbilal elements. These spacebased navigation systems were developed for terrestrialbased users, and transmitting antennas point toward the
earth. Therefore, these systems provide measurement
coverage for lower altitude satellites (h <20,000 km for
GPS and <loo0 km for TRANSIT). Past studies have
consideredGPSforsatellitesataltitudesabove20,000km
by receiving signals from satellites on the far side of the
earth as shown by Figure 18. The GPS signal structure
was developed to be tolerant of intentional and nonintentional interference,and signal features appear in the
literature Ref. 131. The DoD policy to phase in GPS as the
primary navigation aid and phase out TRANSIT and
TACAN systems discourage development of satellite
navigators dependent upon these systems.
Figure 18. GPS reception at GEO.

As currently used, TDRSS provides a bent pipe between
user satellite and ground terminal. Measurement types
provided are one-way Doppler, and two-way range and

Doppler. Past studies have examined one-way range and
Doppler from the ground station through TDRSS with
measurement processing onboard the user satellite [Ref.
141. There are two active TDRSS satellites in geostationary orbit at 41 and 171 degrees west longitude which
q u i r e ground support for ephemeris generation. The
TDRSS ephemeris error is larger than GPS, and accuracy
achievable withTDRSScannot match that of GPS. Moreover, TDRSS satellites do not broadcast ephemeris information. The satellites possess steerable antennas so they
may be capable of supporting higher altitude users, although relatively low geometrical dynamics between
TDRSS and higher altitude users may limit achievable
accuracy. A major limitation of TDRSS for military
applications is the lack of survivability of the TDRSS
Inertially-referend line-of-sightmeasuremenrs to ground

landmarkscan be used for satellite navigation Two types
of tracking algorithms have been reported in the literature
[Ref. 151. Known landmarks are earth-fixed ar a known
location. Unknown landmarktrackmgalsoassumesearthfmed landmarks but rather than require a prwri location
knowledge, landmark coordinatesare includedin the state
vector and estimated by the navigation algorithm. Optical
sighting to ground landmarks possess several disadvantages. At visible wavelengths,landmarksareobscurcdby
clouds and depend upon the availabilityof natural lighting
(i.e., daylight only). Longer wavelengthssuch as infrared
(IR) penwate clouds, but have lower resolution. The
image must be recognizableand uniquely identiftableand
at some altitude (which will vary with tracker design)
detectionand discrimination will become infeasible. Some
limitations of passive optical measurements may be overcome with active (cooperative)systems. For example, an
onboard IR laser could operate with ground beacons
(retroreflectors) or an onboard laser could operate with
orbiting satellite retroreflectors. Again, U. S. satellites
using ground beacons should rely only upon those within
the continental U.S.for reasons of survivability.
Astronomical measurements of the angle between a star
and near body can provide navigational updates. For nearearth orbiting sate.llites. the body is typically the earth
horizon or moon. A related technique is observation of
stellar refraction or dispersion as stars set behind the
atmospherichorizon of the earth as Seen from the satellite.
Optical sightings from the satellite are highly coupled
with satellite attitude, and auitude knowledgeerrors contribute to navigation error for these systems. Radiometric
measurements made by satellitescan provide higher accuracy by at least an order of magnitude and are essentially
decoupled from vehicle attitude.
Passive satellites in highly stableorbits which are predictable over long time periods can also provide a navigation
aiding source. These laser retroreflector satellites are
simplein design and highly reliable. Thelasergeodynamic
satellite (LAGEOS) launched in May 1976 to an altitude
of 5900 km is such a satellite. This measurement source
has the potential for high accuracy and immunity to
jamming since narrow beamwidths can be used and only
space-to-spacelinks are involved. Small, mobile ground
tracking stations are sufficient for tracking the laser passive network.
With a constellation containing an adequate number and
distribution of satellites, the member satellites can make
inter-satellitemeasurementsusing communicationsequipment to provide relative navigation information. This
403
type of navigation has been considered for the Strategic

Defense Initiative space-based assets [Ref. 161 based
upon direct vehicle-to-vehicle ranging techniques. The
GPS Block IIR program plans to use a UHF time slot
broadcast technique [Ref. 17) for crosslink measurements.
Studies have also examined the ejection of a navigation
aidsatellitehm theprimaryusersatellite intoanearorbit
for tracking. [Ref. 181 Radar measurements of range,
range-rate, and angles were considered as well as laser
measurements of range and angles. The method is predicated upon reference of the relative state measurement to
an inertial coordinate system as established with an IMU
and/or star trackers.
1.23 Models
To achieve the accuracy required, onboard navigator
models provide compensation for dominant disturbing
accelerations acting upon the satellite. Since modeling
errors are bounded by incorporation of measurements of
the navigation state, reduced order gravity models should
prove adequate for many satellite applications. Low
altitude satellites may also need to incorporate an atmospheric drag model. High altitude systems may also need
to model the effects of solar radiation pressure and lunarsolar attraction.
1.2.4 State Vector
Satellite navigation state is typically defined as inertiallyreference Cartesian position and velocity vectors. Although this particular state vector has a convenient interpretation in termsof the physical quantitiesof interest. the
state vector elements suffer from a wide dynamic range
due primarily to the central force gravity term. A satellite
at a IO00 km orbital altitude around the equator of a
sphericalearth will experience a variation off 7378 km in
position over the 100 minute orbital period and, correspondingly approximately f7.4 k d s in velocity. This
large dynamic range imposes a significant computational
burden on the state propagation technique by requiring
extremely small integration step sizes and a high level of
numerical accuracy.
The constrained natureof the on-orbit navigation problem
suggests the use of a state vector which is time-invariant
for undisturbed two-body motion. A set of six orbital
parametersdefinesthe motion of a satellite in its orbit. For
the special case of a true Keplerianorbit, these parameters
will be constants. In general, these time-varying orbital
parameters are functions of the perturbing accelerations.
and personal compum (pc) support tools for 1750A

software development have become available. More
recently, the Generic VHSIC Spaceborne Computer
(GVSC) program is developingradiation-hardened1750A
computers using VHSIC technology. These computers
are expectedto provide three to four mega-instructionsper
second throughput capability.
7.2.6 Clocks
Quartz crystal, rubidium, and cesium oscillators are candidates for onboard clocks. Both quartz crystal and
rubidium oscillators exhibit advantages with respect to
size, weight, power, and cost. Rubidium cloclrs provide
good performance in the presence of a hostile environment. Cesium clocks provide excellent long-term performance at a higher cost, weight, power, and volume penalty.
8. TRADE-OFFS
The first trade-offinvolves determining whether naviga-
tional statedataisrequuedonboardthesatellite,and,ifso,
whether the data should be provided to the vehicle by
ground-based orbit determination techniques or onboard
navigation. Ground-based techniques offer advantages
for vehicle weight, power, volume, and complexity.
Onboard approaches offer advantages of autonomy, SUIvivability, and accuracy. Figure 19 [Ref. 1, Paper No. 241
shows the potential accuracy improvement with onboard
GPS navigation as compared with ground-based orbit
determination. This figure illustrates that ground-based
techniques achieve best performance in the 2,000 to
20,000km altituderegion. At loweraltitudesthecoverage
achievableby ground stations is limited and atmospheric
drag effects are most severe. At higher altitudes, there is
little change in geometry between the satellite and ground
stations.
Source: AGARD Tarsid *pplicetkmr
d Space Systems. AGARDCP.460.

E
mc
SATELLITE W S I T O N
ACCURACY ACHIEVABLE
ON THE GROUND
BY SCN WITHIN THE
NEXTONE ORBIT
(ammaUy in the
in tradr direction)
REALTIME POSITION ACCURACY

ACHIEVABLE USING AN ON BOARD
GPS RECEIVER
t
MO
400
lo00 xxM3ooo
10.o00m,ooo40.o00
SATELLITE ORBIT ALTITUDE. km
1.2.5 Flight Compulers

State-of-the-artmilitary satellite computers in the early
1980s were examined by Reference 19. Since then,
commercial advances in digital data processing hardware
Figure 19. Onboard navigation accuracy with GPS as

compared with ground-based orbit determination
techniques.
capability have repeatedly increased performance such as

throughput and memory size while simultaneously decreasing size and weight. Although the full extent of these
advances has not been realized for satellite computers,
some advances have been made here as well. Processor
development using the MIL-STD-1750A standard instruction set architecture (ISA) has been actively pursued
by several vendors, and was selected by the MILSTAR
program. A 1750A Adam compiler has been validated
Passive optical measurementsfrom satellite to ground are

subject to cloud coverage, lighting conditions, seasonal
effects, and sensor pointing errors. For survivability
reasons, the landmarks should be located within the continental U. S. (CONUS) which may significantly limit
measurementopportunities. These observation types are
more applicableto lower altitude satellites since accuracy
depends upon ground resolution. Some problems associ-
404
ated with natural landmarks (e.g.. seasonal and wealher

effects, correlation with reference landmarks) can be
ameliorated with arlificial landmarks such as search beacons. These landmarks then require maintenance to
ensure availability.
Cooperativetechniquesfrom space-to-groundwould have
lasers onboard the satellites and ground-located optical
beacons (retroreflectors). Laser beams are subject to
atmospheric effects and may require two frequencies for
refraction compensation. Survivability against blinding
and beacon denial requires beacons be CONUS located
which will constrain measurement opportunities. Cooperative space-to-spacesystemswith laser rangers onboard
the user satellitesand comercube relroreflecting satellites
are not subject to atmospheric effects, but require ephemeris maintenance of the retroreflector satellites although
highly stable and, therefore, predictable orbits can be
used. This techniqueappearsto have the potential for very
high survivability. The narrow beamwidth of lasers may
require stringent pointing capability for the user.
I
I
Sightings to earth horizon, sun, and m m from the user

satellite also offer high survivability as it is difficult to
deny these phenomenologies. The accuracy achievable
with horizon sensors is limited by knowledge of the
horizon and is subject to seasonal variations as well as
local variations. The measurement isan angle between the
local vertical and another celestial body such as sun,
moon, or star. As an angle measurement. the position
accuracy degrades with orbit altitude so at geosynchronous altitude the variations in horizon are less detectable
but achievableaccuracy is decreased. The sun and moon
sightingsare limited by the ability to determine the center
of m a s of these bodies. The use of star sightings with an
earth horizon offers the best measurement availability
since sun and moon sightings are highly geometry specific.
Radio frequencymeasurementsemanatingfrom the ground
and received by the satellite would enable a relatively
simplereceiver since transmitter location could be known
and fixed. Anti-jamming and anti-spoofing techniques
need to be applied and the transmitter should be CONUS
located for systems with survivabilityrequirements. This
technique is essentially independent of satellite pointing
errors. The transmitters could be sufficiently powerful so
only a low power omni-directional antenna would be
required by the satellite. If the signal were modulated with
position and time information, this system would be the
inversion of GPS. For accuracy, two frequencies are
required to correct for ionospheric effectson signal propagation. Recc.ption from GPS satellites directly provides
greater measurement opportunity for low altitude satellites and does not involve signal propagation through the
atmosphere. Groundbeaconscould still be used to supplement GPS if so desired. The survivability of satellites
using GPS is dependent upon the survivability of the GPS
system. The GPS signal is specially designed for protection against electronic jamming and spoofing. This is not
the case for TDRSS which is a "soft" system. GPS offers
thepotential for high accuracy without demanding onboard
resources in terms of power, weight, and computational
processing.
Before concluding, we considex the use of radar for

satllite navigdon from space to ground. Radar can
measure range and range rate to the surface Lhrough uncooperative radar return from the surface or cooperative
retm from microwave reflectors. Radar provides an allweather, daybight measurement capability. Real aperture radars consume substantial power and are terrain
model sensitive. The support requirements for real aperture radar (primarily size, weight, and power) with range
capability in excess of several hundreds of kilometers are
not compatible with satellite support capabilities. Additionally, since power requirements increase substantially
with range, onboard radar for anything but the lowest
altitude satellites is not practical. The need for large
antennasalso is a problem. Generally, radarsare inappropriate for satellite navigation. The role of radars to date
has been for rendezvous with other spacecraft and as
altimeters during surface landing. Synthetic aperture
radar ( S A R ) offers antenna size advantages over real
apertures but use of the data onboard would be extremely
demanding of signal and data processing resources to
obtain a robust correlation of S A R images with onboard
reference images.
9. SUMMARY
In this section we have considered technical requirements
and genericnavigation suites for spacecraftwith emphasis
upon military satellites. Both ground-based and satellitebased navigation techniques were addressed. Navigation
suites were considered in terms of the phenomenology
providing the measurement source. We also considered
the other major components of the navigation system such
as algorithms, computers, and clocks.
ACKNOWLEDGEMENT
This effort was partially sponsored by the Strategic Defense InitiativeOffice under contract SDIO84-88-CMX)9.
DISCLAIMER
Review of this material does not imply United States
Department of Defense indorsement of factual accuracy
or opinion.
References
1. AGARD Tactical Applications of Space Systems,
AGARD CP-460, October 1989.
2. Jerardi, T. W., Orbital Mechanics, Anthology, (undated).
3. Taff. L. G.. Celestial Mechanics, A Computational

Guidefor the Practitioner, John Wiley & Sons, Inc., New
York. 1985.
4. Moulton,F. R., Celestial Mechanics, Macmillan, New
York, 1902.
5. Battin, R. H.. Astronautical Guidance. McGraw-Hill,

New York, 1964.
6. Ashkenazi, V.and Moore, T., The Navigation ofNavigation Safellites, J. Navigation (GB), Vol. 39, No. 3,
September 1986, pp. 377-93.
405
7. Van Dierendonck, A. J.,Melton, W. C.,Birnbaum, M.,

and Harkens, M.D., The Approach to SatelliteEphemeris
Determinationfor the NAVSTAR Global Positioning Sysrem. Navigation, Vol. 23. No. 1. Spring 1976, pp. 76-86.
8. Department of Defense World Geodetic System 1984,
Its Definition and Relationship with Local Geodetic Syst e m , DMA TR 8350.2.30 September 1987 and Supplements DMA TR 8350.2-A Part I. Methods. Techniques,
and Data Used in WGS 84 Development, DMA TR
8350.2-BPart 11, Paramelers, Formulas, and Graphics
for the Practical Application of WGS 84.
9. The AstronomicalAlmanac, (USNGB), US Govemment Printing Office, Washington.

10. ExplanutorySupplementfo theAstronautica1Ephemeris and American Ephemeris andNautica1Almanac, Her
Majesty's Stationery Office, London, 1961.
11. Fruejauf, H., Precision Oscillators and Their Role
and Performance in Navigation Systems, PLANS 1982.
12. Mi1ler.J.L. andDaly.K.C.,AutonomousNavigatwn
for Orbital Satellites, Journal of Defense Research, Vol.
17, No. 3. Fall 1985, pp. 242-268.
13. Spilker, Jr., J. J., Signal Structure and Performance
Characteristics. ION GPS Papers, Vol. I, pp. 29-54.
14. Elrod. B. D., Low Earth Orbit Navigation in the
Tracking and Dora Acquisifion Sysfem Era, AIAA J.
Guidance. Vol. 8, No. 4, July-August 1985.
15. Kau. S.P..AutonomousSatellite Orbital Navigation

Using Known and Unknown Landmarks, AIAA Guidance and Control Conference, AIAA 751096,1975.
16. Trede1.A. J.,AulonomousNavigalion - When Will We
HaveIt?,Navigation, ION,Vol. 34, No. 2, Summer 1987,
pp. 93-114.
17. Ananda, M. P., Bernstein, H., Bruce, R. W.,
Cunningham. K.E..Feess. W. A., J0rgensen.P. S..Menn,
M., and Price C. M., Auionomous Navigation of the
Global Positioning System, AIAA 84- 1863-CP. 1984.
18. Muller, E. and Kachmar,P.,A New Approach to OnBoard Orbit Navigation, CSDL-E-2566,
February 1971.
19. Gilley, G., Digital Hardware for Use in Spacecraft
Control Applications, AAS 80-031, Guidance and Control 1980, Vol. 42, pp. 631-50.
407
SECTION VI
TEST METHODOLOGY
Coy L. Hunt
46th Test SquadrodEGTN
1644 Vandergrift Road
Holloman AFB NM 88330-7850
USA
6. TEST METHODOLOGY
6.1 Introduction
This section of the document describes the various methods
employed to evaluate the performance of aerospace navigation
systems. Included are discussions regarding test equipment, test
data, reference systems, environmental conditions, laboratory
tests, flight test profiles and procedures, and statistical methods
used to measure system performance. The test methods detailed in
this document are somewhat specific and may need to be adapted
to particular systems and their specifications.
6.1.1 Test System
The basic component of the test system is an inertial navigation
unit (WU) and may be integrated into a system utilizing a
multi-sensor configuration. All sensor components should include
their own monitoring units to facilitate real-time validation of
operational performance.
6.1.2 Data Acquisition System
The data acquisition equipment should be capable of recording
the necessary parameters output by the integrated system as well
as the navigation parameters from the individual sensors. The
required data parameters, sample rates, and timing requirements
are specified in Section 6.2.
6.1.3 Reference System
The requirements for reference system data acquisition equipment
are quite similar to those in Section 6.2.2 above. Additional
discussion related to the reference system is presented in Section
6.3.
6.1.4 Laboratory Test Equipment
The required laboratory test equipment includes test tables
capable of positioning the test system in any number of specific
orientations. Equipment capable of subjecting the test system to
various environmental conditions (vibration, heat, cold, humidity,
etc.) will also be required. Additionally, equipment for testing
Electromagnetic Interference (EMI) should be available.
6.2 Test Data
Test log data should include complete identification of the test
article and all test equipment and accessories as well as basic data
accuracies. The log data should include the actual test sequence
used, ambient test conditions recorded periodically during the test,
and performance data recorded periodically during the test.
Laboratory data should include as a minimum: ambient
temperature, humidity, latitude, longitude, altitude, cooling air
temperature, cooling air flow rates, and power. It is recommended
that aircraft data on temperature, humidity, vibration, cooling air
temperature, cooling air flow rates, and power be included to help
isolate system performance problems or anomalies. Performance
data should include, as a minimum, indicated position and
velocity components. As a minimum, the data recording rate for
static tests should be one-per-minute. For dynamic tests, the

recording rate should be one-per-second.
6.3 Reference System for Flight Tests
63.1 Description
An independent reference system is required for use as a truth
model against which the test system performance will be
compared. The accuracy and reliability of the reference system
must have been previously established in order to reliably
establish the performance of the test system.
63.2 Accuracy
It is highly desirable that the accuracy of the reference system
exceed the estimated accuracy of the test system by a factor of ten.
This will insure that the significant portion of the error attributed
to the test system is in fact generated by the test system and not by
the reference system.
6.3.3 Data Rate
Data from the reference system should be available at least
once-per-second. If the test aircraft is performing maneuvers that
involve high dynamics, then the sample rate should be higher than
one-per-second to prevent the introduction of interpolation errors.
6.3.4 Restrictions
Flight test profiles should be designed to fly in the area where the
reference system can provide the most reliable tracking. If this is
not possible, then it should be noted that the reference system is
likely producing data with degraded accuracy when flying in
areas where reference sensor coverage is marginal.
6.4 Environmental Conditions
6.4.1 Restrictions
Discretion must be exercised in connection with the testing of any
navigation system to insure that the environmental limitations of
the test system are not exceeded. These limits are defined by the
manufacturer and should be noted prior to the initiation of any
type of testing.
6.5 Laboratory Tests
6.5.1 Assumption
This test procedure assumes that a series of tests has previously
been conducted on the individual components of which the
integrated system is comprised. Therefore, the purpose of all
laboratory tests will be to determine integrated system
performance for a variety of static and dynamic modes while
experiencing diverse environmental conditions. These tests are
primarily designed for the unit containing the inertial navigation
system whether or not other systems are integrated into the same
unit.
6.5.2 Static Tests
These tests should be conducted to assure that the system
functions within designated limits for the desired operational
408
intervals. W Otests should be conducted for each type of ground

alignment which the system will be expected to perform
(Gyrocompass, Stored Heading, Enhanced-Intempted, Best
Available True Heading, etc.). Data from the static tests will be
used to establish baseline performance for the navigation system.
65.3 Random Vibration Tests
The objectives of these tests are to evaluate the effects of
endurance level vibration and to determine if performance is
satisfactory following the prescribed vibration. The system should
be mounted with a north heading on a vertical exciter, at ambient
temperature. A normal gyrocompass alignment will be performed
after which the system is switched to the navigate mode. After
five minutes of static navigation, random vibration will be applied
to the Z-axis for 1 hour. The acceleration spectral density will be
0.001 g-squared/& from 10 to 1000 Hz, with a 6 dB/octave roll
off to 2000 Hz, 1.22 g rms (see figure 1). When the vibration
input is removed, static navigation should continue for an
additional hour. One test should be conducted.
65.4 Electrical Power Variation Tests
These tests should be conducted to assure that the navigation
system will function under power conditions expected in a flight
environment. These tests consist of six parts for a system powered
by AC or two for a system powered by DC. One test should be
conducted for each part. The voltage and frequency levels
specified in these tests are typical of levels used to test navigation
systems. However, these levels should be interpreted with respect
to the specifications of the particular system to be tested.
a. Normal and Emergency AC Power, Power Factor, and Power
Consumption. With the system in the navigate mode, the three
power phases are simultaneously varied to 113, 104, 113, 108,
113, 119, 113, 122, and 113 volts AC, remaining at each
voltage level for 10 minutes. The individual phase power
factor and power consumption are to be recorded.
b. AC Frequency Variation - Navigation Mode. While operating

in the navigation mode, the power frequency is varied to 400,
420,400,380, and 400 Hz,remaining at each frequency for 10
minutes.
c. Three Phase Voltage Transients - Navigation Mode. The
system is operated in the navigation mode for 42 minutes after
which simultaneous three-phase voltage transients are applied
at 10-minute intervals as follows:
Volts (RMS)
e. AC Power Variation - Alignment Mode. The voltage is

changed to 380 Hz, 108 VAC during the alignment mode.
After switching to the navigation mode, the system is allowed
to navigate for 90 minutes.
f. AC Voltage Transients -Alignment Mode. With the,AC
voltage and frequency set to 115 VAC, 400 Hz,the transients
of part c are imposed during the alignment mode at 1-minute
intervals. After switching to the navigation mode, the system is
allowed to navigate for 90 minutes.
g. DC Power Variation - Alignment Mode. The system should be
mounted with a north heading and gyrocompass aligned with
the DC power level at 10 percent below nominal. Following
alignment, the system should be allowed to navigate for 1
hour. This test should be repeated with the DC power level at
10 per cent above nominal.
h. DC Power Variation - Navigation Mode. The system should
be mounted with a north heading and gyrocompass aligned.
The following input voltages with specified durations should
be applied to the system during the navigation segment,
starting 1 minute after the completion of alignment. One test
should be performed with a total navigation time of 67
minutes. Nominal power is 28 volts.
Volts
Duration (minutes)
28
24
10
28
25
10
28
26
10
28
27
10
28
29
10
28
30
10
28
Duration (Seconds)
122
132
122
2.0
122
152
122
0.8
122
180
122
0.1
104
104
2.0
The system is operated in the navigation mode for an

additional 30 minutes following the transients.
d. AC Power Variation - Alignment Mode. The voltage is
changed to 420 Hz, 118 VAC during the alignment mode.
After switching to the navigation mode, the system is allowed
to navigate on normal power for 90 minutes.
For testing systems whose nominal power differs from 28

volts, the table above should be used as a guideline to vary the
voltages by similar percentages.
6.55 Electromagnetic Interference ( E M ) Tests
Electromagnetic Interference testing will be conducted to measure
the effects of conducted or radiated radio frequency waves on
system performance. All tests should be conducted in an anechoic
chamber. Four different types of EMI tests should be conducted.
Additional details regarding these tests are contained in
MIL-STD-461A, Notice 3 and MIGSTD-462, Notice 2, test
methods CS06, RS02, and RS03. The voltage and frequency
levels specified for the EMI tests are typical of levels used to test
navigation systems. However, these levels should be interpreted
409
with respect to the specifications of the particular system to be

tested.
a. Radiated Susceptibility-Magnetic Induction Field Test. These
tests consist of two parts. In part one, a wire is wrapped around
the cables connecting the Inertial Navigation Unit (INU), the
data acquisition system, and the control-display unit. This wire
is wrapped around the cables spiraling at two turns per meter
and running the length of the cables to within 15 centimeters of
each end connector. The wire is powered by a 400 Hz
generator and terminated with a 20 amp load. The 400 Hz is
applied to the wire-wrapped harness during both the alignment
and navigation segments. With the same configuration, 400
volts AC at 10 pulses-per-second are applied; positive pulses
starting at the beginning of alignment, then switching to
negative pulses for the first 45 minutes of the navigation
segment. The system is then allowed to navigate for an
additional 45 minutes. For the second part of the test, the wire
is wrapped around the system itself. The 400 Hz and 20 amp
load and the 400 volt pulses are again applied as stated above.
A total of four tests should be conducted.
b. Conducted Susceptibility, Spike, Power Test. R o tests should
be conducted. In the first test, positive 100 volt, 1 kHz pulse
modulated pulses of 10 microsecond duration should be
applied to the three input phase voltages by way of pigtails
installed parallel to the three power lines. Positive pulses are
applied during the alignment segment, and negative pulses are
applied during the fmt 15 minutes of the navigation segment.
The system is then allowed to navigate for an additional 75 minutes. For the second test, pulses are applied to the phase A
line fmt, then to phases B and C.
c. Radiated Susceptibility, Electric Field, 14kHz to 10 GHz. The
system is installed on the ground plane in the anechoic
chamber with the connector end facing the radiating antenna
which is placed one meter from the system in a vertical
orientation. After entering the navigation mode, the electric
field intensity is brought up to 20 voltslmeter. The frequency
sweep is then initiated from low to high in the given
bandwidth. The antenna is then rotated 90 degrees to the
horizontal plane and the sweep is repeated. The sweep is
computer controlled to produce discrete frequency steps. As
frequency is increased, the step width is also increased. Time at
each step is dependent upon the radio frequency generator
stabilizing at that frequency as well as allowing for the power
level to adjust. The nominal duration is two seconds per step.
The nine tests which should be conducted are detailed in the
following table.
Frequency
Field
Modulation
10 kHz-25 MHz
Vertical
AM
25 - 200 MHz
Vertical
Pulse
25 - 200 MHz
Horizontal
Pulse
1 - 2 GHz
Horizontal-Vertical
Pulse
2 - 4 GHz
Vertical-Horizontal
Pulse
4 - 8 GHz
Horizontal-Vertical
Pulse
8 - 10 GHz
Vertical-Horizontal
Pulse
500MHz-1GHz
Horizontal
Pulse
200 - 500 GHz
Vertical-Horizontal
Pulse
d. Transient Impulse (Relay Chatter) Test. A wire from the relay

test box is located parallel to the system cable bundle, then
attached to three sides of the N U before returning to the relay.
R o double throw switches are used to provide four
combinations of current flow conditions. One set of conditions
provides forward and reverse current to the relay coil, and the
second set provides forward and reverse outputs of the relay.
Each combination is used in sequence for 5 minutes and is
initiated 1 minute after entering the navigation mode. The
system is allowed to navigate for an additional 70 minutes. The
four combinations provided by the switches are:
Fore
Aft
Aft
Fore
Fore
Fore
Aft
Aft
6.5.6 Combined Environments Test

The purpose of the combined environments test is to determine if
the system performs satisfactorily while operating in a simulated
service environment. Three types of environments will be
simulated; arctic, desert, and tropic. Altitude, temperature, relative
humidity, and vibration will be varied while the system is
operating in the navigational mode following a precision
alignment. Only the tropic tests require precise humidity control.
Each test cycle should consist of a 65-minute temperature soak, a
simulated flight, a 15-minute system-off period, and another
simulated flight for a total of 480 minutes. These tests yield 40
410
hours of simulated flight (10 simulated flights) as indicated in the

table below.
Cycle
Temperature
Soak (deg C)
Type Cycle
Relative
Humidity
71
Desert
0 - 50%
50
Desert
0 - 50%
50
Tropic
-40
Arctic
0 - 50%
Arctic
0 - 50%
91-100%
For each simulated flight, the system will be precision aligned and
placed in the navigation mode. The profiles for temperature,
altitude, relative humidity (tropic only), and vibration are shown
in figures 2 and 3. A tabulated description of the profiles is
provided in table 1. For each test, the system shall be provided
with cooling air at laboratory room ambient temperature (23
pludminus 10 degrees C). Cooling air will be turned on just prior
to system turn-on and turned off immediately after the system is
turned off.
65.7 Moving Base Alignment
This test will be performed to measure the effects of base motion
during gyrocompass alignment and the resulting effects on
navigation accuracy. The system will be mounted on a single axis
vibration table, temperature stabilized for 2 hours, and
gyrocompass aligned with a north heading. While in the
alignment mode, the table will be driven in the east-west direction
at a 1 Hz,2.54 cm peak-to-peak (equivalent to 0.05 g
(zero-to-peak) rate. An additional motion, a 2 cm step in 0.5
seconds, will be added 5 seconds prior to the completion of
alignment. At the completion of the alignment, the system will be
placed in the navigation mode for a period of 60 minutes. This
test will be repeated with an initial alignment heading of 180
degrees.
6.6 Flight Tests
6.6.1 Discussion
The goal of flight testing is to subject the test navigation system to
the type of flight environment which will be experienced when
the system becomes operational. This involves testing the system
in the various aircraft (cargo, fighter, helicopter) in which the
system is designed to be flown and in a manner which will
provide a meaningful indication of the systems performance
capabilities in both benign and dynamic flight environments. The
flight profiles and procedures detailed in the next four sections are
designed to determine the systems performance while being
subjected to realistic operational conditions.
6.6.2 Cargo Aircraji Flight Tests
The purpose of these tests is to determine navigation system
performance in both the aided and unaided modes of operation
during benign and dynamic flights. To satisfy the objectives, a
flight profile composed of medium (less than 5 gs) and low
dynamic (less than 2 gs) maneuvers will be used. This profile will
include maneuvers such as tight turns with changes in altitude and
benign straight-and-level flying. This flight test profile will begin
with a low dynamic racetrack type pattern, establishing a baseline

for maneuvers. This will be followed by a medium dynamic
profile designed to test the performance of the integrated system.
Next will be a cardinal heading profile with the system operated
in the unaided mode. Finally, a return cardinal heading will be
flown in the fully aided mode, testing the recovery capabilities of
the system. The chart below illustrates the timeline for this cargo
flight profile. Table 2 along with figure 4 are provided as a
description of a sample maneuver profile to be used with these
tests.
Take
Off
1.5
Hours
3.0
Hours
4.0
Hours
I. . . Aided
. . . . . .I. . . .Unaided
. . . . . . . .I . . . .Aided
..... I
I Maneuvers I CardinalHdg I Cardinal Hdg I
Land
A minimum of six flights should be flown to achieve the desired

level of statistical confidence in the estimate of the systems
performance. References 3 and 4 present detailed discussions
regarding the required number of flights.
Since the position and velocity errors may be non-zero at the
beginning of the unaided segment, it would be necessary to
compensate these parameters during the entire unaided interval by
the amount of the error present at the beginning of the interval.
Only by performing this compensation will a true picture of the
systems actual performance during this segment be obtained.
The systems ability to perform in-flight alignments (FA) should

also be tested. A total of four flights should be flown; two benign
flights and two maneuver flights. Each flight will consist of three
IFAs,each followed by 30 minutes of navigation to verify the
validity of the F A . The decision regarding the operational mode
(aided or unaided) of the system should be determined by the test
objectives. The IFAs for the benign flights and for the maneuver
flights should be divided into separate ensembles.
6.63 Helicopter Flight Tests
The purpose of this series of tests is to determine the navigation
system performance in the fully aided mode of operation during
benign and maneuver flights with the system subjected to the
vibration environment of the helicopter. To satisfy this objective,
two different flight profiles will be flown.
The first profile is a cardinal heading cruise profile. This profile
consists of a cardinal heading flight at 3,000 feet above-groundlevel (AGL) for a duration of 90 minutes.
The second profile is the tactical maneuver flight profile and
consists of a clockwise, 360 degree turn at takeoff followed by a
20-minute cardinal heading cruise leg, 45 minutes of
maneuvering, and a 25-minute return leg. Specific tactical
maneuvers performed are a rapid spiraling descent (3,000 feet to
500 feet AGL with two 360 degree turns), a counterclockwise
pinwheel turn, a series of four S-turns (45-degree bank with 60
degree heading changes), a series of two shallow S-turns
(30-degree bank with 30-degree heading changes) followed by
two vertical S-turns (100 feet up, 200 feet down, 100 feet up),
terrain following, and a high performance climb-out to 3,000 feet
AGL. Six flights of each profile should be flown.
41 1
6.6.4 Fighter Flight Tests

The purpose of fighter aircraft flight testing is to determine
navigation system performance in the fully aided mode during
benign and high dynamic flights. The fighter test bed may vary
with test requirements, but in most cases, a high performance
fighter aircraft should be used as the test bed.
Three basic types of flight profiles should be flown during fighter
testing: benign flight, ordnance delivery maneuvers, and air
combat maneuvers. Each flight will be 90 minutes in duration,
with any maneuvers conducted at the end of the initial benign leg.
Six flights will consist of straight-and-level, out-and-back benign
profiles, six will include air combat maneuvers, and six will
include ordnance delivery maneuvers. The maneuver flights will
consist of a 40-minute benign leg followed by 10 minutes of
maneuvers, then a @-minute benign return leg. The profiles, with
the exception of the maneuver segments, should be flown at an
altitude of 35,000 feet MSL. Throughout the fighter testing, the
pilot should attempt to conduct the flights in a repeatable manner,
noting deviations from the intended profiles as they occur. Entry
headings, altitudes, and air speed should be consistent for all
maneuvers whenever possible.
6.65 Terrain Referenced Navigation Flight Tests
The purpose of these tests is to evaluate the accuracy of the terrain
referenced navigation (TRN) system while flying over various
terrain types at different altitudes. The TRN system produces
trajectory data by comparing the measured terrain profile from the
radar altimeter with the digital terrain data base. The performance
accuracy of the TRN system is determined by comparing its
trajectory data with time-correlated reference data.
In order to adequately characterize the performance of a TRN
system, it should be flown over flat terrain, gently rolling hills,
moderately rough terrain, and rough terrain. Additionally, the test
flights should be flown at both 1,000 and 2,000 feet above
ground level (AGL) due to the altitude-sensitive nature of radar
altimeters. Six flights for each ensemble should be flown. The
duration of these flights will be determined by the type of terrain
in the immediate area of the testing headquarters.
In addition to determining the navigation accuracy of the TRN
system, other items which need to be determined are as follows:
a. How long does it take the system to acquire the proper track?
b. How vulnerable is the system to acquiring a false track?
c. How effectively does the system recover from tracking errors?
d. How well does the system perform in the coasting mode over
flat terrain?
6.7 Data Reduction Procedures
6.7.1 Generation of Error Outputs
Time-correlated system test data and reference data should be

recorded during each test. Data rates are prescribed in sections B
and C. Errors are calculated as a function of time by comparing
system data with reference data. Error outputs are generated by
linearly interpolating the reference data to system times and
comparing the interpolated reference data to system data. Latitude,
longitude, and altitude position errors are output as well as north,
east, and vertical velocity errors. The horizontal position and
velocity errors are root-sum-squared (RSS)to produce radial

position and velocity errors. Provisions should be made for
suspending the comparison process during intervals where the
reference system data are of marginal value. In these cases, gaps in
position and velocity error data will occur.
6.7.2 Plots to be Presented
It is useful for comparison purposes to represent the error plots of

several tests with a single plot. This is accomplished in one of two
ways. The first type is obtained by presenting the radial error
curves from an ensemble of tests on a composite plot. These
ensembles contain only those tests which come from similar.test
profiles. The statistical analysis of these ensembles will reflect the
systemsoverall performance and will establish certain confidence
limits with respect to that performance. This analysis should
output time histories of the mean, standard deviation, median,
SOth, and 95th percentiles of the radial position and velocity
errors. Plots of any or all of these parameters may be presented as
an indication of system performance. For details regarding
statistical methods, refer to section 6.9.
6.7.3 Tabular Data to be Presented
For the purpose of comparing one system against another, it is
useful to specify a systems performance with a single number.
For inertial navigation systems, where level-axis gyro drifts
generating ramps in position error have been the dominant error
source, the straight-lindzero-intercept fit to the radial position
error has been the most popular performance parameter. However,
with the advent of hybrid navigation systems, which produce
errors that are largely time-invariant, this is no longer the case. It
is much more meaningful to present the mean and standard
deviation of the 50th and 95th percentiles of radial position and
velocity errors for an ensemble of flights. These values should be
presented as indications of the position accuracy of a hybrid
navigation system. To obtain a measure of velocity accuracy, the
radial velocity error data for all flights in an ensemble should be
merged, then the composite RMS computed for the radial velocity
error.
6.8 Statistical Methods
6.8.1 Introduction
For the purpose of comparing one navigation system with

another, it is highly desirable to specify a systems performance in
a simple, straightforward manner, preferably with a single
quantity stated as a function of time or distance. However, there
are a variety of methods for describing the performance accuracy
of a navigation system as well as numerous assumptions which
can be made regarding a systems error characteristics. These
dissimilarities may introduce inconsistencies in comparing the
performance of one navigation system with another. The goal of
this document is to standardize the method of presenting
navigation system accuracies.
412
6.8.2 Dejnitions
The statistical parameters most frequently used to quantify a
systems radial position and velocity errors are mean, standard
deviation, root-mean-square (RMS), geometric mean (GM), 50th
percentile (Circular Error Probable, CEP), and the 95th percentile.
The mean, standard deviation, RMS, and GM are computed using
simple algorithms. These are defined as follows:
T R i
Mean
i= 1
R =N
(1 1
where Ri is the radial error at time ti and N is the number of data

points.
The 50th percentile (R50) and 95th percentile (R95) are defined
respectively as being 50 or 95 percent probable that the error in a
randomly selected flight from an ensemble of similar flights will
fall below this level, or that the error will be below this level in 50
or 95 percent of flights if the number of flights in the ensemble is
very large.
6.83 Discussion
The process of determining the accuracy of a navigation system
involves the conducting of flight tests. However, due to fiscal and
time constraints, it is rarely ever possible to conduct a large
enough sample of tests to completely characterize a systems
performance. Because of this limitation, a method has been
developed which makes it possible to extrapolate from a limited
number of samples of system performance to an estimation of
system performance for the parent population. This method
involves the ratio of the GM and the RMS values of the radial
position error for an ensemble of tests. The GMRMS algorithm
was developed at Holloman Air Force Base, New Mexico, and is
widely used by facilities involved in the testing of navigation
systems. Derivation of this method is detailed in reference 1, and
further explanation can be found in references 2 and 3.
From the radial errors computed at each time point in an ensemble
of valid flights, calculate the RMS value and the GM of the errors
as defined in section 2 above. Form the ratio GMRMS. Also,
form the ratio R(P)/RMS, where R(P) is the desired percentile
level, usually 50 or 95. The relationship between these two ratios
is a group of functions, one for each percentile level P. These
functions are independent of the actual test data and can be
precomputed. These functions are presented in figure 5. Using the
GMRMS value along the ordinate axis, determine the multiplier

corresponding to the desired percentile P,and multiply this value
by the RMS sample value. The results will be the Pth percentile
level of error at time t for this ensemble of flights. The quantity t
represents the time elapsed since switching to the navigate mode
or from the time of initiation of aiding.
Once the Pth percentile has been computed for all available data
points, this curve is used to determine system performance. For
systems whose radial position errors are functions of time or
distance, the most commonly used method of determining system
performance has been to perform a least-squares fit (constrained to
pass through the origin) to the Pth percentile curve of the radial
position error. This produces the radial position error growth rate.
For systems whose radial position errors are not functions of time
or distance, compute the Mean and Standard Deviation of the Fth
percentile curve. These values will then be used to characterize the
performance of a hybrid navigation system.
413
c,
0
\
m
U
(D
0
0
0
cv
0
0
0
uj
Q,
>
Q,
-I
c
0
*
0
.-
c
a>
3
0
L
a,
LI
0
r
T-
O
0
0
>
0
U
I
(d
414
60
--
35
--
Arnbien?
--
ala
-U 0)
3LL
-0
U 0
STEP NUMBER FROM TABLE 1

S >
9
12
-0
a -X
71
50
1 2
11 6 7 8
IO
I
T I M E (HINUTES)
+.. . . ..
I
U
0
al
I
3
U
al
c
E
al
I-
-25
Desert o r T r o p i c
- - - - - Arctic
. . . . . . . . .- Extreme
*Tropic Only
+
Uncontrol l e d
.>
0.02
d
-
-1
--
Figure 2. Combined Environments Test Profile.
415
>
I-
rn
.04
Z
W
n
J
a~
= I
.02
t j xl'
w z n
v)
za
lo
-6 dB/OCT
-6 dB/OCT
15
1000
2000
Frequency (Hz)
T h i s l e v e l l e to b e r u n d u r i n g a l l v i b r a t i o n - o n t i m e s e x c e p t
for five minutes out o f each 5.000 f t . a l t i t u d e s e g m e n t of

the flight profile.
T h i s level i s to b e r u n f o r f i v e m i n u t e s d u r i n g e a c h 5 , 0 0 0
ft. altitude segment of the flight profile (total test time
i s ten m i n u t e s f o r e a c h 8 - h o u r c y c l e .
Figure 3. Random Vibration Levels for CET.
416
Maneuver C h e c k p o i n t L o c a t i o n s
Checkpoint
6oooa
Longitude
Latitude
-106 15.8
-106 05.6
-106 05.6
-106 15.8
-106 09.8
-106 09.8
-106 12.0
-106 12.3
-106 12.0
-106.11-5
33
33
33
33
33
33
33
33
33
33
soooa
5
40000
moos
2000c
l00OC
- 1000(
I
-20000
I
- 10000
10000
20000
Meters
Figure 4 .
C a r g o M a n e u v e r Profile
13.8
13.8
00.0
00.0
13.8
00.0
12.3
08.5
05.2
02.0
417
2.8
.................. .....................................
...........
................... ......
2.4
.............. ...........
.................. .....................................
.........
280
................
-95%
.................. ..................................
......
.............................................
....................................p.v90g....... >.;
.....
................... ....................
.....................................
1.2
...........................................
. . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
.....
0.8
I
.... I
(
.....:................
.................. .....................................
.....
.......................
1
0,4
...........
. . . .i . . . .; . . .
......... ...................
1,6
....... ......
..................
..
...................
080
3
0.2
0.'4
0.6
0,'s
1 :o
GM/RMS
PERCENTILES OF THE MAXIMUM LIKELIHOOD

E S T I M A T E OF THE D I S T R I B U T I O N OF
R A D I A L ERROR
1:2
418
419
Table 2.
TRK
POS
TRK A LT
DEG M S L
LONGITUDE
Aircraft Controller Data.
LATITUDE
KTAS MANEUVER
1A
153
-106 15.8
33 13.8
350
20 DEG BANK RIGHT
2A
180
150
-1 06 05.6
33 13.8
350
END TURN
3A
180
150
-106 05.6
33 00.0
350
20 DEG BANK RIGHT
4A
150
-106 15.8
33 00.0
350
END TURN
1B
150
-106 15.8
33 13.8
350
32 DEG BANK RIGHT
SA
180
150
-106 09.8
33 13.8
350
END TURN
6A
180
150
-106 OS.8
33 00.0
350
32 DEG BANK RIGHT
48
150
-106 15.8
33 00.0
350
END TURN, SLOW TO 325 KTAS
1c
150
-106 15.8
33 13.8
325
28 DEG BANK RIGHT
58
180
150
-1C6 09.8
33 13.8
325
40 DEG BANK RIGHT BEGIN CLIMB TO

FL 170 BEGIN SLOW TO 275 KTAS
7A
270
150 *
-1 06 12.0
33 12.3
325
'40 DEG BANK LEFT
8A
90
1sO@
-106 12.3
33 0 8 5
300
'40 DEG BANK RIGHT
9A
270
170
-1 06 12.0
33 05.2
275
40 DEG BANK LEFT BEGIN DIVE TO FL

lhACCELERATE TO 350 KTAS
10A
90
16043
-106 11.5
33 02.0
300
'40 DEG BANK RIGHT
6B
180
150
-106 09.8
33 00.0
325
'28 DEG BANK RIGHT
4c
150
-106 15.8
33 13.2
350
END TURN BEGIN CLIMB TO F L 200
1D
Iso@
-106 15.8
33 13.8
353
20 DEG BANK RIGHT
28
180
17-
-1 06 05.6
.33 1 x 8
350
END TURN
38
180
18043
-1 06 05.6
33 00.0
350
20 DEG BANK RIGHT
4D
1go@
-1 06 15.8
33 00.0
350
END TURN
1E
200
-1 06 15.8
33 13.8
350
32 DEG BANK RIGHT
5C
180
200
-106 09.8
33 13.8
350
END TURN
6C
180
200
-1 06 09.8
33 00.0
350
32 DEG BANK RIGHT
4E
200
-106 15.8
33 00.0
350
1F
200
-1OG 15.8
33 13.8
325
28 DEG BANK RIGHT
5D
180
200
-106 09.8
33 13.8
325

78
270
200
-1 06 12.0
33 12.3
325
'40 DEG BANK LEFT
8B
90
2100
-106 12.3
33 08.5
300
'40 DEG BANK RIGHT
9B
270
220
-106 12.0
33 05.2
275
200 ACCELERATE TO 350 KTAS
420
Table 2.
TRK
POS
TRK ALT
DEG M S L
Aircraft Controller Data (Continued).
LONGITUDE
LATITUDE
KTAS MANEUVER
2100
-106 11.5
33 02.0
300
'40 DEG BANK RIGHT
180
200
-106 09.8
33 00.0
325
'28 DEG BANKBIGHT
4E
200
-106 15.8
33 13.2
350
END TURN BEGIN CLIMB TO FL 250
1G
2 1 0
-106 15.8
33 13.8
350
20 DEG BANK RIGHT
2c
180
220@
-1 06 05.6
33 13.8
350
END TURN
3c
180
2 3 0
-1 06 05.6
33 00.0
350
20 DEG BANK RIGHT
4F
24w
-106 15.8
33 00.0
350
END TURN
1H
250
-106 15.8
33 13.8
350
32 DEG BANK RIGHT
5E
180
250
-106 09.8
33 13.8
350
END TURN
6E
180
250
-106 09.8
33 00.0
350
32 DEG BANK RIGHT
4G
250
-106 15.8
33 00.0
350
11
250
-106 15.8
33 13.8
325
28 DEG BANK RIGHT
5F
180
250
-1 06 09.8
33 13.8
325

7C 270
250
-106 12.0
33 12.3
325
'40 DEG BANK LEFT
8C
90
26-
-10612.3
33 08.5
300
'40 DEG BANK RIGHT
9C
270
270
-106 12.0
33 05.2
275

250 ACCELERATE TO 350 KTAS
1oc
90
26-
-10611.5
33 c2.0
300
'40 DEG BANK RIGHT
6F
180
250
-10609.8
33 00.0
325
'28 DEG BANK RIGHT
4H
250
-106 15.8
33 13.2
350
END TURN
1OB
90
6D
@ lndlcates approxlmate altitude MSL
lndlcates approxlmate true air speed
42 1
6.9 References
1. Mason, F.J. and Bodwell, C.A. Maximum Likelihood
Estimation of the Distribution of Radial Errors, Directorateof
Guidance Test, Air Force Missile Development Center,
Holloman Air Force Base, New Mexico, Working Paper
WP-MDSGA45-4, Rev 1 , l Oct 1969, DTlC &A950
316.
2. Air STD 53/16A, The Specification for Evaluation of the
Accuracy of Hybrid Navigation Systems, Air Standardization
Coordinating Committee, 6 April 1988.
3. Sakran, F.C. Jr., Considerations in Use of the GM/RMS

Estimation Method For Navigation System Position Error,
Force Warfare Aircraft Test Directorate, Patuxent River,
Maryland, Technical Memorandum TM 91-03 FW,17 July
1991.
4. Calculation of Confidence Intervals for Percentile Levels,
Memo for The Record dtd 20 May 1983, Holloman Air Force
Base, New Mexico.
1. Recipients Reference
2. Originators Reference
AGARD-AG-33 1
5. Originator
3. Further Reference
ISBN 92-836-101 8-0
4. Security Classification
of Document
UNCLASSIFIED/
UNLIMITED
Advisory Group for Aerospace Research and Development

North Atlantic Treaty Organization
7 rue Ancelle, 92200 Neuilly-sur-Seine, France
6. Title

7. Presented at/sponsored by
the Guidance and Control Panel of AGARD

8. Author(s)/Editor(s)
9. Date
June 1995
various
10. AuthorsEditors Address
11. Pages
432
various
12. Distribution Statement
There are no restrictions on the distribution of this document.

Information about the availability of this and other AGARD
unclassified publications is given on the back cover.
13. KeyworddDescriptors
Air navigation
Inertial navigation
Satellite navigation
Radar navigation
Multisensors
Rotary wing aircraft
Spacecraft
Space navigation
Doppler navigation
Magnetic navigation
Global positioning system
Integrated systems
Fixed wing aircraft
Kalman filtering
14. Abstract
The need for an up to date, comprehensive treatise on aerospace navigation systems has been
recognized by the navigation community. Such a document will serve as a lasting reference as did
previously published material prepared in the late 60s and early 70s. These earlier documents,
though remarkably prescient in their technical forecasts, have been overtaken by very rapidly
advancing technology.
It is anticipated that the target reader of this AGARDograph will be an individual who has
responsibility for the integration of navigation equipment aboard an aerospace vehicle. He may not
have, but is desirous of obtaining, an experts perspective on the capabilities and limitations of the
various navigation sensor and integration techniques.
The AGARDograph is organized into six sections with the initial section providing the motivation for
establishing the requirements to assure that the development of an aerospace navigation system will
meet its operational requirement. The second section of this document reviews navigation coordinate
frames with a discussion of inertial, terrestrial and geodetic coordinate systems. The third section of
the document describes the navigation sensor technologies that are employed in modem aerospace
navigation systems.
The fourth section of this document addresses system analysis and synthesis methods. Examination
and trade-off of the technology, for each sensing, processing and control and display element,
generates alternative preliminary designs. The fifth section of the document describes representative
state-of-the-art navigation system implementations in fixed wing, rotary wing aircraft and spacecraft.
The sixth and concluding section of the document describes the various test methods employed to
verify the performance of aerospace navigation systems used by NATO countries.
NATO
- @ ~OTAN
7 RUE ANCELLE 92200 NEUILLY-SUR-SEINE

FRANCE
DIFFUSION
DES PUBLICATIONS
AGARD NON CLASSIFIEES
TMcopie (1)47.38.57.99 rn Telex 610 176
CENTRES DE DIFFUSION NATIONAUX

ALLEMAGNE
ISLANDE
Fachinformationszentrum,
Director of Aviation
Karlsruhe
c/o Flugrad
Reykjavik
D76344 Eggenstein-Leopoldshafen2
BELGIQUE
ITALE
CoordoMateur AGARD-VSL
Aeronautrca Militare
Etat-major de la Force akienne
Ufficio del Delegato Nazionale all 'AGARD
wartier Reine Elisabeth
Aero ono Pratica di Mare
Rue d'Evere. 1140 Bruxelles
Pomezia (Roma)
CANADA
LUXEMBOURG
Directeur. Services d'information scientifique
Voir Beleiaue
I.
Ministkre de la Defense nationale
NORVEGE
Ottawa. Ontario KIA O K 2
Norwegian Defence Research Establishment
DANEMARK
Attn: Biblioteket
Danish Defence Research Establishment
P.O. Box 25
Ryvangs All6 1
N-2007 Kjeller
P.O.Box 2715
PAYS-BAS
DK-2100 Copenhagen 0
Netherlands Delegation to AGARD
National Aeros ace Laboratory NLR
P.O. Box 9050g
1006 BM Amsterdam
PORTUGAL
ForGa A h a Portuguesa
Cenlm de DocumentqEo e InformaSHo
Alfragide
2700 Amadora
FRANCE
ROYAUME-UNI
O.N.E.R.A. (Direction)
Defence Research Information Centre
29, Avenue de la Division Leclerc
Kentigern House
92322 Ch&illon Cedex
65 Brown Street
Glasgow G2 8EX
GRECE
Hellenic Air Force
TURQrn
Air War College
Milli Savunma Bqkanli$ (MSB)
Scientific and Technical Library
ARGE Dairesi Bagkanliki (MSB)
Dekelia Air Force Base
06650 Bakanliklar-Ankara
Dekelia, Athens TGA I010
Le cenbp de distribution national des Eiats.Unis ne d6tieul PAS de stocks des publicatlorn de I'AGARD.
D6ventuelles demandes de photocopies doivent etre formulks directement auprhs du NASA Center for AemSpace Information (CASI)
h I'adresse ci-dessnus. Toute notification de changement d'adresse doit 2tre fait 6galement aupks de CASI.
~~
owl
AGENCES DE VENTE
NASA Center for
ESAnnformation Retrieval Service
The British Libran,
Aerospace Information (CASI)
European Space Agency
Vocumenl Suppl) 'Dnisiun
800 Elkridge Landing Road
IO, N e Mario Nikis
Boston Spa, Wctherby
Linthicum Heights, MD 21090-2934
75015 Paris
Wesl Yorkshire LS23 7BQ
Etats-Unis
France
Royaume-Uni
Les demandes de microfiches ou de photocopies de documents AGARD (y compris les d e m d e s faites aupks du C A S ) doivent
componer la dknomination AGARD, aiusi que le numtro de s6rie d'AGARD (par exemple AGARD-AG-315). Des informations
analogues, telles que le titre et la date de publication sont souhaitables. Veuiller noter qu'il y a lieu de specifier AGARD-R-nnn et
AGARD-AR-nnn lors de la commande des rapports AGARD et des rapports consultatifs AGARD respectivement. Des rftnnces
bibliographiqnes completes ainsi que des risumts des publications AGARD figurent dans les jouruaux suivants:
Scientific and Technical Aerospace Reports (STAR)
Government Repons Announcements and Index (GRAM)
publie par la NASA Scientific and Technical
publit par le National Technical Information Service
Information Division
Springfield
NASA Headquarters (JlT)
Virginia 22161
Washington D.C. 20546
Etats-Unis
Etats-Unis
(accessible Cgalement en mode interactif dans la base de
donnks bibliographiques en lime du NTIS, et sur CD-ROM)
~~~
ImprimC par le Groupe Communication Canada

45, boul. Sacrc?-Caur, Hull (QukbecJ, C a n a h KIA OS7
~~~~~~
7 RUE ANCELLE
g22oo
NEUIC
FRANCE
AAG-
AGARD PUBLICATIONS
Telehu (1)47.38.57.89 Telex 610 176

w v 0fthcablif.tiOns ttut
Lcetun saisr and spadrl
and WahogoFatp npons pubusha bom 1993 onward. Pordotaila. write or&
hsld in 1993 a Irter. .nd of
a tclefax to !
baddress giVfQ rbovs.
BELGIUM
coonlonnatmr AGARD-VSL.
NATIONAL DISTRIBUTION CBNTRES

LUXEMBOURG
See Belgium
de la Face aedieme
QuutiaRsiasBLislbcth
Rue dBvao.1140 Bm~.llea
!&t-lmjor
Nclhrxlands Dek@ion to AGARD

NatiwalAeros.ccLsbaatary,NLR
P.O. Box 9034
1006 BM Amacadam
NORWAY
Defeoec Rearch Emblishment
CANADA
Dirsctm Sdmitic Informston Suvias
Dqlf of N a t i d WulW
Ottawa, Ontario KIA OK2
DENMAXK
D.aisbDefsaoeRcscuchHstsblisbmnt
Es%t&Xt
P.O. Box 25
N-2Ol Kjellcr
!%%%%:
DK-2100 Copmhagsn 0
FR4NcE
ON.E&A. @inctiop!
29 Avmnc de la D i w m Lcdcn:
923aZchntilIon~
PORTUGAL
Foqa A k a Pmgncsa
cent rode^ e m##$&ora
SPAIN
INTA (AGARD Publications)
pintorRoslllcs34
21uylg Madrid
l1l1l

AGARDAG6331

Uploaded by

Copyright:

Available Formats

AGARDAG6331

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

AGARDAG6331

Uploaded by

Copyright:

Available Formats

- ' -111

Aerospace Navigation Systems

NORTH ATLANTIC TREATY ORWIZATION

Published June 1995

Aerospace Navigation Systems

North Atlantic Treaty Organization

The Mission of AGARD

scientific and technical information;

The content of this publication has been reproduced

Copyright 0 AGARD 1995

Printed by Canada Communication Group

Guidance and Control Panel

Chairman: Mr. S. Leek

Mail from Europe:

Guidance and Control Panel

SECTION III- NAVIGATION TECHNOLOGY

Doppler Radar Navigation

Terrain Referenced Navigation

An Overview of OMEGA Radio Navigation System

AGARDograph on Advanced Astroinertial Navigation Systems

Magnetic Heading References

SECTION IV - SYSTEMS ANALYSIS, DESIGN AND SYNTHESIS METHODOLOGIES

An Overview of a Generic Multi-Sensor Integrated Navigation System Design

Deep Integration of GPS, INS, SAR and Other Sensor Information

Federated Filter for Fault-Tolerant Integrated Navigation

GPSDnertial Integration Overview

Design Considerations for a Suboptimal Kalman Filter

Vertical Channel Design Considerations

SECTION V - REPRESENTATIVEVEHICLE IMPLEMENTATIONS

Aerospace Navigation Systems Requirements for Fixed Wing Aircraft

Rotary Wing Aircraft

Spacecraft Navigation Requirements

MOTIVATION AND OVERVIEW

The need for an up to date,

the requirements to assure

partitioning of the navigation

considerably complicates the

The need for a navigation

system must provide data to

Requirements for aerospace

Thus an integrated navigation

during which intermittent

Following the penetration

During each mission segment,

The requirements for the

An important aspect of all

The requirements on the

The technical requirement for

for an integrated navigation

The functionality and

The maximum acceptable

The maximum cost of the

- The duration of the

The nature of external

The purpose of an integrated

The means to rigorously specify navigation accuracy has been

Fried, W.R., Avionics Navigation System,

Couperthwaite, W.J. 61 Stokes, r.F., Suggestion concerning

Accurate navigation requires an understanding of the three coordinate frames

the coordinate frames or systems are